Computational Gas-Solids Flows and Reacting Systems - Theory, Methods and Practice (2011)

Computational Gas-Solids
Flows and Reacting

Systems:
Theory, Methods and Practice
Sreekanth Pannala
Oak Ridge National Laboratory, USA
Madhava Syamlal
National Energy Technology Laboratory, USA
Thomas J. O'Brien
EnginEEring sciEncE rEfErEncE

Hershey • New York
Director of Editorial Content: Kristin Klinger
Director of Book Publications: Julia Mosemann
Acquisitions Editor: Lindsay Johnston
Development Editor: Christine Bufton
Typesetter: Michael Brehm
Production Editor: Jamie Snavely
Cover Design: Lisa Tosheff
Published in the United States of America by

Engineering Science Reference (an imprint of IGI Global)
701 E. Chocolate Avenue
Hershey PA 17033
Tel: 717-533-8845
Fax: 717-533-8661
E-mail: cust@igi-global.com
Web site: http://www.igi-global.com
Copyright © 2011 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in
any form or by any means, electronic or mechanical, including photocopying, without written permission from the publisher.
Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or com-
panies does not indicate a claim of ownership by IGI Global of the trademark or registered trademark.
Library of Congress Cataloging-in-Publication Data
Computational gas-solids flows and reacting systems : theory, methods and

practice / Sreekanth Pannala, Madhava Syamlal and Thomas J. O'Brien, editors.
p. cm.
Summary: "This book provides various approaches to computational gas-solids
flow and will aid the researchers, graduate students and practicing engineers
in this rapidly expanding area"--Provided by publisher.
Includes bibliographical references and index.
ISBN 978-1-61520-651-3 (hardcover) -- ISBN 978-1-61520-652-0 (ebook) 1.
Two-phase flow. 2. Gas-solid interfaces. 3. Computational fluid dynamics.
I. Pannala, Sreekanth, 1971- II. Syamlal, Madhava. III. O'Brien, Thomas J.
(Thomas Joseph), 1941-
TA357.5.M84C646 2010
620.1'064--dc22
2009053472
British Cataloguing in Publication Data

A Cataloguing in Publication record for this book is available from the British Library.
All work contributed to this book is new, previously-unpublished material. The views expressed in this book are those of the
authors, but not necessarily of the publisher.
List of Reviewers
Aibing Yu, UNSW, Australia
Aparna Baskaran, Syracuse, New York, USA
Badri Velamur Asokan, ORNL, USA
Berend van Wachem, Imperial College, UK
Christine Hrenya, Boulder, Colorado, USA
Dale Snider, CPFD, USA
Dimitri Gidaspow, IIT, USA
George Bergantz, Seattle, Washington, USA
Janine Galvin, NETL, USA
Jeff Dietiker, NETL, USA
Jennifer Sinclair, Florida, USA
Jim Dufty, Florida, USA
Judith Hill, ORNL, USA
Juray De Wilde, UCL, Belgium
Kate Evans, ORNL, USA
Lothar Reh, ETH, Switzerland
Peter Witt, CSIRO, Australia
Ray Cocco, PSRI, USA
Richard Mills, ORNL, USA
Rodney Fox, Iowa State, USA
Ron Breault, NETL, USA
Shankar Subramaniam, Iowa State, USA
Sofiane Benyahia, NETL, USA
Sourabh Apte, Oregon State, USA
Sreekanth Pannala, ORNL, USA
Stuart Daw, ORNL, USA
Todd Pugsley, University of Saskatchewan, Canada
Wei Ge, CAS, China
Table of Contents
Foreword ............................................................................................................................................ xiv
Preface ................................................................................................................................................ xvi
Section 1
Theory
Chapter 1
Multiphase Continuum Formulation for Gas-Solids Reacting Flows..................................................... 1
Madhava Syamlal, National Energy Technology Laboratory, USA
Sreekanth Pannala, Oak Ridge National Laboratory, USA
Chapter 2
Hydrodynamic Equations from Kinetic Theory: Fundamental Considerations.................................... 66
James W. Dufty, University of Florida, USA
Aparna Baskaran, Syracuse University, USA
Chapter 3
Kinetic Theory for Granular Materials: Polydispersity ...................................................................... 102
Christine M. Hrenya, University of Colorado, USA
Chapter 4
Interfacial Interactions: Drag .............................................................................................................. 128
Wei Ge, Chinese Academy of Sciences, China
Ning Yang, Chinese Academy of Sciences, China
Wei Wang, Chinese Academy of Sciences, China
Jinghai Li, Chinese Academy of Sciences, China
Chapter 5
Mass and Heat Transfer Modeling ...................................................................................................... 178
Ronald W. Breault, National Energy Technology Laboratory, USA
Section 2
Numerical Methods
Chapter 6
Coupled Solvers for Gas-Solids Flows ............................................................................................... 204
Chapter 7
Quadrature-Based Moment Methods for Polydisperse Gas-Solids Flows ......................................... 221
Alberto Passalacqua, Iowa State University, USA
Prakash Vedula, University of Oklahoma, USA
Rodney O. Fox, Iowa State University, USA
Chapter 8
Direct Numerical Simulation of Gas-Solids Flow Based on the Immersed Boundary Method ......... 245
Rahul Garg, Iowa State University, USA
Sudheer Tenneti, Iowa State University, USA
Jamaludin Mohd. Yusof, Los Alamos National Laboratory, USA
Shankar Subramaniam, Iowa State University, USA
Chapter 9
The Multiphase Particle-in-Cell (MP-PIC) Method for Dense Particle Flow .................................... 277
Dale M. Snider, CPFD, USA
Peter J. O’Rourke, CPFD, USA
Section 3
Practice
Chapter 10
Circulating Fluidized Beds ................................................................................................................. 316
S.B. Reddy Karri, PSRI, USA
Ted Knowlton, PSRI, USA
Chapter 11
CFD Modeling of Bubbling Fluidized Beds of Geldart A Powders ................................................... 357
T. Pugsley, The University of Saskatchewan, Canada
S. Karimipour, The University of Saskatchewan, Canada
Z. Wang, The University of Saskatchewan, Canada
Chapter 12
Computational Modeling of Gas-Solids Fluidized-Bed Polymerization Reactors ............................. 373
Ram G. Rokkam, Iowa State University, USA
Michael E. Muhle, Univation Technologies, USA
Chapter 13
Validation Approaches to Volcanic Explosive Phenomenology ......................................................... 398
Sébastien Dartevelle, Los Alamos National Laboratory, USA
Compilation of References .............................................................................................................. 430
About the Contributors ................................................................................................................... 467
Index ................................................................................................................................................... 471

Detailed Table of Contents
Foreword ............................................................................................................................................ xiv
Preface ................................................................................................................................................ xvi
Section 1
Theory
Chapter 1
Multiphase Continuum Formulation for Gas-Solids Reacting Flows..................................................... 1
Madhava Syamlal, National Energy Technology Laboratory, USA
Sreekanth Pannala, Oak Ridge National Laboratory, USA
This chapter describes the formulation of multiphase continuum models for gas-solids flows with
chemical reactions. A typical formulation of the equations is presented here, following the equations in
the open-source software MFIX (http://mfix.netl.doe.gov) so that interested users may look up details
of the numerical implementation, study the solutions, or experiment with the numerical implementation
of alternative formulations. We will first provide a brief overview of the significance of gas-solids react-
ing flows and the challenges in modeling these systems along with various efforts undertaken by differ-
ent groups over the last 2–3 decades. We will then summarize the methods used to derive multiphase
continuum models and to formulate constitutive equations. We will later provide information on the
formulation for mass, momentum, granular energy, energy, and species balance equations for gas and
multiple solids phases. We will discuss the constitutive equations required in each of the balance equa-
tions; a detailed discussion of certain constitutive equations, such as the gas-solids drag and granular
stresses (derived from kinetic theory), will be presented by other authors in later chapters. We will point
out the differences between different approaches and direct the reader to references that discuss those
approaches in detail. We will end the chapter with the example problem of the simulation of a bubbling
fluidized bed to illustrate some of the modeling options — physical models, numerical discretization
schemes, and grid resolution – that need to be considered to accurately simulate gas-solids systems.
Chapter 2
Hydrodynamic Equations from Kinetic Theory: Fundamental Considerations.................................... 66
James W. Dufty, University of Florida, USA
Aparna Baskaran, Syracuse University, USA
In this chapter, a theoretical description is provided for the solid (granular) phase of the gas-solids
flows that are the focus of this book. Emphasis is placed on the fundamental concepts involved in de-
riving a macroscopic hydrodynamic description for the granular material in terms of the hydrodynamic
fields (species densities, flow velocity, and the granular temperature) from a prescribed “microscopic”
interaction among the grains. To this end, the role of the interstitial gas phase, body forces such as
gravity, and other coupling to the environment are suppressed and retained only via a possible non-
conservative external force and implicit boundary conditions. The general notion of a kinetic equation
is introduced to obtain macroscopic balance equations for the fields. Constitutive equations for the
fluxes in these balance equations are obtained from special “normal” solutions to the kinetic equation,
resulting in a closed set of hydrodynamic equations. This general constructive procedure is illustrated
for the Boltzmann-Enskog kinetic equation describing a system of smooth, inelastic hard spheres. For
weakly inhomogeneous fluid states the granular Navier-Stokes hydrodynamic equations are obtained,
including exact integral equations for the transport coefficients. A method to obtain practical solutions
to these integral equations is described. Finally, a brief discussion is given for hydrodynamics beyond
the Navier-Stokes limitations.
Chapter 3
Kinetic Theory for Granular Materials: Polydispersity ...................................................................... 102
Christine M. Hrenya, University of Colorado, USA
Kinetic-theory-based models of rapid, polydisperse, solids flows are essential for the prediction of
a wide range of practical flows found in both nature and industry. In this work, existing models for
granular flows are critically compared by considering the techniques used for their derivation and the
expected implications of those techniques. The driving forces for species segregation, as predicted by
kinetic theory models, are then reviewed. Although the rigor associated with the development of such
models has improved considerably in the recent past, a systematic assessment of model validity and
computational efficiency is still needed. Finally, a rigorous extension of such models to gas-solids flows
is discussed.
Chapter 4
Interfacial Interactions: Drag .............................................................................................................. 128
Wei Ge, Chinese Academy of Sciences, China
Ning Yang, Chinese Academy of Sciences, China
Wei Wang, Chinese Academy of Sciences, China
Jinghai Li, Chinese Academy of Sciences, China
The drag interaction between gas and solids not only acts as a driving force for solids in gas-solids
flows but also plays as a major role in the dissipation of the energy due to drag losses. This leads to
enormous complexities as these drag terms are highly non-linear and multiscale in nature because of the
variations in solids spatio-temporal distribution. This chapter provides an overview of this important
aspect of the hydrodynamic interactions between the gas and solids and the role of spatio-temporal het-
erogeneities on the quantification of this drag force. In particular, a model is presented which introduces
a mesoscale description into two-fluid models for gas-solids flows. This description is formulated in
terms of the stability of gas-solids suspension. The stability condition is, in turn, posed as a minimiza-
tion problem where the competing factors are the energy consumption required to suspend and trans-
port the solids and their gravitational potential energy. However, the lack of scale-separation leads to
many uncertainties in quantifying mesoscale structures. The authors have incorporated this model into
computational fluid dynamics (CFD) simulations which have shown improvements over traditional
drag models. Fully resolved simulations, such as those mentioned in this chapter and the subject of a
later chapter on Immersed Boundary Methods, can be used to obtain additional information about these
mesoscale structures. This can be used to formulate better constitutive equations for continuum models.
Chapter 5
Mass and Heat Transfer Modeling ...................................................................................................... 178
Ronald W. Breault, National Energy Technology Laboratory, USA
This chapter focuses on the important topic of mass and heat transfer models and closures for con-
tinuum gas-solids reacting flows. The previous three chapters have primarily focused on the hydrody-
namics of gas-solids flows, However, in addition, accurate models for heat and mass transfer must be
constructed to allow predictive simulations of reacting gas-solids flows. As mentioned before, the goal
of an ideal reactor is to establish the best temperature, reactant species distribution and residence time
conditions for reactions. This, obviously, requires an accurate understanding the heat and mass transfer
within the gas and also across the gas-solids interface. This chapter provides an overview of this topic
for dilute and dense gas-solids systems. Specifically, it covers diffusional mass transfer, turbulent dis-
persion, and convective heat transfer between the different phases and, also, at the boundaries. Again in
this case, there is a lack of scale-separation. Thus, there are large uncertainties in the models available
in the literature. Fully resolved simulations can lead to a reduction of this uncertainty.
Section 2
Numerical Methods
Chapter 6
Coupled Solvers for Gas-Solids Flows ............................................................................................... 204
In recent years, the application of coupled solver techniques to solve the Navier-Stokes equations has
become increasingly popular. The main reason for this, is the increased robustness originating from
the implicit and global treatment of the pressure-velocity coupling. The drawback of a coupled solver
are the increase in memory requirement and the increased complexity of implementation. However,
fully coupled methods are reported to have an overall favorable computational cost when a suitable
pre-conditioner and algorithm for solving the resulting set of linear equations are employed. In solving
multi-phase flow problems, the coupled solver approach is even more advantageous than in single-
phase, due to the presence of large source terms arising from the coupling of the phases. In this chapter,
various strategies for the fully coupled approach are discussed. These strategies include employing ar-
tificial compressibility, applying physically consistent cell face interpolation, and applying momentum
weighted cell face interpolation. The idea behind the strategies is outlined and their advantages and
disadvantages are discussed. The treatment of source terms and volume fraction in coupled methods is
also shown. Finally, a number of examples of implementations and calculations are presented.
Chapter 7
Quadrature-Based Moment Methods for Polydisperse Gas-Solids Flows ......................................... 221
Alberto Passalacqua, Iowa State University, USA
Prakash Vedula, University of Oklahoma, USA
Classical Euler-Euler two-fluid models based on the kinetic theory of granular flows assume the particle
phase to be dominated by collisions, even when the particle volume fraction is low and hence collisions
are negligible. This leads to erroneous predictions of the particle-phase flow patterns and to the inabil-
ity of such models to capture phenomena like particle trajectory crossing for finite Stokes numbers. To
correctly predict the behavior of dilute gas-particle flows a more fundamental approach based on solv-
ing the Boltzmann kinetic equation is necessary to treat non-zero Knudsen-number and finite Stokes-
number conditions. In this chapter an Eulerian quadrature-based moment method for the direct solution
of the Boltzmann equation is adopted to describe the particle phase, and it is fully coupled with an
Eulerian fluid solver to account for the two-way coupling between the phases. The solution algorithm
for the moment transport equations derived in the quadrature-based moment method and the coupling
procedure with a fluid solver are illustrated. The predictive capabilities of the method are shown con-
sidering a lid-driven cavity flow with particles at finite Stokes and Knudsen numbers, and comparing
the results with both Euler-Euler two-fluid model predictions and with Euler-Lagrange simulations.
Chapter 8
Direct Numerical Simulation of Gas-Solids Flow Based on the Immersed Boundary Method ......... 245
Rahul Garg, Iowa State University, USA
Sudheer Tenneti, Iowa State University, USA
Jamaludin Mohd. Yusof, Los Alamos National Laboratory, USA
Shankar Subramaniam, Iowa State University, USA
In this chapter, the Direct numerical simulation (DNS) of flow past particles is described. DNS is a
first-principles approach for modeling interphase momentum transfer in gas-solids flows that does not
require any further closure as the flow around the particles is fully resolved. In this chapter, immersed
boundary method (IBM) is described where the governing Navier-Stokes equations are modeled with
exact boundary conditions imposed at each particle surface using IBM and the resulting three dimen-
sional time-dependent velocity and pressure fields are solved. Since this model has complete descrip-
tion of the gas-solids hydrodynamic behavior, one could extract all the Eulerian and Lagrangian statis-
tics for validation and development of more accurate closures which could be used at coarse-grained
simulations described in other chapters.
Chapter 9
The Multiphase Particle-in-Cell (MP-PIC) Method for Dense Particle Flow .................................... 277
Dale M. Snider, CPFD, USA
Peter J. O’Rourke, CPFD, USA
This chapter is a review of the multiphase particle-in-cell (MP-PIC) numerical method for predicting
dense gas-solids flow. The MP-PIC method is a hybrid method such as IBM method described in the
previous chapter, where the gas-phase is treated as a continuum in the Eulerian reference frame and the
solids are modeled in the Lagrangian reference frame by tracking computational particles. The MP-PIC
is a derivative of the Particle-in-Cell (PIC) method for multiphase flows and the method employs a
fixed Eulerian grid, and Lagrangian parcels are used to transport mass, momentum, and energy through
this grid in a way that preserves the identities of the different materials associated with the particles.
The main distinction with traditional Eulerian-Lagrangian methods is that the interactions between the
particles are calculated on the Eulerian grid. One of the main advantages of PIC methods is the accu-
racy of the convection algorithms (Lagrangian advection is non-diffusive) and this can be important
aspect of gas-solids flows where the overall dynamics is dictated by the instabilities in the system due
to sharp interfaces between the phases. Additional details about this method along with examples are
provided in this chapter. In the following chapter, the application of this method to Circulating Fluid-
ized Beds (CFBs) is described.
Section 3
Practice
Chapter 10
Circulating Fluidized Beds ................................................................................................................. 316
S.B. Reddy Karri, PSRI, USA
Ted Knowlton, PSRI, USA
In the last 20 years, significant improvements in the computational fluid dynamics (CFD) modeling
have been made that allow the simulation of large-scale, commercial CFBs. Today, commercial codes
are available that can model some of this behavior in large-scale, commercial units in a reasonable
amount of time. However, the hydrodynamics in a riser or fluidized bed are complex with both micro
and macroscale features. From particle clustering to large streamers to the core-annulus profile, the par-
ticle behavior in these unit operations rarely behaves as a “continuous fluid.” Even the role of particle
size distribution is often neglected and models that do consider particle size distribution don’t always
consider the role of particle size on granular temperature. Many models use insufficient boundary con-
ditions by assuming uniform or symmetric profiles, which is rarely the case. Furthermore, grid sizing
is usually based on computer limitations instead of model limitations, and many models of commercial
systems extend beyond the capability of the constitutive equations being used. Successful application
of today’s CFD models requires a good understanding of the equations behind the code, the assump-
tions used for those equations and the capability or limitations of the code. CFD is nothing more than a
guess without an understanding of the fundamentals, underlying assumptions and code limitations that
are part of every model.
Chapter 11
CFD Modeling of Bubbling Fluidized Beds of Geldart A Powders ................................................... 357
T. Pugsley, The University of Saskatchewan, Canada
S. Karimipour, The University of Saskatchewan, Canada
Z. Wang, The University of Saskatchewan, Canada
Fluidized beds have applications in a range of industrial sectors from oil refining and coal combustion
to pharmaceutical manufacture and ore roasting. In spite of more than 80 years of industrial experience
and a tremendous amount of academic attention, the fundamental understanding of fluidized bed hydro-
dynamics is still far from complete. Advanced modeling using computational fluid dynamics is one tool
for improving this understanding. In the current chapter the focus is on the application of CFD to a par-
ticularly challenging yet industrially relevant area of fluidization: fluidized beds containing Geldart A
powders where interparticle forces influence the bed behavior. A critical step in modeling these systems
is proper representation of the interfacial drag closure relations. This is because the interparticle cohe-
sive forces lead to the formation of clusters that reduce the drag below that for non-cohesive particles.
The influence on the predictions of the macroscale fluidized behavior due to mesoscale phenomena
such as clustering is discussed. At present in the literature, approaches to representing the reduced drag
arising from mesoscale phenomena is ad hoc. There is a pressing need for an improved understanding
of cluster formation and for robust models describing it that can be incorporated into coarse grid models
of industrial scale fluidized beds.
Chapter 12
Computational Modeling of Gas-Solids Fluidized-Bed Polymerization Reactors ............................. 373
Ram G. Rokkam, Iowa State University, USA
Michael E. Muhle, Univation Technologies, USA
Gas-solids flows have numerous industrial applications and are also found in natural processes. They
are involved in industries like petrochemical, polymer, pharmaceutical, food and coal. Fluidization is a
commonly used gas-solids operation and is widely used in production of polyethylene. Polyethylene is
one of the most widely used thermoplastics. Over 60 million tons are produced worldwide every year
by both gas-phase and liquid-phase processes. Gas-phase processes are more advantageous and use
fluidized-bed reactors (e.g., UNIPOLTM PE PROCESS and Innovene process) for the polymerization
reactions. In this work a chemical-reaction-engineering model incorporating a given catalyst size distri-
bution and polymerization kinetics along with the quadrature method of moments is used to predict the
final polymer size distribution and temperature. An Eulerian-Eulerian multi-fluid model based on the
kinetic theory of granular flow is used to solve the fluidized-bed dynamics and predict behavior such as
particle segregation, slug flow and other non-ideal phenomena.
Chapter 13
Validation Approaches to Volcanic Explosive Phenomenology ......................................................... 398
Sébastien Dartevelle, Los Alamos National Laboratory, USA
Large-scale volcanic eruptions are inherently hazardous events, hence cannot be described by detailed
and accurate in situ measurements. As a result, volcanic explosive phenomenology is poorly under-
stood in terms of its physics and inadequately constrained in terms of initial, boundary, and inflow
conditions. Consequently, little to no real-time data exist to validate computer codes developed to
model these geophysical events as a whole. However, code validation remains a necessary step, par-
ticularly when volcanologists use numerical data for assessment and mitigation of volcanic hazards
as more often performed nowadays. We suggest performing the validation task in volcanology in two
steps as followed. First, numerical geo-modelers should perform the validation task against simple and
well-constrained analog (small-scale) experiments targeting the key physics controlling volcanic cloud
phenomenology. This first step would be a validation analysis as classically performed in engineering
and in CFD sciences. In this case, geo-modelers emphasize on validating against analog experiments
that unambiguously represent the key-driving physics. The second “geo-validation” step is to compare
numerical results against geophysical-geological (large-scale) events which are described ―as thor-
oughly as possible― in terms of boundary, initial, or flow conditions. Although this last step can only
be a qualitative comparison against a non-fully closed system event —hence it is not per se a validation
analysis—, it nevertheless attempts to rationally use numerical geo-models for large-scale volcanic
phenomenology. This last step, named “field validation or geo-validation”, is as important in order to
convince policy maker of the adequacy of numerical tools for modeling large-scale explosive volca-
nism phenomenology.
Compilation of References .............................................................................................................. 430
About the Contributors ................................................................................................................... 467
Index ................................................................................................................................................... 471

xiv
Foreword
This collection of 13 chapters on gas-solids flow describes the new theories, numerical methods and
some new applications (e.g. polymerization reactors and volcanic eruptions) that have emerged in
the last two decades. This book is similar to “Fluidization”, edited by J.F. Davidson and D. Harrison,
Academic Press, 1971. Both books review various subjects in depth and can be read by both beginners
and experts. Hence this book, like the Davidson and Harrison volume, should be useful to researchers,
graduate students and professors. I used the Fluidization book in class in a graduate course at IIT in the
mid- seventies and still use some chapters to-date in my consulting work.
The Fluidization book was followed by a second edition, edited by J.F. Davidson, R. Clift and D,
Harrison in 1985. Similarly, this book needs to be followed up by a second volume treating subjects,
such as detailed treatment of heat transfer, nanoparticles, cohesive particles, erosion and new applica-
tions, such as blood flow.
A comparison between the Davidson and Harrison’s books and the present volume shows the tremen-
dous progress made in the last two decades due to the development of the kinetic theory of multiphase
flow and fast computers. In my opinion, many multiphase reactors as documented in this book, such as
coal gasifiers and fluidized bed silicon production reactors can now be designed and optimized using
multiphase computational dynamics codes. This is a cost- and time-effective approach than the construc-
tion of billion dollar synthetic fuel demonstration plants in the 1980s.
The Morgantown group has been a leader in funding innovative research in multiphase flows for the
last three decades. The 1982 Coal Gasification modeling workshop proceedings, edited by M. Ghate and
J.W. Martin describe the first gas-solids computational fluid dynamics code. I have reviewed this study
in my 1994 book. Madhava Syamlal and Tom O’Brien have been supporters of this transformational
research for the last two decades and it is now bearing fruit.
The first chapter, in the “Theory Section,” by Syamlal and Pannala describes the mass, momentum,
energy and chemical species balances and their approximations found in the DOE NETL MFIX computer
code. This code and similar commercial codes are now being used in design and scale-up of gas-solids
reactors all over the world.
The second chapter by Dufty and Baskaran reviews the kinetic theory as a basis for hydrodynamics.
It combines the classical physics approach with some recent numerical simulations.
The third chapter by Hrenya reviews some of the recent polydisperse kinetic theories. In a mixture of
gases, the temperatures of the components are equal. But in a mixture of particles the granular tempera-
tures differ greatly due to inelastic collisions. This leads to unequal particle viscosities and segregation
phenomena which as yet remain unexplained.
xv
The fourth chapter by Ge, Yang, Wang and Jinghai Li reviews the interfacial drag relations. The
unique energy minimization multiscale model of Li and Kwauk gives some trends similar to those be-
ing developed by Sundaresan using different methods. Corrections to standard drag relations are as old
as fluidization, but still require further research to quantitatively predict some flow regimes, such as the
dense and the dilute regions in turbulent fluidization, where Li and Kwauk’s corrections were the key
to prediction. The fifth chapter reviews the heat and mass transfer relations. It has been known for half
a century that the Nusselt and Sherwood numbers for fluidization of fine particles are orders of magni-
tude below the conduction or diffusion limit of two. It is only recently that we are finding out that this
is probably due to the formation of clusters.
In chapter one, part two, “Numerical Methods”, van Wachem reviews the coupled solvers for gas-
solids flow. His example is for a single phase cavity driven flow.
In chapter two, Passalacqua, Vedula and Fox describe the quadrature based moment methods for
polydisperse gas-solids flows. In this method local isotropy is not assumed. In addition to granular tem-
perature, all velocity moments are computed. In principle, convergence should be obtained. Hence, this
is a more advanced kinetic theory based method. The third chapter on numerical methods by Garg, Ten-
neti, Yusof and Subramaniam describes the immersed boundary method. This approach has reproduced
known results for the average drag in Stokes flow past ordered arrays of particles. The fourth chapter
by Snider and O’Rourke describe their multiphase numerical method for predicting dense particulate
flow. In this method the fluid is treated as a continuum and the solids are modeled by Lagrangian com-
putational particles.
The first chapter under “Practice” by Cocco, Karri and Knowlton gives a thorough review of ap-
plications, hydrodynamics, axial and radial solids profiles, clusters, segregation and CFD modeling of
circulating fluidized beds. The authors favor the Barracuda code described by Snider and O’Rourke for
polydisperse particles. The second chapter under “Practice” by Pugsley, Karimipour and Wang describes
CFD modeling of bubbling fluidized beds of Geldart A powders. The authors used the MFIX code to
model a stripper. The third chapter by Rokkam, Fox and Muhle describes computational modeling of
gas-solids fluidized bed polymerization reactors. The authors are using ANSYS FLUENT 6.3 software
to predict slugs. The last chapter by Dartevelle describes explosive modeling of volcanoes. Videos of the
simulations and actual pyroclastic flows have been shown on public television all over the world. In this
context, multiparticle size simulation is described by A.Neri, et al in Journal of Geophysical Research
(2003) using a code based on Syamlal’s early computer code and multisize particle model.
A comparison between the 1971 and 1985 Fluidization volumes and this collection of chapters shows
the transition we have made in the last three decades from an almost empirical science to one based
on computation. Hopefully the industry will use this new science to improve the efficiency of energy
conversion and other processes.
In summary, this book will be a valuable resource to the gas-solids community of graduate students,
academicians and researchers and serves as a good complement to existing books in this field.
Dimitri Gidaspow
Distinguished Professor
Illinois Institute of Technology
xvi
Preface
An important problem facing the world today is finding sustainable, cost-effective and ecologically
friendly energy sources. In order to meet a growing global energy demand, especially in developing na-
tions, the increased supply of such sources must augment continuing attention to the efficient utilization
of energy. Reacting gas-solids flows play an important role in many aspects of efficient fuel production
and energy production and energy utilization, e.g:
• Fuel production and processing: catalytic crackers, H2 production, S removal, coal gasification,
gas clean-up (SOx, NOx, Hg, CO2), biomass (cellulosic) pyrolysis and gasification, nuclear fuel
production
• Energy production: fuel cells, coal and biomass combustion, CO2 separation, nuclear reactors
and separation, silicon production and coating for photovoltaic applications, novel combustion
technologies like oxy-combustion and chemical looping
• Energy utilization and efficiency: polymerization reactors, catalytic reactors, iron and steel mak-
ing, cement production
These energy sources have to satisfy the current demand but must also meet the projected demand in
2050 (doubled) and at the end of the century (tripled). Considering the inflexibility of the energy markets
to even slight fluctuations in demand, this implies an enormous challenge to the scientific community
and society in general. This is complicated by the fact that the prevalent technology development process
is based on experimentation, using Edisonian approaches which require several iterations to develop
the end product. Computational science and engineering, using high performance computers, can ac-
celerate this development process by providing an integration framework which allows scaling between
the laboratory experiments and commercial units, facilitates cooperation between various science and
engineering disciplines. The topical area of computational gas-solids reacting flows, which is critical to
developing new, clean and efficient energy alternatives, is still in its infancy. This book’s objective is to
bring together the recent advances in all the interdisciplinary fields which impact this area. We believe
that this volume provides an intersection of all aspects necessary for understanding computational gas-
solids flows and will enable researchers from academia/industry and designers from industry to push
the state-of-the-art and find rapid solutions for urgent research and design problems.
The objective of reactor design is to create the right conditions for reactions. The temperature and
reactant species distribution, appropriate residence time and removal of products must be considered.
Including the effect of a catalyst may be necessary. A comprehensive understanding of all the competing
and interacting mechanisms is required to arrive at better designs and improved processes. In particular,
xvii
gas-solids reacting flows involve, not only complex interactions of granular materials with gas flow, but
also phase-change, heterogeneous and homogeneous reactions, heat and mass transfer. Moreover, the
spatial and temporal scales may vary over many orders of magnitude. Thus modeling gas-solids react-
ing flows requires the integration of the best physics and chemistry models from various science and
engineering fields with the most advanced computational algorithms. These algorithms must be scalable
to large high-performance computers in order to bear on this important topic. In addition, because of
the nascent stage of this field, these theory approaches and numerical techniques are constantly evolv-
ing. This book attempts to bring the latest advances into one collection. Specifically, it presents the
multiphase continuum formulation for gas-solids flows along with the theory for closure relations for
granular stresses, heat and mass transfer correlations, models for chemical reactions, as well as numerical
algorithms appropriate for of high-performance computing. The book also illustrates the applications
of these computational tools to general gas-solids flows with a focus on energy conversion processes.
This book consists of a comprehensive collection of authoritative works on computational gas-solids
flows, encompassing theory, numerical methods and practice. It is distinguished from other books on
multiphase flows since it is entirely devoted to all aspects of computational gas-solids flows and also
serves as a bridge to the various disciplines critical to further advances of this topic, physics, chemis-
try, computational science, applied mathematics and various applied engineering fields. A book of this
nature is very much needed since computational science is fast becoming the “third pillar” of science
and engineering, joining theory and experimentation, due to the rapid advances in both computational
hardware and algorithms. It should serve as a commanding reference for researchers in computational
gas-solids flows. It is also meant as a text book to train advanced graduate students in the concepts and
applications of computational science and engineering that is beginning to transform the study and
design of gas-solids reacting systems.
The primary objectives of the authors and editors are to provide:
• A comprehensive book on computational gas-solids reacting flows, which presents the most recent
developments in theory and numerical techniques
• Examples of applications where these computational tools have been employed
• A convenient reference book for practicing engineers and researchers in computational gas-solids
flows
• A text book for advanced graduate level course in engineering and science
The primary audience of this book is intended to be:
• Advanced graduate students in interdisciplinary science and engineering subject areas or compu-
tational science
• Researcher workers and practicing engineers
• Users of multiphase flow software, such as the open-source MFIX or the commercial codes FLU-
ENT®, CFX®, etc.
xviii
OrganizatiOn
This book has thirteen chapters in total, divided into three sections: Theory, Numerical Methods, and
Practice. Section one, Theory, has five chapters. The first chapter discusses the development of the con-
tinuum theories and an overview of the various closures needed to solve gas-solids reacting flows. The
second and third chapters provide kinetic theory based closures for granular stresses for mono-disperse
and poly-disperse systems. The fourth chapter covers wide-ranging issues related to interfacial drag
closures and finally the fifth chapter discusses closures for heat and mass transfer. Section two consists
of four chapters and they cover various methods to solve the theoretical descriptions provided in the
first section. The first chapter in this section introduces fully coupled implicit solvers for gas-solids
flows and how these methods contrast with segregated solvers on staggered grids. The second chapter
deals with the use of DQMOM (direct quadrature method of moments) for polydisperse gas-solids
flows. The third chapter covers the topic of direct numerical simulations for gas-solids flows where
the particle collisions and flow over the particles are resolved. The fourth chapter in this section delves
into the mixed Eulerian-Lagrangian hybrid methods based on MP-PIC (Multiphase particle-in-cell).
The third section comprises of four chapters that give four illustrative examples of the application of
computational gas-solids flows. The first chapter in this section deals with circulating fluidized beds and
the recent success of MP-PIC methods in simulating large-scale risers. The second chapter details the
simulation of bubbling fluidized beds for group A particles. The third chapter is on the polymerization
reactors and illustrates the effectiveness of population balance methods to model particle-size changes.
The final chapter details the application of the methods discussed in this book for large-scale simulations
of volcanic flows with particular emphasis on verification and validation. Below we provide an overview
of these thirteen chapters in this book:
theory
Multiphase Continuum Formulation for Gas-Solids Reacting Flows

Madhava Syamlal and Sreekanth Pannala
This introductory chapter lays out the formulation of multiphase continuum models for gas-solids flows,
including heat and mass transfer with chemical reactions. It presents the basis for the different continuum
formulations for mass, momentum, granular energy, thermal energy, and species balance equations for
interpenetrating gas and (multiple) solids phases along with the constitutive relationships needed to close
the coupled system of equations. This chapter is primarily derived from the authors’ extensive involve-
ment in developing an open-source computational platform for gas-solids flows, MFIX (http://mfix.
netl.doe.gov). A detailed history of the development of this field over the last 2-3 decades is presented
along with extensive references for the interested reader to explore the literature in greater depth. Finally
provides an illustrative example explores the effects of exercising the different model and simulation
options. The chapter not only serves as a good guide for a widely accepted simulation software but one
could also obtain the software along with the example problem to get a “hands-on” experience and, pos-
sibly, to contribute to this growing field. The chapter also provides a common context for later chapters.
xix
Hydrodynamic Equations from Kinetic Theory: Fundamental Considerations

James Dufty and Aparna Baskaran
This chapter provides a theoretical description for the dynamics of the granular phase only, with empha-
sis on the derivation of the macroscopic hydrodynamic fields for the granular phase (species densities,
flow velocity, and the granular temperature) for a prescribed microscopic interaction among the grains.
This is achieved by introducing the general notion of a kinetic equation to obtain macroscopic balance
equations for the fields. The constitutive equations for the fluxes appearing in these balance equations
are obtained through specialized solutions to the kinetic equation, resulting in a closed set of hydrody-
namic equations. This is demonstrated using the Boltzmann-Enskog kinetic equation for a system of
smooth, inelastic hard spheres that leads to granular Navier-Stokes hydrodynamic equations for weakly
inhomogeneous fluid states. For gas-solids flows, the role of interstitial gas is important. This effect can
be included, along with other body forces, as source terms in the kinetic equations. This chapter also
includes a brief discussion about hydrodynamics beyond the Navier- Stokes limit.
Kinetic Theory for Granular Materials: Polydispersity

Christine M. Hrenya
In this chapter, existing kinetic theory based models for granular flows are critically evaluated in the
context of polydisperse granular flows. The species segregation predicted by these kinetic theory models
is reviewed and rigorous extensions of these models for polydisperse gas-solids flows are discussed.
There has been significant recent progress in advancing the state-of-the-art for polydisperse, kinetic-
theory models for granular flows and many of the simplifications associated with earlier models (e.g.,
equipartition of energy) are now treated in a more rigorous fashion. However, for gas-solids flows, the
incorporation of gas effects into polydisperse models is just beginning and further work is needed to
improve their predictive capabilities.
Interfacial Interactions: Drag

Wei Ge, Ning Yang, Wei Wang and Jinghai Li
The drag interaction between gas and solids not only acts as a driving force for solids in gas-solids flows
but also plays as a major role in the dissipation of the energy due to drag losses. This leads to enormous
complexities as these drag terms are highly non-linear and multiscale in nature because of the variations
in solids spatio-temporal distribution. This chapter provides an overview of this important aspect of the
hydrodynamic interactions between the gas and solids and the role of spatio-temporal heterogeneities
on the quantification of this drag force. In particular, a model is presented which introduces a mesoscale
description into two-fluid models for gas-solids flows. This description is formulated in terms of the
stability of gas-solids suspension. The stability condition is, in turn, posed as a minimization problem
where the competing factors are the energy consumption required to suspend and transport the solids
and their gravitational potential energy. However, the lack of scale-separation leads to many uncertain-
ties in quantifying mesoscale structures. The authors have incorporated this model into computational
fluid dynamics (CFD) simulations which have shown improvements over traditional drag models. Fully
resolved simulations, such as those mentioned in this chapter and the subject of a later chapter on Im-
mersed Boundary Methods, can be used to obtain additional information about these mesoscale structures.
This can be used to formulate better constitutive equations for continuum models.
xx
Mass and Heat Transfer Modeling

Ronald W. Breault
This chapter focuses on the important topic of mass and heat transfer models and closures for continuum
gas-solids reacting flows. The previous three chapters have primarily focused on the hydrodynamics of
gas-solids flows, However, in addition, accurate models for heat and mass transfer must be constructed
to allow predictive simulations of reacting gas-solids flows. As mentioned before, the goal of an ideal
reactor is to establish the best temperature, reactant species distribution and residence time conditions
for reactions. This, obviously, requires an accurate understanding the heat and mass transfer within the
gas and also across the gas-solids interface. This chapter provides an overview of this topic for dilute
and dense gas-solids systems. Specifically, it covers diffusional mass transfer, turbulent dispersion, and
convective heat transfer between the different phases and, also, at the boundaries. Again in this case, there
is a lack of scale-separation. Thus, there are large uncertainties in the models available in the literature.
Fully resolved simulations can lead to a reduction of this uncertainty.
numerical Methods
Coupled Solvers for Gas-Solids Flows

Berend van Wachem
Segregated solvers commonly used in most gas-solids flow software converge one variable at a time and
ensure overall convergence through non-linear iterations through the variables. On the contrary, the main
advantage of collocated, non-Cartesian, coupled solvers, the approach discussed in this chapter, is an
increased robustness due to the implicit treatment of the pressure-velocity coupling or in general all the
variables. Although the equations describing multiphase flows appear similar to those of single-phase
flows, their solution is usually much more difficult since the coupling between equations is often more
important than the coupling between terms in each equation, which is the default case for single-phase
flow. This is due to the presence of volume fractions which are constrained to sum to unity, and large
source terms, as well as their gradients. This makes the availability of a robust solution method extremely
desirable. A number of approaches for a fully coupled solution approach are discussed in this chapter
and application of coupled solvers for gas-solids flows is still in its initial phase.
Quadrature-Based Moment Methods for Polydisperse Gas-Solids Flows

Alberto Passalacqua, Prakash Vedula, and Rodney Fox
Common Eulerian-Eulerian two-fluid models assume that the particle phase to be dominated by colli-
sions and employ kinetic theory even when the particle volume fraction is low. Since collisions do not
dominate in the extremely dilute regime, this assumption leads to erroneous predictions of the particle
phase flow patterns and the models cannot capture phenomena like particle trajectory crossing for finite
Stokes numbers. In this chapter, an Eulerian quadrature-based moment method (QMOM) for the direct
solution of the Boltzmann kinetic equation for the particle phase coupled with Eulerian fluid solver is
described. This is a more fundamental approach to treat non-zero Knudsen-number and finite Stokes-
number conditions for dilute gas-particle flows. The chapter details the derivation of QMOM method
for moment transport equations and the coupling procedure with the fluid solver. The predictions of the
method are shown for a lid-driven cavity flow with particles, at finite Stokes and Knudsen numbers,
and compared with both Eulerian-Eulerian two-fluid model predictions and with Eulerian-Lagrangian
simulations.
xxi
Direct Numerical Simulation of Gas-Solids Flows Based on the Immersed Boundary

Method
Rahul Garg, Sudheer Tenneti, Jamaludin Mohd.-Yusof, and Shankar Subramaniam
boundary method (IBM) is described where the governing Navier-Stokes equations are modeled with
exact boundary conditions imposed at each particle surface using IBM and the resulting three dimensional
time-dependent velocity and pressure fields are solved. Since this model has complete description of
the gas-solids hydrodynamic behavior, one could extract all the Eulerian and Lagrangian statistics for
validation and development of more accurate closures which could be used at coarse-grained simula-
tions described in other chapters.
The MP-PIC Method for Dense Particle Flows

Dale M. Snider and Peter J. O’Rourke
is a derivative of the Particle-in-Cell (PIC) method for multiphase flows and the method employs a
fixed Eulerian grid, and Lagrangian parcels are used to transport mass, momentum, and energy through
this grid in a way that preserves the identities of the different materials associated with the particles.
The main distinction with traditional Eulerian-Lagrangian methods is that the interactions between the
particles are calculated on the Eulerian grid. One of the main advantages of PIC methods is the accuracy
of the convection algorithms (Lagrangian advection is non-diffusive) and this can be important aspect
of gas-solids flows where the overall dynamics is dictated by the instabilities in the system due to sharp
interfaces between the phases. Additional details about this method along with examples are provided
in this chapter. In the following chapter, the application of this method to Circulating Fluidized Beds
(CFBs) is described.
Practice
Circulating Fluidized Beds

Ray Cocco, S.B. Reddy Karri and Ted Knowlton
In this chapter, the general features of the circulating fluidized beds (CFBs) are presented. In particular
the complex hydrodynamic behavior in CFBs is described where one can observe particle clustering,
streaming, core-annulus concentration profiles and segregation. This chapter also describes the vari-
ous applications of CFBs including FCC (fluid catalytic crackers), coal/biomass gasification, chemical
looping. This chapter also details the significant improvements made over the last two decades in the
computational fluid dynamics (CFD) modeling that have made it possible to simulate large-scale, com-
mercial CFBs. The practical application of CFD models to study commercial CFBs is described where
sometimes one has to compromise with coarse grid because of compute power limitations or incomplete
boundary conditions because of lack of detailed measurements or even insufficient models to describe
the multiscale behavior of the CFBs adequately.
xxii
CFD Modeling of Bubbling Fluidized Beds of Geldart A Powders

Todd Pugsley, S. Karimipour and Z. Wang
In this chapter, the role of fluidized beds in a range of industrial sectors, from oil refining and coal com-
bustion to pharmaceutical manufacture and ore roasting is described. Even after 80 years of experience
using these systems in the industry and extensive experimental research at academic/research institu-
tions, the fundamental understanding of fluidized bed hydrodynamics is still far from complete. This
chapter details the role of advanced modeling using CFD as one tool for improving this understanding
and focuses on the application of CFD to fluidized beds containing fine Geldart A powders. One of the
main features of these beds is the formation of clusters that reduce the drag below that for non-cohesive
particles. This chapter discusses the role of mesoscale clustering on the predictions of the macroscale
fluidized behavior and emphasizes the fact that approaches to reduce drag such as the one described in
this chapter are ad hoc. Thus, there is a pressing need for an improved understanding of cluster forma-
tion (possibly through methods such as IBM described before) and for robust models describing their
effect that can be utilized in coarse grid simulations of industrial scale fluidized beds
Computational Modeling of Gas-Solids Fluidized Bed Polymerization Reactors

Ram G. Rokkam, Rodney Fox and Michael Muhle
This chapter discusses the fluidization process used for production of polyethylene, one of the most
widely used thermoplastics. The gas-phase process is based on fluidized-bed reactors (e.g., UNIPOLTM
PE PROCESS and Innovene process) for the polymerization units. In this chapter, a detailed CFD
model that incorporates the catalyst size distribution along with polymerization kinetics is employed. A
quadrature method of moments (QMOM) is employed to predict the final polymer size distribution and
temperature. The chapter details the Eulerian-Eulerian model based on the kinetic theory of granular
flow that is used to solve the fluidized-bed dynamics and the predictions of particle segregation, slug
formation and other non-ideal phenomena.
Validation Approaches to Volcanic Explosive Phenomenology

Sébastien Dartevelle
This chapter deals with natural large scale gas-solids flows unlike the commercial/laboratory devices at
smaller scales described before. This chapter describes the fact that volcanic explosive phenomenology is
poorly understood in terms of its fundamental physical processes as it is difficult to collect data because
of the inherent uncertainties about the location and occurrence of volcanic eruptions and any placement
of measurement devices. In addition, the accuracy of any simulation tools is severely constrained be-
cause of the limitations on the initial, boundary, and inflow conditions. As such it is important to conduct
detailed verification and validation to increase the benefits of simulations. This chapter presents a two
step approach for validating volcanology models. The first part involves validation against simple and
well-constrained analog (small scale) experiments targeting the key physics controlling volcanic cloud
phenomenology. The second “geo-validation” step requires comparing against well characterized (as
much as possible) geophysical-geological (large scale) events. The author points out that this last step
can only be qualitative as the natural system is not fully characterized. However, it is a necessary step
to have reasonable confidence that the models describe large scale volcanic phenomenology and be of
use to policy-makers and others to make decisions based on these models.
xxiii
In conclusion, this book brings together the latest work on theory, numerical techniques and applica-
tions for gas-solids reacting flows from the experts in the various interdisciplinary fields. This book is
meant for interdisciplinary students and researchers from diverse fields, like physics, chemistry, applied
mathematics and various engineering disciplines. It also articulates unresolved issues within these fields
which should be addressed in order to advance the entire area of computational science and engineer-
ing for gas-solids reacting flows. This would have great relevance to the design of industrial processes,
particularly in energy related industries. To our knowledge, this will be the first book on this topic which
brings together literature from different fields and, thus, act as a bridge between these fields, serving as
a vehicle of knowledge dissemination. This book is also unique in that it comprehensively relates the
subject of computational gas-solids flows with modern high-performance computing and the general
area of computational science.
Section 1
Theory
1
Chapter 1
Multiphase Continuum
Formulation for Gas-
Solids Reacting Flows
Madhava Syamlal
Sreekanth Pannala
Oak Ridge National Laboratory, USA
abstract
This chapter describes the formulation of multiphase continuum models for gas-solids flows with chemi-
cal reactions. A typical formulation of the equations is presented here, following the equations in the
open-source software MFIX (http://mfix.netl.doe.gov) so that interested users may look up details of
the numerical implementation, study the solutions, or experiment with the numerical implementation of
alternative formulations. The authors will first provide a brief overview of the significance of gas-solids
reacting flows and the challenges in modeling these systems along with various efforts undertaken by
different groups over the last 2–3 decades. They will then summarize the methods used to derive mul-
tiphase continuum models and to formulate constitutive equations. They will later provide information
on the formulation for mass, momentum, granular energy, energy, and species balance equations for
gas and multiple solids phases. They will discuss the constitutive equations required in each of the bal-
ance equations; a detailed discussion of certain constitutive equations, such as the gas-solids drag and
granular stresses (derived from kinetic theory), will be presented by other authors in later chapters.
The authors will point out the differences between different approaches and direct the reader to refer-
ences that discuss those approaches in detail. They will end the chapter with the example problem of
the simulation of a bubbling fluidized bed to illustrate some of the modeling options — physical models,
numerical discretization schemes, and grid resolution – that need to be considered to accurately simulate
gas-solids systems.
DOI: 10.4018/978-1-61520-651-3.ch001
Copyright © 2011, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
1. intrOductiOn
Gas-solids reactors, which are critical components in many energy and chemical conversion processes,
cannot be designed reliably using correlation-based scale-up methods, and a physics-based method is
needed for increasing the reliability of the design. There are many examples: coal gasifiers that react
coal with oxygen and steam to produce synthesis gas (syngas)—a mixture of hydrogen and carbon mon-
oxide; circulating fluidized-bed combustors that burn coal to generate heat and electric power; or fluid
catalytic cracking (FCC) risers that crack heavy oil with the help of hot catalyst particles, producing
light hydrocarbons such as gasoline. Increasing the conversion of the reactants (e.g., coal in a gasifier)
or improving the selectivity of the products (e.g., gasoline in FCC) or ensuring the reliability of the reac-
tors or reducing the size of the reactors are design objectives that would help increase the profitability
of existing processes or ensure the economic viability of a novel process. For example, reliability is the
single most important technical limitation to be overcome to enable widespread deployment of gasifica-
tion technology (Clayton, Stiegel, & Wimer, 2002). For the rapid development of post-combustion CO2
capture technology, solid-sorbent based fluidized bed reactors need to be designed to reduce capital cost
and energy consumption (Ciferno, Fout, Jones, & Murphy, 2009). The design of such gas-solids reactors
currently relies on data from laboratory-scale batch reactors or continuous pilot-scale units. Although
many processes have been successfully scaled-up in this manner, some notable failures have occurred
(Krambeck, Avidan, Lee, & Lo, 1987; Squires, Kwauk, & Avidan, 1985). The hydrodynamic behavior
of laboratory-scale units could be different from commercial-scale units that are 20–100 times larger,
making scale up that involves large size changes unreliable. Reducing the uncertainty by building and
testing pilot-scale units at several intermediate scales is both expensive and time consuming. In spite of
the decades of empirical experience, scale up of fluidized-bed reactors and transfer systems is even today
characterized as “not an exact science” and as a “daunting task” by engineers well versed in the design
of such systems (Karri & Knowlton, 2005). A physics-based approach, multiphase computational fluid
dynamic (CFD) modeling, that has emerged during the last three decades is rapidly gaining acceptance
as a tool for designing, scaling up, and trouble-shooting gas-solids devices and reactors.
Developments in multiphase CFD have been trailing those in single phase CFD, which is by now a
well-recognized tool for the analysis and design of single-phase devices and reactors (e.g. D. Davidson,
2001). Although multiphase CFD has been successfully used for practical applications (e.g. Guenther
in Syamlal, 2006), there are several theoretical and numerical challenges that must be overcome. Mul-
tiphase flows are much more difficult to analyze than single-phase flows primarily because “the phases
assume a large number of complicated configurations” (Hanratty, et al., 2003). In multiphase devices,
the particles collide, shear, and interact; the particles and gas exchange momentum and interact with the
device boundaries; the particles and gas exchange heat and mass; and heterogeneous and homogeneous
chemical reactions occur at greatly different scales. The multiscale processes involved in these devices
span a wide-range of spatio-temporal scales and have been characterized as the following (Syamlal,
2006): “… the granular flow in a fluidized bed may range from incompressible to hypersonic, while
the granular media may undergo a phase change similar to a gas-to-solid transition, all within the same
reactor. The volume fraction, stress, and energy typically fluctuate spatially and temporally with am-
plitudes comparable to the mean.”
There are at least five approaches used for describing gas-solids flows, each making a tradeoff between
the modeling effort and the computational cost as discussed below. Greater physical resolution reduces
the modeling effort (the effort required to develop constitutive equations) but increases the computational
2
cost. At one end of the spectrum is direct numerical simulation (DNS) method, which fully resolves the
flow around individual particles by solving Navier-Stokes equations and tracks the particle motion by
solving Newton’s equations of motion (Hao, Pan, Glowinski, & Joseph, 2009; Pan, Joseph, Bai, Glo-
winski, & Sarin, 2002; Pan, Joseph, & Glowinski, 2001) and is the topic of chapter 3.3. This method
is the cheapest in modeling effort and the most expensive computationally. The size of the system as
well as the physics that can be described by this method is limited; for example, in DNS calculations
of multiphase flows, chemical reactions cannot be included because the spatiotemporal scales that need
to be resolved span over 10 orders of magnitude, making it impossible to perform such simulations
even on the (two) petascale computers available today. A (computationally) less expensive approach is
lattice-Boltzmann method (LBM), which resolves the flow around particles by solving lattice-Boltzmann
equations and tracks the particle motion by solving Newton’s equations of motion (e.g. Beetstra, van
der Hoef, & Kuipers, 2007a, 2007b; Hill, Koch, & Ladd, 2001a, 2001b; Ladd, 1994; Sankaranarayanan,
Shan, Kevrekidis, & Sundaresan, 1999; Sankaranarayanan & Sundaresan, 2008; van der Hoef, Annaland,
Deen, & Kuipers, 2008; Van der Hoef, Beetstra, & Kuipers, 2005; Wylie, Koch, & Ladd, 2003; Yin &
Sundaresan, 2009). Much computational expense can be avoided by not resolving the flow field around
the particles, which leads to the discrete element method (DEM) (e.g. Gera, Syamlal, & O’Brien, 2004;
Goldschmidt, Beetstra, & Kuipers, 2004; Tsuji, Kawaguchi, & Tanaka, 1993), but a price in modeling
effort needs to be paid for not resolving the flow field around the particles in terms of developing con-
stitutive relations for the gas-solids drag. The DEM approach quite effectively accounts for the transfer
of momentum between colliding particles in fleeting contact or sliding particles in enduring contact and
for the effect of particle size and shape. It is possible to track several millions of particles with the cur-
rently available computational capabilities. It is expected that in a decade the computational capabilities
will allow simulations with around 1 billion particles in a realistic 3D geometry (e.g. Tuzun & Cleary,
2006). Even that large a number of particles is not sufficient for modeling industrial gas-solids reactors
because the number of particles in such reactors is typically several orders of magnitude greater than
1 billion; e.g., a commercial FCC reactor contains over a trillion sub-millimeter sized particles (Cocco
and Hrenya in Syamlal, 2006).
Much of the computational time required in DEM simulations is for particle contact detection and
integrating through the contacts (Williams & O’Connor, 1999). The computational effort for contact
detection can be reduced by probabilistic detection of the collisions between sampled particles (rather
than all the individual particles) as in direct simulation Monte Carlo (DSMC, G. A. Bird, 1994; Her-
rmann & Luding, 1998) or altogether avoided by obtaining the collisional stresses from an Eulerian
grid as in Multiphase particle in cell methods (MPPIC, Chapter 2.4, Snider, 2001; Snider, O’Rourke, &
Andrews, 1998) or by not tracking individual particles and treating their collective motion as that of a
fluid. The third option leads to a continuum gas-solids model formulation, which is the subject of this
chapter and several other chapters in this book. Continuum models are well accepted for fluids; our ev-
eryday perception of fluids like water or air is that of a continuous fluid rather than that of a collection of
molecules. Although there are words in English language, such as sand (collection of grains) and gravel
(collection of pebbles), that attribute fluid-like properties to a collection of particles, physical models
based on the fluid-like behavior of a collection of particles have been developed only relatively recently
compared to ordinary fluids. When the equations of motion of discrete particles are averaged, the result-
ing continuum-solids phase co-locates with the fluid phase, leading to an interpenetrating continuum
model (also called a two-fluid model or an Eulerian-Eulerian model). By not having to keep track of
particle contacts, computational expense is considerably reduced. An increased price in the modeling
3
effort, however, must be paid, the effort for developing the complex constitutive relations that go into
the continuum equations—a challenging topic of ongoing research and the subject of several chapters
of this book. The constitutive relations are developed from empirical correlations, analytical solutions
and numerical solutions.
Even the continuum models may not be fast enough for simulating large industrial-scale reactors. A
thumb-rule for grid independent results has been that the grid size should be less than about ten times
the particle diameter (Andrews, Loezos, & Sundaresan, 2005; S. Benyahia, Syamlal, & O’Brien, 2007;
C. Guenther & Syamlal, 2001). In the case of an FCC reactor with a trillion particles, the number of
computational cells required for a truly grid-independent simulation is greater than a billion, which is
two orders of magnitude larger than the grid sizes solvable with current computational capabilities. The
computational challenge is compounded by the curse-of-time—to resolve the transient dynamics using
small time-steps (especially at high spatial resolutions) requiring prohibitive computational resources.
Practical computations of commercial reactors, therefore, rely on coarse grids, and the lack of resolu-
tion is compensated by ad hoc physical models. For example, coarse grids cannot account for the effect
of particle clusters that are smaller than the grid size. Such clusters could have a profound effect on the
gas-solids drag, typically reducing the effective drag. This effect can be accounted for by correcting the
gas-solids drag coefficient (M. Y. Liu, Li, & Kwauk, 2001; B. Lu, et al., 2007; B. N. Lu, Wang, & Li,
2009; Syamlal & O’Brien, 2003). There is an effort underway to systematically determine the correc-
tion in the drag coefficient (and other model parameters) by developing filtered continuum models for
coarse grids, which use fine-grid simulations for generating the required closure relations (Igci, Andrews,
Sundaresan, Pannala, & O’Brien, 2008).
The previous paragraphs discussed the reduction in the computational expense as the physical resolu-
tion is coarsened from DNS to filtered continuum equations and the attendant increase in the complexity
of the constitutive equations or closure relations. This has led to research into multiscale approaches to
reduce the complexity of the closure relations, while also reducing the computational effort. For example,
LBM simulations could be used to generate accurate gas-solids drag correlations (e.g., Beetstra, et al.,
2007a; S. Benyahia, Syamlal, & O’Brien, 2006; Hill, et al., 2001a, 2001b; Van der Hoef, et al., 2005)
that can be used in continuum models. Rather than conducting simulations at one degree of resolution,
generating constitutive relations, and conducting simulations at a higher degree of resolution, there is
also ongoing research into efficiently bridging the approaches at different degrees of resolution. For
example, Mishra et al. (2008) used wavelets to bridge models operating at different spatio-temporal
scales; Van der Hoef et al. (2008) and Lu et al. (2007) have used different approaches to obtain drag
closures, which were linked with CFD calculations.
We will now briefly discuss the developmental history of continuum gas-solids flow models. Con-
tinuum gas-solids models have been under development for over five decades. During the early part of
that development, there was no attempt made to solve the rather formidable set of partial differential
equations constituting the model because of the lack of powerful computers. The models were primar-
ily used to study the stability of the different regimes and explore the details of the bubble motion
(Anderson & Jackson, 1967a, 1967b; Darton, Lanauze, Davidson, & Harrison, 1977; J.F. Davidson &
Harrison, 1963; J. F. Davidson, Harrison, & Carvalho, 1977; Glasser, Kevrekidis, & Sundaresan, 1996,
1997; Glasser, Sundaresan, & Kevrekidis, 1998; Jackson, 2000; Pigford & Baron, 1965; Ruckenstein
& Muntean, 1967; Soo, 1967; Sundaresan, 2003).
When high-speed computers became available, several groups started to solve the multiphase CFD
equations numerically. In the late 1970s, the U.S. Department of Energy (DOE) supported two projects
4
to develop computer models of coal gasifiers based on multiphase CFD. The CHEMFLUB code from
Systems, Science, and Software, Inc., was developed to solve continuum equations for describing gas-
solids flow in fluidized-bed gasifiers (Blake & Chen, 1980; Garg & Pritchett, 1975; Richner, Minoura,
Pritchett, & Blake, 1990). The FLAG code from JAYCOR, Inc., was developed to describe gas-solids
flow using the DEM approach (Scharff, et al., 1982). In parallel to those efforts, Professor Gidaspow’s
group at the Illinois Institute of Technology (IIT) began to develop computer codes for describing flu-
idized beds by adopting numerical techniques developed at Los Alamos National Laboratory (Harlow
& Amsden, 1975a, 1975b) and incorporated in the K-FIX program (Rivard & Torrey, 1977)—a code
originally developed for describing steam-water flow, especially for simulating loss of coolant accidents
in nuclear power plants. A Modified version of K-FIX was used for modeling bubbling fluidized beds
(e.g. D. Gidaspow, 1994; Dimitri Gidaspow & Ettehadieh, 1983; D. Gidaspow & Jiradilok, 2009).
Aided by the widespread availability of computers, their increased computational power, and the
availability of graphical processing tools, the last three decades saw a great increase in the application of
multiphase CFD to gas-solids flows that occur in both engineered and natural systems. Based on these
studies, much progress has been made towards the development and application of comprehensive com-
puter codes for describing gas-solids flows (S. Benyahia, Arastoopour, & Knowlton, 2002; S. Benyahia,
Arastoopour, Knowlton, & Massah, 2000; Bouillard, Gidaspow, & Lyczkowski, 1991; Dartevelle, 2004;
Dartevelle, Rose, Stix, Kelfoun, & Vallance, 2004; Ding & Gidaspow, 1990; D. J. Patil, A. V. Annaland,
& J. A. M. Kuipers, 2005a; D. J. Patil, M. V. Annaland, & J. A. M. Kuipers, 2005b; Sinclair & Jackson,
1989; Syamlal & O’Brien, 2003; Syamlal, O’Brien, Benyahia, Gel, & Pannala, 2008; Syamlal, Rogers,
& O’Brien, 1993; van Wachem & Almstedt, 2003; van Wachem, Schouten, van den Bleek, Krishna, &
Sinclair, 2001).
We conclude this introduction by noting that the development of the gas-solids continuum model has
resulted in some notable successes, although there is much that remains to be done to improve the physical
fidelity of the models and to increase the computational speed of the codes. The following are notable
examples from our research group: The first example is that of the pilot-scale coal transport gasifier at
the Power Systems Development Facility, Wilsonville, Alabama. National Energy Technology Labora-
tory (NETL) researchers have conducted simulations of the pilot-scale gasifier and predicted the effect
of the design changes to prevent oxygen breakthrough, increase mixing and residence times in the riser
section of the gasifier (these predictions were later confirmed by test data) in a two-week effort costing
about $10,000 (excluding the initial validation effort), whereas the actual modification of the gasifier
took about 3–4 months of downtime at a cost of around $6 million (Gel, et al., 2009; C Guenther, et al.,
2002; C Guenther, Syamlal, Longanbach, & Smith, 2003; Shi, Guenther, & Orsino, 2007; Syamlal, 2006;
Syamlal, Guenther, Cugini, et al., 2010; Syamlal, Guenther, Gel, & Pannala, 2009, 2010). The success of
these predictions gave the impetus to use simulations for the design of a commercial scale transport gas-
ifier. In these simulations, the focus was on the operating conditions of the coal feed nozzles. The design
engineers used a number of parametric variations to determine the effect of coal jet penetration into the
riser on the syngas composition and on the local conditions near the coal nozzle. Second, Guenther and
coworkers have used simulations to evaluate 17 design variations of a hydrogasifier for Arizona Public
Service in about 3–4 weeks, whereas the actual construction of the selected design took about 3 months
(Syamlal, Guenther, Cugini, et al., 2010). In this case, evaluating all these design options by hardware
modifications would have taken approximately 3 months for each of the 17 variations, resulting in a
total of 51 months. In an entirely different area, the development of a fluidized-bed process for coating
nuclear fuel particles used to take 20–30 years to achieve an optimal design, through a trial-and-error
5
process with extensive experimentation (Heit, 1986; Noren & Develasco, 1992). The physics-based
simulations conducted at Oak Ridge National Laboratory (ORNL), using MFIX, led to the selection of
a single reactor design in a year. The recommended design was built at BWXT, Lynchburg, Virginia, and
is currently producing high-quality particles. The knowledge gained through the simulations in support
of experimental development was crucial to the success of this project (Daw, Finney, & Pannala, 2006;
Pannala, Daw, Boyalakuntla, & Finney, 2006; Pannala, et al., 2007). Some more illustrative examples
are described in the applications part of this book.
Yet practical applications of multiphase CFD have lagged behind those of single-phase CFD perhaps
because there are several challenges that need to be overcome to develop a fully predictive capability.
The underlying obstacles are many and the reader is referred to the following review articles (Campbell,
1990; D. A. Drew, 1983), perspective articles (Sundaresan, 2000, 2001) and reports on working groups
on multiphase flows (Prosperetti & Tryggvason, 2003; Sundaresan, Eaton, Koch, & Ottino, 2003; Sy-
amlal, 2006). Broadly speaking, the accuracy of the model’s predictions may be limited for a variety of
reasons: incomplete or inconsistent formulation of the governing equations, insufficient knowledge of
the constitutive and closure relations, unsatisfactory numerical treatment of the governing partial dif-
ferential equations, insufficient information on initial and boundary conditions, and lack of resolution
in the computational grid, usually as a result of limited availability of computational resources. Often,
trends predicted by the model are more useful than absolute values of various quantities. Therefore,
much care is needed in designing simulations and interpreting their results.
This chapter will give the governing equations for gas-solids reacting flows and serve as an overview
for the later chapters in the book that delve into details on various topics mentioned here. Since the
topic of this chapter is very broad, we will try to bring out the important aspects (especially through our
experience with the development and application of MFIX) and provide relevant references for further
reading. Finally, we will describe an application that represents the theory covered in this chapter and
illustrates the power and limitations of such simulations in understanding gas-solids reacting flows.
2. gas-sOlids cOntinuuM MOdel
The motion of individual particles can be conceptually averaged and replaced by the motion of a con-
tinuum, as illustrated by the two levels of resolution shown in figure 1. This leads to the mathematical
description of interpenetrating continua, a gas-phase continuum interpenetrating one or more solids
phase continua. By doing so, as mentioned earlier, we hope to gain computational speed at the cost of
reduced resolution in the description of the physical processes. Two approaches can be used to arrive at
the multiphase flow equations: an averaging approach and a mixture theory approach. In the averaging
approach, the equations are derived by space, time, or ensemble averaging of the local, instantaneous
balances for each of the phases (Anderson & Jackson, 1967a, 1967b; D. A. Drew, 1983; Ishii & Mishima,
1984; Jackson, 1997, 2000; Joseph, Lundgren, Jackson, & Saville, 1990; Prosperetti, 2007; D. Z. Zhang &
Prosperetti, 1994a, 1994b). In the mixture theory approach, the multiphase flow equations are postulated,
and restrictions on the constitutive relations are derived from general principles of continuum mechanics
(Atkin & Craine, 1976; Bedford & Drumheller, 1983; R.M. Bowen, 1976; Truesdell C., 1957). Both ap-
proaches yield a set of balance equations for mass, momentum, and energy. These equations, regardless
of how they are derived, are incomplete in the sense that they must be closed by specifying constitutive
relations such as fluid-solids drag relation or the equations for the solids phase stress.
6
Figure 1. Gas-solids flow at two levels of resolution. On the left are photographs of riser flow at two
resolutions: 305 mm x 305 mm in the top frame, 6 mm x 8 mm in the bottom frame (courtesy F. Shaffer,
NETL) and on the right are MFIX simulation results from models inspired by that physical picture. The
top frame shows clusters calculated using a continuum model (solids volume fraction scale: red – high,
blue – low), and the bottom frame shows particles around a “bubble” in a fluidized bed and their ve-
locity vectors calculated using a discrete element model. The four frames are unrelated and are shown
together only to illustrate different physical and model resolutions.
The crux of multiphase model development is the formulation of constitutive relations, which may
be based on techniques ranging from correlation of empirical information (e.g., the usage of Ergun
equation for drag; D. Gidaspow, 1994), to phenomenological models (e.g., frictional stress equations
derived by generalizing Coulomb’s friction law; D. A. Drew, 1983; Jackson, 1983; Rao & Nott, 2008;
Schaeffer, 1987), and to micromechanical theories (e.g., viscous regime stresses derived from the ki-
netic theories for granular materials; Garzo & Dufty, 1999; Garzo, Dufty, & Hrenya, 2007; Garzo,
Hrenya, & Dufty, 2007; D. Gidaspow, 1994; Jenkins, 1982; Jenkins & Mancini, 1989; Jenkins & Zhang,
2002). The differences between multiphase theories primarily originate from the closure relations em-
ployed, some of which are the subject of much debate (e.g., closure relations for the quasi-static and
plastic granular regimes). A typical set of governing equations is discussed here, closed with constitutive
relations derived from various sources.
2.1. Mixture theory
The mixture theory, as we commonly refer to it, originates in the seminal work by Truesdell (1957). In that
paper, Truesdell assigned motion, density, body force term, partial stress tensor, partial internal energy,
partial heat flux, and partial heat supply density to each constituent of the mixture. These equations of
mass, momentum, and energy balances are derived from irreversible thermodynamics, and a branch of
mechanics termed as rational thermodynamics, and most of the seminal papers related to this work can
be found in the journal titled “Archive for Rational Mechanics and Analysis.” Bowen (1967) pointed out
that some of the incorporation of partial stress tensors, partial heat fluxes, and partial heat supplies leads
to the special case of ideal mixtures and made further generalizations to the theory of thermodynamics
and mechanics of mixtures without invoking the strong assumption that the total quantity is the sum of
7
the partials. More details of this mixture theory can be found in the authoritative text book by Bowen
(R.M. Bowen, 1976). Passman (1977) suggested another approach to multiphase mixtures where one can
retain the general notion of continuum but attribute a more complex structure for each particle to account
for the discrete nature of the interactions. This work was later extended to include chemical reactions
and diffusion (Nunziato & Walsh, 1980). The mixture theory equations of balance can be postulated as
follows (Joseph, et al., 1990) for incompressible constituents:
(2.1)
(2.2)
(2.3)
(2.4)
and
(2.5)
where the subscripts g and m stand for gas and solids phase respectively; ε stands for volume fraction;
U stands for velocities; ρ stands for density; b stands for body forces; m stands for force of interaction
between the constituents; T stands for stress tensor and S stands for an interaction stress tensor. The
above equations can be considered the basic equation set for mixture theories.
These equations have to be supplemented by constitutive relationships to be able to describe the flow
of multiphase mixtures. The standard restrictions imposed on these constitutive equations are the entropy
inequality (to satisfy second law of thermodynamics), material frame-indifference, and material symmetry.
2.2. averaging Methods
The other widely adopted way to obtain continuum equations is to employ averaging techniques (e.g.,
Jackson, 2000). Here we describe two such averaging techniques, namely, ensemble averaging and
volume averaging. In both of these averaging approaches, the formalism of interpenetrating continua
requires field variables for quantifying the phasic volume fractions or the fraction of the averaging
volume occupied by each phase. These are denoted by εg (or ε0) for the fluid phase (also known as the
void fraction) and εm for the mth solids phase. The subscript g helps us to distinguish the gas-phase con-
8
servation equations, whose form in some cases is different from that of the solids phase. The volume
fractions are assumed to be continuous functions of space and time. By definition, the volume fractions
of all of the phases must sum to one:
(2.6)
where M is the total number of solids phases. The effective (macroscopic) density or the bulk density
of the mth phase is defined as
(2.7)
The ensemble averaging technique has been widely used to derive two-fluid equations (D. A. Drew,
1983; Joseph, et al., 1990; Lundgren, 1972; Saffman, 1971). In the ensemble averaging technique, an
indicator function is defined as follows (Joseph, et al., 1990):
(2.8)
In addition, let us define an averaging operator ‹› that represents the ensemble average over many
independent trials: consider conducting an experiment many times (N) starting from the same initial
conditions and recording a certain quantity at the same location and time evolution. The operator ‹›
stands for the average of that quantity as we let N → ∞. Using this definition for the indicator function
and the averaging operator, we arrive at the average volume fraction for the gas:
(2.9)
and of the solids (considering only one solids-phase—M=1):
. (2.10)
Similarly, we can obtain the averaged gas velocity:
(2.11)
and solids velocity as
(2.12)
9
where U(x,t) is the true velocity of the mixture. Given these definitions, one can arrive at the following
equations for multiphase mixtures for incompressible constituents (Joseph, et al., 1990):
(2.13)
(2.14)
(2.15)
(2.16)
The definitions of all the quantities are same as the ones introduced in the mixture theory approach
above. The ensemble averaging process introduces the explicit form for the interfacial drag term ‹δ(x)
t› where δ(x) is the Dirac delta-function for the solid-fluid interface at x, and t is traction vector dotted
with the outward normal vector at x.
Before commenting on the differences between the ensemble-averaged and mixture theory based
equations, we want to introduce the equations derived using volume averaging. In the volume averaging
approach, the point variables are averaged over a region that is large compared with the particle spac-
ing but much smaller than the flow domain. This involves the implicit assumption that there is a clear
separation of scales between the particle scale, the averaging scale and the device scale (Nigmatulin,
1979). In practice, however, this might be difficult to realize because of the extreme configuration states
within complex flows we are modeling, and mostly what we realize is a mixed average (a combination
of volume and ensemble average) as pointed out by Prosperetti and termed as “representative elementary
volume” (Prosperetti, 2007).
Let d be the particle diameter and L is the domain of interest at the macroscopic scale such that L>>d;
i.e., the domain of interest is much larger than the particle size. Let us choose a normalized weighting
function g(r) a monotonically decreasing function with separation distance, r, between any pair of points
in space. The function g(r) can be given as:
(2.17)
with additional restrictions that the function is differentiable as many times as we need based on the
requirements of the solution, and one could define the function width or radius, w:
10
(2.18)
As long as w is chosen such that L >> w >> d, the choice of the function should not affect the aver-
ages significantly (Anderson & Jackson, 1967a; Jackson, 1997, 2000). Given this functional, we can
define local spatial averages of interest to construct the volume averaged equations. For example, we
can write down the volume fraction of the gas at spatial location, x, as
(2.19)
where Vg stands for the volume of gas phase. Using this averaging convention, we can define averages
for other point properties, such as gas velocities, as follows:
(2.20)
Here, Ũg(y) stands for the true velocity at any arbitrary location in space while Ug(x) stands for the
spatially averaged velocity.
One could also construct averages for the space and time-derivatives (Anderson & Jackson, 1967a;
Jackson, 1997). The volume averages for the solids phases are obtained through a similar process except
that the domain of integration is over the volume occupied by the suspended particles. The averages
corresponding to the gas phase counterparts given above are
(2.21)
(2.22)
Given these definitions for the averaging process, one can arrive at the averaged equations of the
following form (Jackson, 1997):
(2.23)
(2.24)
11
(2.25)
(2.26)
In both these momentum equations, the first term on the right-hand side stands for drag interaction
term, the second term stands for body forces, and the final term for stresses. The expression above for
solids-phase average is different from the particle-phase equations given in (Jackson, 1997). The equa-
tions resulting from all the three different approaches on surface look similar. However, depending on
the closure terms for stresses, the mixture and averaged formulations can differ in the interaction terms
(the first term on the RHS) in the equations (2.3-2.5 for mixture formulation; 2.15-2.16 for ensemble
averaged and 2.25-2.26 for volume averaged). However, these terms for the ensemble and volume
averaging approaches agree with each other (Jackson, 1997; Joseph, et al., 1990). In order to solve the
above equations, one would need closures for the drag interaction term and the stress terms. We will
now discuss each of the balance equations and give a typical set of closure relations. The details of other
types of closures are given in the later chapters in this book.
3. cOnservatiOn Of Mass
The conservation of mass (or continuity equation) for the gas phase is as follows:
(3.1)
The first term on the left-hand side in equations (3.1) accounts for the rate of mass accumulation per
unit volume, and the second term is the net rate of convective mass flux. The term on the right accounts
for interphase mass transfer because of chemical reactions or physical processes, such as evaporation.
Rgn is the rate of generation of gas-phase species n of the total of Ng gas-phase species.
There are M solids-phase continuity equations, one for each solids phase. We will use index m going
from 1 to M to represent different solids phases and use the index 0 synonymously as g to represent gas
phase. The solids phase continuity equation is
12
(3.2)
Rmn is the rate of generation of solids-phase-m, species n, and there are Nm species in solids phase-m.
3.1. equation of state
The fluid phase continuity equation is supplemented by an equation of state, such as for example the
ideal gas law
(3.3)
or the fluid is considered incompressible.

The solids phase densities are typically considered to be a constant. If the density changes because of
chemical reactions, an additional equation of state is required to relate the solids density to its chemical
composition.
4. cOnservatiOn Of MOMentuM
The gas-phase momentum balance is expressed as follows by expanding the gas-phase stress tensor in
the averaged formulation introduced earlier:
(4.1)
where τgij is the gas-phase stress tensor, Igmi is the momentum transfer between the gas phase and the mth
solids phase, fgi is a general body force (e.g., pressure drop caused by a porous media) and εgρggi is the
body force due to gravity. The momentum equation for the mth solids phase is
(4.2)
where τmij is the gas-phase stress tensor, Imli is the momentum transfer between the mth and lth phases
(l=0 being the gas phase) and εmρmgi is the body force due to gravity. The first term on the left-hand side
in these momentum equations represents the net rate of momentum increase, and the second term, the
net rate of momentum transfer by convection. Now we will discuss the various constitutive equations
required to close this balance equation.
13
4.1. fluid-solids Momentum transfer
In the momentum conservation equations, (4.1) and (4.2), the term Igm accounts for the interaction force,
or momentum transfer, between the gas phase and the mth solids phase. The mechanisms and formulation
of the interaction forces have been reviewed in detail by (Johnson, Nott, & Jackson, 1990; Massoudi,
2003). From studies of the motion of a single particle in a fluid (e.g. Maxey & Riley, 1983) several dif-
ferent mechanisms have been identified: drag force, caused by velocity differences between the phases;
buoyancy, caused by the fluid pressure gradient; virtual mass effect, caused by relative acceleration
between phases; Saffman lift force, caused by fluid-velocity gradients; Magnus force, caused by particle
spin; Basset force, originating from the history of the particle’s motion through the fluid; Faxen force, a
correction applied to the virtual mass effect and Basset force to account for the effect of fluid-velocity
gradients; and forces caused by temperature and density gradients.
Equations (4.1) and (4.2) explicitly express a buoyancy force as -εm ∇Pg, which in a simplified form
leads to an ill-posed set of equations. When the equations are simplified to describe one-dimensional
two-phase flow, we get a set of equations consisting of two continuity and two momentum equations,
the momentum equations containing only the buoyancy and drag terms.
(4.3)
where εg and εm have been replaced by ε and (1-ε) respectively. If the two matrices on the left-hand side
of equation (4.3) are denoted as A and B, the characteristic equation is given by |Aλ − B|; i.e.,
(4.4)
where λ is the characteristic value. Equation (4.4) gives the following quadratic equation for λ:
14
(4.5)
Clearly, the roots of equation (4.5) are complex, if Ug ≠ Um. On physical grounds one would expect
the above equation set to be hyperbolic in nature, possessing real characteristics. Because the equation
set possesses imaginary characteristics, initial-value problems based on such equations are ill-posed (D.
Gidaspow, Lyczkowski, Solbrig, Hughes, & Mortensen, 1973). Any consistent numerical scheme for
these equations is unconditionally unstable; i.e., for any constant ratio Δt/Δx, geometrically growing
instabilities will always appear if Δx is made sufficiently small (D. Gidaspow, 1994; Lyczkowski, Gida-
spow, Solbrig, & Hughes, 1978; Prosperetti, 2007; Stewart & Wendroff, 1984). Although questions about
the ill-posed equations remain unsettled, equations with the pressure-gradient term as in equations (4.1)
and (4.2) are widely used in multiphase-flow studies (Anderson & Jackson, 1967a; D. Drew & Lahey,
1993; Kashiwa & Rauenzahn, 1994; Wallis, 1969a, 1969b). In practice, physical diffusion terms (e.g.,
solids stresses) appear in the system of equations making them parabolic in nature. A solids pressure
term, invariably needed to prevent the void fraction from reaching low, unphysical values, makes the
equations well-poised for a range of void fractions (D. Gidaspow, 1994). Also, in discretized equations
there are numerical diffusion terms (Stewart, 1979) that have a mitigating effect. For example, Kashiwa
and Rauenzahn (1994) have shown that real characteristics can be ensured for the forward-time stepping
process “by keeping the volume fraction fixed for any given time increment.” An alternative is to drop
the fluid-pressure gradient term in the solids-momentum equation and account for buoyancy by writing
the body force term as (ρm - ρg)g (Bouillard, Lyczkowski, Folga, Gidaspow, & Berry, 1989). This model
is called “model B” and the model described above, “model A”. Model B is well-posed (D. Gidaspow,
1994; Hudson & Harris, 2006) and is entirely consistent with the derivation of the solids stresses from
kinetic theory (D. Gidaspow, 1994; D. Gidaspow & Lyczkowski, 2010).
A drawback of Model B is that it accounts for the buoyant force under equilibrium conditions only.
As an illustration, consider the settling of solids in a centrifugal force field. In such a case, the buoyant
force on the particles has the form , where vθ is the azimuthal component of veloc-
ity and r is the radial coordinate. Such an expression for the buoyant force cannot be included in the
solids momentum equation, similarly to (ρm - ρg)g; the equations would fail to be coordinate invariant.
Consequently, Model B will not correctly calculate the centrifugal acceleration of the solids. This limi-
tation of Model B exists not only for rotary motion but also for any non-rectilinear flow, e.g., swirling
flows in a circulating fluidized bed, flow through a bend, etc. It appears that the simplification that led
to the equation set (4.3) is somewhat arbitrary, and, hence, modifying the full set of equations to make
such a simplified set well-posed is unwarranted; typically ‘model A’ is used in practice.
A term missing from Equations (4.1) and (4.2) must be explained. The equations do not include a
term that accounts for the momentum transfer caused by mass transfer, often included in formulations
found in the literature. For example, Bowen (1976, eq. 1.4.2), Kashiwa and Rauenzahn (1994), Jakobsen
(2008), and Gidaspow (1994, eq. 1.13) include in their equations a term of the form .
Kashiwa and Rauenzahn (1994) do not give a closure for the average velocity <vmn>. Jakobsen (2008)
states (on p.629) that <vmn> is clearly zero for catalytic interfaces and is assumed to be zero for gas-
liquid interfaces as well. This is not an additional assumption; it is a requirement in the context of the
continuum equations presented in this chapter as we will argue later. Bowen (1976) and Gidaspow (1994)
15
set <vmn>=vm. Although this looks arbitrary at first and appears to violate momentum conservation be-
cause the resultant force in the mixture momentum equation may not go to zero as it should, there is no
such problem in Bowen (1976) because the momentum transfer terms are not closed. In fact, Bowen
(1976) later asserts (equation 1.4.31) that closure relations must be such that the sum of momentum
transfer terms between different phases goes to zero. Gidaspow (1994), on the other hand, closes the
momentum transfer terms, and, therefore, his equation (1.17), which sets the sum of the momentum
transfer terms to zero, should have included the terms for momentum transfer caused by mass transfer
as well.
To close this term, the particle-scale phenomena need to be modeled. Let us start by assuming that
the products of chemical reactions will issue as discrete jets from a particle, an assumption made for
convenience with little bearing on the conclusions. For example, jets of volatiles (gases) will issue when
coal particles are heated. There is no way to determine the direction of those jets within the context of
the continuum equations presented in this chapter, however. Because the equations do not contain a field
variable that tracks the direction of the jets, we cannot but assume that the jets issue uniformly in all
directions. In that case the jets cause a net-zero force on the particles—two jets in diametrically opposite
directions cancelling out the force exerted by each on the particle. This conclusion remains unchanged
even when the starting assumption is relaxed, allowing uniform outflow of the products of chemical
reactions from the particles or the transfer of mass from gas to the particles. Therefore, the momentum
transfer caused by mass transfer must be zero. This is a limitation of the present formulation, which does
not contain field equations to track the direction of such mass transfer. Nevertheless, it is also a fairly
good approximation. Even when individual particles show a preferred direction for emitting gases, a
sufficiently large number of particles would display an average behavior with no preferred direction for
mass transfer and, hence, no force on the continuum resulting from the mass transfer.
There is, however, an effect on the pressure drop due to the products accelerating (or decelerating)
the gas-phase, which comes naturally from the combination of the continuity and momentum equations.
Also, the mass transfer affects the gas-solids drag, which is discussed later.
The drag force is the most significant interphase momentum transfer term. Several well-known for-
mulas accurately represent the drag force on a single-sphere as a function of the Reynolds number (e.g.,
Khan, Pirie, & Richardson, 1987). In a multi-particle system, the nearness of other particles makes the
drag force on each particle significantly greater than that given by the single-particle drag formula. Thus,
the formula for multi-particle drag force must include at least the fluid-volume fraction as an additional
parameter to account for the effect of neighboring particles. The drag formula must also be a linear
function of the volume fraction of the solids phase, the linearity ensuring that the equations correctly
satisfy the following requirement (Syamlal, 1985). A solids phase consisting of identical particles can be
represented either as a single-solids phase of volume fraction εs or as M distinct solids phases (although
of identical particle diameter and density), whose respective volume fractions (εm) would sum to εs:
(4.6)
The number of sets of solids-phase momentum equations would be one in the first case and M in the
second case. Because the division into M phases is arbitrary, the sum of the M momentum equations
16
must be equal to the single momentum equation of the first case. This is possible if and only if the drag
relations are a linear function of the solids volume fraction (εm).
A multi-particle drag coefficient is defined as follows, to ensure that the above requirements are
satisfied:
(4.7)
where CDm is only a function of the Reynolds number and fluid volume fraction εg.
The drag coefficients are usually obtained from experimental correlations, primarily based on two
types of experimental data. One type, valid for high values of the solids volume fraction, is the packed-
bed pressure drop data expressed in the form of a correlation, such as the Ergun (Ergun, 1952) equation.
Because the Ergun equation does not have the correct dependence on the Reynolds number in the limit
εg → 1 (single-particle limit), it is often supplemented with a drag correlation, such as the Wen and Yu
equation, applicable in that limit (D. Gidaspow, 1986). A formulation of this type is now sometimes
called the Gidaspow drag model (D. Gidaspow, 1994):
(4.8)
The other type of experimental data is the terminal velocity in fluidized or settling beds. The data
are used to correlate the dimensionless ratio of the terminal velocity of a multi-particle system to that
of a single-particle with the void fraction and the Reynolds number (based on the terminal velocity of
the single particle.) A widely known formula of this type is that of (Richardson & Zaki, 1954). Syamlal
and O’Brien (1987) showed that a formula for the multi-particle drag can be derived from such terminal
velocity correlations as follows:
(4.9)
where CDs(Rem/Vrm) is a single-sphere drag formula expressed as a function of Rem/Vrm (rather than of
Rem), and Vrm is the Richardson-Zaki-type correlation for the ratio of the terminal velocity of a multi-
particle system to that of a single-particle. The above relationship is derived by noting that under terminal
settling conditions for single and multi-particle systems, the equality of the drag force and the buoyant
weight of the particle lead to equations with identical right-hand sides.
17
Force balance under terminal settling for single particle:
(4.10)
and for multi-particle
(4.11)
where
(4.12)
Equations (4.10) and (4.11) are combined to get equation (4.9), where Vrm = Ret/Rets. Since equations
(4.10) and (4.11) are valid for the terminal settling condition, equation (4.8) must be valid at least for the
terminal settling condition; that is, it is valid for a particular magnitude of the drag force, a magnitude
equaling the buoyant weight of a particle. But in deriving equations (4.10) and (4.11), the magnitude of
the drag force is not explicitly considered other than in terms of the drag coefficient. Therefore, equation
(4.9) should be valid for all values of drag force.
The form of the (Richardson & Zaki, 1954) formula is such that an iterative technique is required for
calculating the drag values with the above procedure (Syamlal & O’Brien, 1987). An explicit formula
for drag can be derived from a similar correlation reported by (Garside & Aldibouni, 1977), however:
(4.13)
where
(4.14)
(4.15)
and the Reynolds number of the mth solids phase is given by
18
(4.16)
There are numerous formulas available for CDs (Khan, et al., 1987), including the following simple
formula proposed by (Dalla Valle, 1948):
(4.17)
To use this formula in equation (4.9), note that Re must be replaced with Rem/Vrm.
Until recently, the drag formula could only be determined from experimental data as discussed above,
which led to correlations with limited theoretical underpinnings (e.g., J. Z. Yang & Renken, 2003).
Ironically, closure formulations for secondary constitutive laws such as the granular stress, are much
more firmly based in theory, thanks to the kinetic theory of granular materials (e.g., D. Gidaspow, 1994).
There is now a growing body of work on deriving accurate drag correlations from numerical experiments
based on LBM (e.g., Beetstra, et al., 2007a; Hill, et al., 2001a, 2001b; Hölzer & Sommerfeld, 2009;
Kandhai, Derksen, & Van den Akker, 2003; Sankaranarayanan & Sundaresan, 2008; Van der Hoef, et al.,
2005). Benyahia et al. (2006), for example, derived a drag law based on information reported by Hill et
al. (2001a, 2001b), which extends over a large range of Reynolds number and solids volume fractions,
ensuring that the function goes to known limiting forms of the drag function and taking care to avoid
discontinuities in the function.
In spite of the paramount importance of gas-solids drag in multiphase calculations, numerous attempts
reported in the literature to develop accurate correlations, and our current ability to develop accurate
information from LBM numerical experiments, developing a drag correlation universally applicable to all
practical multiphase systems is not an achievable goal. The main reason for this disappointing assertion
is quite simple: practical systems contain size, shape, and roughness distributions that simply cannot be
characterized with just the parameters Reynolds number and phase-volume fractions. Parameterizing
drag with anything beyond those variables would be an enormous challenge and does appear to have
been undertaken for multi-particle systems. Instead of developing a universal drag correlation with
several parameters, it may be easier to develop drag correlations (as functions of the Reynolds number
and the volume fractions) for classes of materials (e.g., coal-char, biomass) or even for the specific ma-
terial being studied by conducting physical or numerical experiments for calibrating the drag formula.
The experimentally determined lattice structure of the material could be used in LBM simulations for
determining the drag function. For example, Manz et al. (1999) and Mantle et al. (2001) conducted LBM
simulations with the lattice derived from the three-dimensional image of the structure of packed beds
determined with magnetic resonance imaging (MRI).
Syamlal and O’Brien (2003) reports a method to improve the fidelity of bubbling fluidized-bed cal-
culations, for which the typical accuracy of general drag correlations is unacceptably large. Wen and Yu
(1966) report a standard deviation of ±34% in the comparison between experimental data and a minimum
fluidization velocity (Umf) formula based on the Ergun equation. However, even a 5% difference between
the experimental and predicted Umf could make a staggering difference in the predicted characteristics of
a bed of Geldart Group B particles (Geldart, 1973; Geldart, Harnby, & Wong, 1984) when the fluidization
velocity is close to Umf (the gas velocity at which the pressure drop equals the weight of the bed); the
19
bed may remain packed if the gas velocity is less than Umf or vigorously bubbling, if the gas velocity is
even slightly greater than Umf. Syamlal and O’Brien (2003) circumvented this problem by calibrating the
drag formula such that the calculated drag force at minimum fluidization conditions exactly matches the
experimental drag force. From equation (4.16)Umf can be written as a function of the Reynolds number,
the εg in the numerator accounting for the fact that Umf is defined as a superficial velocity:
(4.18)
Now equations (4.9) and (4.12)–(4.17) give an explicit formula for the Reynolds number at minimum
fluidization:
(4.19a)
where
(4.19b)
Ar and A are defined in equations (4.12) and (4.14). We modify equation (4.15) by introducing two
calibration parameters (c and d), to get the following formula for Bnew:
(4.20)
To ensure that the function Bnew is continuous, d must be related to c as follows:
(4.21)
Now with the above formulas we can adjust the parameter c such that the predicted Umf is exactly
equal to the experimental value. This ensures that the calculated drag force at minimum fluidization
condition exactly matches the experimental value. Also, the formulation ensures that the drag force
agrees with the single-particle drag force at the other extreme of very low solids volume fraction. The
single-particle drag force, currently given by equation (4.17), may also be changed, if a more accurate
formula for the given particles is known.
20
We discussed the near impossibility of finding a universally applicable drag formula, especially when
it is parameterized only by the Reynolds number and the phase volume fraction. Formulas using such
a parameterization would not be able to handle particles found in practical fluid-solids systems that are
rough, non-spherical, and have a size distribution. These and other difficulties in accurately modeling
gas-solids drag are discussed in the following paragraphs.
A narrow particle-size distribution may be characterized by an average size, typically a surface area-
weighted diameter or Sauter mean diameter. However, that is accurate only for large Reynolds numbers.
Loth et al. (2004) showed that the Sauter mean diameter is the effective mean diameter for particles in
the inertial-dominated regime (e.g., Rep>2000), regardless of particle shape, particle size distribution,
particle density distribution, or net volume fraction. They showed that for particles in the creeping flow
regime (Rep<< 1) the effective mean diameter is the volume-width diameter.
The above conclusions about the mean diameter are based on the assumption that the size distribu-
tion of the flowing material remains unchanged, which is true only when the size distribution in narrow.
Particles with a broad size (or density) distribution will segregate. One way to account for segregation
is to discretize the distribution into two or more size (or density) fractions, each characterized by an
average particle size (or density), yielding multiple solids phases. A standard drag formula could be ap-
plied to each solids phase. This will account for the drag experienced by solids phase from its relative
velocity with the fluid and may be called the diagonal component of drag. But each solids phase will
also feel an off-diagonal component of drag because of the relative velocity of other solids phases with
the fluid (, not to be confused with the direct transfer of momentum through collisions between particles
of one phase with those of another.) Yin and Sundaresan (2009) conducted LBM simulations of binary
mixtures, showed the importance of the off-diagonal components, and developed explicit constitutive
equations for the off-diagonals.
There are several studies that account for the effect of particle shape (Ganser, 1993; Haider & Lev-
enspiel, 1989; Hölzer & Sommerfeld, 2009; Tran-Cong, Gay, & Michaelides, 2004) and roughness
(Crawford & Plumb, 1986) in single-particle drag formulas. Haider and Levenspiel (1989), for example,
proposed using sphericity, the ratio between the surface area of the volume equivalent sphere and that of
the particle, in addition to Reynolds number to parametrize the single-particle drag formula. In addition
to sphericity and Reynolds number, Hölzer and Sommerfeld (2008) included two other parameters that
consider the particle orientation with respect to the flow: the crosswise sphericity and the lengthwise
sphericity. Hölzer and Sommerfeld (2009) conducted LBM simulations to determine drag, lift, and
torque acting on non-spherical single particles. They showed that the drag depends strongly on particle
shape, the angle of incidence, and particle rotation at a particle Re greater than 240. This underscores
the difficulty in modeling shape effects in that the parameterization of drag by geometric factors alone
(e.g., mean size, sphericity) may not be sufficient because the drag would also depend upon the particle
orientation with respect to the flow and particle rotation.
The drag formulas are based on the difference in the mean velocity of the gas-phase and the solids
phase. However, all the particles do not move with the mean velocity, and the particle velocity fluctua-
tions will affect the gas-solids drag (Wylie, et al., 2003). Formulas accounting for this effect are needed
especially in the high-shear regions of the flow.
When there is mass transfer between phases, the drag force must be modified to account for its ef-
fect (e.g. Montlucon, 1975). For example, during coal devolatilization or combustion gases generated
from the solids phase enter the gas phase. We discussed earlier that the momentum transfer caused by
the mass transfer is zero in the context of the continuum equations presented here. There is another ef-
21
fect on the drag caused by the modification of the flow field near the particles. This effect, sometimes
called a “blowing correction,” can be approximately accounted for by using formulas, such as from
Bird et al. (2006).
Finally, the gas-solids drag is significantly affected by spatial in-homogeneity in the volume fractions,
which is often not accounted for in the models. Usually the lack of resolution in the numerical solution
is what causes such in-homogeneities to be smeared out. The size of the numerical grid will determine
how well the flow structures are resolved. Lack of resolution has a significant effect on the calculated
drag because particle clusters (resolvable on a fine grid), which allow the gas to flow more easily around
them, experience significantly less drag than particles uniformly distributed (as approximated by a
coarse grid). It has been well established that finer and finer structures emerge as the grid size is refined,
length-scale of the fine structures being as small as ten particle diameters (Agrawal, Loezos, Syamlal, &
Sundaresan, 2001; Andrews, et al., 2005; Igci, et al., 2008; Duan Z. Zhang & VanderHeyden, 2002). For
the simulation of commercial-scale reactors it is currently not possible to use grid sizes of ten-particle
diameter resolution (e.g., Syamlal, et al., 2009), however. O’Brien and Syamlal (1993), Boemer et al.
(1998), McKeen and Pugsley (2003), Yang et al. (2004), and Heynderickx et al. (2004) have proposed
corrections to the drag coefficient to account for the missing effect of clustering on drag in coarse grid
simulations of commercial-scale reactors.
A method being developed to account for the subgrid-scale structures is the filtered continuum mod-
els (Andrews, 2007; Igci, et al., 2008). In this approach, highly resolved numerical simulations based
on the full set of equations are conducted for a domain at the size of a coarse-scale grid. The results of
the fine-scale simulations are then analyzed to derive a coarse-grid correction for drag (as well as for
solids-phase pressure and viscosity) as a function of the grid size (Igci, et al., 2008).
It is possible that there are subgrid-scale structures that cannot be captured even with highly resolved
simulations. One reason is that particle clustering is caused by inter-particle forces, such as van der
Waals forces not accounted for in the model (e.g., M. H. Zhang, Chu, Wei, & Yu, 2008). Obviously, no
amount of grid refinement will capture that phenomenon. There is another plausible reason that is not
yet debated in the literature or widely accepted. The averaging required to approximate the particles as
a granular continuum renders the hydrodynamic equations incapable of resolving the wake-dominated
micro-hydrodynamics near the particles that, under certain conditions, could cause the particles to form
clusters (Fortes, Joseph, & Lundgren, 1987). O’Brien and Syamlal (1993) have argued that the effect of
such structures must be explicitly accounted for in the fluid-solids interaction constitutive relation. It is
not clear that the fine-scale structures, such as for example shown by Igci et al. (2008), originates from
the micro-hydrodynamics. In equation (4.2) we assume that fluid-particle and particle-particle forces
can be separated into two terms. It is not yet established that such a representation of the forces would
capture the interaction between particles mediated by the intervening fluid. Lu, Wang, and Li (2009)
point out that even a grid-independent calculation may not capture the real behavior for this reason and
that the meso-scale structures (particle cluster size, shape, and orientation) could profoundly affect the
macro-scale hydrodynamics. They then account for the missing information by using an energy minimi-
zation principle or the energy minimization multi-scale (EMMS) model, which is the subject of another
chapter in this book. In the EMMS approach, a structure consisting of particle-rich dense cluster phase
and a gas-rich dilute phase is characterized by eight parameters (N. Yang, et al., 2004). The effective
drag on the system is then determined by solving the local mass and momentum balances subject to
the minimization of the energy consumption for suspending and transporting the particles (which is a
22
system stability condition). A comparison of the filtered continuum and EMMS approaches has been
reported by Benyahia (2009).
4.2. solids-solids Momentum transfer
Compared to fluid-solids momentum transfer, much less is known about solids-solids momentum trans-
fer. As in the case of fluid-solids drag, the drag between the phases is caused by velocity differences.
Arastoopour, Lin, and Gidaspow (1980) observed that such a term is necessary to correctly predict
segregation among particles of different sizes in a pneumatic conveyor. Arastoopour, Wang, and Weil
(1982) studied this effect experimentally in a pneumatic conveyor. Equations to describe such interac-
tions have been derived or suggested by several researchers: Soo (1967), Nakamura and Capes (1976),
Syamlal (1985, 1987), and Srinivasan and Doss (1985).
In the present work the solids-solids momentum transfer, Iml, is represented as
, (4.22)
A simplified version of kinetic theory was used by Syamlal (1987) to derive an expression for the
drag coefficient Fml,
(4.23)
where the last term was added by (Gera, et al., 2004) to account for a “hindrance effect” caused by
particles in enduring contact. eml and Cfml are the coefficient of restitution and coefficient of friction,
respectively, between the lth and mth solids-phase particles and scoef is an empirical coefficient. The radial
distribution function at contact, g0lm, is that derived by (Lebowitz, 1964) for a mixture of hard spheres:
(4.24)
Solids-solids drag formulas based on polydisperse kinetic theories can be found in papers such as,
for example, Jenkins and Mancini (1989), Mathiesen et al. (1999), and Huilin et al. (2003).
4.3. fluid-Phase stress tensor
The stress tensor for the fluid phase, either gas or liquid, is given by the standard formula
(4.25)
23
where, Pg is the pressure. The viscous stress tensor is assumed to be of the Newtonian form
(4.26)
where is the identity tensor and is the strain rate tensor for the fluid phase, given by
(4.27)
4.4. solids-Phase stress tensor
The development of continuum closure laws for particle-particle interactions over the entire dilute to
dense range of granular flows is a difficult task, and the research in that area is still in its infancy. The
difficulty arises from the fundamentally different modes for particle-particle force transmission that is
responsible for granular material to act as a gas, a liquid or a solid. Jaeger, Nagel and Behringer (1996b;
1996a) have reviewed the special properties of granular materials that distinguish them from liquids,
solids and gases, but allow them to behave like these three phases depending on the conditions imposed.
Liu and Nagel (1998) have proposed a phase diagram for jamming (see Figure 2) that is a function of
reciprocal of the density, load and temperature. In this depiction, as these three variables are reduced,
Figure 2. Phase diagram for jamming (Adapted from A. J. Liu & Nagel, 1998)
24
the granular material would reach the jammed state. This is a static picture of the phase diagram and the
dynamics do play an important role in these transitions (Walker & Tordesillas, 2010).
In the phase diagram of Liu and Nagel, one could consider that reciprocal density is related to void
fraction, load is related to shear rate and temperature is a derived quantity which is a function of these
two parameters. We can use these two parameters to describe regime maps for the transition from the
dilute to the dense regimes of the granular flows. In Figure 3, the dilute to dense/static granular flows
is schematically described as a function of shear rate and a recent study from Tardos’s group (Langroudi,
Turek, Ouazzi, & Tardos, 2010) using shear cell experiments mapped these transitions carefully. In this
scenario, as one increases the shear rate, the granular material undergoes transition from static to quasi-
static regime followed by uncertain/intermediate regimes to eventually the inertial regime. One reason
for the uncertainty is the usage of one parameter to characterize the regime, in contrast to the three pa-
rameters used by Liu and Nagel. One could also view the regime change as a function of void fraction
alone as schematically shown in Figure 4, which is the approach used in this chapter. In the dilute flow
regime (also referred to as viscous or kinetic/collisional), particle-particle contacts are fleeting and
momentum transfer occur during collisions; in the dense flow regime (also referred to as plastic) parti-
cle-particle contacts are enduring and momentum transfer occurs through frictional sliding contact (see
Figure 4). For the dilute regime, a formulation based on the kinetic theory of granular flows (e.g. Lun,
Savage, Jeffrey, & Chepurniy, 1984) seems to be adequate for predicting solid stresses, although this
field continues to develop as discussed in later chapters. On the other hand, empirical models from soil
mechanics are used for dense plastic flows (Jackson, 1983; Rao & Nott, 2008; Schaeffer, 1987). Because
many gas-solids flow applications (like fluidized beds, spouted beds, and hopper discharges) involve
spatiotemporal transition between dilute and dense flow regimes, a model that accounts for regime
transition is required.
Figure 3. Regime map for the rheology of granular materials as a function of shear rate (adapted from
S. Turek and A. Ouazzi, http://www.featflow.de, http://www.mathematik.uni-dortmund.de/LS3, and
Langroudi, et al., 2010)
25
Figure 4. Schematic of the plastic/transition/viscous granular rheology regime map as a function of

void fraction
Over the last two decades, different approaches have been proposed to address the dilute-dense mode
transition. Johnson and Jackson (1987) proposed direct addition of the collisional and frictional contri-
butions over the entire range of solids volume fraction. Syamlal et al. (1993) proposed a switch function
where the plastic stresses replace the collisional stresses above a critical solids volume fraction. Srivas-
tava and Sundaresan (2003) have suggested an additive stress model similar to that of Johnson and
Jackson (1987), but modified the quasi-static formulation of Schaeffer (1987) to account for strain rate
fluctuations (Savage, 1998). Recently, Makkawi et al. (2006) have proposed an additive approach for
collisional-frictional stress in the intermediate dense regions based on the results of Patil et al. (2005a;
2005b) and Tardos et al. (2003).
A method for describing the continuous transition from the collisional to the frictional regime is to
replace the step-function used by Syamlal et al. (1993) with a continuous function (Pannala, et al., 2008;
Pannala, et al., 2007; Reuge, et al., 2008; Xie, Battaglia, & Pannala, 2008a, 2008b). This model is based
on using a weighted averaging function that smoothly bridges the limiting modes. To better understand
the physical motivation, consider the schematic in Figure 4 where the viscous regime (dilute) and the
plastic regime (dense) are depicted as functions of void fraction, εg. Below a critical void fraction, ε*, in
the plastic (dense) regime, the large number of long-lasting particle contacts produce quasi-static particle
assemblies that can be modeled using Schaeffer’s theory (Schaeffer, 1987). Under viscous (dilute) re-
gime, the particle contacts are fleeting and the granular stresses can be approximated with kinetic theory
(Lun, et al., 1984). In the intermediate (transition) regime, where we can expect co-existing regions of
freely moving particles interspersed with regions of quasi-static assemblies. Over time these co-existing
regions interact with each other by exchanging particles and may grow or shrink in relative size. In ad-
dition, one can postulate that the time-averaged amounts of each regime can be parameterized by void
fraction and that this transition process would happen over a narrow void-fraction range. Even though
the transition is abrupt at the particle level, one can assume that this is smooth at macroscopic scales
because of the effects of statistical averaging over time and space. Far away from the critical solids
26
fraction one can expect the limiting behaviors (plastic and collisional) to be fully recovered without any
contribution from the other regime.
This general formulation for solids stresses can be written as
(4.28)
where the function ϕ(εg) is a blending function that has a smooth but rapid variation across the transi-
tion point, and ε* is the cut-off point for the transition (Pannala, et al., 2008). In the case of the switch
function (Syamlal, et al., 1993), the function ϕ(εg) is a Heaviside step function. The viscous stresses are
represented by σviscous, while the quasi-static and plastic stresses are represented by σplastic.
Plastic Regime
Syamlal et al. (1993) developed constitutive equations for the plastic regime using Schaeffer’s frictional
theory (Schaeffer, 1987). This model is usually called the Schaeffer model in the literature. Benyahia
(2008a) has compared this model with the model of Srivastava and Sundaresan (2003). When the solids
volume fraction exceeds the maximum packing limit, a functional form typically used in plastic flow
theories (Jenike, 1987) is used to relate the solids pressure to the void fraction:
(4.29)
An expression for the shear viscosity given by (Schaeffer, 1987) is adapted as follows:
(4.30)
where ϕ is angle of internal friction and μf,max is some stipulated maximum granular viscosity,
(4.31)
and the rate of strain tensor
(4.32)
27
With this model, we need to specify the critical void fraction below which only plastic stresses
are included.
The viscosity given by expressions such as (4.30) becomes unbounded when the rate of strain goes
to zero, for example, at the corner of an L-valve where the solids could become packed and stagnant.
In computational models, the viscosity is bounded by an upper limit (μf,max) to prevent the model from
diverging. Physically this signifies a regime change or fluid-glass transition (e.g. Silbert, 2005) into an
elastic regime. The plastic regime equations can no longer describe such systems (e.g., the corner of
an L-valve or a heap of sand), and currently no continuum theory exists that can bridge the plastic and
elastic regimes. But models that preserve the discrete nature of the particles, such as DPM for example,
are able to describe such regime transitions adequately.
Viscous Regime
The closure relationships for the stresses in the solids phase in the dilute region are typically derived
from kinetic theory of granular gases. This is analogous to the kinetic theory of gases (Agrawal, et al.,
2001; Campbell, 1990; D. Gidaspow, 1994; Jackson, 2000; Lun, et al., 1984; Peirano & Leckner, 1998;
Sinclair & Jackson, 1989). Kinetic theory provides a framework to systematically derive the continuum
representations for the particle positions and velocities. Recent developments in granular kinetic theory
are given in two chapters later in the book. Here we only summarize the relevant equations to close the
above set of equations. The collisional stress tensor from the kinetic theory can be represented as fol-
lows. Although a subscript m is used, the following equations are valid only for a single solids phase.
(4.33)
where the strain rate tensor can be written as:
(4.34)
and the normal solids pressure is given as:
(4.35)
Solids viscosity:
(4.36)
28
where ; ; ; , g0 stands for radial
distribution function and e stands for coefficient of restitution.

Here, Θm stands for granular temperature (also referred to as pseudo-thermal energy) and can be
evaluated through a transport equation:
(4.37)
The first term on the right-hand side stands for the granular temperature diffusion or conduction, the
second term represents the production term, the third term includes viscous dissipation and effect of
gas-particle slip, and the final term stands for the collisional dissipation.
The solids conductivity is given as:
(4.38)
where
and
The collisional dissipation term is given as and the exchange terms
are
If the transport of the granular energy is not important (e.g., in bubbling beds the granular energy
produced is mostly dissipated locally), one can construct the algebraic form by equating the production
to dissipation and derive the following algebraic equation:
(4.39)
29
(4.40)
(4.41)
(4.42)
(4.43)
A kinetic theory for describing poly-dispersed systems can be found in a later chapter as well as in
Gidaspow and Jiradilok (2009). Also see Benyahia (2008b) for a comparison of four different kinetic
theories for poly-dispersed systems.
5. cOnservatiOn Of sPecies Mass
The gas and solids phases may contain an arbitrary number of chemical species, Ng. The mass conserva-
tion equation for the gas phase species n is
(5.1)
where Xgn is the mass fraction and Rgn is the rate of formation of gas species n. The species conservation
equation for solids phase m, species n is
(5.2)
where Xmn is the mass fraction and Rmn is the rate of formation of solids phase m, species n. The above
equations consider the rate of accumulation, convection, diffusion and generation of species mass.
30
5.1. reaction Kinetics
To model gas-solids reactors, reaction kinetic expressions need to be supplied to close the species mass
balance equations. Such expressions will depend upon the specific chemistry being described. As an
example, consider coal (idealized as carbon) combustion reaction:
2C + O2 ⇨ 2CO.
A common way of determining a rate expression for this reaction is by assuming a shrinking core
mechanism, as depicted in figure 5, which considers the three resistances: external film diffusion, dif-
fusion through the ash layer, and the reaction at the surface of the unreacted core (C.Y. Wen, Chen, &
Onozaki, 1982; Yoon, Wei, & Denn, 1978). A rate expression is then derived by assuming a pseudo-steady
state; that is, the time constant for the shrinking of the core is much larger than that for the transport of
oxygen to the core. The rate of formation of oxygen is then given by the following (O2 is gas species 1,
CO is gas species 2, and C is solid species 1):
(5.3)
where Po2 is the partial pressure of oxygen. The film resistance is given by
(5.4)
Figure 5. Shrinking core model for coal combustion
31
where Do2 is the diffusion coefficient and Ro2 is the gas constant for oxygen, Tfm is an average film
temperature, and the Sherwood number is given by (Gunn, 1978):
(5.5)
where the Schmidt number is defined as
(5.6)
and the ash layer resistance is given by
(5.7)
where, De is an effective ash diffusivity given by (C.Y. Wen, et al., 1982)
(5.8)
and the ratio of core diameter to particle diameter
(5.9)
can be related to the solids mass fraction as
(5.10)
where, Xs1 is the carbon-mass fraction, Xs4 is the ash-mass fraction, and superscript 0 indicates the initial
values of those quantities. Wen et al. (1982) obtained the ash porosity from
(5.11)
The surface reaction resistance is given by (Desai & Wen, 1978)
(5.12)
32
From equation 5.3, the other formation rates can be obtained as
(5.13)
and (5.14)
6. cOnservatiOn Of internal energy
Following the MFIX implementation, the internal energy balances are presented here in terms of the
temperatures. The equation for the fluid phase is
(6.1)
where, in the first term on the right hand side is the fluid-phase conductive heat flux, the second
term is the fluid-solids interphase heat transfer, the third term accounts for a simplified model for the
heat transfer due to radiation, the fourth term accounts for the interfacial flow work term (Arnold, Drew,
& Lahey, 1989, 1990; Lyczkowski, Solbrig, & Gidaspow, 1982) and ΔHg is the heat of reaction, its
value being negative for exothermic reactions. The equation for the mth solids phase is
(6.2)
where, qmi in the first term on the right hand side is the solids-phase conductive heat flux, the second term
is the fluid-solids interphase heat transfer, the third term accounts for the heat transfer due to radiation,
and ΔHm is the heat of reaction.
Two simplifying assumptions have been made in the formulation of internal energy equations (6.1)
and (6.2):
1. The irreversible rate of increase of internal energy due to viscous dissipation and interphase mo-
mentum transfer has been neglected. Such terms are negligible except in the case of very large
relative velocities.
2. The direct heat transfer between different solids phases has been neglected.
33
We will now derive the energy equations as shown in (6.1) and (6.2). The gas phase internal energy
balance is
(6.3)
where the terms on the right-hand side account for heat diffusion, interphase transfer, radiative transfer,
work done by gas because of changes in the gas volume fraction, and the energy transfer between gas
and solids phases because of mass transfer (D. Gidaspow, 1994, p.9). Heating due to viscous dissipation
and interphase drag has been ignored.
The solids phase energy equation is
(6.4)
By definition the enthalpies are given by
(6.5)
(6.6)
Substituting (6.5) and (6.6) in equations (6.3) and (6.4), and subtracting out the respective continuity
equations (3.1) and (3.2) multiplied by the enthalpy we get
(6.7)
and
(6.8)
34
Now
(6.9)
where
(6.10)
which yields
(6.11)
where
(6.12)
Similarly
(6.13)
where
(6.14)
(6.15)
Substituting (6.11) in the transient (first) term of equation (6.7), we get
(6.16)
The spatial gradients in the convection term of (6.7) can be similarly determined:
35
(6.17)
Now substituting equations (6.16) and (6.17) into (6.7), we get
(6.18)
Comparing (6.18) to (6.1) we get
(6.19)
which can be simplified as follows by using the non-conservative form of species balance equation (i.e.,
Eq. 5.1 - Xgn* Equation 3.1):
(6.20)
Similarly
(6.21)
6.1. Heat transfer caused by Mass transfer
In Section 4.1 we showed that the momentum transfer caused by mass transfer in the context of the
present theory is zero. In contrast, the energy transfer caused by mass transfer cannot be taken as zero.
First we note that
(6.22)
so that the terms accounting for enthalpy transfer between gas and solids phases due to mass transfer
cancel out, as they should, when the equations (6.3) and (6.4) are summed to form the mixture energy
36
equation. Therefore, the closure for the enthalpy of the material exchanged between the phases (Hmm′)
would not change the overall energy balance; it will only change the partitioning of the energy among
the phases, which will need to be modeled because the location of the gas-solids reaction front is not
tracked in continuum models. For example, in char combustion a carbon monoxide flame front may
reside at the char surface, in the ash layer covering the char, or in the gas boundary layer surrounding
the ash-layer, depending upon process conditions (Arri & Amundson, 1978). Therefore, the heat of
combustion could cause the heating of the particles or the gas or a combination of both. Thus, to parti-
tion the heat of the coal combustion reaction C + O2 ⇨ CO2, Syamlal and Bissett (1992) assigned the
heat of reaction for the step C + ½ O2 ⇨ CO to the solids phase and the heat of reaction for the step CO
+ ½ O2 ⇨ CO2 to the gas phase. Equations (6.20) and (6.21) show that the term Hmm′ accounts for such
partitioning of the heat of reaction.
We will first postulate the closure relations for Hmm′ and then show that the forms correctly predict
the phenomena in certain limiting cases.
The mass exchanged between two solids phases is assumed to reach the destination phase in the form
and state of the material existing in the source phase. So the model for enthalpy transfer accompanying
mass transfer is postulated as follows:
(6.24)
where Rmm′n is the rate of mass transfer of species n from phase m to phase m’, its positive value indicat-
ing that mass is being added to phase m’.
In the case of mass exchange between gas and a solids phase, it appears that an asymmetry in the
transfer needs to be respected. At the scale of particles, the material leaves or reaches the particles as
a gas. Therefore, a simple model would be to set the specific enthalpy of the transferred material to be
that at the conditions that exist in the gas phase:
(6.25)
where Rgmn is the rate of mass transfer of species n from gas phase to phase m, positive value indicating
that mass is being added to phase m.
We will now validate the above closures with certain thought experiments.
Coalescence of Solids Phases
Say two solids phases (phase 1 and phase 2) at two different temperatures are flowing under vacuum.
There will be no heat transfer between the phases because of the absence of a mediating gas to effect
the particle-to-gas-to-particle heat transfer. Also, take particle-to-particle heat transfer caused by colli-
sions or radiation as negligible. Say phase 1 (source phase) is getting absorbed into phase 2 (destination
phase). A physical example would be that of fine droplets (phase 1) of an ionic liquid, whose vapor
pressure is practically zero, coalescing with large droplets (phase 2) of the same liquid at a different
temperature. Then, energy balances for the two phases will show that the temperature of phase 1 would
remain unchanged and the temperature of phase 2 would change; T2 decreases if initially it is greater
37
than T1, otherwise it increases. We will show below that the closure given by equation (6.24) correctly
captures this effect.
In this case each phase contains only one species, say A, and the following holds true:
(6.26)
Then
(6.27)
and
(6.28)
(6.29)
(6.30)
From equations (6.2) and (6.29), we deduce that T1 will remain a constant as required. Furthermore,
(6.31)
which shows that if T1 > T2 then T2 will increase and otherwise, decrease.
Condensation and Evaporation
We will now show that the above model correctly captures empirical observation of evaporating and
condensing droplets. The “wet-bulb effect” during evaporation is quite well known: a thermometer
covered with a wick saturated in a liquid (wet bulb) will show a temperature lower than a thermometer
directly exposed to the gas (dry bulb); water stored in an earthenware pot will be cooler than ambient
air because of transpiration from the surface of the pot. Likewise, water droplets suspended in a stream
of dry air will cool down as the water evaporates. The water vapor leaving the droplet surface has a
specific enthalpy equal to that of water vapor at the temperature of the air, which includes the latent heat
of vaporization. The latent heat of vaporization is supplied by the droplet, causing it to cool. When the
droplet temperature drops to the wet-bulb value, the gas-to-droplet heat transfer maintains the droplet
temperature at the wet-bulb value. Conversely, the temperature of a droplet rises as its size increases
because of condensation (e.g., Fladerer & Strey, 2003). We will now show that the closure given by
equation (6.25) will ensure that the continuum equations reproduce the above observations.
For evaporating or condensing droplets, air, being insoluble in water, does not get transferred from
gas to droplet:
38
(6.32)
The gas and droplet phase water species sources are of same magnitude and opposite sign:
(6.33)
The gas to droplet mass transfer of water species and gas-phase water species source are of the same
magnitude and opposite sign:
(6.34)
(6.35)
then
(6.36)
and
(6.37)
where λvap is the latent heat of vaporization. When the wet-bulb temperature is reached, ignoring conduc-
tion and radiation, the energy equations become
(6.38)
and
(6.39)
Note that equation (6.39) is identical to equation (22.3-32) of Bird, Stewart, and Lightfoot (2006), the
energy balance equation for the wet bulb. For evaporating droplets, RgH2O > 0, and equation (6.39) will
yield a wet-bulb temperature Tm lesser than the dry-bulb temperature Tg. Thus, the closure postulated in
equation (6.25) allows us to recover the “wet-bulb” effect. Suppose we had set Hgm = 0 then the equa-
tions would have led to the unphysical prediction Tm > Tg.
39
Similarly, for condensing droplets RgH2O < 0 and equation (6.39) will yield a wet-bulb temperature
Tm greater than the dry-bulb temperature Tg.
6.2. fluid-solids Heat transfer
The heat transfer between the fluid and solids is assumed to be a function of the temperature difference:
(6.40)
where, γgm is the heat transfer coefficient between the fluid phase and the mth solids phase. γgm is deter-
mined from the heat transfer coefficient in the absence of mass transfer, , corrected for interphase
mass transfer by using the following formula derived from film theory (R. B. Bird, et al., 2006, p. 658):
(6.41)
Note that in the limit , the heat transfer coefficient is related to the particle
Nusselt number Num:
(6.42)
where Num is the Nusselt number for the individual particles constituting the mth solids phase.
The Nusselt number is typically determined from one of the many correlations reported in the lit-
erature for calculating the heat transfer between particles and fluid in packed or fluidized beds (e.g.,
Gelperin, Einstein, & Toskubay.In, 1971; Gunn, 1978; Zabrodsky, 1966). Syamlal and Gidaspow (1985)
used a set of correlations presented by Zabrodsky (1966). Kuipers, Prins, and Van Swaaij (1992), used
the following correlation proposed by Gunn (1978) applicable for a porosity range of 0.35–1.0 and a
Reynolds numbers up to 105:
(6.43)
where the Prandtl number is defined as
(6.44)
40
6.3. conductive Heat flux in fluid Phase
The conductive heat flux within the fluid phase, , is described by Fourier’s law:
(6.45)
where kg is the gas thermal conductivity.
6.4. conductive Heat flux in solids Phase
In a simulation of the heat transfer from a fluidized bed to a wall, Syamlal and Gidaspow (1985) found
it necessary to consider solids-phase conductive heat flux to be able to calculate bed-to-wall heat trans-
fer coefficients comparable to experimental measurements. The conductive heat flux in the solids phase,
, is assumed to have a form similar to that in the fluid phase:
(6.46)
where ksm is the particle phase conductivity.

Syamlal and Gidaspow (1985) used a model proposed by Zehner and Schlunder (Bauer & Schlunder,
1978) to determine the solids phase conductivity. Kuipers, Prins, and van Swaaij (1992) improved that
model to determine the solids-phase conductivity. Their model accounts for direct conduction through the
fractional contact area ζ and indirect conduction through a wedge of gas trapped between the particles.
The Zehner and Schlunder model has been simplified by neglecting the radiation between the particles
and the resistance to heat transfer due to inhibition of the normal movement of gas molecules between
the particles (Smoluchowski effect). Following Kuipers, Prins, and van Swaaij (1992), we also delete
the contribution of gas conductivity from the Zehner and Schlunder model to obtain
(6.47)
where
(6.48)
(6.49)
41
and, for spherical particles,
(6.50)
The contact area fraction has the value ϕK = 7.26 x 103. By using this model for fluidized beds, we
are clearly extending its applicability beyond the packed-bed range where enduring contact between
particles occurs. We also assume that the model can be extended to describe conduction in multiparticle
systems. As a simpler alternative, ksm can be assumed to be a small multiple of kg, by noting that for
typical values of kpm and the void fraction, the ratio of ksm to kg is between 1 and 5 (Syamlal & Gidaspow,
1985). Lathouwers and Bellan (2001a, 2001b) have followed a different path to obtain the solids-phase
energy equation. They have combined the kinetic description of the particles for momentum with that
of energy and employed a more elaborate model for heat transfer between two colliding particles.
7. initial and bOundary cOnditiOns
The equations described so far have to be supplemented with initial and boundary conditions (BCs) to be
able to solve them. The initial conditions (ICs) serve as a starting point for the simulations, and usually
the solutions after the initial transients should not depend on the choice of initial conditions. However,
a good set of initial conditions is often needed to get good convergence. The solutions that satisfy the
governing equations are subjected to the boundary conditions, making the boundary conditions very
critical to the accuracy of the solution.
7.1. initial conditions
The initial values of all the field variables (ε, Pg, Tg, Tm, , , Xgn, Xmn) must be specified for the
entire computational domain. As mentioned before, the initial transients are usually not of interest, and
the solutions satisfy the governing equations constrained by the boundary conditions. In that case the
initial conditions need only be accurate enough to allow convergence. In fluidized beds, for example,
the solids velocity is usually set to zero, and the gas velocity is given some uniform unidirectional
value.
7.2. boundary conditions
Inflow Boundary
An inflow BC should be specified at a location where uniform flow is expected. All the field variables
need to be specified at the boundary. Two types of inflow boundary conditions are possible, constant
pressure or constant mass flux. The constant mass flux condition is more commonly used.
42
Outflow Boundary
Specified constant pressure with zero-gradient velocity at the exit is the most common condition for the
fluid outflow boundary and for this condition to be valid, the outflow needs to be reasonably uniform
with no flow reversal. If the exit has flow reversal (e.g., shedding vortices), one needs to employ non-
reflecting boundary conditions, which are not generally implemented for multiphase flows.
Walls
Boundary conditions for all the variables have to be specified at the walls. Since in most devices, walls
play an important role in the overall reactor performance, much attention must be paid to the choice of
the wall boundary conditions. It is typical to set the gas-phase normal component of velocity to zero
(no-penetration) and the gradient of the tangential component to zero (free-slip) or the tangential com-
ponent itself to zero (no-slip) at the wall. A more detailed treatment for the wall boundary condition for
gas-phase velocities was suggested by Sinclair and Jackson (1989) for a fully-developed pipe flow but
such a treatment is not generalized or commonly used. The choice of the gas-phase velocity boundary
condition has minimal effect on the simulation of most reactors with dense solids flow.
On the other hand, the boundary conditions for the solids velocities could have profound effect on the
simulation results (e.g., S. Benyahia, Syamlal, & O’Brien, 2005). The boundary conditions developed
by Johnson and Jackson (1987) are widely used for solids velocity and the granular temperature. These
boundary conditions rely on the two parameters, specularity coefficient and the particle-wall coefficient
of restitution. Jenkins (1992) has proposed a different form of the boundary condition for the frictional
particles, which requires parameters related to the frictional properties of the particles and nature of the
surface of the walls. These boundary conditions were further extended for both small and large sliding
cases (Jenkins & Louge, 1997). Benyahia et al. (2005) have performed detailed evaluations of these
boundary conditions and concluded that for a case they studied the behavior observed fell between the
predictions based on the large- and small-friction boundary conditions of Jenkins and Louge or the
prediction based on the Johnson and Jackson boundary condition with a small specularity coefficient.
The wall heat transfer in a fluidized bed can be predicted by using a sufficiently fine grid near the
walls (Syamlal & Gidaspow, 1985). This approach, however, is too expensive for practical computa-
tions. Therefore, the boundary conditions for the energy equations are set such that the walls are non-
conducting, and typically the wall temperature or flux or mixed-boundary condition is prescribed to
account for heating or cooling of the walls. A zero-flux boundary condition for the species is typically
valid for impermeable walls.
In conclusion, the boundary conditions are quite important for the accurate integration of the gov-
erning equations, an area that is least researched. For example, there is hardly any work on including
the effect of the interstitial gas on the particle boundary conditions described above. As the governing
equations and their closures are set on a firmer footing, one can expect that there will be considerable
research on improving the accuracy of the boundary conditions.
43
8. deMOnstratiOn PrObleM: bubbling fluidized bed
In this section, we will present results from the simulation of a bubbling fluidized bed with the geometry
shown in Figure 6. This simple example is introduced in 2D to specifically illustrate the various model-
ing options — physical models, numerical discretization schemes, and grid resolution – that need to
be considered to conduct accurate simulations. The setup is derived from a recent experiment that has
details of both the particle velocities and granular temperature (Jung, Gidaspow, & Gamwo, 2005). This
step of comparing simulation results with experimental data is called model validation, an essential step
before such models can be used for the design of fluidized bed reactors. The physical properties of the
particles and the fluid, and the geometric grid parameters used in the simulations are shown in table 1.
The simulations were conducted with the open source code MFIX (http://mfix.netl.doe.gov).
The different simulation cases are listed in Table 2, and these cases have been constructed to use two
different forms of the granular energy equation (a PDE version given by equation 4.37 or its simplified
form given by equation 4.39 and supplemented by Johnson and Jackson boundary conditions for the
solids at the walls), two different grid resolutions (Table 1) and two different discretization schemes
Figure 6. Schematic of the bubbling bed. The boundary conditions (unless otherwise mentioned) specified
are a uniform mass-inflow at the bottom boundary, no-slip walls for both gas and solids phases on the
side boundaries, and a specified pressure at the top boundary. The initial condition is that of a uniform
distribution of void fraction at the bottom (0.4) and top (1.0) sections of the fluidized bed.
44
(first order upwinding or the second order Superbee scheme; see C. Guenther & Syamlal, 2001). In the
following discussion we will denote different cases with the case names given in table 2 (e.g., FHP,
where F stands for Fine Resolution, H stands for High Order, and P for the PDE version of the transport
equation for granular energy). The predicted hydrodynamic features of the bubbling bed are shown in
Figure 7 for the FHP case. The figure shows a nearly circular bubble close to the bed surface and bubble-
like structures deeper in the bed. The core of the bubbles have low solids fraction and bed-material
concentrates around the bubbles. Gas bypasses through the bubbles and percolates through the emulsion
phase of the bed. The particles flow upward in the wake of the bubbles and flow downward in regions
away from the bubbles. The bubbles bursting into the freeboard eject particles and the particles return
to bed under gravity. These are in qualitative agreement with experimental observations as shown in
numerous computational studies reported in the literature (e.g., Boemer, et al., 1998; D. Gidaspow, 1994;
D. Gidaspow & Jiradilok, 2009; McKeen & Pugsley, 2003; Patil, et al., 2005a)
Figure 8 compares the temporal variation of the lateral solids velocity, vertical solids velocity and
the granular temperature for the FHA/FHP cases at a location near the free board section at the center
of the bed. These two cases compare the role of the algebraic versus the transport form of the granular
Table 1. Particle, flow, and geometric parameters for different 2D bubbling bed simulations
Model Parameter Value Units

Particle diameter, dp 530 μm
Particle density, ρs 2500 kg m-3
Coefficient of restitution 0.99
Angle of internal friction 30
Fluid viscosity 1.8e-5 kg m-1s-1
Coarse Resolution (dx, dy) 0.01 m
Fine Resolution (dx, dy) 0.005 m
Table 2. Different simulation cases. The case names are three-letter acronyms with the letters corre-
sponding to grid resolution, discretization scheme, and granular temperature calculation method. For
example, in the name FHP, F stands for Fine Resolution, H stands for High Order and P stands for the
PDE (partial differential equation) version of the transport equation for granular energy.
Granular Temperature
Case Name Grid Resolution Discretization Scheme Calculation Method
CLA Coarse Resolution Low Order Algebraic
CHA Coarse Resolution High Order Algebraic
FLA Fine Resolution Low Order Algebraic
FHA Fine Resolution High Order Algebraic
CLP Coarse Resolution Low Order PDE Transport
CHP Coarse Resolution High Order PDE Transport
FLP Fine Resolution Low Order PDE Transport
FHP Fine Resolution High Order PDE Transport
45
Figure 7. Hydrodynamics features of the bubbling bed at 5s for the FHP case: (a) Void fraction (EP_g)
distribution by itself, (b) with superimposed gas velocity vectors, and (c) with superimposed solids
velocity vectors
Figure 8. Time series of the lateral solids velocities, vertical solids velocities, and granular temperature
for the different granular stress models.
temperature equation while keeping all the other parameters the same. The FHA case with the alge-
braic form shows higher fluctuations of the plotted quantities and a significantly larger average granular
temperature. The algebraic formulation assumes that locally generated granular energy is dissipated
without any transport (advection/conduction) while the PDE form allows for the transport of granular
46
Figure 9. Time series of the lateral solids velocities, vertical solids velocities and granular temperature
for the different order of the spatial discretization schemes.
energy. At the observation location close to the freeboard the solids are agitated quite vigorously, and
the approximation that local balance of production/dissipation of granular energy might not be appropri-
ate, which explains the differences between the two cases.
Figure 9 shows the temporal variation of the lateral solids velocity, vertical solids velocity, and the
granular temperature for the FLP/FHP cases at the same location as in figure 8. Here again everything
is the same except for the order of spatial discretization. The frequency of oscillations seems to be rela-
tively close in these two cases, but the magnitude and steepness of the oscillations seem to be larger in
the case with high-order scheme. This can be explained by the lower numerical dissipation caused by
the high-order scheme as compared to the first-order upwind scheme used in the FLP case.
Figure 10 shows the temporal variation of the lateral solids velocity, vertical solids velocity, and the
granular temperature for the CHP/FHP cases at the same location near the freeboard as in the results
presented above. Here again everything is kept the same except the resolution of the spatial grid. The
frequency of oscillations is larger in the case with finer resolution but the magnitude of the oscillations
is higher in the case with coarse resolution. The coarse description is allowing for larger bubbles, which
form slowly leading to the above observations.
The multiphase reactors are highly dynamical systems, and the comparison of the transients (e.g.,
CHP and FHP) is not all that meaningful as the transient trajectories are expected to diverge, even when
started from the same initial condition. Above, we have compared the transients in part to illustrate this
point and in part to qualitatively compare the statistical properties of the two solutions. For this reason
local instantaneous values of field variables are typically not compared with experimental data, and only
recently have researchers started making such comparisons (e.g. Pannala, et al., 2007). Typically, time-
or space- averaged quantities or statistical properties of the predicted behavior are compared with ex-
47
Figure 10. Time series of the lateral solids velocities, vertical solids velocities, and granular temperature
for the different spatial resolution.
perimental data for validating the model. The analyst must consider several factors while making such
comparisons, which we will discuss in the rest of this section. We will first make several observations
about a comparison of the predicted and experimental frequencies and then explain the subtleties that
must be considered in making such comparisons.
In Table 3 we compare the predicted dominant (peak) frequency of the power spectrum of lateral and
vertical solids velocities with the experimental data of Jung et al. (2005). We list the cases in the order
of the agreement of the predicted frequency (vertical) with experimental data. The frequency (verti-
cal) was chosen because that frequency is directly linked to the passage of bubbles, a phenomenon of
much practical significance in the design of fluidized bed reactors. The FHP case with the finer grid
resolution, higher order of discretization, and more general physical model (algebraic model being a
simplified form of the PDE model) than other cases gives results closest to the experimental data. We
may be tempted to think that this is to be expected, an unfounded expectation as will be discussed later.
As if proving the point, the CLA case, which we would have expected to perform the worst, is not at
the bottom and predicts a frequency in much better agreement than FHA case. The predicted dominant
frequency (lateral), except for one case, is in better agreement with the data as compared to dominant
frequency (vertical). We observe no consistencies with respect to the modeling options used: the high-
order discrtization scheme (H) does not consistently perform better than the low-order discretization
scheme (L); the fine grid (F) does not consistently perform better than the coarse grid (C); the PDE
model (P) does not consistently perform better than the algebraic model (A). We will now discuss the
reason for this apparent inconsistency.
48
Table 3. Predicted dominant frequency based on lateral and vertical solids velocities for the various
simulation cases compared against the experimental observation.
Case Name Dominant Frequency (Lateral) Dominant Frequency (Vertical)

Experiments (Jung et al., 2005) 2.63 2.28
FHP 2.62 1.77
FLP 2.01 1.64
CHA 2.38 1.28
CLP 0.67 0.79
CLA 0.91 0.67
CHP 1.34 0.54
FLA 0.31 0.31
FHA 0.79 0.3
When making comparisons such as in Table 3 we are trying to answer the question “do the model
equations, programmed in the computational software, accurately describe physical reality?”, a question
that needs to be answered in the affirmative before the model can be used with confidence for design.
This phase is called model validation. However, in this section the comparison between the predictions
of different cases and the experimental data is done only for illustration; we already know that FHP is
the best case. For example, we are not asking whether the algebraic equation (A) or the PDE (P) granu-
lar energy model is better. We already know that P is more exact than A because A is a simplified form
of P. (The only reason for choosing A over P would be for reducing computational cost). Although P is
more exact than A, it is not necessary that P should be in better agreement with experiments than A. But
the fact that P is in better agreement with experiments (just focusing on cases FHP and FHA for now)
gives us confidence in the accuracy of the components of the model other than the granular energy
equation. If A were to give better predictions than P, then we have to suspect a cancellation of errors,
somehow the errors in the other components of the model being cancelled out by the simplification that
led to A.
Based on the results for FHP and FHA discussed in the previous paragraph, it appears that the model
is physically realistic. Then why does case CHA give a better prediction than case CHP? That brings us
to a discussion of the model verification phase, which must be completed before model validation. In
this phase we need to answer the question “are the equations solved correctly by the computational soft-
ware?” The developers of modern computational software subject it to a hierarchy of tests of increasing
complexity to partially answer this question. But the final steps of verification must be undertaken by
the users of the software. One step is to ensure that the iterations have converged and that the conver-
gence tolerance used is adequate. Although this is stating the obvious, there are published papers that
try to draw (erroneous) conclusions from unconverged results. Also, users often overlook the fact that
a low residual achieved upon the convergence of the iterations by itself does not guarantee the degree
of the accuracy of the solution. Only by decreasing the residual further and confirming that the solution
doesn’t change, can we be sure of the accuracy of the solution. For a given grid accept only that part of
the solution, which did not change when the convergence tolerance is reduced by an order of magnitude.
The second step is the verification that the solution is grid independent. This is done by successively
refining the grid until the solution does not change with respect to the grid size. In grid convergence
49
study, it is usually necessary to use a high-order discretization technique because the grid convergence
of low order methods is slow and could give the false impression of grid convergence. It is known that
coarse grid solution could even be qualitatively incorrect (e.g., S. Benyahia, et al., 2007). So the use
of coarse grid appears to be why an approximate model (A) is giving more accurate prediction than an
exact model (P). Furthermore, on a coarse grid the use of a high order technique does not consistently
give the expected better agreement (CHP vs. CLP), whereas on a fine grid FHP (high-order) is in better
agreement with experiment than FLP (low order). This then illustrates the paramount importance of
checking grid convergence.
Finally, we will consider the over 20% discrepancy between experimental and predicted dominant
frequency (vertical) for the FHP case. At first blush it casts doubt over the fidelity of the model. Is this
a reasonable agreement or not? To answer this question properly it is necessary to do uncertainty quan-
tification. First, we will need an estimate of the uncertainty in the experimental data, which originates
from the measurement techniques as well as the method used to determine the dominant frequencies from
the measurements. This information needs to be supplied by the experimentalists. Second, we need to
quantify the uncertainties in the predictions, originating from the uncertainties in the model parameters,
from constitutive relations such as drag correlations, from discretization errors, from convergence errors,
and from the method used for determining the dominant frequencies. Recall that an uncalibrated drag
correlation itself could have over 20% uncertainty (e.g., C. Y. Wen & Yu, 1966). If such an uncertainty
exists in the constitutive relations and the predictions are sensitive to the uncertainty in those constitu-
tive relations, then there is no hope for improving the fidelity of the predicted frequency by improving
the theory (say, P instead of A). So it is important to identify the sources of uncertainties in the predic-
tions, so that the relevant components of the model can be improved (e.g., drag correlations). This is a
difficult undertaking, seldom attempted in multiphase flow calculations, but an essential one for using
multiphase flow simulations with confidence for design.
9. cOnclusiOn
A typical, complete set of governing equations for reacting gas-solids flows is presented here. The
equations are still actively being developed, and some of the recent developments are described in later
chapters. The goal is not to cover the field comprehensively but to provide a broad overview of the cur-
rent status, enunciating the challenges and providing references to later chapters and relevant literature
for further study. The equation set presented here is available in its entirety in the open-source software
MFIX, and the readers are encouraged to explore various modeling options with that software. Given the
broad applicability of the gas-solids reacting flows, especially in the context of clean energy, we hope
that additional research will be carried out in this important area to make the simulations highly accurate.
10. references
Agrawal, K., Loezos, P. N., Syamlal, M., & Sundaresan, S. (2001). The role of meso-scale structures
in rapid gas-solid flows. Journal of Fluid Mechanics, 445, 151–185. doi:10.1017/S0022112001005663
50
Anderson, T. B., & Jackson, R. (1967a). A Fluid Mechanical Description of Fluidized Beds. Industrial
& Engineering Chemistry Fundamentals, 6(4), 527. doi:10.1021/i160024a007
Anderson, T. B., & Jackson, R. (1967b). Hydrodynamic Stability of a Fluidized Bed. Industrial & En-
gineering Chemistry Fundamentals, 6(3), 478. doi:10.1021/i160023a032
Andrews, A. T. (2007). Filtered models for gas-particle flow hydrodynamics. Princeton, NJ: Princeton
University.
Andrews, A. T., Loezos, P. N., & Sundaresan, S. (2005). Coarse-grid simulation of gas-particle flows in
vertical risers. Industrial & Engineering Chemistry Research, 44(16), 6022–6037. doi:10.1021/ie0492193
Arastoopour, H., Lin, D., & Gidaspow, D. (1980). Hydrodynamic Analysis of Pneumatic Transport of a
Mixture of Two Particle Sizes. In Veziroglu, T. N. (Ed.), Multiphase Transport (Vol. 4, pp. 1853–1871).
New York: Hemisphere Publishing Corporation.
Arastoopour, H., Wang, C. H., & Weil, S. A. (1982). Particle-Particle Interaction Force in a Dilute Gas-
Solid System. Chemical Engineering Science, 37(9), 1379–1386. doi:10.1016/0009-2509(82)85010-0
Arnold, G. S., Drew, D. A., & Lahey, R. T. (1989). Derivation of Constitutive-Equations for Interfacial
Force and Reynolds Stress for a Suspension of Spheres Using Ensemble Cell Averaging. Chemical
Engineering Communications, 86, 43–54. doi:10.1080/00986448908940362
Arnold, G. S., Drew, D. A., & Lahey, R. T. (1990). An Assessment of Multiphase Flow Models Us-
ing the 2nd Law of Thermodynamics. International Journal of Multiphase Flow, 16(3), 481–494.
doi:10.1016/0301-9322(90)90077-V
Arri, L. E., & Amundson, N. R. (1978). Analytical Study of Single-Particle Char Gasification. AIChE
Journal. American Institute of Chemical Engineers, 24(1), 72–87. doi:10.1002/aic.690240109
Atkin, R. J., & Craine, R. E. (1976). Continuum Theories of Mixtures: Basic Theory and Historical De-
velopment. The Quarterly Journal of Mechanics and Applied Mathematics, 29(2), 209–244. doi:10.1093/
qjmam/29.2.209
Bauer, R., & Schlunder, E. U. (1978). Effective Radial Thermal-Conductivity of Packings in Gas-Flow.
2. Thermal-Conductivity of Packing Fraction without Gas-Flow. International Chemical Engineering,
18(2), 189–204.
Bedford, A., & Drumheller, D. S. (1983). Theories Of Immiscible And Structured Mixtures. International
Journal of Engineering Science, 21(8), 863–960. doi:10.1016/0020-7225(83)90071-X
Beetstra, R., van der Hoef, M. A., & Kuipers, J. A. M. (2007a). Drag force of intermediate Reynolds
number flow past mono- and bidisperse arrays of spheres. [Article]. AIChE Journal. American Institute
of Chemical Engineers, 53(2), 489–501. doi:10.1002/aic.11065
Beetstra, R., van der Hoef, M. A., & Kuipers, J. A. M. (2007b). Numerical study of segregation using a
new drag force correlation for polydisperse systems derived from lattice-Boltzmann simulations. [Ar-
ticle]. Chemical Engineering Science, 62(1-2), 246–255. doi:10.1016/j.ces.2006.08.054
51
Benyahia, S. (2008a). Validation Study of Two Continuum Granular Frictional Flow Theories. Industrial
& Engineering Chemistry Research, 47(22), 8926–8932. doi:10.1021/ie8003557
Benyahia, S. (2008b). Verification and validation study of some polydisperse kinetic theories. Chemical
Engineering Science, 63(23), 5672–5680. doi:10.1016/j.ces.2008.08.016
Benyahia, S. (2009). On the Effect of Subgrid Drag Closures. Industrial & Engineering Chemistry Research.
Benyahia, S., Arastoopour, H., & Knowlton, T. M. (2002). Two-dimensional transient numerical simu-
lation of solids and gas flow in the riser section of a circulating fluidized bed. Chemical Engineering
Communications, 189(4), 510. doi:10.1080/00986440212093
Benyahia, S., Arastoopour, H., Knowlton, T. M., & Massah, H. (2000). Simulation of particles and gas
flow behavior in the riser section of a circulating fluidized bed using the kinetic theory approach for the
particulate phase. Powder Technology, 112(1-2), 24–33. doi:10.1016/S0032-5910(99)00302-2
Benyahia, S., Syamlal, M., & O’Brien, T. J. (2005). Evaluation of boundary conditions used to model
dilute, turbulent gas/solids flows in a pipe. Powder Technology, 156(2-3), 62–72. doi:10.1016/j.pow-
tec.2005.04.002
Benyahia, S., Syamlal, M., & O’Brien, T. J. (2006). Extension of Hill-Koch-Ladd drag correlation
over all ranges of Reynolds number and solids volume fraction. Powder Technology, 162(2), 166–174.
doi:10.1016/j.powtec.2005.12.014
Benyahia, S., Syamlal, M., & O’Brien, T. J. (2007). Study of the ability of multiphase continuum mod-
els to predict core-annulus flow. AIChE Journal. American Institute of Chemical Engineers, 53(10),
2549–2568. doi:10.1002/aic.11276
Bird, G. A. (1994). Molecular Gas Dynamics and the Direct Simulation of Gas Flows. Oxford, UK:
Clarendon.
Bird, R. B., Stewart, W. E., & Lightfoot, E. N. (2006). Transport Phenomena (2nd ed.). New York: John
Wiley & Sons.
Blake, T. R., & Chen, P. J. (1980). Computer Modeling of Fluidized-Bed Coal-Gasification Reactors.
[-Inde.]. Abstracts of Papers of the American Chemical Society, 180(Aug), 17.
Boemer, A., Qi, H., & Renz, U. (1998). Verification of Eulerian simulation of spontaneous bubble
formation in a fluidized bed. Chemical Engineering Science, 53(10), 1835–1846. doi:10.1016/S0009-
2509(98)00044-X
Bouillard, J. X., Gidaspow, D., & Lyczkowski, R. W. (1991). Hydrodynamics of Fluidization - Fast-Bubble
Simulation in a 2-Dimensional Fluidized-Bed. Powder Technology, 66(2), 107–118. doi:10.1016/0032-
5910(91)80092-W
Bouillard, J. X., Lyczkowski, R. W., Folga, S., Gidaspow, D., & Berry, G. F. (1989). Hydrodynamics
of Erosion of Heat-Exchanger Tubes in Fluidized-Bed Combustors. Canadian Journal of Chemical
Engineering, 67(2), 218–229. doi:10.1002/cjce.5450670208
52
Bowen, R. M. (1967). Toward a thermodynamics and mechanics of mixtures. Archive for Rational
Mechanics and Analysis, 24(5), 370–403. doi:10.1007/BF00253154
Bowen, R. M. (1976). Theory of Mixtures. In Eringen, A. C. (Ed.), Continuum Physics (Vol. 3). New
York: Academic Press.
Campbell, C. S. (1990). Rapid Granular Flows. Annual Review of Fluid Mechanics, 22, 57–92. doi:10.1146/
annurev.fl.22.010190.000421
Ciferno, J. P., Fout, T. E., Jones, A. P., & Murphy, J. T. (2009). Capturing Carbon from Existing Coal-
Fired Power Plants. Chemical Engineering Progress, 105(4), 33–41.
Clayton, S. J., Stiegel, G. J., & Wimer, J. G. (2002). U.S. DOE’s Perspective on Long-Term Market
Trends and R&D Needs in Gasification. Paper presented at the 5th European Gasification Conference:
Gasification – The Clean Choice.
Crawford, C. W., & Plumb, O. A. (1986). The Influence of Surface-Roughness on Resistance to Flow
through Packed-Beds. Journal of Fluids Engineering-Transactions of the Asme, 108(3), 343–347.
doi:10.1115/1.3242584
Dalla Valle, J. M. (1948). Micromeritics. London: Pitman.
Dartevelle, S. (2004). Numerical modeling of geophysical granular flows: 1. A comprehensive approach
to granular rheologies and geophysical multiphase flows. Geochemistry Geophysics Geosystems, 5.
Dartevelle, S., Rose, W. I., Stix, J., Kelfoun, K., & Vallance, J. W. (2004). Numerical modeling of geo-
physical granular flows: 2. Computer simulations of plinian clouds and pyroclastic flows and surges.
Geochemistry Geophysics Geosystems, 5.
Darton, R. C., Lanauze, R. D., Davidson, J. F., & Harrison, D. (1977). Bubble-Growth Due to Coalescence
in Fluidized-Beds. Transactions of the Institution of Chemical Engineers, 55(4), 274–280.
Davidson, D. (2001). The enterprise-wide application of CFD in the chemicals industry. Paper presented
at the 6th World Congress of Chemical Engineering, Melbourne, Australia.
Davidson, J. F., & Harrison, D. (1963). Fluidized Particles. London: Cambridge University Press.
Davidson, J. F., Harrison, D., & Carvalho, J. R. F. G. D. (1977). Liquid-Like Behavior of Fluidized-Beds.
Annual Review of Fluid Mechanics, 9, 55–86. doi:10.1146/annurev.fl.09.010177.000415
Daw, C. S., Finney, C., & Pannala, S. (2006). Process Modeling Phase II Summary Report for the Ad-
vanced Gas Reactor Fuel Development and Qualification Program (No. INL MPO 00056009). ORNL.
Desai, P. R., & Wen, C. Y. (1978). Computer Modeling of the MERC Fixed Bed Gasifier. Morgantown,
WV: Morgantown Energy Technology Center.
Ding, J., & Gidaspow, D. (1990). A Bubbling Fluidization Model Using Kinetic-Theory of Granular Flow.
AIChE Journal. American Institute of Chemical Engineers, 36(4), 523–538. doi:10.1002/aic.690360404
Drew, D., & Lahey, R. (1993). Particulate two-phase flow. Boston: Butterworth-Heinemann.
53
Drew, D. A. (1983). Mathematical-Modeling of 2-Phase Flow. Annual Review of Fluid Mechanics, 15,
261–291. doi:10.1146/annurev.fl.15.010183.001401
Ergun, S. (1952). Fluid Flow Through Packed Columns. Chemical Engineering Progress, 48(6), 89–94.
Fladerer, A., & Strey, R. (2003). Growth of homogeneously nucleated water droplets: a quantitative
comparison of experiment and theory. Atmospheric Research, 65(3-4), 161–187. doi:10.1016/S0169-
8095(02)00148-5
Fortes, A. F., Joseph, D. D., & Lundgren, T. S. (1987). Nonlinear Mechanics of Fluidization of Beds
of Spherical-Particles. Journal of Fluid Mechanics, 177, 467–483. doi:10.1017/S0022112087001046
Ganser, G. H. (1993). A rational approach to drag prediction of spherical and nonspherical particles.
Powder Technology, 77(2), 143–152. doi:10.1016/0032-5910(93)80051-B
Garg, S. K., & Pritchett, J. W. (1975). Dynamics of Gas-Fluidized Beds. Journal of Applied Physics,
46(10), 4493–4500. doi:10.1063/1.321421
Garside, J., & Aldibouni, M. R. (1977). Velocity-Voidage Relationships for Fluidization and Sedimenta-
tion in Solid-Liquid Systems. Industrial & Engineering Chemistry Process Design and Development,
16(2), 206–214. doi:10.1021/i260062a008
Garzo, V., & Dufty, J. (1999). Homogeneous cooling state for a granular mixture. Physical Review E:
Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics, 60(5), 5706–5713. doi:10.1103/
PhysRevE.60.5706
Garzo, V., Dufty, J. W., & Hrenya, C. M. (2007). Enskog theory for polydisperse granular mixtures. I.
Navier-Stokes order transport. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 76(3).
doi:10.1103/PhysRevE.76.031303
Garzo, V., Hrenya, C. M., & Dufty, J. W. (2007). Enskog theory for polydisperse granular mixtures. II.
Sonine polynomial approximation. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics,
76(3). doi:10.1103/PhysRevE.76.031304
Gel, A., Guenther, C., Pannala, S., Benyahia, S., Galvin, J., & Syamlal, M. (2009). (Manuscript submitted
for publication). Accelerating clean and efficient coal gasifier designs with high performance computing.
Journal of Computational Science.
Geldart, D. (1973). Types of gas fluidization. Powder Technology, 7(5), 285–292. doi:10.1016/0032-
5910(73)80037-3
Geldart, D., Harnby, N., & Wong, A. C. (1984). Fluidization of Cohesive Powders. Powder Technology,
37(Jan-), 25-37.
Gelperin, N. I., & Einstein, V. G., & Toskubay.In. (1971). Heat Transfer Coefficient between a Surface
and a Fluid Bed. British Chemical Engineering and Process Technology, 16(10), 922.
Gera, D., Syamlal, M., & O’Brien, T. J. (2004). Hydrodynamics of particle segregation in fluidized beds.
International Journal of Multiphase Flow, 30(4), 419–428. doi:10.1016/j.ijmultiphaseflow.2004.01.003
54
Gidaspow, D. (1986). Hydrodynamics of fluidization and heat transfer: supercomputer modeling. Ap-
plied Mechanics Reviews, (39): 23.
Gidaspow, D. (1994). Multiphase Flow and Fluidization: Continuum and Kinetic Theory Descriptions.
Boston: Academic Press.
Gidaspow, D., & Ettehadieh, B. (1983). Fluidization in two-dimensional beds with a jet. 2. Hydrodynamic
modeling. Industrial & Engineering Chemistry Fundamentals, 22(2), 193–201. doi:10.1021/i100010a008
Gidaspow, D., & Jiradilok, V. (2009). Computational Techniques: The Multiphase CFD Approach to
Fluidization and Green Energy Technologies. New York: Nova Science Publishers.
Gidaspow, D., Lyczkowski, R. W., Solbrig, C. W., Hughes, E. D., & Mortensen, G. A. (1973). Charac-
teristics of Unsteady One-Dimensional 2-Phase Flow. Transactions of the American Nuclear Society,
17(Nov), 249–250.
Glasser, B. J., Kevrekidis, I. G., & Sundaresan, S. (1996). One- and two-dimensional travelling wave solu-
tions in gas-fluidized beds. Journal of Fluid Mechanics, 306, 183–221. doi:10.1017/S0022112096001280
Glasser, B. J., Kevrekidis, I. G., & Sundaresan, S. (1997). Fully developed travelling wave solutions
and bubble formation in fluidized beds. Journal of Fluid Mechanics, 334, 157–188. doi:10.1017/
S0022112096004351
Glasser, B. J., Sundaresan, S., & Kevrekidis, I. G. (1998). From bubbles to clusters in fluidized beds.
Physical Review Letters, 81(9), 1849–1852. doi:10.1103/PhysRevLett.81.1849
Goldschmidt, M. J. V., Beetstra, R., & Kuipers, J. A. M. (2004). Hydrodynamic modelling of dense
gas-fluidised beds: comparison and validation of 3D discrete particle and continuum models. Powder
Technology, 142(1), 23–47. doi:10.1016/j.powtec.2004.02.020
Guenther, C., Shahnam, M., Syamlal, M., Longanbach, J., Cicero, D., & Smith, P. (2002, September
23-27). CFD modeling of a transport gasifier. Paper presented at the Proceedings of the 19th Annual
Pittsburgh Coal Conference, Pittsburgh, PA.
Guenther, C., & Syamlal, M. (2001). The effect of numerical diffusion on simulation of isolated bubbles
in a gas-solid fluidized bed. Powder Technology, 116(2-3), 142–154. doi:10.1016/S0032-5910(00)00386-7
Guenther, C., Syamlal, M., Longanbach, J., & Smith, P. (2003, September 15-19). CFD modeling of a
transport gasifier Part II. Paper presented at the Proceedings of the 20th Annual Pittsburgh Coal Confer-
ence, Pittsburgh, PA.
Gunn, D. J. (1978). Transfer of Heat or Mass to Particles in Fixed and Fluidized-Beds. International
Journal of Heat and Mass Transfer, 21(4), 467–476. doi:10.1016/0017-9310(78)90080-7
Haider, A., & Levenspiel, O. (1989). Drag coefficient and terminal velocity of spherical and nonspherical
particles. Powder Technology, 58(1), 63–70. doi:10.1016/0032-5910(89)80008-7
Hanratty, T. J., Theofanous, T., Delhaye, J.-M., Eaton, J., McLaughlin, J., Prosperetti, A., et al. (2003).
Workshop on Scientific Issues in Multiphase Flow, Report to Program on Engineering Physics of the
Department of Energy.
55
Hao, J., Pan, T. W., Glowinski, R., & Joseph, D. D. (2009). A fictitious domain/distributed Lagrange
multiplier method for the particulate flow of Oldroyd-B fluids: A positive definiteness preserving ap-
proach. Journal of Non-Newtonian Fluid Mechanics, 156(1-2), 95–111. doi:10.1016/j.jnnfm.2008.07.006
Harlow, F. H., & Amsden, A. A. (1975a). Flow of Interpenetrating Material Phases. Journal of Compu-
tational Physics, 18(4), 440–464. doi:10.1016/0021-9991(75)90096-0
Harlow, F. H., & Amsden, A. A. (1975b). Numerical-Calculation of Multiphase Fluid-Flow. Journal of
Computational Physics, 17(1), 19–52. doi:10.1016/0021-9991(75)90061-3
Heit, W. (1986). SiC in Nuclear Technology Gmelin Handbook of Inorganic Chemistry, Silicon Suppl.
(Vol. B3, pp. 478-500): Springer.
Herrmann, H. J., & Luding, S. (1998). Modeling granular media on the computer. Continuum Mechanics
and Thermodynamics, 10(4), 189. doi:10.1007/s001610050089
Heynderickx, G. J., Das, A. K., De Wilde, J., & Marin, G. B. (2004). Effect of Clustering on Gas-Solid
Drag in Dilute Two-Phase Flow. Industrial & Engineering Chemistry Research, 43(16), 4635–4646.
doi:10.1021/ie034122m
Hill, R. J., Koch, D. L., & Ladd, A. J. C. (2001a). The first effects of fluid inertia on flows in ordered and
random arrays of spheres. Journal of Fluid Mechanics, 448, 213–241. doi:10.1017/S0022112001005948
Hill, R. J., Koch, D. L., & Ladd, A. J. C. (2001b). Moderate-Reynolds-number flows in ordered and
Hölzer, A., & Sommerfeld, M. (2008). New simple correlation formula for the drag coefficient of non-
spherical particles. Powder Technology, 184(3), 361–365. doi:10.1016/j.powtec.2007.08.021
Hölzer, A., & Sommerfeld, M. (2009). Lattice Boltzmann simulations to determine drag, lift and torque act-
ing on non-spherical particles. Computers & Fluids, 38(3), 572–589. doi:10.1016/j.compfluid.2008.06.001
Hudson, J., & Harris, D. (2006). A high resolution scheme for Eulerian gas-solid two-phase isentropic
flow. Journal of Computational Physics, 216(2), 494–525. doi:10.1016/j.jcp.2005.12.010
Huilin, L., Yurong, H., & Gidaspow, D. (2003). Hydrodynamic modelling of binary mixture in a gas
bubbling fluidized bed using the kinetic theory of granular flow. Chemical Engineering Science, 58(7),
1197–1205. doi:10.1016/S0009-2509(02)00635-8
Igci, Y., Andrews, A. T., Sundaresan, S., Pannala, S., & O’Brien, T. (2008). Filtered two-fluid models
for fluidized gas-particle suspensions. AIChE Journal. American Institute of Chemical Engineers, 54(6),
1431–1448. doi:10.1002/aic.11481
Ishii, M., & Mishima, K. (1984). 2-Fluid model and hydrodynamic constitutive relations. Nuclear En-
gineering and Design, 82(2-3), 107–126. doi:10.1016/0029-5493(84)90207-3
Jackson, R. (1983). Some mathematical and physical aspects of continuum models for the motion of
granular materials. In Meyer, R. E. (Ed.), Theory of Dispersed Multiphase Flow. New York: Academic
Press.
56
Jackson, R. (1997). Locally averaged equations of motion for a mixture of identical spherical particles
and a Newtonian fluid. Chemical Engineering Science, 52(15), 2457–2469. doi:10.1016/S0009-
2509(97)00065-1
Jackson, R. (2000). The dynamics of fluidized particles. Cambridge, UK: Cambridge University Press.
Jaeger, H. M., Nagel, S. R., & Behringer, R. P. (1996). Granular solids, liquids, and gases. Reviews of
Modern Physics, 68(4), 1259. doi:10.1103/RevModPhys.68.1259
Jaeger, H. M., Nagel, S. R., & Behringer, R. P. (1996). The physics of granular materials. Physics Today,
49(4), 32–38. doi:10.1063/1.881494
Jakobsen, H. A. (2008). Chemical Reactor Modeling. Berlin, Heidelberg: Springer.
Jenike, A. W. (1987). A Theory of Flow of Particulate Solids in Converging and Diverging Channels Based
on a Conical Yield Function. Powder Technology, 50(3), 229–236. doi:10.1016/0032-5910(87)80068-2
Jenkins, J. T. (1982). Rapid Deformations of Granular-Materials. Journal of Rheology, 26(6), 583–583.
Jenkins, J. T. (1992). Boundary-Conditions for Rapid Granular Flow - Flat, Frictional Walls. Journal of
Applied Mechanics-Transactions of the Asme, 59(1), 120–127. doi:10.1115/1.2899416
Jenkins, J. T., & Louge, M. Y. (1997). On the flux of fluctuation energy in a collisional grain flow at a
flat, frictional wall. Physics of Fluids, 9(10), 2835–2840. doi:10.1063/1.869396
Jenkins, J. T., & Mancini, F. (1989). Kinetic-Theory for Binary-Mixtures of Smooth, Nearly Elastic
Spheres. Physics of Fluids. A, Fluid Dynamics, 1(12), 2050–2057. doi:10.1063/1.857479
Jenkins, J. T., & Zhang, C. (2002). Kinetic theory for identical, frictional, nearly elastic spheres. Physics
of Fluids, 14(3), 1228–1235. doi:10.1063/1.1449466
Johnson, P. C., & Jackson, R. (1987). Frictional Collisional Constitutive Relations for Antigranulocytes-
Materials, with Application to Plane Shearing. Journal of Fluid Mechanics, 176, 67–93. doi:10.1017/
S0022112087000570
Johnson, P. C., Nott, P., & Jackson, R. (1990). Frictional Collisional Equations of Motion for Particu-
late Flows and Their Application to Chutes. Journal of Fluid Mechanics, 210, 501–535. doi:10.1017/
S0022112090001380
Joseph, D. D., Lundgren, T. S., Jackson, R., & Saville, D. A. (1990). Ensemble Averaged and Mixture
Theory Equations for Incompressible Fluid Particle Suspension. International Journal of Multiphase
Flow, 16(1), 35–42. doi:10.1016/0301-9322(90)90035-H
Jung, J., Gidaspow, D., & Gamwo, I. K. (2005). Measurement of two kinds of granular temperatures,
stresses, and dispersion in bubbling beds. Industrial & Engineering Chemistry Research, 44(5), 1329–
1341. doi:10.1021/ie0496838
Kandhai, D., Derksen, J. J., & Van den Akker, H. E. A. (2003). Interphase drag coefficients in gas-solid
flows. AIChE Journal. American Institute of Chemical Engineers, 49(4), 1060–1065. doi:10.1002/
aic.690490423
57
Karri, S. B. R., & Knowlton, T. M. (2005). The effect of riser termination geometry on close-coupled
cyclone performance. Circulating Fluidized Bed Technology VIII, 899-906 1024.
Kashiwa, B. A., & Rauenzahn, R. M. (1994). A multimaterial formalism. Paper presented at the Ameri-
can Society of Mechanical Engineers (ASME) Fluids Engineering Division summer meeting. Retrieved
from http://www.osti.gov/energycitations/servlets/purl/10139551-7iWb9X/native/
Khan, A. R., Pirie, R. L., & Richardson, J. F. (1987). Hydraulic Transport of Solids in Horizontal Pipe-
lines - Predictive Methods for Pressure-Gradients. Chemical Engineering Science, 42(4), 767–778.
doi:10.1016/0009-2509(87)80036-2
Kuipers, J. A. M., Prins, W., & Vanswaaij, W. P. M. (1992). Numerical-Calculation of Wall-to-Bed Heat-
Transfer Coefficients in Gas-Fluidized Beds. AIChE Journal. American Institute of Chemical Engineers,
38(7), 1079–1091. doi:10.1002/aic.690380711
Ladd, A. J. C. (1994). Numerical simulations of particulate suspensions via a discretized boltzmann-equation.
1. Theoretical foundation. Journal of Fluid Mechanics, 271, 285–309. doi:10.1017/S0022112094001771
Langroudi, M. K., Turek, S., Ouazzi, A., & Tardos, G. I. (2010). An investigation of frictional and col-
lisional powder flows using a unified constitutive equation. Powder Technology, 197(1-2), 91–101.
doi:10.1016/j.powtec.2009.09.001
Lathouwers, D., & Bellan, J. (2001a). Modeling of dense gas-solid reactive mixtures applied to biomass
pyrolysis in a fluidized bed. International Journal of Multiphase Flow, 27(12), 2155–2187. doi:10.1016/
S0301-9322(01)00059-3
Lathouwers, D., & Bellan, J. (2001b). Yield optimization and scaling of fluidized beds for tar production
from biomass. Energy & Fuels, 15(5), 1247–1262. doi:10.1021/ef010053h
Lebowitz, J. L. (1964). Exact Solution of Generalized Percus-Yevick Equation for a Mixture of Hard
Spheres. Physical Review, 133(4A), A895. doi:10.1103/PhysRev.133.A895
Liu, A. J., & Nagel, S. R. (1998). Nonlinear dynamics: Jamming is not just cool any more. Nature,
396(6706), 21–22. doi:10.1038/23819
Liu, M. Y., Li, J. H., & Kwauk, M. S. (2001). Application of the energy-minimization multi-scale method
to gas-liquid-solid fluidized beds. Chemical Engineering Science, 56(24), 6805–6812. doi:10.1016/
S0009-2509(01)00318-9
Loth, E., O’Brien, T., Syamlal, M., & Cantero, M. (2004). Effective diameter for group motion of poly-
disperse particle mixtures. Powder Technology, 142(2-3), 209–218. doi:10.1016/j.powtec.2004.04.033
Lu, B., Wang, W., Li, J. H., Wang, X. H., Gao, S. Q., & Lu, W. M. (2007). Multi-scale CFD simulation
of gas-solid flow in MIP reactors with a structure-dependent drag model. Chemical Engineering Science,
62(18-20), 5487–5494. doi:10.1016/j.ces.2006.12.071
Lu, B. N., Wang, W., & Li, J. H. (2009). Searching for a mesh-independent sub-grid model for CFD
simulation of gas-solid riser flows. Chemical Engineering Science, 64(15), 3437–3447. doi:10.1016/j.
ces.2009.04.024
58
Lun, C. K. K., Savage, S. B., Jeffrey, D. J., & Chepurniy, N. (1984). Kinetic Theories for Granular Flow
- Inelastic Particles in Couette-Flow and Slightly Inelastic Particles in a General Flowfield. Journal of
Fluid Mechanics, 140(Mar), 223–256. doi:10.1017/S0022112084000586
Lundgren, T. S. (1972). Slow flow through stationary random beds and suspensions of spheres. Journal
of Fluid Mechanics, 51(JAN25), 273-&.
Lyczkowski, R. W., Gidaspow, D., Solbrig, C. W., & Hughes, E. D. (1978). Characteristics and Stability
Analyses of Transient One-Dimensional 2-Phase Flow Equations and Their Finite-Difference Approxi-
mations. Nuclear Science and Engineering, 66(3), 378–396.
Lyczkowski, R. W., Solbrig, C. W., & Gidaspow, D. (1982). Forced-Convection Heat-Transfer in Rect-
angular Ducts - General-Case of Wall Resistances and Peripheral Conduction for Ventilation Cooling
of Nuclear Waste Repositories. Nuclear Engineering and Design, 67(3), 357–378. doi:10.1016/0029-
5493(82)90065-6
Makkawi, Y., & Ocone, R. (2006). A model for gas-solid flow in a horizontal duct with a smooth merge
of rapid-intermediate-dense flows. Chemical Engineering Science, 61(13), 4271–4281. doi:10.1016/j.
ces.2006.01.017
Mantle, M. D., Sederman, A. J., & Gladden, L. F. (2001). Single- and two-phase flow in fixed-bed reac-
tors: MRI flow visualisation and lattice-Boltzmann simulations. Chemical Engineering Science, 56(2),
523–529. doi:10.1016/S0009-2509(00)00256-6
Manz, B., Gladden, L. F., & Warren, P. B. (1999). Flow and dispersion in porous media: Lattice-Boltzmann
and NMR studies. AIChE Journal. American Institute of Chemical Engineers, 45(9), 1845–1854.
doi:10.1002/aic.690450902
Massoudi, M. (2003). Constitutive relations for the interaction force in multicomponent particulate flows.
International Journal of Non-linear Mechanics, 38(3), 313–336. doi:10.1016/S0020-7462(01)00064-6
Mathiesen, V., Solberg, T., Arastoopour, H., & Hjertager, B. H. (1999). Experimental and computational
study of multiphase gas/particle flow in a CFB riser. AIChE Journal. American Institute of Chemical
Engineers, 45(12), 2503–2518. doi:10.1002/aic.690451206
Maxey, M. R., & Riley, J. J. (1983). Equation of Motion for a Small Rigid Sphere in a Nonuniform Flow.
Physics of Fluids, 26(4), 883–889. doi:10.1063/1.864230
McKeen, T., & Pugsley, T. (2003). Simulation and experimental validation of a freely bubbling bed of
FCC catalyst. Powder Technology, 129(1-3), 139–152. doi:10.1016/S0032-5910(02)00294-2
Mishra, S. K., Muralidharan, K., Pannala, S., Simunovic, S., Daw, C. S., & Nukala, P. (2008). Spatio-
temporal Compound Wavelet Matrix Framework for Multiscale/Multiphysics Reactor Simulation: Case
Study of a Heterogeneous Reaction/Diffusion System. [Journal]. International Journal of Chemical
Reactor Engineering, 6(A28). doi:10.2202/1542-6580.1715
Montlucon, J. (1975). Heat and mass transfer in the vicinity of an evaporating droplet. International
Journal of Multiphase Flow, 2(2), 171–182. doi:10.1016/0301-9322(75)90006-3
59
Nakamura, K., & Capes, C. E. (1976). Vertical Pneumatic Conveying of Binary Particle Mixtures. In
Keairns, D. L. (Ed.), Fluidization Technology (pp. 159–184). Washington, DC: Hemisphere Publishing
Corp.
Nigmatulin, R. I. (1979). Spatial Averaging in the Mechanics of Heterogeneous and Dispersed Systems.
International Journal of Multiphase Flow, 5(5), 353–385. doi:10.1016/0301-9322(79)90013-2
Noren, R., & Develasco, R. (1992). Evolution of Coating Gas Distributors (No. PC- 000345). General
Atomics.
Nunziato, J. W., & Walsh, E. K. (1980). On ideal multiphase mixtures with chemical reactions and dif-
fusion. Archive for Rational Mechanics and Analysis, 73(4), 285–311. doi:10.1007/BF00247672
O’Brien, T. J., & Syamlal, M. (1993). Particle Cluster Effects in the Numerical Simulation of a Circulat-
ing Fluidized Bed. Paper presented at the 4th International CFB Conference.
Pan, T. W., Joseph, D. D., Bai, R., Glowinski, R., & Sarin, V. (2002). Fluidization of 1204 spheres: simu-
lation and experiment. Journal of Fluid Mechanics, 451, 169–191. doi:10.1017/S0022112001006474
Pan, T. W., Joseph, D. D., & Glowinski, R. (2001). Modelling Rayleigh-Taylor instability of a sediment-
ing suspension of several thousand circular particles in a direct numerical simulation. Journal of Fluid
Mechanics, 434, 23–37. doi:10.1017/S002211200100369X
Pannala, S., Daw, C. S., Boyalakuntla, D., & Finney, C. (2006). Process Modeling Phase I Summary
Report for the Advanced Gas Reactor Fuel Development and Qualification Program (No. TM-2006/520).
ORNL.
Pannala, S., Daw, C. S., Finney, C. E. A., Benyahia, S., Syamlal, M., & O’Brien, T. J. (2008). Modeling
the Collisional-Plastic Stress Transition for Bin Discharge of Granular Material
Pannala, S., Daw, C. S., Finney, C. E. A., Boyalakuntla, D., Syamlal, M., & O’Brien, T. J. (2007). Simu-
lating the dynamics of spouted-bed nuclear fuel coaters. Chemical Vapor Deposition, 13(9), 481–490.
doi:10.1002/cvde.200606562
Passman, S. L. (1977). Mixtures of Granular-Materials. International Journal of Engineering Science,
15(2), 117–129. doi:10.1016/0020-7225(77)90027-1
Patil, D. J., Annaland, A. V., & Kuipers, J. A. M. (2005a). Critical comparison of hydrodynamic models
for gas-solid fluidized beds - Part II: freely bubbling gas-solid fluidized beds. Chemical Engineering
Science, 60(1), 73–84. doi:10.1016/j.ces.2004.07.058
Patil, D. J., Annaland, M. V., & Kuipers, J. A. M. (2005b). Critical comparison of hydrodynamic models
for gas-solid fluidized beds - Part I: bubbling gas-solid fluidized beds operated with a jet. Chemical
Peirano, E., & Leckner, B. (1998). Fundamentals of turbulent gas-solid flows applied to circulating
fluidized bed combustion. Progress in Energy and Combustion Science, 24(4), 259–296. doi:10.1016/
S0360-1285(98)00002-1
60
Pigford, R. L., & Baron, T. (1965). Hydrodynamic Stability of a Fluidized Bed. Industrial & Engineer-
ing Chemistry Fundamentals, 4(1), 81. doi:10.1021/i160013a014
Prosperetti, A. (2007). Averaged equations for multiphase flow. In Prosperetti, A., & Tryggvason, G. (Eds.),
Computational Methods for Multiphase Flow. Cambridge: Cambridge University Press. doi:10.1017/
CBO9780511607486.009
Prosperetti, A., & Tryggvason, G. (2003). Appendix 3: Report of study group on computational physics.
International Journal of Multiphase Flow, 29(7), 1089–1099. doi:10.1016/S0301-9322(03)00081-8
Rao, K. K., & Nott, P. R. (2008). An Introduction to Granular Flow. Cambridge, UK: Cambridge Uni-
versity Press. doi:10.1017/CBO9780511611513
Reuge, N., Cadoret, L., Coufort-Saudejaud, C., Pannala, S., Syamlal, M., & Caussat, B. (2008). Multi-
fluid Eulerian modeling of dense gas-solids fluidized bed hydrodynamics: Influence of the dissipation
parameters. Chemical Engineering Science, 63(22), 5540–5551. doi:10.1016/j.ces.2008.07.028
Richardson, J. F., & Zaki, W. N. (1954). Sedimentation and Fluidization: Part I. Transactions of the
Institution of Chemical Engineers, 32, 22.
Richner, D. W., Minoura, T., Pritchett, J. W., & Blake, T. R. (1990). Computer-Simulation of Isothermal
Fluidization in Large-Scale Laboratory Rigs. AIChE Journal. American Institute of Chemical Engineers,
36(3), 361–369. doi:10.1002/aic.690360306
Rivard, W. C., & Torrey, M. D. (1977). K FIX: A Computer Program for Transient, Two-Dimensional,
Two-Fluid Flow (No. LA NUREG 6623).
Ruckenstein, E., & Muntean, O. (1967). On Mechanism of Bubble Formation in a Fluidized Bed. Ca-
nadian Journal of Chemical Engineering, 45(2), 95. doi:10.1002/cjce.5450450207
Saffman, P. G. (1971). Boundary Condition at Surface of a Porous Medium. Studies in Applied Math-
ematics, 50(2), 93.
Sankaranarayanan, K., Shan, X., Kevrekidis, I. G., & Sundaresan, S. (1999). Bubble flow simulations
with the lattice Boltzmann method. Chemical Engineering Science, 54(21), 4817–4823. doi:10.1016/
S0009-2509(99)00199-2
Sankaranarayanan, K., & Sundaresan, S. (2008). Lattice Boltzmann Simulation of Two-Fluid Model
Equations. Industrial & Engineering Chemistry Research, 47(23), 9165–9173. doi:10.1021/ie800283b
Savage, S. B. (1998). Analyses of slow high-concentration flows of granular materials. Journal of Fluid
Mechanics, 377, 1–26. doi:10.1017/S0022112098002936
Schaeffer, D. G. (1987). Instability in the Evolution-Equations Describing Incompressible Antigranulo-
cytes Flow. Journal of Differential Equations, 66(1), 19–50. doi:10.1016/0022-0396(87)90038-6
Scharff, M. F., Chan, R. K. C., Chiou, M. J., Dietrich, D. T., Dion, D. D., Klein, H. H., et al. (1982).
Computer Modeling of Mixing and Agglomeration in Coal Conversion Reactors, Vol. I & II (No. DOE/
ET/10329-1211).
61
Shi, S. P., Guenther, C., & Orsino, S. (2007, July 17-19, 2007). Numerical study of coal gasification using
Eulerian-Eulerian multiphase model. Paper presented at the ASME Powder 2007, San Antonio, Texas.
Silbert, L. E. (2005). Temporally heterogeneous dynamics in granular flows. Physical Review Letters,
94(9). doi:10.1103/PhysRevLett.94.098002
Sinclair, J. L., & Jackson, R. (1989). Gas-Particle Flow in a Vertical Pipe with Particle-Particle Inter-
actions. AIChE Journal. American Institute of Chemical Engineers, 35(9), 1473–1486. doi:10.1002/
aic.690350908
Snider, D. M. (2001). An incompressible three-dimensional multiphase particle-in-cell model for dense
particle flows. Journal of Computational Physics, 170(2), 523–549. doi:10.1006/jcph.2001.6747
Snider, D. M., O’Rourke, P. J., & Andrews, M. J. (1998). Sediment flow in inclined vessels calculated
using a multiphase particle-in-cell model for dense particle flows. International Journal of Multiphase
Flow, 24(8), 1359–1382. doi:10.1016/S0301-9322(98)00030-5
Soo, S. L. (1967). Fluid Dynamics of Multiphase Systems. Waltham, MA: Blaisdell Publishing Corp.
Srinivasan, M. G., & Doss, E. D. (1985). Momentum-Transfer Due to Particle Particle Interaction in Dilute
Gas Solid Flows. Chemical Engineering Science, 40(9), 1791–1792. doi:10.1016/0009-2509(85)80044-0
Srivastava, A., & Sundaresan, S. (2003). Analysis of a fractional-kinetic model for gas-particle flow.
Powder Technology, 129(1-3), 72–85. doi:10.1016/S0032-5910(02)00132-8
Stewart, H. B. (1979). Stability of 2-Phase Flow Calculation Using 2-Fluid Models. Journal of Compu-
tational Physics, 33(2), 259–270. doi:10.1016/0021-9991(79)90020-2
Stewart, H. B., & Wendroff, B. (1984). 2-Phase Flow - Models and Methods. Journal of Computational
Physics, 56(3), 363–409. doi:10.1016/0021-9991(84)90103-7
Sundaresan, S. (2000). Modeling the hydrodynamics of multiphase flow reactors: Current status and
challenges. AIChE Journal. American Institute of Chemical Engineers, 46(6), 1102–1105. doi:10.1002/
aic.690460602
Sundaresan, S. (2001). Some outstanding questions in handling of cohesionless particles. Powder Tech-
nology, 115(1), 2–7. doi:10.1016/S0032-5910(00)00423-X
Sundaresan, S. (2003). Instabilities in fluidized beds. Annual Review of Fluid Mechanics, 35, 63–88.
doi:10.1146/annurev.fluid.35.101101.161151
Sundaresan, S., Eaton, J., Koch, D. L., & Ottino, J. M. (2003). Appendix 2: Report of study group
on disperse flow. International Journal of Multiphase Flow, 29(7), 1069–1087. doi:10.1016/S0301-
9322(03)00080-6
Syamlal, M. (1985). Multiphase Hydrodynamics of Gas-Solids Flow. Chicago: Illinois Institute of
Technology.
Syamlal, M. (1987). The Particle-Particle Drag Term in a Multiparticle Model of Fluidization. Mor-
gantown: DOE.
62
Syamlal, M. (2006). Report on Workshop on Multiphase Flow Research (No. DOE/NETL-2007/1259).

Syamlal, M., & Bissett, L. A. (1992). METC Gasifier Advanced Simulation (MGAS) Model. Morgan-
town: DOE METC.
Syamlal, M., & Gidaspow, D. (1985). Hydrodynamics of Fluidization - Prediction of Wall to Bed Heat-
Transfer Coefficients. AIChE Journal. American Institute of Chemical Engineers, 31(1), 127–135.
doi:10.1002/aic.690310115
Syamlal, M., Guenther, C., Cugini, A., Ge, W., Wang, W., & Yang, N. (2010). Impact of computational
science on technology development. Chemical Engineering Progress.
Syamlal, M., Guenther, C., Gel, A., & Pannala, S. (2009). Advanced coal gasifier design using large-scale
simulations. Journal of Physics: Conference Series, 180, 012034..doi:10.1088/1742-6596/180/1/012034
Syamlal, M., Guenther, C., Gel, A., & Pannala, S. (2010). High performance computing: clean coal gasifier
designs using hybrid parallelization. Paper presented at the Fluidization XIII 2010, Gyeong-ju, Korea.
Syamlal, M., & O’Brien, T. J. (1987). A Generalized Drag Correlation for Multiparticle Systems. Un-
published Report U.S. Department of Energy, Office of Fossil Energy, Morgantown Energy Technology
Center.
Syamlal, M., & O’Brien, T. J. (2003). Fluid dynamic simulation of O3 decomposition in a bubbling fluid-
ized bed. AIChE Journal. American Institute of Chemical Engineers, 49(11), 2793–2801. doi:10.1002/
aic.690491112
Syamlal, M., O’Brien, T. J., Benyahia, S., Gel, A., & Pannala, S. (2008). Open source development
experience with a computational gas-solids flow code. Modelling and Simulation in Engineering, 10,
Article ID 937542, 937510 pages.
Syamlal, M., Rogers, W., & O’Brien, T. J. (1993). MFIX Documentation: Theory Guide (No. DOE/
METC-94/1004 (DE94000087)): Morgantown Energy Technology Center.
Tardos, G. I., McNamara, S., & Talu, I. (2003). Slow and intermediate flow of a frictional bulk powder
in the Couette geometry. Powder Technology, 131(1), 23–39. doi:10.1016/S0032-5910(02)00315-7
Tran-Cong, S., Gay, M., & Michaelides, E. E. (2004). Drag coefficients of irregularly shaped particles.
Powder Technology, 139(1), 21–32. doi:10.1016/j.powtec.2003.10.002
Truesdell, C. pp. 33–38 158–166. (1957). Sulle basi della thermomeccanica. Rand Lincei 22 (Series 8)
(33-38), 9.
Tsuji, Y., Kawaguchi, T., & Tanaka, T. (1993). Discrete Particle Simulation of 2-Dimensional Fluidized-
Bed. Powder Technology, 77(1), 79–87. doi:10.1016/0032-5910(93)85010-7
Tuzun, U., & Cleary, P. (2006). Discrete Element Modelling (DEM). Paper presented at the 5th World
Congress of Particle Technology. Retrieved from http://www.csiro.au/resources/DiscreteElementMo-
dellingTutorial.html van der Hoef, M. A., Annaland, M. V., Deen, N. G., & Kuipers, J. A. M. (2008).
Numerical simulation of dense gas-solid fluidized beds: A multiscale modeling strategy. Annual Review
of Fluid Mechanics, 40, 47-70.
63
Van der Hoef, M. A., Beetstra, R., & Kuipers, J. A. M. (2005). Lattice-Boltzmann simulations of low-
Reynolds-number flow past mono- and bidisperse arrays of spheres: results for the permeability and
drag force. [Article]. Journal of Fluid Mechanics, 528, 233–254. doi:10.1017/S0022112004003295
van Wachem, B. G. M., & Almstedt, A. E. (2003). Methods for multiphase computational fluid dynam-
ics. Chemical Engineering Journal, 96(1-3), 81–98. doi:10.1016/j.cej.2003.08.025
van Wachem, B. G. M., Schouten, J. C., van den Bleek, C. M., Krishna, R., & Sinclair, J. L. (2001).
Comparative analysis of CFD models of dense gas-solid systems. AIChE Journal. American Institute
Walker, D. M., & Tordesillas, A. (2010). Topological evolution in dense granular materials: A complex
networks perspective. International Journal of Solids and Structures, 47(5), 624–639. doi:10.1016/j.
ijsolstr.2009.10.025
Wallis, G. B. (1969a). Annular 2-Phase Flow. 2. Additional Effects. Mechanical Engineering (New York,
N.Y.), 91(10), 74.
Wallis, G. B. (1969b). Annular 2-Phase Flow. I. A Simple Theory. Mechanical Engineering (New York,
N.Y.), 91(10), 73.
Wen, C. Y., Chen, H., & Onozaki, M. (1982). User’s Manual for Computer Simulation and Design of
the Moving Bed Coal Gasifier. Morgantown: DOE.
Wen, C. Y., & Yu, Y. H. (1966). A Generalized Method for Predicting Minimum Fluidization Velocity.
AIChE Journal. American Institute of Chemical Engineers, 12(3), 610. doi:10.1002/aic.690120343
Williams, J., & O’Connor, R. (1999). Discrete element simulation and the contact problem. Archives of
Computational Methods in Engineering, 6(4), 279–304. doi:10.1007/BF02818917
Wylie, J. J., Koch, D. L., & Ladd, A. J. C. (2003). Rheology of suspensions with high particle inertia
and moderate fluid inertia. Journal of Fluid Mechanics, 480, 95–118. doi:10.1017/S0022112002003531
Xie, N., Battaglia, F., & Pannala, S. (2008a). Effects of using two- versus three-dimensional computational
modeling of fluidized beds - Part I, hydrodynamics. Powder Technology, 182(1), 1–13. doi:10.1016/j.
powtec.2007.07.005
Xie, N., Battaglia, F., & Pannala, S. (2008b). Effects of using two- versus three-dimensional computational
modeling of fluidized beds: Part II, budget analysis. Powder Technology, 182(1), 14–24. doi:10.1016/j.
powtec.2007.09.014
Yang, J. Z., & Renken, A. (2003). A generalized correlation for equilibrium of forces in liquid-solid
fluidized beds. Chemical Engineering Journal, 92(1-3), 7–14. doi:10.1016/S1385-8947(02)00084-0
Yang, N., Wang, W., Ge, W., Wang, L., & Li, J. (2004). Simulation of Heterogeneous Structure in a Cir-
culating Fluidized-Bed Riser by Combining the Two-Fluid Model with the EMMS Approach. Industrial
& Engineering Chemistry Research, 43(18), 5548–5561. doi:10.1021/ie049773c
Yin, X. L., & Sundaresan, S. (2009). Drag Law for Bidisperse Gas-Solid Suspensions Containing Equally
Sized Spheres. Industrial & Engineering Chemistry Research, 48(1), 227–241. doi:10.1021/ie800171p
64
Yoon, H., Wei, J., & Denn, M. M. (1978). Model for Moving-Bed Coal-Gasification Reactors. AIChE
Zabrodsky, S. S. (1966). Hydrodynamics and Heat Transfer in Fluidized Beds. Cambridge, MA: The
M.I.T. Press.
Zhang, D. Z., & Prosperetti, A. (1994a). Averaged Equations for Inviscid Disperse 2-Phase Flow. Journal
of Fluid Mechanics, 267, 185–219. doi:10.1017/S0022112094001151
Zhang, D. Z., & Prosperetti, A. (1994b). Ensemble Phase-Averaged Equations for Bubbly Flows. Physics
of Fluids, 6(9), 2956–2970. doi:10.1063/1.868122
Zhang, D. Z., & VanderHeyden, W. B. (2002). The effects of mesoscale structures on the macroscopic
momentum equations for two-phase flows. International Journal of Multiphase Flow, 28(5), 805–822.
doi:10.1016/S0301-9322(02)00005-8
Zhang, M. H., Chu, K. W., Wei, F., & Yu, A. B. (2008). A CFD-DEM study of the cluster behavior in
riser and downer reactors. Powder Technology, 184(2), 151–165. doi:10.1016/j.powtec.2007.11.036
65
66
Chapter 2
Hydrodynamic Equations
from Kinetic Theory:
Fundamental Considerations
James W. Dufty
University of Florida, USA
Aparna Baskaran
Syracuse University, USA
abstract
In this chapter, a theoretical description is provided for the solid (granular) phase of the gas-solidss
flows that are the focus of this book. Emphasis is placed on the fundamental concepts involved in deriv-
ing a macroscopic hydrodynamic description for the granular material in terms of the hydrodynamic
fields (species densities, flow velocity, and the granular temperature) from a prescribed “microscopic”
interaction among the grains. To this end, the role of the interstitial gas phase, body forces such as
gravity, and other coupling to the environment are suppressed and retained only via a possible non-
conservative external force and implicit boundary conditions. The general notion of a kinetic equation
is introduced to obtain macroscopic balance equations for the fields. Constitutive equations for the
fluxes in these balance equations are obtained from special “normal” solutions to the kinetic equation,
resulting in a closed set of hydrodynamic equations. This general constructive procedure is illustrated
for the Boltzmann-Enskog kinetic equation describing a system of smooth, inelastic hard spheres. For
weakly inhomogeneous fluid states the granular Navier-Stokes hydrodynamic equations are obtained,
including exact integral equations for the transport coefficients. A method to obtain practical solutions
to these integral equations is described. Finally, a brief discussion is given for hydrodynamics beyond
the Navier-Stokes limitations.
DOI: 10.4018/978-1-61520-651-3.ch002
Hydrodynamic Equations from Kinetic Theory
intrOductiOn
Activated granular materials occur ubiquitously in nature and practical realizations in industry. Many of
the phenomena occur on length scales that are large compared to the size of constituent particles (grains)
and time scales long compared to the time between collisions among the grains. In this case a descrip-
tion of the system in terms of the values for hydrodynamic fields in cells containing many particles,
analogous to molecular fluids, can apply for granular fluids. The hydrodynamic fields for molecular
fluids are the densities associated with the globally conserved quantities. In the absence of reactions,
these are the species densities, total momentum density, and the energy density. More commonly the
momentum density is replaced by a corresponding flow field, and the energy density is replaced by a
related temperature. The time dependence of such a macroscopic description (hydrodynamics) follows
from the exact conservation equations for these fields, supplemented by “constitutive equations” provid-
ing a closed description in terms of the fields alone.
The key difference between granular and molecular fluids is that the former involves collisions be-
tween macroscopic grains. These collisions conserve momentum but dissipate energy since part of the
kinetic energy of the grains goes into micro-deformations of the surface and exciting other internal modes
of the grains. Even so, a hydrodynamic description for a fluid of grains can be given under appropriate
conditions, following closely the approach developed in the context of molecular fluids, starting from
the exact “balance equations” for the densities of interest. The objective of this chapter is to provide
an overview of how general constitutive equations can be obtained from a fundamental basis in kinetic
theory. The discussion does not make specific reference to a particular fluid state or kinetic theory. This
overview is followed by a practical illustration for the special case of Navier-Stokes hydrodynamics for
weakly non-uniform states, derived from the generalized Enskog kinetic theory (van Beijeren & Ernst
1973, 1979) extended to granular systems (Brey, Dufty & Santos 1997; see also Appendix A of Garzo,
Dufty & Hrenya 2007). Extensive references to previous work on Navier-Stokes constitutive equations
from Boltzmann and Enskog kinetic theories can be found in the review of Goldhirsch 2003, the text of
Brilliantov & Poschel 2004, and in the recent articles Garzo, Dufty, & Hrenya 2007 and Garzo, Hrenya
& Dufty 2007.
The balance equations are local identities expressing the change in hydrodynamic fields of a cell due
to their fluxes through the boundaries of that cell and local sources within the cell. The central problem
is to represent the fluxes and sources in terms of these hydrodynamic fields and their gradients. In many
cases the form of these constitutive equations is known from experiments (e.g., Fick’s diffusion law,
Newton’s viscosity law). An important advantage of kinetic theory as the basis for constitutive equations,
in contrast to such phenomenology generalized from experiment, is that both quantitative and qualita-
tive predictions follow as mathematical consequences of the theory. Thus the form of the hydrodynamic
equations, the values of their parameters, and the validity conditions for applications are provided as
one unit. In practice, most applications to granular fluids have focused on low density conditions and
moderate densities at low dissipation (e.g., the Boltzmann and Enskog kinetic theories) (Jenkins &
Mancini 1989, Jenkins 1998, Lun 1991). However, the approach emphasized here is more general and
provides a means to describe quite general complex fluid states such as those that occur more generally
for granular fluids. The aim of this chapter is to provide a pedagogical overview of the basis for hydro-
dynamics as arising from kinetic theories (for a similar analysis based on the low density Boltzmann
equation see Dufty & Brey 2005). With this goal in mind, attention is restricted to the simplest case of
smooth grains interacting through pair-wise additive short ranged interactions. Other important effects
67
such as those due to the interstitial fluid phase, gravity etc. are included only at the level of an external
body force acting on the grains, and are addressed briefly in the next section.
The layout of the chapter is as follows. Section 2 provides an overview of role of kinetic theory and
hydrodynamics in the context of gas-solids flows, highlighting the advantages and limitations of each. In
the next section the notion of a kinetic theory as a “mesoscopic” theory is introduced in its most general
form. Next, the balance equations for the hydrodynamic fields are obtained from the kinetic theory with
explicit expressions for the fluxes in terms of the solution to the kinetic equation. Finally, the notion of
constitutive equations is introduced for special “normal” solutions to the kinetic equation. Together, the
balance equations supplemented with the constitutive equations yield the closed hydrodynamic descrip-
tion of the fluid in terms of the local fields. These general considerations are formally exact and provide
the basis for specialization to particular applications and practical approximations. The remainder of
this chapter is then focused on the important case of states with small spatial and temporal variations,
for which the Navier-Stokes hydrodynamic equations are obtained. The force law for particle-particle
collisions in the kinetic theory is idealized to that of smooth, inelastic hard spheres, and the collision
operator is specialized to a practical form (revised Enskog kinetic equation) appropriate for a wide range
of space and time scales, and densities. As an important illustration, the normal solution is described for
weakly inhomogeneous states as an expansion in the small spatial and temporal gradients, leading to
the explicit constitutive equations and expressions for the associated transport coefficients. Finally, the
need to go beyond Navier-Stokes hydrodynamics for many granular states is discussed. Applications of
the Enskog kinetic theory to uniform shear flow is noted as an important example.
The scope of topics covered is quite broad and a complete citation of all the important literature is
not practical. Instead, in many cases reference will be given to reviews in which extensive bibliographies
appear. Generally, it is hoped that the material presented is self-contained in the sense that the logical
presentation can be followed even though the full details of calculations are left implicit.
cOntext: gas-sOlidss flOws
Before embarking on the details of a kinetic theory and its basis for hydrodynamics, it is useful to review
the context of each in the description of the complex flows encountered in gas-solidss systems (e.g.
gas-fluidized beds). The complexity arises from a number of sources, e.g. gravitational field, geometry
(boundary conditions), and formation of heterogeneous structures (macroscopic bubbles or high density
clusters) (van Swaaij 1990, Gidaspow 1994). Since most features of interest occur at the laboratory scale,
hydrodynamics has been a primary tool in attempts to model gas-solidss flows (Jackson 2000, Kuipers,
Hoomans & van Swaaij 1998). On this large scale, both gas and particle subsystems are described by
Navier-Stokes continuum equations with sources coupling the two self-consistently (the Two Fluid
Model, Anderson & Jackson 1967). However, the parameters of these equations such as the particle-gas
force and transport properties must be supplied phenomenologically and their detailed forms can make
a significant difference in the flows predicted. To overcome this limitation, a more detailed description
on smaller length scales is required.
One approach is to describe the particle dynamics by numerical simulation of the associated Newton’s
equations of motion, while retaining a hydrodynamic description for the gas. This is the Discrete Particle
Model (also referred to as the Discrete Element Method) (Hoomans, Kuipers, Briels, van Swaaij, 1996;
Deen, van Sint Annaland, van der Hoef & Kuipers, 2007, Zhu, Zhou, Yang & Yu, 2007). This provides
68
a detailed and accurate description of the particle trajectories for given forces. In addition to gravity,
pressure gradients, particle-particle and particle-wall forces, an essential force in Newton’s equation is
that of the gas on the fluid particle. Implementation of the DPM can lead to different results depending
on the nature and treatment of the forces on the particle (Feng & Yu, 2004, Leboriero, Joseph & Hrenya
2008, van Wachem et al., 2007). Although this drag force on the particle is localized at the surface of
the particle, it can depend on the details of the gas fluid state including the indirect influence of other
particles on this state. For example, it can be different for dense monodisperse and polydisperse gas-solids
systems (van der Hoef, Beetstra & Kuipers, 2005). Recently, accurate modeling of the gas-particle drag
force has been possible using lattice gas Boltzmann methods that allow an accurate simulation of the
gas on a lattice smaller than the particle size, accounting for both the momentum transfer to the particles
and incorporating details of the boundary conditions. In this way lattice gas Boltzmann simulations
on the smallest scale provide the needed input for DPM on the mesoscopic scale (Hill, Koch & Ladd,
2001; Benyahia, Syamlal & O’Brien, 2006; Yin & Sundaresan, 2009; van der Hoef, van Sint Annaland
& Kuipers, 2004).
A kinetic theory description for the particles provides an alternative to the DPM on the same scale
of the particle positions and velocities. There are two main advantages of kinetic theory. First, it does
not have the practical limitations of discrete particle simulations to small (compared to laboratory)
systems of particles. Second, as described below, the transition to hydrodynamics and identification of
its parameters is straightforward from kinetic theory, but very much less so for DPM. It requires the
same input forces, both for collisions and for coupling to the gas phase (still described by Navier-Stokes
equations), and so can benefit from the recent developments for DPM. On the other hand, for very dense
clusters and glassy structures the form of the particle-particle collisions in kinetic theory is only known
semi-phenomenologically at this point. Applications of kinetic theory to gas-solids flows are mainly in
the context of providing the form and parameters of the two fluid model (see however Minier & Peirano
2001). Early examples of this approach include Sinclair & Jackson 1989, and Koch (1990); for a review
and references see Gidaspow, Jung, & Singh 2004. Recent improvements in the kinetic theory and its
systematic application for the normal solution have led to a more accurate solid phase hydrodynamics,
as described below. The coupled sets of continuum equations in the two fluid model then constitute a
problem in computational fluid dynamics, often including additional assumptions for the gas phase to
describe turbulent conditions, as described elsewhere in this book.
The derivation of hydrodynamics (specifically, constitutive equations) generally entails necessary
conditions, e.g., sufficiently small Knudsen numbers for Navier-Stokes hydrodynamics. As emphasized
below, the term “hydrodynamics” includes more general fluid states with correspondingly more com-
plex constitutive equations. In any case, such closures have associated validity conditions that must
be checked before application to a given problem. The complication arising from the particle-gas drag
force can affect these validity conditions significantly. The kinetic equation remains valid more gener-
ally, just as DPM, but the possibility of a hydrodynamic description might be precluded. Under these
conditions, the particle hydrodynamic description must be replaced with a more general solution to the
kinetic equation itself.
The objective of the current chapter is focused on the method of deriving systematically hydrodynamic
equations from a given kinetic equation. The specific kinetic equation considered and the complexity of
the resulting hydrodynamic description can depend on the flow conditions, geometry, degree of hetero-
geneity, etc. and such applications are the subject of other chapters in this book. The illustration of this
method given here is for an ideal granular fluid of inelastic hard spheres described by the generalized
69
Enskog equation. It is expected that this is a good compromise between accuracy and practical utility, not
limited by conditions of Knudsen number, Reynolds number, density heterogeneity, or geometry. In most
respects it has both the generality of DPM and the advantages of kinetic theory. The application of that
kinetic theory to Navier-Stokes hydrodynamics described here, however, has the additional more severe
limitations to small Knudsen numbers. While this is the most common hydrodynamics currently in use
for the two fluid description of gas-solids flows, it is clear from the derivation here that failure should
be expected for many conditions of interest (e.g., bubbles, plugs, rheology). Nevertheless, the systematic
derivation of Navier-Stokes hydrodynamics provides accurate results under its validity conditions for both
the form of those equations as well as quantitative values for the transport coefficients. For example, the
physical mechanisms that govern polydisperse mixing and separation processes (Duran, Rajchenbach
and Clement 1993) are not well understood yet; the recent results described here have been applied to
a controlled, quantitative means to study one of these mechanisms, thermal diffusion (Garzo 2009).
Kinetic tHeOry as a basis fOr HydrOdynaMics
Kinetic theory
Consider a mixture of s species of smooth spherical particles with masses {mi;1..s}. Their sizes and
material composition can be suppressed at this point as they enter only through the force laws for the
particle-particle and particle-gas interactions. These are taken to be short ranged (compared to relevant
cell sizes for the macroscopic description), conserve momentum, but dissipate energy. The hydrodynamic
fields of interest describe a few densities at each spatial point in the system. A more complete mesoscopic
description is given by the distribution of particles in the six dimensional phase space defined by the
points r,v, where r is the position and v is the velocity of a particle. For a system with s different spe-
cies, there is a set of distribution functions for all the species {fi(r, v; t); i = 1, …, s}. If these functions
are normalized to unity, then each fi(r,v;t) is the probability density of finding a particle of species i at
position r with velocity v at the time t. In the following, the normalization is chosen instead to be the
species densities so the interpretation is that of a number density of species i at r,v,t.
The species densities {ni(r,t)}, energy density e(r,t), and momentum density p(r,t) are defined in
terms of the distribution functions by
ni (r, t ) = ∫ dvf (r, v; t ) ,

i
i = 1,..., s (1)
s
1
e (r, t ) = ∑ ∫ dv mi v 2 f i (r, v; t )
i =1 2
1 s
+ ∑ ∫ dvdv ' dr ' Vij (r − r ') f ij (r, v; r ', v '; t ), (2)
2 i,j=1
p (r, t ) = ∫ dvm vf (r, v; t ) .

i i
(3)
70
Here, Vij(r − r′) is the potential energy for a pair of particles associated with the conservative part of
the force between them, and fij is the joint distribution function for two particles. It is usual to introduce
a flow field U in place of the momentum density according to
p (r, t ) ≡ ρ(r, t ) U (r, t ) . (4)
Similarly, the energy in the local rest frame is represented by a corresponding kinetic temperature T
3 1
e(r, t ) ≡ n(r, t )T (r, t ) + ρ(r, t )U 2 (r, t ), n(r, t )
2 2 (5)
= ∑ ni (r, t ), ρ(r, t ) = ∑ mi ni (r, t ).
i i
In the following, {ni(r,t)}, T(r,t), and U(r,t) will be referred to as the independent hydrodynamic fields.
These definitions express the dynamics of the hydrodynamic fields in terms of the more fundamental
dynamics of the distribution functions. Their evolution follows exactly from Newton’s equations (Ernst
2000)
(∂ t ) ( )
+ v1 ⋅ ∇r f i (r1, v1; t ) + ∇v ⋅ mi−1F0i (r1, v1 ) f i (r1, v1; t )
1 1
s (6)
= −∇ v ⋅ ∑ ∫ dr2 dv 2 mi−1Fij (r12 , v12 ) f ij (r1, v1; r2 , v 2 ; t ),
1
j =1
The left side describes the evolution of a distribution without interparticle interactions, in the presence
of a (possibly non-conservative) external force F0i due to confinement, gravity, and coupling to the gas
(in gas-solids flows). The right side represents the changes in the distribution for species i due to a force
from a particle of species j. The occurrence of that other particle at any point with any velocity is given
by the joint distribution for two particles fij(r1,v1; r2, v2; t). The force law Fij is chosen to depend on the
relative distances r12 = r1 − r2 and relative velocities v12 = v1 − v2, representing a central force conserv-
ing total momentum, but not conserving energy. Model interactions that belong to this class include the
Hertzian contact force model (Campbell 1990)
 3/ 2 
Fij (r, v ) = ˆrΘ(σ − r)aij (σ − r ) − cijˆr ⋅ v , (7)
 
where the first term describes response to elastic deformation and the second term describes the dis-
sipation of energy during this deformation. In general there is an additional tangential component of the
force as well, describing the roughness of the grains. More general non–central force laws representing
the shape of the particles can be included, but will not be considered here. Below, a limiting form of
equation (7) representing smooth, inelastic hard spheres with constant or velocity dependent coefficient
of restitution will also be introduced. The rest of the chapter is focused on developing the theory in the
context of these two model interactions.
Equation (6) is exact and applicable to quite general state conditions. However, it couples the dis-
tribution fi(r1,v1; t) to the two particle distributions fij(r1,v1; r2, v2; t). These obey similar equations but
71
are coupled to still higher order multi-particle distributions. The resulting set of equations is known as
the Born, Bogoliubov, Green, Kirkwood, Yvon (BBGKY) hierarchy (McLennan, 1989; Résibois & De
Leener, 1977; Ferziger & Kaper, 1972). In contrast, a kinetic theory is a closed equation for the set of
single particle functions {fi(r1,v1; t)} alone. Such closed equations result from equation (6) if the two
particle functions can be expressed as functionals of the one particle functions
( { })
f ij (r1, v1; r2 , v 2 ; t ) = ij r1, v1; r2 , v 2 | f k (t ) . (8)
If such a functional can be found, then (6) becomes a closed kinetic equation
(∂ t ) ( )
+ v1 ⋅ ∇r f i (r1, v1; t ) + ∇ v ⋅ mi−1F0i (r1, v1 ) f i (r1, v1; t )
1 1
(9)
(
= Ci r1, v1 | f k (t ) ,{ })
with the “collision operator”
( {
Ci r1, v1 | f k (t ) })
s (10)
(
= −∇ v ⋅ ∑ ∫ dr2 dv 2 mi−1Fij (r12 , v12 ) ij r1, v1; r2 , v 2 | { f k (t )} .
1
j =1
)
This constitutes the most general definition of a kinetic theory.
The discovery of the functional (i.e., a non-local dependence on the fields at all points) in equation
(8) and the corresponding collision operator is the point at which the difficult many-body problem is
confronted. It is common to all theoretical descriptions of macroscopic systems and there is a long his-
tory for molecular fluids (Bogoliubov, 1962; Cohen, 1962). Some analyses are systematic when there
is a small parameter. For example, at low density the granular Boltzmann equation can be recovered in
this way (Dufty, 2001). More generally, there is a mixture of analysis and phenomenology combined
with feedback from comparison of predictions with experiments. This is an area of active current inves-
tigation for dense granular systems. It will not be discussed further here beyond emphasizing that this
notion of a kinetic theory does not preclude the description of quite complex granular states, far outside
the limitations of Boltzmann kinetic theory.
Macroscopic balance equations
The macroscopic balance equations are those for the time derivatives of the hydrodynamic fields. They
follow directly from their definitions above in terms of integrals over fi(r, v; t) and the kinetic equation.
The balance equations for the number densities, energy density, and momentum are obtained in their
familiar forms
∂ t ni (r, t ) + mi−1∇ ⋅ ji (r, t ) = 0, (11)
72
∂ t e (r, t ) + ∇ ⋅ s (r, t ) = −w (r, t )

s (12)
+∑ ∫ dvF0i (r, v ) ⋅ vf i (r, v, t ),
i =1
s
∂ t pβ (r, t ) + ∇ γ tγβ (r, t ) = ∑ ∫ dvF0iβ (r, v ) f i (r, v, t ). (13)
i =1
The left sides are the expected forms of a time derivative for the density plus the divergence of a
flux. The right sides describe the sources and external forces. For example w(r,t) is the energy density
loss rate due to the non-conservative collisions among particles. In obtaining these expressions from
the kinetic equation, the mass fluxes {ji}, energy flux s, and momentum flux tB are obtained as explicit
linear integrals over the {fk(t)} and {Fij} To describe them, the contributions from pure convection are
first identified,
ji = j0i (r, t ) + ρi (r, t )U(r, t ), (14)
sβ (r, t ) = qβ (r, t )
3 1 
+  n(r, t )T (r, t ) + ρ(r, t )U 2 (r, t )U β (r, t ) + Pβγ (r, t )U γ (r, t ), (15)
 2 2 
tγβ (r, t ) = Pβγ (r, t ) + ρ (r, t )U β (r, t )U γ (r, t ) . (16)
The first terms on the right sides represent the corresponding flux in the local rest frame for each cell:
the diffusion fluxes j0i(r, t), the heat flux q(r, t), and the pressure tensor Pβγ(r, t) Their explicit forms
from the kinetic theory are (using the Hertzian force model (7) as an illustration of the force law (for
the case of hard spheres see Lutsko 2004))
j0i (r1, t ) ≡ mi ∫ dv1V1 f i (r1, v1, t ), (17)
s
1
q (r1, t ) ≡ ∑ ∫ dv1 mV 2 f (r , v , t )
i =1 2 i 1 i 1 1
1
1 s  2
ˆ (Gij ⋅ σ
ˆ ) aij (σ − λ ) + (µ ij − µ ji ) cij (g ⋅ σ
3/ 2
+ ∑ ∫ dv1 ∫ dv 2 ∫ d λλ 3 ∫ d σ ˆ)  (18)
2 i, j =1 0
 
1
(
×∫ d κij r1 − (1 − κ) λσ ˆ, v 2 | f k (t ) ,
ˆ, v1; r1 + κλσ { })
0
s
1
Pγβ (r1, t ) = ∑ ∫ dv1 mV V f (r , v , t )
i =1 2 i 1β 1γ i 1 1
1
1 s  3/ 2 
+ ∑ ∫ dv1 ∫ dv 2 ∫ d λλ 3 ∫ d σσ ˆ β  aij (σ − λ ) + cij (g ⋅ σ
ˆˆ γσ ˆ ) (19)
2 i, j =1 0
 
1
(
×∫ d κij r1 − (1 − κ) λσ ˆ, v 2 | f k (t ) ,
ˆ, v1; r1 + κλσ { })
0
73
where V1 = v1 − U(r, t) is the velocity in the local rest frame, Gij = μijV1 + μji V2 is the center of mass
velocity of the two colliding particles where μij= mi/(mi + mj) is the reduced mass of species i with re-
spect to species j and g = v1 − v2 is the relative velocity of the two particles. Similarly, the energy loss
rate is found to be
s 1
( { })
2
w (r, t ) = ∑∫ ˆcij (g ⋅ σ
dv1 ∫ dv 2 ∫ d λλ 2 ∫ d σ ˆ ) ij r1, v1; r1 − λσ
ˆ, v 2 | f k (t ) . (20)
i , j =1 0
The first terms on the right sides are the fluxes due simply to the motion of the particles (kinetic
fluxes), while the second terms in equations (18), (19), and (20) are due to the forces between particles
(collisional transfer).
Substituting equations (14) - (16) into (11) - (13) gives the balance equations in the desired form
Dt ni + ni ∇ ⋅ U + mi−1∇ ⋅ j0i = 0, (21)
3 3 s s
nDtT + Pγβ∂ r U β + ∇ ⋅ q − T ∑ mi−1∇ ⋅ j0i = −w + ∑ ∫ dvF0i ⋅ ( v − U) f i , (22)
2 γ
2 i =1 i =1
s
ρDtU β + ∂ r Pγβ = ∑ ni (r, t ) ∫ dvF0iβ f i , (23)
γ
i =1
where Dt = ∂ t + U ⋅ ∇ is the material derivative. These balance equations for the hydrodynamic fields
are an exact consequence of Newton’s equations. They have the same form as those for a molecular
fluid (McLennan 1989), except for the source w in the temperature equation due to non-conservative
forces. The fluxes {J0i}, q, Pγβ, and the energy source w are not given in terms of the fields so these
equations are not “closed”, i.e. they are not self-determined by the fields themselves. The terms involv-
ing the averages of the velocity dependent external force are also required. However, all of these un-
knowns are given in terms of the solution to the kinetic equation through equations (17) - (20) which
provides the controlled means for discovering the appropriate forms of these fluxes in terms of the fields.
In this general context, the form of the balance equations is independent of the specific particle-particle
interaction and the coupling to its environment.
“normal” states and Hydrodynamics
A true macroscopic description is obtained when equations (21) - (23) can be solved for the fields from
their given initial and boundary conditions. This requires a “closure” whereby the fluxes and source {J0i},
q, Pγβ, and w are expressed as functionals of the hydrodynamic fields through constitutive equations.
This is obtained directly from equations (17) - (20) by constructing solutions to the kinetic equations
that are expressed as functionals of the fields. Such solutions are called normal solutions (McLennan
1989). They are characterized by the fact that all space and time dependence of the {fi(r, v; t)} occurs
only through functionals of the hydrodynamic fields (denoted in the following collectively by {yα(t)})
74
( {
f i (r, v; t ) → f i r, v; yα (t ) , }) (24)
so that, for example, the space and time derivatives become
( {
∂ f v | y r, t 
 t i α( ) }) (
δf i r, v | yα (t ) ∂ y (r ', t )  {
 .
})
∇f i v | yα (r, t )  ∫
  = d r ' ∑  t β (25)
 ( { }) β δyβ (r ', t ) 
 ∇ ' y β(
r ', t ) 

Generally, an initial preparation of the system will not have this normal form. However, as for mo-
lecular fluids, it is expected that there is a short “kinetic” stage during which particles in each cell have
their velocities relax toward a universal form (e.g., Maxwellian for molecular fluid), but with values for
the hydrodynamic fields different for each cell. On a longer time scale the normal form can be supported.
For such solutions the functional Fij becomes normal as well,
(
Fij r1, v1; r2 , v 2 | f k ({ y (t )})) → G (r , v ; r , v | { y (t )}).
α ij 1 1 2 2 α
(26)
Consider some arbitrary property Ai(r, t) for species i defined as the average of a(v). For normal
states, all space and time dependence of this property occurs as a functional of the hydrodynamic fields
Ai (r, t ) = ∫ dva ( v) f (r, v | { y (t )}) = A (r | { y (t )}).

i α i α
In this way the averages defining the fluxes and energy source, equations (17) - (20), become the
desired constitutive equations, which for the Hertzian model equation (7) are
( { })
j0i r | yα (t ) = mi ∫ dvVf i r, v | yα (t ) , ( { }) (27)
s
1
( { })
q r | yα (t ) = ∑ ∫ dv
i =1 2
mV
i
2
(
Vf i r, v | yα (t ) { })
σ
1 s  2
ˆ (Gij ⋅ σ
ˆ ) aij (σ − λ ) + (µ ij − µ ji ) cij (g ⋅ σ
3/ 2
∑ dv dv ˆ) 
2 i, j =1 ∫ 1 ∫ 2 ∫0
+ d λλ 3 ∫ d σ (28)
 
1
(
×∫ d κij r − (1 − κ) λσ ˆ, v 2 | yα (t ) ,
ˆ, v1; r + κλσ { })
0
s
( { })
Pγβ r | yα (t ) = ∑ ∫ dvmV V f r, v | yα (t )
i β γ i
i =1
( { })
s σ
 3/ 2 
+ ∑ ∫ dv1 ∫ dv 2 ∫ d λλ 3 ∫ d σσ ˆ β aij (σ − λ ) + cij (g ⋅ σ
ˆˆ γσ ˆ ) (29)
i , j =1
 
0
1
(
×∫ d κij r − (1 − κ) λσ ˆ, v 2 | yα (t ) ,
ˆ, v1; r + κλσ { })
0
and
75
s σ
( { }) ∑ ∫ dv ∫ dv ∫ d λλ ∫ d σˆc (g ⋅ σˆ)  (r, v ; r − λσˆ, v | { y (t )}).

2
w r | yα ( t ) = 1 2
3
ij ij 1 2 α
(30)
i , j =1 0
The balance equations (21) - (23) together with the constitutive equations (27) - (30) constitute the
most general definition of hydrodynamics for a molecular or granular fluid.
It is appropriate at this point to pause and discuss the choice of independent hydrodynamic fields. The
fundamental idea is that such fields should represent the dominant dynamics on large space and time scales.
If they are local conserved densities this property is assured since they ultimately approach constants
as the system becomes uniform. Hence, for a molecular fluid the species densities, momentum density
(or flow velocity), and the energy density (or temperature) are the clear choices. Note that the species
energy densities or partial temperatures are not conserved and are therefore not appropriate choices for
independent hydrodynamic fields. For granular fluids, the species densities and momentum density are
still conserved and are proper choices for fields. However, now the total energy density (temperature),
is not conserved due to the inelastic collisions, and it is not clear that its dynamics should dominate
other kinetic modes on the long time scale. For the moment it will be assumed that this dominance still
applies, and further discussion is provided below. In any case, just as for molecular fluids, the partial
temperatures are not independent fields. For notational simplicity it may be useful to introduce partial
temperatures through the definition,
3 1
nT ≡
2 i i ∫ dv 2 mV i
2
( {
f i r, v | yα (t ) . }) (31)
However, these should be viewed simply as measures of the second moments of the species distribu-
tions and, as the last equality emphasizes, these partial temperatures are determined as functions of the
hydrodynamic fields chosen here. The definition of T in equation (5) then implies the identity
∑ n T ({n } , T )
i i i
T = i
. (32)
∑n i
i
In general, for mechanically different species, the partial temperatures are all different and not equal
to T (Garzo & Dufty, 1999).
This concludes the characterization of the formal basis of macroscopic hydrodynamics of a multicom-
ponent granular fluid as arising from a more mesoscopic description of the system given by a general
kinetic theory of the associated one particle distribution functions.
uniform fluid Hydrodynamics
Solutions to the hydrodynamic equations cannot be addressed until the details of the constitutive equa-
tions are specified and suitable initial and boundary values given. An exception is the simplest case of an
isolated, uniform fluid. The spatial variations of the fields vanish, and with stationary uniform boundary
76
conditions and no external forces a molecular fluid would be in its equilibrium state. For a granular fluid,
however, equations (21) - (23) become
∂ t ni = 0 = ∂ tU β , ( )
∂ tT = −ζ {ni } , T T , (33)
where the “cooling rate” ζ has been introduced in place of the energy loss rate
(
2 w {ni } , T ).
( )
ζ {ni } , T ≡
3nT
(34)
The solution represents a uniform fluid at rest with a monotonically decreasing temperature, and is
known as the homogeneous cooling state (HCS). This state was first discussed by Haff (Haff, 1983),
and subsequently studied via low density kinetic theory (Brey, Ruiz-Montero & Cubero, 1996; van
Noije & Ernst, 1998), and more generally by molecular dynamics simulations (Deltour & Barrat 1997;
Goldhirsch, Tan & Zanetti, 1993; McNamara & Young, 1996). The corresponding normal solution to the
kinetic equation in this case is the homogeneous cooling solution, obtained by substituting the normal
form equation (24) for a homogeneous state into the kinetic equation (9)
( ) ( { }) ( { ({ y (t)})}).
−ζ {ni } , T T ∂ T f i v | yα (t ) = Ci v | f i α
(35)
As will become clear in the next section, a local HCS for each cell with the local values for the
hydrodynamic fields is the reference state about which the hydrodynamic description of an isolated
granular fluid is constructed. In this sense, the local HCS plays the same role for a granular fluid that
the local equilibrium state plays for molecular fluids, and hence the HCS is an important state to study
and characterize. Further discussion of the HCS is given below in the specific context of the hard sphere
granular fluid.
navier stOKes HydrOdynaMics fOr tHe Hard sPHere fluid
The analysis of the above section shows that the derivation of hydrodynamics from kinetic theory has
a very general context. There are two main difficulties in implementing this generic prescription. The
( { ({ y (t )})}) in equation (8) which provides

first is the determination of the functional ij r1, v1; r2 , v 2 | f k α
the kinetic equation. The second is finding the normal solution to the given kinetic equation to obtain
the constitutive equations. In the remainder of this chapter, the application of a normal solution is il-
lustrated for the special case of a granular fluid modeled as a mixture of smooth, inelastic hard spheres.
A practical kinetic theory in this case, applicable over a wide range of densities, is given by the general-
ized Enskog equation (van Beijeren & Ernst, 1973, 1979; Brey, Dufty & Santos, 1997; Garzó, Dufty, &
Hrenya, 2007). The normal solution for this kinetic equation is obtained by the Chapman-Enskog
method (Ferziger & Kaper, 1972; Brey, Dufty, Kim, Santos, 1998; Garzó & Dufty, 1999) for states with
77
small spatial variations of the hydrodynamic fields over distances of the order of the mean free path.
The resulting hydrodynamic equations are partial differential equations with spatial derivatives up to
degree two, known as the Navier-Stokes equations for a granular fluid. Recent applications of hydro-
dynamics and kinetic theory for granular fluids in the context of the Enskog equation and its low den-
sity limit, the Boltzmann equation, can be found in Pöschel & Luding 2001, Pöschel & Brilliantov, 2003,
and Brilliantov & Pöschel, 2004. The presentation here for the generalized Enskog equation follows the
recent work of Garzo, Dufty, & Hrenya 2007.
enskog Kinetic theory
As noted above, the derivation of a kinetic equation requires confrontation of the difficult many-body
problem in nonequilibrium statistical mechanics. The analysis for molecular fluids is most complete for
the idealized force law of hard, elastic spheres. This is a realistic quantitative model as well because the
real short ranged repulsion is ≈ (σ/r)P, where σ is the particle–particle force range (particle size), and p is
an integer. Since p is large for repulsive interactions the hard sphere limit (p → ∞) is a good idealization.
A detailed analysis of the hard sphere limit for molecular fluids is given in (Dufty 2002; Dufty & Ernst
2004). Similarly for granular fluids, the repulsive part of the force in (7) is characterized by parameters
aij that determine the rigidity of the interaction. For rigid particles the fractional compression is small
and the velocity changes occur in a very short time. During this time, however, some energy is lost so
the normal component of the asymptotic relative velocity is decreased. This type of collision is captured
by the idealization of inelastic hard spheres that replace the short collision time by an instantaneous col-
lision and the energy loss is captured by a single parameter for each pair of species, namely a coefficient
of restitution αij. For this model interaction, the velocities of a pair of particles undergoing a collision
event change instantaneously according to
mj
v1′ = v1 −
mi + m j
(1 + α ij )(σˆ ⋅ g12 ) σˆ,
(36)
mi
v ′2 = v 2 +
mi + m j
(1 + α ij )(σˆ ⋅ g12 ) σˆ.
The prime denotes the velocities after collisions, g12 ≡ v1 − v2, and ŝ is a unit vector directed along
the line of the centers from 2 to 1. The particles are characterized by their masses and diameters {mi,
σi}, and a set of restitution coefficients for collisions among the same and different species {αij}. The
mµ
( )
2
change in kinetic energy for the pair in equation (6) is ∆Eij = − 1 − α ij 2 i ji (g12 ⋅ σ ˆ ) . Thus the
4
collisions are elastic for αij = 1 and inelastic for 0 < αij < 1. It is easily verified that the total momentum
is conserved for these collisions in both cases. In general these coefficients of restitution should depend
on the normal component of the relative velocity for the pair. However, for sufficiently activated par-
ticles it is possible to consider a simpler model for which the αij are constants. For notational simplicity
below, it is useful to introduce a substitution operator bij that changes the velocities of a function accord-
ing to the rule equation (36)
78
X ( v1′, v ′2 ) = bij X ( v1, v 2 ) . (37)
The form equation (36) implies no change in the tangential components of the velocities (perpen-
dicular to ŝ ) and so represents smooth hard spheres. More generally, it is possible to include the effects
of rough spheres (Lun 1991) and hard, non-spherical shapes but that will not be considered here.
The forces between hard spheres are singular (impulsive) and so the formal collision operator in
equation (10) must be changed accordingly to (Brey, Dufty, & Santos, 1997; Dufty & Baskaran, 2005;
Ernst 2000; Lutsko, 2004; van Noije & Ernst, 2001)
( { })
Ci r1, v1 | f i (t ) = −∑ ∫ dr2 dv 2Tij (r12 , v12 ) ij r1, v1; r2 , v 2 | f k (t )
j =1
( { }) (38)
The action of the force has been replaced by a binary scattering operator Tij (r12 , v12 )
Tij (r12 , v12 ) = σij2 ∫ d σ

ˆΘ (σ
ˆ ⋅ g12 ) (σ (
ˆ ⋅ g12 ) α −ij 2δ (r2 − r1 + σ) bij−1 − δ (r2 − r1 − σ) . ) (39)
The delta functions represent the fact that velocity changes only for particles at contact and on the
hemisphere with particles directed at each other. Further details of the dynamics of hard particles are
not required here and the interested reader is referred to the literature (see for example and references,
Appendix A of Baskaran, Dufty, & Brey 2008).
The “revised” Enskog kinetic theory results from an approximation to Fij(r1, v1; r2, v2 | {fk(t)}) as
follows,
(
ij r1, v1; r2 , v 2 | f k (t ) { }) (40)
(
= χij r1, r2 | {n }) f (r , v , t ) f (r , v , t ) .
k i 1 1 j 2 2
Where {nk} are the nonequilibrium densities (i.e., the velocity integrals of {fk}). To interpret this
approximation, consider the integral ofFij(r1, v1; r2, v2 | {fk(t)}) overall velocities
∫ dv dv  (r , v ; r , v | { f (t )}) =∫ dv dv f (r , v ; r , v , t )
1 2 ij 1 1 2 2 k 1 2 ij 1 1 2 2
(41)
= n (r , t ) n (r , t ) g (r , r , t )
i 1 j 2 ij 1 2
which defines the nonequilibrium pair distribution function for positions, gij(r1, r2, t). This is being re-
placed in the Enskog approximation by a universal functional of the nonequilibrium density χij r1, r2 | {nk } . ( )
For a molecular fluid the functional is fixed by the requirement that gij(r1, r2, t) should be the known
functional at equilibrium. A similar requirement can be imposed here by the HCS solution (Lutsko 2001,
2002). Finally, it is noted that while the above Enskog approximation retains correlations of positions
it completely neglects velocity correlations. This is consistent with the equilibrium state for a molecular
fluid, but not for nonequilibrium states in general. For a granular fluid it does not apply exactly even for
the HCS, and so it is an uncontrolled assumption here that the effects of such velocity correlations are
relatively small for many properties of interest.
79
The revised Enskog equation now follows from equations (9), (38), and (39)
(∂ t 1
)
+ v1 ⋅ ∇r f i (r1, v1; t ) + ∇ v
1
(42)
( )
⋅ m F (r1, v1 ) f i (r1, v1; t ) = CiE r1, v1 | f k (t ) ,
−1
i 0i ( { })
with the collision operator
( { })
CiE r1, v1 | f k (t ) ≡ −∑ σij2 ∫ dv 2 ∫ d σ
ˆΘ (σ
ˆ ⋅ g12 ) (σ
j =1
ˆ ⋅ g12 )
(43)
( )
×[χ ij r1, r1 − σ | {nk } f i (r1, v1′′; t ) f j (r1 − σ, v ′′2 ; t )
−χ ij (r , r + σ | {n }) f (r , v ; t ) f (r + σ, v ; t )].
1 1 k i 1 1 j 1 2
Here v ′′α ≡ bij−1v α represents the velocities due to restituting collisions (the inverse of equation (36)).
It is easily seen that the granular Boltzmann kinetic equation (Brey, Dufty, Kim, Santos 1998) is recov-
ered in the low density limit for which Xij → 1 and f j (r1 ± σij , v 2 ; t ) → f j (r1, v 2 ; t ) since σij is small
compared to the mean free path. At higher densities both the effects of Xij ≠ 1 and the delocalization of
colliding pairs become important, and are included in the revised Enskog theory. As a historical note,
the original Enskog equation has Xij as a constant (Ferziger & Kaper 1972) rather than a functional of
{ni}, the actual nonequilibrium densities (van Beijeren & Ernst 1973, 1979). While the earlier version
is adequate to describe transport in a one component fluid, it is inconsistent with nonequilibrium ther-
modynamics for mixture transport (López de Haro, Cohen & Kincaid 1983). The revised theory resolves
this problem, and also provides an exact description of the equilibrium fluid and solid phases for hard
spheres (Kirkpatrick, Das, Ernst & Piasecki 1990). Thus the revised kinetic equation has the capacity
to describe complex states of fluids with solid-like clusters which can occur more commonly in granu-
lar matter. Applications to molecular fluids suggests that predictions have quantitative accuracy up to
moderately dense states (nσ3 ≤ 0.2) and semi-quantitative accuracy for some properties up to the freez-
ing density (Alley, Alder, & Yip, 1983; Boon & Yip, 1991). Similar accuracy is observed for granular
fluids as well, but further conditioned by the degree of inelasticity.
In summary, the revised Enskog kinetic theory is a remarkably simple yet accurate description of the
complex dynamics under conditions where many-body effects are expected to be important. Theoretical
descriptions beyond the Enskog approximation are necessary for very dense, glassy metastable states
for molecular fluids (for early references see (Alley, Alder, & Yip, 1983; Boon & Yip, 1991; Dorfman
& Kirkpatrick, 1980; Sjogren, 1980; Alley, Alder, & Yip, 1983)). Its qualitative limitations for granular
fluids are still under investigation.
Hcs solution
For spatially homogeneous states, the normal solution to the Enskog equation is the HCS and has the form
 
 v − U h 
(
f hi v, {nk } , Th (t ) ) −3
= n v ϕ i  ; {nh } . (44)
i 0
(
 v0 Th (t ) 
 )
80
s
where v0 (Th ) = 2Th m is a “thermal velocity” associated with the temperature Th and m = ∑ mi s
i =1
is the average mass. This defines the dimensionless distribution φi. The uniform densities {nhi}, uniform
temperature Th(t), and flow velocity Uh in equation (44) are those for the HCS. This scaling property is
special to the hard sphere interaction and follows from dimensional analysis since there is no intrinsic
energy scale in this case. The functional forms of the HCS distributions are obtained from equation (35)
1 (0)
− ζ h∇ V ⋅ ( Vf hi ) = ∑ χij J ij (V | f hi ), (45)
2 j
(0)
and J ij (v 1
| f hi ) are the Boltzmann collision operators for a low density granular mixture
(0)
J ij( v | f ) ≡ σ ∫ dv ∫ d σˆΘ (σˆ ⋅ g )(σˆ ⋅ g )[α
1 hi
2
ij 2 12 12
−2
ij
f ( v1′′, t ) f j ( v ′′2 , t )
i
(46)
− f ( v , t ) f ( v , t )].
i 1 j 2
(the low density Boltmann operators occurs here because for the local HCS the Enskog and Boltzmann
operators differ only by the factors Xij). Use has been made of translational and rotational invariance to
( )
reduce χ ij r1, r1 − σij | {nk } to a constant function of {ni}, denoted simply by Xij. Aside from normal-
ization, all other density dependence of the {fhi} occurs through the {Xij}. In the elastic limit, αij → 1,
the solutions are all Maxwellians at the constant temperature T, as required. Otherwise the fhi have a
quite different functional form.
It was noted above that the partial temperatures for each species do not constitute hydrodynamic
fields, but rather are functions of T and {ni}. Still, they may be useful properties to characterize the
solutions to equation (45). The partial temperature Ti is related to the species kinetic energy analogous
to equation (5), and the associated cooling rate is defined by Ti−1∂ tTi = −ζi . Multiplying equation (45)
by miV2 and integrating gives ζ = ζi for all species i. Thus, in the HCS the common feature among spe-
cies is their cooling rates rather than their temperatures. In fact, it can be shown that for mechanically
different species the partial temperatures are different (Garzó & Dufty, 1999). Equipartition of energies
is a property of equilibrium states, implying a common temperature for all degrees of freedom, but vio-
lated for nonequilibrium states including all granular states. This has been verified directly via MD
simulation (Dahl, Hrenya, Garzo, Dufty, 2002).
Methods of solution
The focus of this chapter is the derivation of hydrodynamic equations from the kinetic equation. As
described above, this requires a special “normal solution” to the kinetic equation in which all space and
time dependence occurs through the hydrodynamic fields. The explicit construction of such a solution
for the Enskog kinetic equation is the subject of the next section. However, it is appropriate to digress
for a brief overview of some other, more general, methods to solve the Enskog equation. Such solutions
provide a more detailed representation of the fluid from macroscopic length and time scales down to
81
those for the particles. Although more numerically intensive in applications than hydrodynamics, these
solutions are appropriate when validity conditions for a continuum description fail.
The most accurate method for solution to the Enskog equation is that of direct simulation Monte Carlo
(DSMC). Originally developed to solve the Boltzmann equation (Bird 1994, Garcia & Alexander 1997),
it has been extended to the Enskog equation as well (Montanero & Santos 1996) and has a direct analog
for the granular fluid. This is a simple and efficient algorithm applicable to a wide range of conditions.
In principle, the solution to the kinetic equation is equivalent to specifying all moments of the dis-
tribution function. Equations for the moments are obtained by taking moments of the Enskog equation,
leading to a hierarchy of equations coupling lower order moments to those of higher order. Approximate
solutions are obtained by truncating this hierarchy. This method was first applied to the Boltzmann equa-
tion in Grad 1960 and is known as the Grad moment method. An early application of moment methods
to two-phase flows was given in Simonin 1996. A generalized, more accurate form of the Grad moment
method for the Enskog equation has been given in Lutsko 1997 and Lutsko 2004 and applied to granular
shear flow far outside the Navier-Stokes domain. A related method based on information from low order
moments is the Maximum Entropy method (Koopman, 1969), currently being applied to polydisperse
gas-solids flows.
The next section is devoted to finding the normal solution to the Enskog equation according to the
Chapman-Enskog method. This method does not address the most general normal solution, but rather its
form when the spatial gradients of the hydrodynamic fields are sufficiently small to allow a perturbation
expansion around a local HCS. The extension of the Chapman-Enskog method to arbitrary reference
states has been proposed by Lutsko 2006, but will not be considered here.
chapman-enskog solution
Now consider the general case of a spatially inhomogeneous state and return to the problem of finding
the normal solution. The functional dependence of the normal distribution function on the hydrodynamic
fields at all points can be given an equivalent local representation in terms of the fields and all their
derivatives at a given point
( { }) ( {
f i r, v | yα (t ) = f i v, yβ (r, t ), ∇r yβ (r, t ),... . }) (47)
If the system is only weakly inhomogeneous a further Taylor series expansion can be carried out to give
( { })
f i v, yβ (r, t ), ∇r yβ (r, t ),...
(48)
= f ( v, { y (r, t )}) + f ( v, { y (r, t ), ∇ y (r, t )}) + ...
() 0 () 1
i β i β r β
(0) (1)
Here, f i depends on the fields but not the gradients, and f i is both a function of the fields and
linear in their gradients. The dots in the above equation denote second and higher powers of ∇r yα as
well as higher degree derivatives of yα This is an expansion in the small spatial variations of the hydro-
dynamic fields over distances of the order of the mean free path,  mfp | ∇r yα | / yα . Here, the mean free
path is defined as  mfp = 1 / nσ2 , where n is a characteristic number density and σ is a characteristic
82
particle size. The Chapman-Enskog method is a procedure for constructing a normal solution to a given
kinetic equation as an expansion in these small spatial gradients (Ferziger & Kaper 1972). Formally, a
solution of the form equation (48) is substituted into the kinetic equation and terms of each order in the
small parameter are set equal to zero to determine the solution. There are two complications in ordering
the kinetic equation. The first is the choice for any “size” of external force as measured in terms of the
small parameter, and the second is the size of the time derivatives for each term in the distribution func-
tion. The scaling of the force depends on the conditions of interest, such as a gravitational or other ex-
ternal field, such as the coupling to the gas phase in gas-solids flows, whose effect on transport may be
weak or strong when compared to that due to spatial nonuniformities in the fields. A proper assessment
of this force with respect to all dimensionless parameters of both the gas and solid phases is essential
for a correct modeling of gas-solids flows and the justification of the two fluid model resulting from the
Chapman-Enskog solution. In Garzo, Dufty, & Hrenya 2007 the magnitude of the force was arbitrarily
taken to be of first order in the expansion parameter (e.g., of first order in the Knudsen number defined
by the length scale for variation of the hydrodynamic fields). Here, however, to illustrate the Chapman-
Enskog method in its simplest form the analysis is restricted to the case for which no the external
forces are present.
The ordering of the time derivative is accomplished by noting that the time derivatives of fi(r, v |
{yα(t)}) are proportional to the time derivatives of the fields, since the solution is normal. Furthermore, it
is required that these hydrodynamic fields be solutions to the balance equations. The latter relate the time
derivatives of the fields to their spatial gradients, providing the desired ordering of the time derivatives,
(0) (1)
∂ t yα (t ) = ∂ t yα (t ) + ∂ t yα (t ) + ..., (49)
For example, substituting directly the expansion (49) into equations (21) – (23) and identifying con-
tributions to each order in the gradients gives for the first and second order derivatives
(0) (0) (0) (0)

∂ t ni = 0, ∂ t T = −ζ h T , ∂ t U β = 0, (50)
and
(1) (0)
∂ t ni + ∇ ⋅ (ni U) + mi−1∇ ⋅ j0i = 0, (51)
(1)
∂t T = −
2n
(
3 (0)
Pγβ ∂ r U β − ∇ ⋅ q
γ
(0)
)
(52)
s
T (0) (1)
+∑ ∇ ⋅ j0i − ζ h T ,
i =1 nmi
(1) (0)
ρ∂ t U β = −ρU ⋅ ∇U β − ∂ r Pγβ . (53)
γ
83
The fluxes and cooling rate with superscripts 0 or 1 denote the result obtained from equations (27)
(0) (1)
– (30) using f i or f i , respectively. Note that these are not equations for the fields, which are not
(0) (1)
being expanded. Rather they are definitions of ∂ t yα (t ) and ∂ t yα (t ) in terms of the fields and their
(0)
gradients. The fact that ∂ t ≠ 0 (due to collisional cooling) is a new feature of granular fluids. Its exact
incorporation in the Chapman-Enskog expansion assures that the small gradient expansion places no
restrictions on the degree of inelasticity.
(0) (1)
It is now straightforward to implement the Chapman-Enskog procedure to determine f i and f i
in equation (48). The details are described elsewhere (for recent reviews and lists of early references see
(Garzo, Dufty, Hrenya 2007; Lutsko 2005)) and only the results described here. At lowest order the
(0)
Enskog kinetic equation is of the same form as the HCS equation equation (45). Hence, its solution, f i
is the local HCS distribution
 
(0) | v − U |
fi = ni v0−3 (T ) ϕ i  ; {nk } . (54)
 v (T ) 
 0
More specifically, it is the same functional form as in equation (44) except with {nhi}, Th(t), and Uh
replaced by the actual nonequilibrium fields {ni(r,t)}, T(r,t), and U(r,t). It should be emphasized that this
leading order “reference state” is not chosen as the basis for the expansion, but rather is a consequence
of the Chapman - Enskog procedure. There is no flexibility to choose a different reference state for the
chosen small gradient expansion.
To first order in the gradients the solution is
fi
(1)
(
= ∑ im V; T , {nk } ⋅ ∇r ym .
m
) (55)
Here {ym} is a set of independent linear combinations of {ni}, T and components of U. They can be
chosen such that the linear equations determining the coefficients im ( V ) all have the form
((L − λ ) A ) = A .
(m) m
i
i
m
(56)
The linear operator L is defined in terms of the linearized forms of the Enskog operators in equation (46)
1 (0)
(X ) i
=
2
ζ ∇ V ⋅ ( VX ) + ( LX ) ,
i i
(57)
( )
s
= −∑ χij J ij  v1 | X i , f j  + J ij  v1 | f j , X i  .
(0) (0) (0) (0)
( LX ) i    
(58)
j =1
84
The right sides of equation (56), im , are explicit functions determined from { f ( ) (t )} and
i
0
(
ij r1, v1; r2 , v 2 | f k { (0)
(t )}) for the Enskog form equation (40). Hence they are known from the lowest
 1 (0) 1 (0) 1 (0) 1 (0) 
⇔ 0, ζ , − ζ , − ζ , − ζ  are the smallest eigenvalues of L From
(m)
order solution. Finally, λ
 2 2 2 2 
the corresponding eigenvectors and their properties, it is possible to show that the integral equations
(56) have solutions and that they are unique (Garzó & Dufty 1999; Dufty, Garzo, Hrenya 2007).
In summary, the exact normal solution to the generalized Enskog equation has been constructed
up through terms to first order in the gradients of the hydrodynamic fields. In particular, the first two
terms equation (54) and equation (55) together with the equations that determine them, equation (45)
and equation (56) place no limitations on the degree of inelasticity or density, beyond those implied by
the kinetic equation itself.
The results of this subsection require  mfp | ∇r yα | / yα << 1. For finite geometries this implicitly
requires  mfp << D, where D is any relevant system dimension. For domains where this Knudsen
number is not small this small gradient solution fails and the constitutive equations of the next section
do not apply. In some cases this can be corrected by modified boundary conditions (slip conditions).
Otherwise, the solution does not apply and a different solution to the kinetic equation must be con-
structed.
constitutive equations
The constitutive equations required for hydrodynamics are obtained from equations (27)-(30), special-
ized to the case of smooth, inelastic hard spheres. Substitution of the Chapman-Enskog solution equation
(54) and equation (55) into these expressions gives the constitutive equations to first order in the gradi-
ents. There are two types of contributions, arising from the first terms on the right sides evaluated with
fi
(0) (1)
and f i , and those obtained from evaluating the two particle functional ij r1, v1; r2 , v 2 | f k (t ) ( { })
to first order in the potential. The latter are called “collisional transfer” contributions. They vanish at
low densities, but dominate at high densities. Included in the collisional transfer contributions are first
(0) (0)
order gradient terms from products f i f j for a colliding pair depending on the fields due to the dif-
ferent positions of the colliding pair at r and r ± σij. Also, there are collisional transfer contributions
(
from the functional dependence of χ ij r1, r1 ± σij | {nk } on the species densities. )
There are many simplifications due to the assumption of fluid symmetry. For example, to zeroth order
in the gradients the mass and energy fluxes vanish, while the cooling rate and pressure tensor become
π mi m j
(0)
∑ (1 − α ) m (0) (0)
(V ) f ( ) (V ) | V − V
0
ζ = 2
ij
χij σij2 ∫ dv1 ∫ dv 2 f i j
|3 , (59)
12nT i, j i
+ mj 1 2 1 2
 2 mj 
Pγβ = pδ γβ =  nT + π∑ (1 + α ij )
(0) (0)
χij σij3 n j n jTi  δ γβ . (60)
 9 i, j mi + m j 
85
Recall that the { f ( )} are local HCS solutions so that the fields in these equations are the physical
i
0
values {T(r, t), {ni(r, t)}} to be determined from the resulting hydrodynamic equations.
The complete set of constitutive equations to first order in the gradients is
(0)
ζ→ζ + ζU ∇ ⋅ U, (61)
s nj
j0i → −∑ mi m j Dij ∇ ln n j − ρDiT ∇ ln T , (62)
j =1 ρ
s
q → −λ∇T − ∑ T 2 Dq,ij ∇ ln n j , (63)
i , j =1
 2 
Pβµ → pδ βµ − η ∂ βU µ + ∂ µU β − δ βµ ∇ ⋅ U − κδ βµ ∇ ⋅ U. (64)
 3 
These expressions are characterized by transport coefficients that are functions of the temperature
and the species densities, determined from integrals over the solutions to the integral equations (56).
There are (s − 1)(2s + 1) transport coefficients for the mass flux, Dij and DiT , the shear and bulk viscos-
ity for the pressure tensor, η and K, and 2s2 + 1 coefficients for the heat flux, Dq,ij and λ. There remain
only the practical issues of solving the equations (45) and (56), which are discussed below.
approximate evaluation
The above constitutive equations, and definitions of the transport coefficients, are direct consequences
of the generalized Enskog equation. Any limitations must be attributed to approximations leading to that
kinetic equation since the Chapman-Enskog method gives an exact construction (to each order in the
perturbation expansion considered). Applications of these results require the explicit solutions to equa-
(0)
tion (45) for the HCS distribution, f i , and to equations (56) for the functions A(m) to determine the
transport coefficients. In the following approximation methods for each are described briefly.
Consider first the solution to equation (45) by representing it as an expansion in a complete set of as-
sociated Laguerre polynomials (Sonine polynomials, up to a normalization constant) (Ferziger & Kaper
1972). First write equation (54) as
v−U
fi
(0)
(
= ni v0−3 (T ) ϕ i c; {nk } ; c = ) v0 (T )
, (65)
 ∞ 
( ) ( )
ϕ i c; {nk } = ϕ Mi c; {nk } 1 + ∑ bn L1n/ 2 c 2  .
  ( ) (66)
 n=2 
( )
Here ϕ Mi c; {nk } is a Maxwellian chosen to have the exact second moment of ϕ i c; {nk } . Hence ( )
it is parameterized by the kinetic temperature Ti which must be determined self-consistently from the
86
condition ζ = ζi (see discussion of the HCS above). An approximate solution is obtained by truncating
the expansion. The coefficients are determined by substitution of the truncated expansion into equation
(45) and taking moments of that equation with respect to all polynomials retained in the truncation.
There is a complication in this case due to the nonlinearity of equation (45) which implies that the trun-
cated series is not sufficient to determine the truncated set of coefficients {bn}. Consequently, there is
no unique way to determine their best estimate from a given truncation. This situation has been reviewed
recently in Santos &J. Montanero 2008. Fortunately, the simplest truncation at n = 2 and a linearization
of equation (45) for b2 gives a very good approximation (compared to a Monte Carlo solution to the
kinetic equations) except at very strong dissipation (Brey, Ruiz-Montero & Cubero 1996; van Noije &
Ernst 1998).
A similar method applies for the solution to the linear integral equations (56). The solutions are given
a series representation as above
∞ 
( ) n =1 ( )
im (c ) = ϕ Mi c; {ni }  ∑ anm Ln m c 2  .
1/ 2 + λ

(67)
 
The parameter λm takes the value 0, 1 or 2 depending on whether Aim is a scalar, vector, or second
order tensor. Again, truncation of the series defines a particular approximation. Substitution into equa-
tion (56) and taking the corresponding moments now determines the coefficients of that truncation
exactly, since the equations are linear. Again, leading order truncations give very good approximations
except at strong dissipation (Montanero, Santos & Garzó 2005). Recently, it has been shown that
higher order truncations extend the accuracy to strong dissipation (Garzo, Santos & Montanero 2007,
Garzo, Vega Reyes & Montanero 2008).
An instructive test of this approximation scheme and its results is given by a recent study of impurity
diffusion (Garzo & Vega Reyes 2009). The above Enskog analysis is specialized to a binary mixture
in which one of the species has very small concentration, denoted with a subscript 0. The constitutive
equation (62) becomes
m02 mm
j00 = − D0∇n0 − 0 D∇n − ρD T ∇ ln T . (68)
ρ ρ
Consider, for example, the diffusion coefficient D0. The corresponding expression from equation (27) is
ρ
D0 = −
3m0n0 ∫ dvV ⋅  ( V),
0
(69)
where A0 is the solution to an integral equation of the type equation (56)
(0)
LA0 = −Vf 0 . (70)
The first two terms in the expansion of A0 are
87
0 ( V ) = f M (V )  a1V + a2S (V ) + ... , (71)

 
1 5 
where S (V ) = V  m0V 2 − T0  . Then substitution of equation (71) into equation (70) and taking
2 2 
moments with respect to V and S(V) gives the two coupled equations for a1 and a2. Finally, with these
known D0 is determined from equation (69). When there is no strong mechanical dissimilarity of the
impurity from the host fluid, the first Sonine approximation gives excellent agreement with DSMC
simulations of the Enskog equation. However, the second Sonine approximation is required when there
is significant mechanical difference (see the last section below).
To summarize this section on Navier-Stokes hydrodynamics for the hard sphere fluid, the central
approximation has been the choice of the generalized Enskog kinetic theory (as opposed to the simpler
Boltzmann equation or more complex high density phenomenological equations). As for normal fluids,
quantitative predictions of this kinetic theory appear to be limited to low or moderately dense gases.
This depends on the property being calculated, with good results for pressure up to solid densities while
those for transport coefficients begin to fail for volume fractions ϕ = πnσ3 / 6 > 0.1 . The domain of
accuracy for granular fluids is further compromised as the degree of inelasticity increases, α < 0.8.
Nevertheless, the Enskog kinetic theory remains semi-quantitative over a much wider range of densities
and inelasticities. The derivation of the normal solution to this kinetic equation to obtain the Navier-
Stokes equations does not involve further approximation. Rather it requires verification of experimental
conditions for small hydrodynamic variations; the form and coefficients of the first order Chapman-
Enskog expansion are given exactly within the Enskog kinetic theory for all densities and degrees of
inelasticity. The practical evaluation of these coefficients may involve a further approximation, such as
truncation of a series representation. Therefore, comparison of experimental results to Navier-Stokes
hydrodynamics first entails assurance that the conditions for small gradients are met. Having done so
any observed discrepancies can be attributed to a failure to implement accurately the consequences of
the kinetic equation (evaluation of the integral equations) or to the kinetic equation itself.
cOMPlex fluid HydrOdynaMics
As emphasized in the comment following equations (27)-(30) the concept of a macroscopic, hydro-
dynamic description is more general than the Navier-Stokes equations. This is well-known for high
molecular weight fluids that exhibit a wealth of rheological effects not present at the Navier-Stokes
level. The surprising and interesting feature of granular fluids is that such effects occur commonly even
for structurally simple systems such as hard spheres. For such states, their hydrodynamic description
requires a more complete determination of the normal solution equation (24), beyond the leading order
(Navier-Stokes) Chapman-Enskog expansion of the last section. The resulting hydrodynamics will be
referred to as “complex fluid” hydrodynamics.
The simplest example is the Burnett hydrodynamics obtained by retaining the first two terms of the
Chapman-Enskog expansion (the results for weak inelasticity are given by Sela & Goldhirsch 1998, and
Kumaran 2004). Even at this level new physical effects occur, such as non-zero viscometric functions
and complex heat flow. However, in most cases for which the Navier-Stokes conditions fail it is not
88
sufficient to add small corrections (e.g., higher order Chapman-Enskog approximations). Instead, more
complex constitutive equations that are highly nonlinear in the gradients can be expected. Examples
are granular fluids under shear, with constitutive equations that are complex functions of the shear rate.
For simple molecular systems complex fluid behavior is rare since the system must be driven hard
externally to generate the necessary large gradients. Granular fluids are different because, in addition to
external driving forces (e.g., the coupling to a gas phase), there is the internal mechanism of collisional
energy loss. The latter is particularly effective in the formation of steady states that would not be pos-
sible for molecular fluids. Consider a fluid between two parallel plates, each at the same fixed tempera-
ture. A normal fluid has its steady state at a uniform temperature, while that for a granular fluid has a
temperature that is lower in the middle than at the edges due to collisional cooling. Thus the granular
fluid “generates” gradients on its own, independent of applied external forces. These inherent gradients
cannot necessarily be made small in order to assure Navier-Stokes hydrodynamics (Hrenya, Galvin &
Wildman 2008; Galvin, Hrenya & Wildman 2007).
A second example is the same configuration but with the wall in relative parallel motion to shear the
fluid. The shear does work creating viscous heating. For the molecular fluid this is compensated either
by heating the uniform fluid, or by generating a steady state with non-uniform temperature gradient.
However, the granular fluid can support a steady, uniform state with viscous heating compensated by
collisional cooling. In this latter case the system autonomously seeks a steady state temperature that is
determined by the imposed shear rate, a, and the degree of inelasticity, α. The dimensionless gradient
in this case is a* = a/v where v is a characteristic collision frequency that scales for hard spheres as
the square root of the steady temperature. Any attempt to make a* small enough to justify the Navier-
Stokes conditions by decreasing a has the effect of decreasing the steady state temperature in just such
a way that a* remains unchanged. The system again sets its own gradient, and in this case it has been
shown (Santos, Garzo & Dufty 2004) that the Navier-Stokes description of uniform shear is never justi-
fied and the fluid is inherently complex. Nevertheless, a complex fluid hydrodynamics may apply.The
Enskog kinetic theory has been applied to the derivation of constitutive equations in special cases, such
as the state of uniform shear flow and polydisperse conditions, with significant success (Lutsko 2004;
Montanero, Garzo, Alam & Luding 2006). Unfortunately, universal features of constitutive equations
for classes of complex fluid states have not been identified and may not exist.
In addition to these complexities due to the internal mechanism of collisional energy loss for granular
fluids, the Navier-Stokes equations may fail in gas-solids flows due to the local driving forces from
the gas phase. When these drive the formation of bubbles or dense plugs there can be large density and
thermal gradients for which the Navier-Stokes constitutive equations are not justified. In this case a
more general normal solution to the Enskog equation itself must be considered. This remains an open
area of current research.
discussiOn
In this chapter, the conceptual basis of granular hydrodynamics as arising from an underlying kinetic
theory has been reviewed. First, in the most general context, the definition of a kinetic theory was given.
Then, using the exact balance equations for the fields of interest and a normal solution to the kinetic
equation, the constitutive equations were obtained expressing the fluxes and the source in terms of the
fields, and hence a closed hydrodynamic description for the fields was obtained. Finally, the analysis
89
was specialized to the theoretically tractable setting of smooth inelastic hard spheres and the kinetic
theory approximated by the revised Enskog equation. In this setting, the constitutive equations were
obtained explicitly and exactly to first order in the spatial gradients of the fields for the case of a weakly
inhomogeneous system. The resulting hydrodynamic equations constitute the granular Navier-Stokes
hydrodynamics, with the pressure, cooling rate, and transport coefficients defined as averages over
solutions to linear integral equations. Practical application of these results based on the first Sonine
approximation appear in reference (Garzó & Dufty, 1999) for a one component fluid, and in (Garzo,
Dufty, & Hrenya, 2007) for multicomponent systems. These results apply for arbitrary dissipation and
are extensions of the earlier work of Jenkins (Jenkins, 1998), and of Jenkins and Mancini (Jenkins &
Mancini, 1989), respectively, which made the additional approximation of weak dissipation. The reader
is referred to the literature for the explicit expressions for the parameters of the Navier-Stokes equations.
As noted at the outset, one of the primary advantages of deriving the hydrodynamic equations system-
atically is that the approximations made at every step are clearly stated and the validity of the resulting
theory can be tested in unambiguous terms. In the remainder of this closing section examples of some
numerical tests are presented.
Once the specialization to smooth, inelastic hard spheres was made, the subsequent derivation here
included three approximations at various stages in its development. The first is the assumption that the
Enskog kinetic equation (42) is a reliable mesocopic description of a moderately dense system of inelas-
tic hard spheres. This can be tested by molecular dynamics (MD) simulation of Newton’s equations of
motion to compute properties of interest for comparison of their calculation from the Enskog equation.
The singular inelastic hard sphere dynamics is implemented using an event-driven algorithm in which
the particles traverse straight lines until a pair is at contact; the velocities of the pair are then changed
according to the collision rules equation (36). These same collision rules are used in the Enskog equa-
tion which can be solved numerically with no further approximations needed, using an extension of the
Direct Simulation Monte Carlo (DSMC) method developed for the Boltzmann equation and extended
to the Enksog equation (Montanero & Santos 1996). A comparison of the results of these two numeri-
cal studies provides a means to quantify the domain of validity for the Enskog equation. Of course, this
will depend somewhat on the property and state considered. Figure 1 shows the comparison from (Dahl,
Hrenya, Garzo, Dufty 2002) of the temperature ratio for a binary mixture in its HCS as a function of mass
ratio (left panel) and size ratio (right panel), from MD simulation (symbols) and a calculation based on
the Enskog equation (lines). Two values for the restitution coefficient are shown, α = 0.95 and α = 0.8,
and two values of volume fractions, φ = 0.1 and φ = 0.2, where φ = πnσ3/6. It is seen that the Enskog
theory is accurate for both values of α for φ = 0.1 over a wide range of mass and size ratios. The results
are less accurate for φ = 0.2, but still quite good for the α = 0.95 case. Figure 2 shows a comparison of
the temperature ratio and the shear viscosity η as a function of α for a granular fluid undergoing uniform
shear flow with a mass ratio of 4 and φ = 0.1 (Montanero, Garzo, Alam, Luding 2006). There is excel-
lent agreement for the shear viscosity for the range α ≥ 0.7 considered; this degree of accuracy extends
to the overall temperature and pressure as well (not shown). However, the accuracy of Enskog for the
temperature ratio decreases as the degree of dissipation increases.
Next, consider the second approximation made in the derivation presented here, namely the construc-
tion of explicit solutions to equations (45) and (56) using a truncated Sonine polynomial expansion. A
chosen truncation can be tested by comparing properties calculated using the analytic results to those
obtained from DSMC simulations. First, consider the Sonine expansion as applied to the HCS distribu-
tion fh for a one component fluid. In this case the HCS solutions to the dimensionless Enskog and
90
Figure 1. Plot of the temperature ratio T1/T2 as a function of a) Top Panel: the mass ratio m1/m2 for σ1/
σ2 = 1 and b) Bottom Panel: the size ratio for m1/m2 = 1. φ1 = φ2 ≡ φ in both panels. The symbols are
MD simulations and the lines are Enskog predictions. Two different values of α, α = 0.95 (circles) and
α = 0.8 (triangles) and two different values of φ, φ = 0.1 (open symbols, solid line) and φ = 0.2 (solid
symbols, dashed line) are shown (reproduced from Dahl, Hrenya, Garzo, Dufty 2002 ©2002 American
Physical Society. Used with permission.)
Figure 2. Shear viscosity η, and temperature ratio T1/T2 versus the coefficient of α with φ1/φ2 = σ1/σ2 = 1,
m1/m2 = 4, and φ = 0.1. The solid circles correspond to the DSMC results and open triangles correspond
to MD simulations (adapted from Montanero, Garzo, Alam, Luding 2006)
Boltzmann equations are the same. DSMC studies were first performed in reference (Brey, Ruiz-Mon-
tero & Cubero 1996) and extended in (Brilliantov & Pöschel 2006). It is found that the leading Sonine
approximation is quite accurate for α > 0.6 except for very large velocities. For α < 0.6 and for large
velocities at any α < 1 this approximation fails. Next, consider the approximate solution to the integral
equations that determine the various transport coefficients. Figure 3 shows the shear viscosity of a one
component granular fluid as a function of α and as a function of φ from DSMC (symbols) and from the
first Sonine approximation (lines) (Montanero, Santos, & Garzó 2005). The agreement is quite good
over the entire parameter space of density and dissipation. Conditions under which this leading ap-
proximation fails is illustrated in Figure 4 for the case of impurity diffusion described above (Garzo
91
Figure 3. Shear viscosity η of a one component granular fluid a) as a function of α for different values
of φ and b) as a function of φ for different values of α. The symbols correspond to the DSMC results and
the lines correspond to the first Sonine approximation (reproduced from Montanero, Santos and Garzo
2005 ©2005 American Institute of Physics. Used with permission.)
Figure 4. Plot of the reduced diffusion coefficient D0(α)/D0(1) as a function of the coefficient of restitu-
tion for φ = 0.2, with m0/m = 2, σ0/σ = 2 in the top graph and m0/m = 1/5, σ0/σ = 2 in the bottom graph.
The solid lines correspond to the second Sonine approximation and the dashed lines refer to the first
Sonine approximation. The symbols are the results obtained from DSMC (Garzo & Vega Reyes 2009
©2009 American Physical Society. Used with permission)
&Vega Reyes 2009). The impurity diffusion coefficient D0 is shown as a function of α at φ = 0.2 as
determined from DSMC (symbols), from the first Sonine approximation (dashed line), and from the
second Sonine approximation (solid line). The top panel is for an impurity that is heavier and larger than
the host fluid particles, while the bottom panel is for a lighter and smaller impurity. There is excellent
agreement between all three calculations in the former case, while the first Sonine approximation fails
badly in the second case. Thus, mechanical differences among constituent particles can require higher
order approximations.
The third approximation is the restriction to inhomogeneous states with weak spatial gradients of the
hydrodynamic fields, in order to justify truncation of the Chapman-Enskog expansion. For molecular
fluids this is a matter of controlling the experimental conditions responsible for the inhomogeneity (suf-
92
Figure 5. The top figure shows the reduced shear rate a* as a function of α0 as determined from the
DSMC (filled symbols), and MD (open symbols) for three values of the reduced density n* = nσ3. Also
shown is a generalized moment solution to the Enskog equation, GME (lines). The lower three figures
show the reduced shear viscosity η* and the two viscometric functions ψ1 and ψ2 for each of the densi-
ties. Symbols are from MD and lines are from GME (reproduced from Lutsko, 2004 ©2004 American
Physical Society. Used with permission.)
ficiently weak initial perturbations, weak boundary forces). It has been demonstrated that this can be
done for granular systems as well in many cases, justifying a Navier-Stokes descriptions. For example,
it has been found that experimental hydrodynamic profiles of a supersonic granular flow past a wedge
are described well by the Enskog Navier-Stokes hydrodynamic equations (Rericha, Bizon, Shattuck,
Swinney, 2002). Similarly, the hydrodynamic profiles of a three dimensional vibrated granular system
are well described by these same hydrodynamic equations, when sufficiently far away from the vibrat-
ing wall (Huan et al., 2004). These and other experimental studies suggest that there is a wide variety
of experimental conditions that can be reliably captured by a hydrodynamic description using the Na-
vier-Stokes constitutive equations.
As noted in the previous section, granular fluids also support states for which the gradients cannot be
controlled externally and some higher order, more complex hydrodynamics is required. Recent examples
have been discussed in reference (Galvin, Hrenya, Wildman 2007; Hrenya, Galvin, Wildman 2008) for
both simulation and experiment. This raises the question of whether the Enskog kinetic equation also
provides an adequate description for more complex states with large gradients and extreme conditions.
Figure 5 illustrates results for a polydisperse (continuous distribution of size, mass, and restitution
coefficient) granular fluid under shear (Lutsko 2004). The reduced shear rate a*, which scales as the
inverse root steady state temperature, is shown as a function of an average restitution coefficient α0
for three densities n* = nσ3. As described in the previous section, for a given shear rate and restitution
coefficient the system seeks a steady state by balancing viscous heating and collisional cooling. This
figure shows that the DSMC simulation of the Enskog equation gives excellent agreement with MD
simulations. Also shown are the results of an approximate solution to the Enskog equation based on
moments of the distribution. This is not hydrodynamics in the sense described here, but closely related
to it. The rheological properties of the pressure tensor are also well-described by the Enskog equation
for this complex granular fluid under conditions far from the Navier-Stokes weak gradient restrictions.
93
In summary, there are qualitative conditions of dissipation 0.8 ≤ α ≤ 1 and packing fractions 0 < φ <
0.2 for which the Navier-Stokes hydrodynamic description obtained from the Enskog kinetic theory,
with coefficients evaluated in a low order Sonine approximation, can be expected to be reliable, if the
experimental conditions verify the restriction to weak gradients. When this Navier-Stokes description
does fail, it could be attributed to the breakdown of any of the three approximations above, i.e., (1) The
system is in a state such that the assumption of weak gradients does not hold and hence the constitutive
equations are more complex (flow down an inclined plane belongs in this category); (2) The coefficient
of restitution could lie in a region where the polynomial approximation breaks down higher order terms
in the expansions are required; or (3) the Enskog kinetic theory itself fails due to velocity correlations
that have a quantitative importance for the properties of interest. Although the above examples give
some indication, a more complete identification of the domain of validity for the Enskog description
and its Navier-Stokes predictions with approximate evaluations of the associated coefficients is an area
of ongoing investigation in the granular physics community via theory, simulation, and experiment.
Such studies also present a fertile set of questions that need to be answered to move granular hydrody-
namics to its full potential beyond both Navier-Stokes and/or the Enskog kinetic theory.
The derivation of Navier-Stokes hydrodynamics described here requires an approximation for the
kinetic theory at an early stage. There is an alternative route to this same hydrodynamics, resulting in
formally exact expressions for the constants in these equations. This is an approach based in statistical
mechanics and leads, for example, to Green-Kubo time correlation function expressions for the transport
coefficients. One method to evaluate these coefficients is by approximate kinetic theories. The advantage
of this approach is that such approximations are postponed to the last stage. These well-developed linear
response methods for molecular fluids have been extended recently to granular fluids (Baskaran, Dufty
& Brey 2008; Dufty 2009; Dufty, Baskaran, & Brey 2008), and it has been shown that an evaluation of
the Green-Kubo expressions via an approximate kinetic theory neglecting velocity correlations leads to
the same results as those obtained here (Baskaran, Dufty & Brey 2007). These two approaches provide
complementary starting points for exploration of effects beyond the Enskog approximation that are
required at higher densities.
aKnOwledgMent
The authors are grateful to V. Garzo, C. Hrenya, J. Lutsko, and A. Santos for providing their figures and
associated data. AB acknowledges support by the NSF through Grants DMR- 0705105 and DMR-0806511.
references
Alexander, F., & Garcia, A. (1997). Computers in Physics, 11, 588. doi:10.1063/1.168619
Alley, W., Alder, B., & Yip, S. (1983). The Neutron Scattering Function for Hard Spheres. Physical
Review A., 27, 3174. doi:10.1103/PhysRevA.27.3174
Anderson, T. B., & Jackson, R. (1967). A Fluid-mechanical Description of Fluidized Beds. I&EC Fun-
dam., 6, 527. doi:10.1021/i160024a007
94
Baskaran, A., Dufty, J., & Brey, J. (2007). Kinetic Theory of Response Functions for the Hard Sphere
Granular Fluid. Journal of Statistical Mechanics, 12, 12002. doi:10.1088/1742-5468/2007/12/P12002
Baskaran, A., Dufty, J., & Brey, J. (2008). Transport Coefficients for the Hard Sphere Granular Fluid.
Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 77, 031311. doi:10.1103/Phys-
RevE.77.031311
Benyahia, S., Syamlal, M., & O’Brien, T. (2006). Extension of Hill–Koch–Ladd drag correlation over
all ranges of Reynolds number and solids volume fraction. Powder Technology, 162, 166. doi:10.1016/j.
powtec.2005.12.014
Bird, G. (1994). Molecular Gas Dynamics and the Direct Simulation of Gas Flows. Oxford, UK: Clarendon.
Bogoliubov, N. N. (1962). Problems in a Dynamical Theory in Statistical Physics, (English translation
by E. Gora). In Uhlenbeck, G. E., & de Boer, J. (Eds.), Studies in Statistical Mechanics I. Amsterdam:
North Holland.
Boon, J.-P., & Yip, S. (1991). Molecular Hydrodynamics. New York: Dover.
Brey, J., Dufty, J., Kim, C.-S., & Santos, A. (1998). Hydrodynamics for Granular Flow at Low Density.
Physical Review E: Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics, 58, 4638.
doi:10.1103/PhysRevE.58.4638
Brey, J. J., Dufty, J. W., & Santos, A. (1997). Dissipative Dynamics for Hard Spheres. Journal of Sta-
tistical Physics, 87, 1051. doi:10.1007/BF02181270
Brey, J. J., Ruiz-Montero, M. J., & Cubero, D. (1996). Homogeneous Cooling State of a Low Density
Granular Flow. Physical Review E: Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary
Topics, 54, 3664. doi:10.1103/PhysRevE.54.3664
Brilliantov, N., & Pöschel, T. (2004). Kinetic Theory of Granular Gases. New York: Oxford University
Press. doi:10.1093/acprof:oso/9780198530381.001.0001
Brilliantov, N., & Pöschel, T. (2006). Breakdown of the Sonine Expansion for the Velocity Distribution
of Granular Gases. Europhysics Letters, 74, 424. doi:10.1209/epl/i2005-10555-6
Campbell, C. S. (1990). Rapid Granular Flows. Annual Review of Fluid Mechanics, 22, 57. doi:10.1146/
annurev.fl.22.010190.000421
Cohen, E.G.D. (1962). On the Generalization of the Boltzmann Equation to General Order in the Density.
Physica 28, 1025. ibid. 28, 1045. ibid 28 1060.
Dahl, S., Hrenya, C., Garzo, V., & Dufty, J. (2002). Kinetic Temperatures for a Granular Mixture. Physical
Review E: Statistical, Nonlinear, and Soft Matter Physics, 66, 041301. doi:10.1103/PhysRevE.66.041301
Deen, N., van Sint Annaland, M., van der Hoef, M., & Kuipers, J. (2007). Review of discrete particle
modeling of fluidized beds. Chemical Engineering Science, 62, 28. doi:10.1016/j.ces.2006.08.014
Deltour, P., & Barrat, J. L. (1997). Quantitative Study of a Cooling Granular Medium. Journal de Phy-
sique. I, 7, 137. doi:10.1051/jp1:1997130
95
Dorfman, J., & Kirkpatrick, T. (1980). Kinetic Theory of Dense Gases not in Equilibrium. In Garrido,
L. (Ed.), Systems Far from Equilibrium, (Lecture Notes in Physics 132). Berlin: Springer.
Dufty, J. (2009). Granular Fluids. In Meyers, R. (Ed.), Encyclopedia of Complexity and Systems Science.
Heidelberg, Germany: Springer.
Dufty, J., & Baskaran, A. (2005). Hard Sphere Dynamics for Normal and Granular Fluids. In Gottesman,
S. (Ed.), Nonlinear Dynamics in Astronomy and Physics. Annals of the New York Academy of Sciences
1045, 93.
Dufty, J., Baskaran, A., & Brey, J. (2008). Linear Response and Hydrodynamics for a Granular Fluid.
RevE.77.031310
Dufty, J., & Brey, J. (2005). Origins of Hydrodynamics for a Granular Gas. In Pareschi, L., Russo, G., &
Toscani, G. (Eds.), Modelling and Numerics of Kinetic Dissipative Systems. New York: Nova Science.
Dufty, J. W. (2001). Kinetic Theory and Hydrodynamics for a Low Density Granular Gas. Advances in
Complex Systems, 4, 397. doi:10.1142/S0219525901000395
Dufty, J. W. (2002). Shear Stress Correlations in Hard and Soft Sphere Fluids. Molecular Physics, 100,
2331. doi:10.1080/00268970110109934
Dufty, J. W., & Ernst, M. H. (2004). Exact Short Time Dynamics for Steeply Repulsive Potentials.
Molecular Physics, 102, 2123. doi:10.1080/00268970412331292858
Duran, J., Rajchenach, J., & Clement, E. (1993). Arching effect model for particle size segregation.
Physical Review Letters, 70, 2431. doi:10.1103/PhysRevLett.70.2431
Ernst, M. H. (2000). Kinetic Theory of Granular Fluids: Hard and Soft Spheres. In Karkheck, J. (Ed.),
Dynamics: Models and Kinetic Methods for Non-Equilibrium Many Body Systems. NATO ASI Series
371. Dordrecht, Germany: Kluwer.
Feng, Y., & Yu, A. (2004). Assessment of model formulations in the discrete particle simulation of gas-
solids flow. Industrial & Engineering Chemistry Research, 43, 8378. doi:10.1021/ie049387v
Ferziger, J., & Kaper, H. (1972). Mathematical Theory of Transport Processes in Gases. Amsterdam:
North-Holland.
Galvin, J., Hrenya, C., & Wildman, R. (2007). On the Role of the Knudsen Layer in Rapid Granular
Flows. Journal of Fluid Mechanics, 585, 73. doi:10.1017/S0022112007006489
Garzo, V. (2008). Brazil-nut effect versus reverse Brazil-nut effect in a moderately dense granular fluid
[R]. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 78, 020301. doi:10.1103/Phys-
RevE.78.020301
Garzó, V., & Dufty, J. (1999). Homogeneous Cooling State for a Granular Mixture. Physical Review
E: Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics, 60, 5706. doi:10.1103/
PhysRevE.60.5706
96
Garzó, V., Dufty, J., & Hrenya, C. (2007). Enskog Theory for Polydisperse Granular Fluids. I. Navier
Stokes Order Transport. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 76, 031303.
doi:10.1103/PhysRevE.76.031303
Garzó, V., & Dufty, J. W. (1999). Dense Fluid Transport for Inelastic Hard Spheres. Physical Review
E: Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics, 59, 5895. doi:10.1103/
PhysRevE.59.5895
Garzó, V., Hrenya, C., & Dufty, J. (2007). Enskog Theory for Polydisperse Granular Fluids. II. Sonine
Polynomial Approximation. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 76,
031304. doi:10.1103/PhysRevE.76.031304
Garzó, V., Santos, A., & Montanero, J. (2007). Modified Sonine Approximation for the Navier-Stokes
Transport Coefficients of a Granular Gas. Physica A, 376, 94. doi:10.1016/j.physa.2006.10.081
Garzó, V., & Vega Reyes, F. (2009). Mass Transport of Impurities in a Moderately Dense Granular
Gas. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 79, 041303. doi:10.1103/
PhysRevE.79.041303
Garzó, V., Vega Reyes, F., & Montanero, J. (2008). Evaluation of the Navier-Stokes Transport Coefficients
of a Granular Binary Mixture from a Modified Sonine Approximation. arXiv:0806.1858.
Gidaspow, D. (1994). Multiphase Flow and Fluidization: Continuum and Kinetic Theory Description.
San Diego, CA: Academic press.
Gidaspow, D., Jung, J., & Singh, R. (2004). Hydrodynamics of fluidization using kinetic theory: an
emerging paradigm. Powder Technology, 148, 123. doi:10.1016/j.powtec.2004.09.025
Goldhirsch, I. (2003). Rapid Granular Flows. Annual Review of Fluid Mechanics, 35, 267. doi:10.1146/
annurev.fluid.35.101101.161114
Goldhirsch, I., Tan, M. L., & Zanetti, G. (1993). A Molecular Dynamics Study of Granular Fluids 1: The
Unforced Granular Gas in 2 Dimensions. Journal of Scientific Computing, 8, 1. doi:10.1007/BF01060830
Grad, H. (1960). Theory of Rarified Gases. In Rarified Gas Dynamics. Oxford, UK: Pergamon Press.
Haff, P. K. (1983). Grain Flow as a Fluid Mechanical Phenomenon. Journal of Fluid Mechanics, 134,
401. doi:10.1017/S0022112083003419
Hill, R., Koch, D., & Ladd, J. (2001). Moderate-Reynolds-number flows in ordered and random arrays
of spheres. Journal of Fluid Mechanics, 448, 243. doi:10.1017/S0022112001005936
Hoomans, B., Kuipers, J., Briels, J., & van Swaaij, W. (1996). Discrete particle simulation of bubble and
slug formation in a two-dimensional gas-fluidized bed: a hard sphere approach. Chemical Engineering
Science, 51, 99. doi:10.1016/0009-2509(95)00271-5
Hrenya, C., Galvin, J., & Wildman, R. (2008). Evidence of Higher Order Effects in Thermally Driven
Rapid Granular Flows. Journal of Fluid Mechanics, 598, 429. doi:10.1017/S0022112007000079
97
Huan, C., Yang, X., Candela, D., Mair, R., & Walsworth, R. (2004). NMR Measurements on a Three-
Dimensional Vibrofluidized Granular Medium. Physical Review E: Statistical, Nonlinear, and Soft Matter
Physics, 69, 041302. doi:10.1103/PhysRevE.69.041302
Jackson, R. (2000). The Dynamics of Fluidized Particles. Cambridge, UK: Cambridge University Press.
Jenkins, J. (1998). Physics of Dry Granular Media (Hermann, H. J., Hovi, J. P., & Luding, S., Eds.).
Dordrecht, Germany: Kluwer.
Jenkins, J., & Mancini, F. (1989). Kinetic Theory for Binary Mixtures of Smooth, Nearly Elastic Spheres.
Physics of Fluids, A 1, 2050.
Kirkpatrick, T., Das, S. P., Ernst, M. H., & Piasecki, J. (1990). Kinetic Theory of Transport in a Hard
Sphere Crystal. The Journal of Chemical Physics, 92, 3768. doi:10.1063/1.457835
Koch, D. (1990). Kinetic theory for a monodisperse gas-solids suspension. Phys. Fluids A, 2, 1711.
doi:10.1063/1.857698
Kuipers, J., Hoomans, B., & van Swaaij, W. (1998), Hydrodynamic modeling of gas-fluidized beds and
their role for design and operation of fluidized bed chemical reactors. In Proceedings of the Fluidization
IX conference, (p.15), Durango, USA.
Kumaran, V. (2004). Constitutive Relations and Linear Stability of a Sheared Granular Flow. Journal
of Fluid Mechanics, 506, 1. doi:10.1017/S0022112003007602
Leboreiro, J., Joseph, G. G., & Hrenya, C. M. (2008). Revisiting the standard drag law for bubbling,
gas-fluidized beds. Powder Technology, 183, 385–400. doi:10.1016/j.powtec.2008.01.008
López de Haro, M., Cohen, E. G. D., & Kincaid, J. M. (1983). The Enskog Theory for Multicomponent
Mixtures I. Linear Transport Theory. Journal of Chemical Physics, 78, 2746. Koopman, B. (1969).
Relaxed Motion in Irreversible Molecular Statistics. Stochastic Processes in Chemical Physics, 15, 37.
Lun, C. K. (1991). Kinetic Theory for Granular Flow of Dense, Slightly Inelastic, Slightly Rough Spheres.
Journal of Fluid Mechanics, 233, 539. doi:10.1017/S0022112091000599
Lutsko, J. (1997). Approximate Solution of the Enskog Equation Far from Equilibrium. Physical Review
Letters, 78, 243. doi:10.1103/PhysRevLett.78.243
Lutsko, J. (2001). Model for the Atomic Scale Structure of the Homogeneous Cooling State of Granular
Fluids. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 63, 061211. doi:10.1103/
PhysRevE.63.061211
Lutsko, J. (2002). Atomic-Scale Structure of Hard-Core Fluids Under Shear Flow. Physical Review E:
Statistical, Nonlinear, and Soft Matter Physics, 66, 051109. doi:10.1103/PhysRevE.66.051109
Lutsko, J. (2004). Kinetic Theory and Hydrodynamics of Dense, Reacting Fluids Far From Equilibrium.
The Journal of Chemical Physics, 120, 6325. doi:10.1063/1.1648012
Lutsko, J. (2004). Rheology of Dense Polydisperse Granular Fluids Under Shear. Physical Review E:
98
Lutsko, J. (2005). Transport Properties for Dense Dissipative Hard Spheres for Arbitrary Energy Loss
Models. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 72, 021306. doi:10.1103/
PhysRevE.72.021306
Lutsko, J. (2006). Chapman-Enskog expansion about nonequilibrium states with application to the
sheared granular fluid. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 73, 021302.
doi:10.1103/PhysRevE.73.021302
McLennan, J. A. (1989). Introduction to Nonequilibrium Statistical Mechanics. Upper Saddle River,
NJ: Prentice-Hall.
McNamara, S., & Young, W. (1996). Dynamics of a Freely Evolving, Two-dimensional Granular Me-
dium. Physical Review E: Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics,
53, 5089. doi:10.1103/PhysRevE.53.5089
Minier, J.-P., & Peirano, E. (2001). The PDF approach to turbulent polydispersed two-phase flows.
Physics Reports, 352, 1. doi:10.1016/S0370-1573(01)00011-4
Montanero, J., Garzó, V., Alam, M., & Luding, S. (2006). Rheology of Two and Three Dimensional
Granular Mixtures Under Uniform Shear Flow: Enskog Kinetic Theory Versus Molecular Dynamics
Simulation. Granular Matter, 8, 103. doi:10.1007/s10035-006-0001-7
Montanero, J., & Santos, A. (1996). Monte Carlo Simulation Method for the Enskog Equation. Physical
Review E: Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics, 54, 438. doi:10.1103/
PhysRevE.54.438
Montanero, J., Santos, A., & Garzó, V. (2005). DSMC Evaluation of the Navier-Stokes Shear Viscosity
of a Granular Fluid. In Capitelli, M. (Ed.), Rarefied Gas Dynamics 24, AIP Conference Proceedings
762, 797.
Pöschel, T., & Luding, S. (Eds.). (2001). Granular Gases. New York: Springer. doi:10.1007/3-540-44506-4
Pöschel, T., & Luding, S. (Eds.). (2003). Granular Gas Dynamics. New York: Springer.
Rericha, E., Bizon, C., Shattuck, M., & Swinney, H. (2002)... Physical Review Letters, 88, 014302.
doi:10.1103/PhysRevLett.88.014302
Résibois, P., & De Leener, M. (1977). Classical Kinetic Theory of Fluids. New York: Wiley Interscience.
Santos, A., Garzó, V., & Dufty, J. (2004). Inherent Rheology of a Granular Fluid in Uniform Shear
Flow. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 69, 061303. doi:10.1103/
PhysRevE.69.061303
Santos, A., & Montanero, J. (2008). The Second and Third Sonine Coefficients of a Freely Cooling
Granular Gas Revisited. arXiv:0812.3022.
Sela, N., & Goldhirsch, I. (1998). Hydrodynamic Equations for Rapid Flows of Smooth Inelastic Spheres,
to Burnett Order. Journal of Fluid Mechanics, 361, 41. doi:10.1017/S0022112098008660
Simonin, O. (1996). Combustion and Turbulence in Two-phase Flows: Continuum Modelling of Dis-
persed Two-phase Flows. In Lecture Series 1996-02. Belgium: Von Karman Institute for Fluid Dynamics.
99
Sinclair, J., & Jackson, R. (1989). Gas-Particle Flow in a Vertical Pipe with Particle-Particle Interac-
tions. AlChE J., 35, 1473.
Sjogren, L. (1980). Kinetic Theory of Current Fluctuations in Simple Classical Fluids. Physical Review
A 22, 2866. ibid 2883.
van Beijeren, H., & Ernst, M. H. (1973). The Modified Enskog Equation. Physica A 68, 437. Ibid, 70, 225.
van Beijeren, H., & Ernst, M. H. (1979). Kinetic Theory of Hard Spheres. Journal of Statistical Physics,
21, 125. doi:10.1007/BF01008695
van der Hoef, M., Beetstra, R., & Kuipers, J. (2005). Lattice Boltzmann simulations of low Reynolds
number flow past mono- and bidisperse arrays of spheres: results for the permeability and drag force.
Journal of Fluid Mechanics, 528, 233. doi:10.1017/S0022112004003295
van der Hoef, M., van Sint Annaland, M., & Kuipers, J. (2004). Computational Fuid dynamics for dense
gas-solids Fuidized beds: a multi-scale modeling strategy. Chemical Engineering Science, 59, 5157.
doi:10.1016/j.ces.2004.07.013
van Noije, T. P. C., & Ernst, M. H. (1998). Velocity Distributions in Homogeneous Granular Fluids: the
Free and the Heated Case. Granular Matter, 1, 57. doi:10.1007/s100350050009
van Noije, T. P. C., & Ernst, M. H. (2001). Kinetic Theory of Granular Gases. In Pöschel, T., & Luding,
S. (Eds.), Granular Gases. New York: Springer. doi:10.1007/3-540-44506-4_1
van Swaaij, W. (1990). Chemical reactors. In Davidson, J., & Clift, R. (Eds.), Fluidization. London:
Academic Press.
van Wachem, B. G. M., Schouten, J. C., van den Bleek, C. M., Krishna, R., & Sinclair, J. L. (2001).
Comparative analysis of CFD models of dense gas-solids systems. AIChE Journal. American Institute
Yin, X., & Sundaresan, S. (2009). Drag Law for Bidisperse Gas-solids Suspensions Containing Equally
Sized Spheres. Industrial & Engineering Chemistry Research, 48(1), 227. doi:10.1021/ie800171p
Yip, S. (1979). Renormalized Kinetic Theory of Dense Fluids. Annual Review of Physical Chemistry,
30, 547. doi:10.1146/annurev.pc.30.100179.002555
Zhu, H. P., Zhou, Z. Y., Yang, R. Y., & Yu, A. B. (2007). Discrete Particle Simulations of Particulate Sys-
tems: Theoretical Developments. Chemical Engineering Science, 62, 3378. doi:10.1016/j.ces.2006.12.089
100
nOMenclature
fi(r, v; t) number density of species i at r, v, t.

ni(r, t) number density of species i at r, t.
e(r, t) energy density
p(r, t) momentum density
T(r, t) granular temperature
U(r, t) flow velocity
Ci(r1, v1 | {fk(t)}) collision operator for species density i
w(r, t) energy density loss rate
ji(r, t) mass flux for species i
j0i(r, t) mass diffusion flux for species i
s(r, t) total energy flux
q(r, t) total heat flux
tγβ(r, t) total momentum flux tensor
Pβγ(r, t) pressure tensor
μij = mi/(mi + mj) reduced mass for the pair i,j
ζ({ni}, T) cooling rate
αij restitution coefficient for hard sphere collision of species i,j
mi, σi particle mass and diameter for species i
( )
ij r1, v1; r2 , v 2 | { f k } two particle functional for kinetic equation “closure”
C i
E
(r , v | { f }) Enskog collision operator
1 1 k
101
102
Chapter 3
Kinetic Theory for
Granular Materials:
Polydispersity
Christine M. Hrenya
University of Colorado, USA
abstract
Kinetic-theory-based models of rapid, polydisperse, solids flows are essential for the prediction of a wide
range of practical flows found in both nature and industry. In this work, existing models for granular
flows are critically compared by considering the techniques used for their derivation and the expected
implications of those techniques. The driving forces for species segregation, as predicted by kinetic
theory models, are then reviewed. Although the rigor associated with the development of such models has
improved considerably in the recent past, a systematic assessment of model validity and computational
efficiency is still needed. Finally, a rigorous extension of such models to gas-solids flows is discussed.
intrOductiOn
In nature and industry alike, flows involving solid particles are time and again polydisperse – i.e., the
particles differ in size and/or material density. This polydispersity may (i) be a property of the start-
ing material itself, (ii) arise from the need for two different materials in a processing step, or, (iii) be
specified in order to improve system performance. Industrial and natural examples of (i) include coal
feedstock to combustors and terrestrial and lunar soils, respectively. The production of titania is an ex-
ample of (ii), where one step of the synthesis involves both titianium ore and coke, which vary in both
particle size and material density from one another. Finally, as an example of (iii), the addition of fines
to a relatively monodisperse material has been shown to decrease attrition in high-speed conveying
lines (Knowlton, Carson, Klinzing, & Yang, 1994), increase conversion in high-velocity, fluidized-bed
DOI: 10.4018/978-1-61520-651-3.ch003
Kinetic Theory for Granular Materials
reactors (Pell & Jordan, 1988) and improve heat transfer efficiency in a circulating fluidized bed (CFB)
combustor (Lee, 1997).
Not surprisingly, the flow behavior of a polydisperse material is different from that of its mono-
disperse counterpart. For example, recall from the previous chapter that a continuum description of
rapidly-flowing, monodisperse solids flows is possible via a kinetic-theory analogy, which requires the
specification of constitutive quantities such as the solid-phase stress. Similar descriptions are possible
for polydisperse flows, though the solid-phase stress, as well as the other constitutive quantities, now
also depend on the characteristics of the polydisperse distribution – e.g., the diameter ratio of particles
in a binary size distribution, the density ratio of two unlike materials, the standard deviation of a con-
tinuous size distribution, etc.
Beyond the aforementioned expected changes to the constitutive quantities caused by polydispersity,
such flows also display a counter-intuitive phenomenon which has no monodisperse counterpart: spe-
cies segregation or de-mixing. For example, agitation of polydisperse solids via vibration, free-fall, or
flow down an incline leads to segregation among unlike particles. A famous example of this behavior
is the well-known Brazil nut problem, in which a can of mixed nuts, after shaking up and down, is
opened only to find an over-representation of the Brazil (large) nuts at the top of the can (Rosato, Prinze,
Standburg, & Swendsen, 1987). Although such segregation may be beneficial to operations targeting
separation, as found in the mining industry, it may prove detrimental if a well-mixed system is desired,
as is common in the pharmaceutical industry. More specifically, consider a tablet which is made from
two powder substances – the medication and the binder which holds the medication together. If these
two substances are not-well mixed prior to tablet formation, a patient may be over- or under-medicated.
A related example with less serious ramifications involves raisin bran cereals, in which the raisins are
poured on top of the box prior to loading on distribution trucks since the vibration during transport mixes
the raisins throughout the box.
Generally speaking, a predictive understanding of polydispersity and the related segregation phenom-
enon remains a challenging task, as has been highlighted in several recent review articles and perspectives
(Curtis & van Wachem, 2004; Muzzio, Shinbrot, & Glasser, 2002; Ottino & Khakhar, 2000; Sundaresan,
2001). In this chapter, the focus will be on the rapid flow of polydisperse solids, in which the contacts
between particles are approximated as instantaneous and binary in nature. (Such flows are equivalently
referred to as granular gases, collision-dominated flows, high-Stokes flows, or massive-particle flows
in the literature.) The scope here will be further limited to continuum descriptions of such flows, which
are accordingly based on an analogy with the kinetic theory of molecular gases. The foundation for such
models, namely the kinetic equation (Boltzmann or Enskog, as detailed below), can be solved using
analytical or numerical methods. In this chapter, the former, which are commonly referred to as “kinetic
theory models”, will be discussed. In a subsequent chapter, a numerical method known as DQMOM,
or discrete quadrature method of moments, will be covered. (For purposes of clarity, it is worthwhile
to note that “kinetic theory models” referred to in this manner is common in the engineering literature,
whereas the physics community uses the term “kinetic theory” to refer to the starting kinetic equation.)
In sum, the objective of this chapter is provide a critical review of existing kinetic-theory-based
models for polydisperse systems, as well as an outlook for future work. It is worthwhile to note that this
review is focused on differences in the derivation process itself, rather than a direct comparison of the
models with data. The latter, albeit a valuable next step, is beyond the scope of the current effort due to
the high degree of complexity associated with the various models (as even a tabulation of the governing
equations and constitutive relations for all of the existing models would more than double the length of
103
this review). Accordingly, the remainder of the chapter is structured as follows. The initial focus is on
polydisperse granular flows – i.e., flows in which the role of the interstitial fluid is negligible. First, a
detailed comparison of differences in the derivation process of existing kinetic-theory models is given,
along with a discussion of the implications of these differences on model applicability. Second, the physi-
cal mechanisms giving rise to species segregation, as described by kinetic-theory models, is covered.
Next, gas-solids flows are briefly considered, where issues related to the incorporation of fluid-phase
effects into kinetic-theory models are reviewed. Finally, an outlook toward future work is given.
Kinetic-tHeOry-based MOdels fOr POlydisPerse sOlids
Kinetic-theory models for polydisperse solids are analogous to those of monodisperse predecessors (as
an early example, see Lun, Savage, Jeffrey, & Chepurniy, 1984), with the addition of several new flow
variables due to the presence of s solid species (s = 2 for binary mixture, s = 3 for ternary mixture, etc.):
ni, which refers to the number density associated with ith solid phase; Ui, which refers to the average
velocity of the ith solid species; and Ti which refers to the granular temperature of the ith solid species. It
is worthwhile to note that for mixtures, the definition of the granular temperature is generally given as
Ti = 1/3 mi <Vi2>, where Vi = vi – U is the peculiar velocity, vi is the instantaneous velocity of solids
s s
species i, and U is the mass-averaged mixture velocity defined as U = ∑ ρi Ui / ∑ ρi , where ρi = mi
i =1 i =1
ni is the mass density of the ith solid species and mi is the mass of species i. Similarly, a number-averaged
S
mixture temperature is defined as T = ∑ niTi / n , where n is the total number density of all species.
i =1
Note that compared to its monodisperse counterpart, the granular temperatures associated with mixtures
are defined with the species mass included, and thus have dimensions of energy rather than the square
of velocity. It is also worthwhile to mention that not all authors use the definitions given above – for
example, the peculiar velocity is sometimes defined relative to the species velocity (Ui) instead of the
mixture velocity (U). Exceptions will be noted below as needed, since such differences should be ac-
counted for when comparing theories and/or data with different definitions of temperature.
In what follows, the polydisperse mixtures under consideration will be composed of inelastic,
frictionless, spherical particles. The addition of friction and/or non-spherical particles adds non-trivial
complication to derivations which are already characterized by a high degree of complexity, and thus
is outside of the current scope.
granular flows
Conservation Equations and Constitutive Relations
Although kinetic-theory models for mixtures were first proposed in the late 1980’s (Jenkins & Mancini,
1987, 1989), the development and testing of such models continues to be an active area of research
since the approximations used during the derivation process vary widely from model to model. Gener-
ally speaking, the common features of the derivation process mimic those of monodisperse systems: (i)
the starting point is a kinetic equation for the single-particle position and velocity distribution function
104
fi for each species i, (ii) macroscopic balances for quantities of interest (e.g., mass, momentum, and
granular energy) are derived via a manipulation of the kinetic equation, (iii) closures for the constitutive
quantities appearing in the macroscopic balances are obtained via an approximate solution for fi (e.g.,
Chapman Enskog expansion). Further details on this derivation process are given in (Garzó, Dufty, &
Hrenya, 2007). Despite these commonalities, a variety of different approximations are possible during
the derivation process, starting with the kinetic equation itself. In the following sections, these differ-
ences and their implications are reviewed for existing polydisperse models on a point-by-point basis.
Accordingly, this work builds on several previous reviews of the derivation process and comparisons
of polydisperse models (Arnarson & Jenkins, 2000; Benyahia, 2008; Garzó, Dufty et al., 2007; Garzó,
Hrenya, & Dufty, 2007; Jenkins, 1998).
Comparison of Model Treatments

Table 1 provides a comparison of existing polydisperse kinetic-theory models in terms of the differ-
ences in the derivation process and the corresponding range of applicability. Note that the detailed
forms of the conservation balances and constitutive relations for each theory are not given due to space
constraints; the interested reader is referred to the original works for these details. Also, due to their
practical relevance, only three-dimensional (spheres) models are considered here (Arnarson & Willits,
1998; Garzó & Dufty, 2002; Garzó, Dufty et al., 2007; Garzó, Hrenya et al., 2007; Huilin, Gidaspow, &
Manger, 2001; Iddir & Arastoopour, 2005; Jenkins & Mancini, 1987, 1989; V. Mathiesen, T. Solberg,
H. Arastoopour, & B. Hjertager, 1999a; Rahaman, Naser, & Witt, 2003; Serero, Goldhirsch, Noskowicz,
& Tan, 2006; Zamankhan, 1995), though a number of two-dimensional models (disks) have also been
proposed (Alam, Willits, Arnarson, & Luding, 2002; Garzó, Dufty et al., 2007; Garzó, Hrenya et al.,
2007; Jenkins & Mancini, 1987; Willits & Arnarson, 1999). Further details on the categories used for
comparison in Table 1 are given below. Note also that the table contains abbreviations for each of the
models being compared; these will be used extensively below.
Number of Species. To date, most efforts on polydisperse granular flows – theoretical, experimental,
and computational – have been focused on binary mixtures (s=2). The same is true for the kinetic-the-
ory models of JM87, JM89, H01, GD02, R03 and S06, whereas the models of Z95, M99, IA05, and
GHD07 are applicable to any number of s species.
Particle Concentration (Boltzmann vs. Enskog equation). As stated above, the first difference arising
in the polydisperse kinetic-theory models stems from the starting kinetic equation itself, which is given
below for species i:
∂f i 1 ∂f
+ v1 ⋅ ∇f i Fi, ext ⋅ i = Cij (1)
∂t mi ∂v1
where v1 is the instantaneous velocity of particle 1 (of species i), Fi,ext is an external force, and Cij is the
collisional operator:
 
s
 1 '
Cij = ∑ σ 2
∫∫ s ⋅ g12 ) 2 f ij − f ij  d σ̂dv 2
s ⋅ g12 )(ˆ
Θ(ˆ (2)
j =1
ij
 e 
ij
105
Table 1. Comparison of polydisperse, kinetic-theory models
106
Range of Applicability Additional Differences in Derivation
Radial Chapman
Number Particle Hydro Distribution SET Enskog Sonine
of Concent Restitution Mass Velocity dynamic Function at vs Solution Expansion Expansion
Reference Abbrev. Species ration1 Coefficient2 Disparities3 Distribution Variables4 Contact RET Method Order Order
Jenkins & Man- JM87 2 dilute eij ~ 1 no restric- Maxwellian ni, Ui, Ti Mansoori et SET CE NS 1st
cini (1987) & dense tion al. (1971)
(non-
equip.)
Jenkins & JM89 2 dilute eij ~ 1 mi/mj ~ 1 non-Maxwellian ni, U, T Mansoori et RET CE NS 1st
Mancini (1989); & dense (equiparti- al. (1971)
Arnarson & Wil- tion)
lits (1998)
Zamankhan Z95 s dilute eij ~ 1 mi/mj ~ 1 non-Maxwellian 13 mo- Carnahan & RET Grad’s
(1995) & dense ments Starling method
Mathiesen et al. M99 s dilute eij ~ 1 no restric- Maxwellian (unlike) ni, Ui, Ti,M Mathiesen et SET CE NS 1st
(1999) & dense tion non-Maxwellian al. (1999)
(like)
Huilin et al. H01 2 dilute eij ~ 1 no restric- Maxwellian ni, Ui, Ti Mansoori et RET CE NS 1st
(2001) & dense tion al. (1971)
Garzó & Dufty GD02 2 dilute no restric- no restric- non-Maxwellian ni, U, T CE NS 1st
(2002) tion tion
Rahaman et al. R03 2 dilute eij ~ 1 no restric- Maxwellian (unlike) ni, Ui, Ti Mathiesen et SET CE NS 1st
(2003) & dense tion non-Maxwellian al. (1999)
(like)
Iddir & IA05 s dilute eij ~ 1 no restric- Maxwellian (unlike) ni, Ui, Ti,IA Iddir & SET CE NS 1st
Arastoopour & dense tion non-Maxwellian Arastoopour
(2005) (like) (2005)
Serero et al. S06 2 dilute eij ~ 1 mi/mj ~ 1 non-Maxwellian ni, U, Ts CE NS 3rd
(2006)
Garzó, Hrenya & GHD07 s dilute no restric- no restric- non-Maxwellian ni, U, T Boublik RET CE NS 1st
Dufty (2007) & dense tion tion (1970)
where σij = (σi+σj)/2 and σi and σj are the diameters of species i and j, respectively; ŝ is a unit vector
pointing from the center of particle j to the center of particle i at contact; g12 = v1-v2 is the relative (in-
stantaneous) velocity between particles 1 and 2; Θ is the Heaviside function (which assures that the
relative velocities g12 are such that a collision takes place); eij is the restitution coefficient between spe-
cies i and j; fij is the two-particle distribution function. The left-hand-side of the kinetic equation (1)
accounts for changes in the distribution function due to transient, convective, and external-force effects,
while the right-hand-side accounts for changes due to collisions between particles 1 and 2, where par-
ticle 2 can be any species. It is the collisional operator Cij and more specifically fij that is defined differ-
ently for the two commonly used starting kinetic equations – the Boltzmann equation and the Enskog
equation. For the Boltzmann equation, the finite volume of the particles is ignored and the particles are
instead treated as points. Accordingly, the two-particle distribution function is assumed to be the prod-
uct of the two corresponding single-particle distribution functions:
f ij  f i (r1, v1, t ) f j (r2 , v 2 , t ) (3)
where r1 and r2 are position vectors. For the Enskog equation, the Enksog assumption is instead employed
to account for the finite volume of particles:
f ij  gij f i (r1, v1, t ) f j (r1 ± sij , v 2 , t ) (4)
where gij is the radial distribution function (or pair correlation function) at contact. Thus, two differences
arise due to the treatment of particles as finite volumes. First, gij is used to account for the spatial (not
velocity) correlation of particles due to volume exclusion effects (i.e., molecular chaos assumption).
gij increases from unity as the particle concentration increases from zero (dilute limit); this quantity is
further detailed in its own subsection below. Second, the location at which the single-particle velocity
distributions are evaluated in fij also differs. Specifically, note that fi and fj are evaluated at the same
location in Equation (3), but at different locations in Equation (4), namely the distance between the
centers of two touching particles (σij) is accounted for at the point of collisional contact. It is this treat-
ment of non-identical particle locations in fij that leads to a non-zero collisional transfer – or collisional
flux – of quantities that are conserved during collision (e.g., species mass, momentum). Accordingly, a
naïve treatment of setting gij = 1 in kinetic theories derived from the Enskog equation will not be equal
those derived from the Boltzmann equation since the former will also have collisional fluxes and the
latter will not.
From a practical standpoint, the choice of starting kinetic equation will impact the particle concentra-
tions over which the resulting macroscopic theory is applicable. In particular, because the Boltzmann
equation does not account for finite-size effects, it is restricted to the dilute limit. Theories derived from
the Enskog equation, on the other hand, are applicable to both dilute and moderately dense systems
(whereas systems near the packing limit should be avoided due to additional correlations arising from ring
collisions). Accordingly, GD02 and S06 are applicable to dilute systems only since they derive from the
Boltzmann equations, whereas all others listed in Table 1 are applicable to a wide range of concentrations.
Restitution Coefficient (eij=1 vs. HCS expansion state). During the derivation process, an approximate
solution for fi is needed to determine the constitutive quantities in terms of the macroscopic variables.
107
A common approach is the Chapman-Enskog expansion (further details below in section on solution
methods) as used for molecular gases (Chapman & Cowling, 1970; Ferziger & Kaper, 1972). This
expansion is a perturbative expansion about low Knudsen numbers (Kn) or “small gradients”, where
Kn is defined as the ratio of the mean free path to the characteristic length of the mean-flow gradients.
For molecular (elastic) systems, the zeroth order solution to this expansion is a Maxwellian velocity
distribution. For dissipative (inelastic) systems, the zeroth order solution is a homogeneous cooling state
(HCS), which is characterized by zero spatial gradients and a temperature which decays with time. The
majority of the polydisperse theories contained in Table 1, namely JM87, JM89, H01, R03, IA05, and
S06), are based on perturbations about small Kn and small (1-eij), in which the HCS and Maxwellian
distributions coincide; these theories are thus limited to nearly elastic systems (eij ~ 1). The theories of
GD02 and GHD07 are based on perturbations about small Kn only and so do not have such restrictions.
Mass Disparities (Equipartition vs. Non-equipartition). An assumption used in the derivation of
several polydisperse kinetic theories is an equipartition of energy, or T1 = T2 = T3, etc. This assumption
is true for perfectly elastic systems in equilibrium, although numerous simulation (Alam & Luding,
2003; Clelland & Hrenya, 2002; Dahl, Clelland, & Hrenya, 2002; Dahl, Hrenya, Garzó, & Dufty, 2002;
Liu, Metzger, & Glasser, 2007; Montanero & Garzó, 2002; Paolotti, Cattuto, Marconi, & Puglisi, 2003),
experimental (Feitosa & Menon, 2002; Schröter, Ulrich, Kreft, Swift, & Swinney, 2006; Wildman &
Parker, 2002), and theoretical studies (Barrat & Trizac, 2002; Garzó & Dufty, 1999; Montanero & Garzó,
2003a) have indicated that degree of non-equipartition increases with mass disparity and dissipation (i.e.,
lower eij). Accordingly, theories incorporating an equipartition-of-energy assumption (JM89, Z95, S06)
are restricted to mi/mj ~ 1 and eij ~1. Although this assumption results in a more straightforward evalua-
tion of collision integrals (terms involving the collisional operator of Equation (2)), the non-negligible
effect of non-equipartition has been documented in several systems (Brey, Ruiz-Montero, & Moreno,
2005; Galvin, Dahl, & Hrenya, 2005; Garzó, 2006; Liu et al., 2007; Yoon & Jenkins, 2006).
Single-particle velocity distribution (Maxwellian vs. Non-Maxwellian). Another key assumption used
in some polydisperse theories is that of a single-particle velocity distribution (fi) that is Maxwellian.
Similar to an equipartition of energy, this is only valid for perfectly elastic systems in equilibrium. For
perfectly elastic mixtures, Willits and Arnarson (1999) showed that a fully Maxwellian-based theory
significantly underpredicts the shear viscosity whereas a theory which accounts fully for non-Maxwellian
effects is in good agreement. As shown by Galvin, Dahl & Hrenya (2005), this trend persists and indeed
worsens as the systems become dissipative (eij < 1) and the mass disparities increase.
In addition to the polydisperse theories based fully on a Maxwellian (JM87 and H01) or non-Max-
wellian (JM89, Z95, GD02, S06, and GHD07) velocity distribution, several theories in the literature use
a combination of these treatments. In particular, M99, R03 and IA05 use a Maxwellian distribution to
evaluate collision integrals between unlike particles and a non-Maxwellian distribution to evaluate col-
lision integrals between like particles. A recent study by Benyahia (2008) indicates that models utilizing
this hybrid approach result in good predictions of the shear stress compared to molecular dynamics (MD)
simulations, though the impact of this approach on other constitutive relations has not been investigated.
Hydrodynamic Variables (Balance Equations). The choice of hydrodynamic variables, or equivalently
the choice of balance equations to include in the theory, is one which occurs early in the derivation pro-
cess. For monodisperse systems, this choice is straightforward: mass, momentum, and granular energy.
For polydispersity, the choice is not as clear since there is now a mass, momentum, and granular energy
associated with each species.
108
For the case of mixtures of molecular (perfectly elastic) gases, the system is described by species
mass balances, a mixture momentum balance, and a mixture energy balance. This choice traces to the
various time scales in the system: balances are required for variables which are dictated by the hydro-
dynamic (slow) time scales, whereas balances for variables dictated by kinetic (fast) time scales are
superfluous. For example, in the molecular system, each species is expected to have a rapid relaxation
after a few collisions to a temperature distribution around the local equilibrium (mixture temperature),
and this mixture temperature then evolves accordingly to the slow hydrodynamic mode. Consequently,
the species temperatures Ti are enslaved to the mixture temperature T, and thus T is the only relevant
hydrodynamic quantity. For the interested reader, a more detailed analysis of the time scales present in
both the molecular system and the dissipative mixture (summarized below) is given by Garzó, Hrenya
& Dufty (2007).
For mixtures of inelastic particles, as are considered here, the choice of appropriate hydrodynamic
variables is not as obvious. In particular, the added presence of a dissipation rate of granular energy
introduces another time scale, which is not known a priori to be of the kinetic (fast) or hydrodynamic
(slow) type. However, MD simulations by Dahl et al. (2002) confirmed an earlier assumption by Garzó
& Dufty (1999) that a separation of time scales between T and Ti does persist for inelastic systems (i.e.,
the time scale of Ti is much shorter than that of T), and thus the appropriate choice of hydrodynamic
variables mimics that of the molecular system, namely ni, U, and T. Nonetheless, because this separation
of time scales has only fairly recently been established, various sets of hydrodynamic variables have been
used in existing polydisperse kinetic theories, as evidenced by Table 1. Namely, the theories of JM87,
H01, R03, and IA05 utilize ni, Ui, and Ti (species balances for mass, momentum, and energy), whereas
JM89, GD02, S06, and GHD07 utilize ni, U, and T (species mass balances, a mixture momentum bal-
ance and a mixture energy balance).
To better illustrate key differences resulting from the choice of hydrodynamic variables, the balance
equations from two recent theories are given below. Specifically, Iddir and Arastoopour (2005) use ni,
Ui, and Ti as the hydrodynamic variables, whereas Garzó, Hrenya & Dufty (2007) use ni, U, and T. The
corresponding mass, momentum, and energy balances for the two theories are given as:
Iddir & Arastoopour (2005)
∂ni mi
+ ∇ ⋅ (ni mi Ui ) = 0 (5)
∂t
 ∂U  s
ni mi  i + Ui ⋅ ∇Ui  = −∇ ⋅ Pi + ni Fi, ext, cons + ∑ FDip (6)
 ∂t  p =1
3  ∂Ti, IA  s
ni  + Ui ⋅ ∇Ti, IA  = −∇ ⋅ q i + Pi : ∇Ui − γ i − ∑ Ui ⋅ FDip (7)
2  ∂t  p =1
Garzo, Hrenya & Dufty (2007)
∂ni mi
+ ∇ ⋅ (ni mi U) = −∇ ⋅ j0i (8)
∂t
109
 ∂U  s
ρ  + U ⋅ ∇U = −∇ ⋅ P + ∑ ni Fi, ext, cons (9)
 ∂t  i =1
3  ∂T  3 s 1 s
1
n  + U ⋅ ∇T  = −∇ ⋅ q − P : ∇U − γ + T ∑ ∇ ⋅ j0i + ∑ Fi, ext, cons ⋅ j0i (10)
2  ∂t  2 i =1 mi i =1 mi
where the subscript i refers to species i in all quantities, P is the stress tensor, Fi,ext,cons refers to conserva-
tive external forces (e.g., gravity), FDip refers to the collisional source of momentum (which includes a
“solids-solids drag force”), q is the flux of granular energy, γ refers to the collisional dissipation rate of
s s s
granular energy, and j0i is the diffusion mass flux of species i. Also, n = ∑ ni , ρ = ∑ ρi = ∑ mi ni .
i =1 i =1 i =1
In the IA05 theory, constitutive relations for Pi, FDip, qi, and γi are given in terms of the hydrodynamic
variables ni, Ui, and Ti; whereas in the GHD07 theory, closures for j0i, P, q, and γ are specified in terms
of ni, U, and T.
Comparing the above theories on a balance-by-balance basis highlights several differences. First,
a comparison of species mass balances, namely Equation (5) for IA05 and Equation (8) for GHD07,
reveals the presence of the divergence of the diffusion mass flux on the right-hand-side of the latter, but
not the former. This difference traces to the velocity used in the convective term, namely Ui in IA05
and U in GHD07. Since the diffusive mass flux is defined as j0i = nimi(Ui − U), these two forms of the
species mass balances are mathematically equivalent, as expected. (Note that this definition allows Ui to
be determined once the hydrodynamic variables, namely ni, U, and T, are solved for. Similarly, GHD07
includes algebraic equations that allow for the solution of Ti in terms of the hydrodynamic variables.)
Second, with regard to momentum balances, IA05 utilizes a species balance (Equation (6)) whereas
GHD07 utilizes a mixture balance (Equation (9)). Accordingly, all terms in the former refer to quantities
associated with species i, whereas all terms in the latter refer to mixture quantities. The only remaining
difference in the two balances is the presence of the collisional momentum source (particle-particle drag)
FDip occurring in IA05. It is important to mention that this term is zero for like particles (i=p) and nonzero
for unlike particles (i≠p) since overall momentum is conserved during a collision. In other words, when
two particles of a given species collide, no momentum change occurs overall for that species. On the
other hand, when particles of types i and j collide, a gain in the momentum of one species and a loss in
that of the other may occur, even though the net momentum change for both species combined is zero
(since overall momentum is conserved upon collision). With these differences in mind, it is clear to see
that the summation of all species momentum balances (over i) in Equation (6) is consistent with the
mixture momentum balance of Equation (9), since the summation of FDip over all i and p is zero since
overall momentum is conserved.
Lastly, a comparison of the granular energy balances is a bit less intuitive due to the different methods
used by IA05 and GHD07 to derive them. Although the left-hand-sides of Equations. (7) and (10) as
well as the first three terms on the right-hand-sides mimic the differences pointed out for the momentum
balances (species vs. mixture quantities), the remaining terms are a bit more curious. Note that the col-
lisional momentum source (FDip) appears in the IA05 energy balance, but not GHD07. Unlike GHD07,
the IA05 balance was derived by (Iddir, 2004): (i) obtaining the balance for the total kinetic energy via
appropriate manipulation of the starting kinetic equation, (ii) obtaining a balance for the mechanical
energy (that associated with the mean flow) via multiplication of the species momentum balance by
110
Ui, and (iii) subtracting (ii) from (i) to obtain the granular (fluctuating) energy balance. This approach
provides an explanation for the appearance of the collisional momentum source (FDip) in the granular
energy balance. (The interested reader is referred to appendix of Galvin (2007) for further discussion
and correction of some errors of the IA05 balances). On the other hand, the granular energy balance
of GHD07 is derived in a single step via the manipulation of the starting kinetic equation to obtain the
fluctuating energy balance directly. As a result, the final two terms on the right-hand-side of (10) take
a different form than the final term in the IA05 theory (Equation (7)).
Beyond the balance-to-balance differences noted above, perhaps the most important ramification of the
choice of hydrodynamic variables is the total number of balance equations that are required to describe a
given system. For theories which utilize ni, Ui, and Ti as hydrodynamic variables, 3s differential balances
are required: s species mass balances, s species momentum balances, and s species energy balances. For
theories which instead use ni, U, and T as hydrodynamic variables, s+2 differential balances are required:
s species mass balances, 1 mixture momentum balance, and 1 mixture energy balance. For a binary system
(s=2), this translates to 6 differential balances for theories using species balances only (ni, Ui, and Ti as
hydrodynamic variables), and 4 differential balances for theories using mixture balances (ni, U, and T
as hydrodynamic variables). For a ternary system, the corresponding numbers are 9 and 5, respectively.
This disparity continues to increase with the number of species s. Because the differential equations are
highly nonlinear and tightly coupled, an increase in the number of differential equations is expected to
have a non-negligible impact on the computational requirements. A rough gauge of this impact is avail-
able from a study conducted by van Wachem et al. (2001), who reported that the computational costs
associated with the solution of a fluidized bed with a binary mixture (using mixture balances) were an
order of magnitude greater than its monodisperse counterpart. Given that computational requirements
for the solution of transient, three-dimensional monodisperse flows are already high by the standards of
today’s computers, any significant savings that can be made in computational costs for polydisperse flows
would be welcome. Nonetheless, since a back-to-back comparison of the computational requirements
using species-based and mixture-based polydisperse theories has yet to be performed, a quantitative
statement on the resulting difference in computational costs is not currently available.
Radial Distribution Function at Contact (gij). As mentioned above, the radial distribution function
at contact takes into account the spatial correlation of particles with finite volumes. In the dilute limit,
no spatial correlations exists and gij = 1. As the particle concentration increases, a spatial correlation
develops due to volume exclusion effects, and gij increases accordingly. In monodisperse systems, gij
depends only on particle concentration, whereas in polydisperse systems it is also expected to depend
on species concentration and diameters.
Several different forms of the radial distribution function at contact have been proposed for systems
with particles of different sizes. Their explicit forms are given in Table 2, where ϕk refers to the solids
fraction of species k, ϕ refers to the total solids volume fraction of all species, and ϕmax refers to the
solids volume fraction at closest packing. One of the earliest contributions is from Lebowitz (1964),
who obtained an explicit expression for gij via an exact solution of Percus-Yevick equation. An empiri-
cally derived extension was later proposed by both Boublik (1970) and Mansoori, Carnahan, Starling
& Leland (1971). (Note from Table 1 that Z05 cite Carnahan & Starling for their gij, though they do
not provide citation details. Nonetheless, their expression is identical to that given by Boublik (1970)
and Mansoori et al. (1971).) Santos, Yuste & López de Haro (2002) put forth another expression using
a “universality” assumption.
111
Table 2. Expressions for radial distribution functions at contact for mixtures
Reference Radial Distribution Function at Contact

Lebowitz (1964)
Boublik (1970) and

Mansoori et al. (1971)
Santos et al. (2002)
Iddir & Arastoopour (2005)

[modified Lebowitz (1964)]
vanWachem et al. (2001)

[modified Mansoori et al. (1971)]
Mathiesen et al. (1999)
For each of the gij expressions cited in the previous paragraph, a finite value of gij is obtained as the
packing limit is approached. In order to attain an infinite value of gij at closest packing, several addi-
tional expressions have been proposed which also incorporate ϕmax, the volume fraction at closest pack-
ing. Iddir & Arastoopur (2005) proposed a modification of the Leibowitz (1964) expression, whereas
van Wachem et al. (2001) used Mansoori et al. (1971) as a basis for their modification. Finally, Mathie-
sen et al. (1999b) modify the monodisperse expression of Bagnold (1954), which unlike the other
polydisperse expressions for gij does not depend on σi or σj. It is worthwhile to note that the value of ϕmax
112
Figure 1. Comparison of various expressions in Table 2 for radial distribution functions at contact (gij)
in the monodisperse limit
used in each of these expressions is expected to depend on the composition of the mixture, though no
expression for this quantity was reported.
A rudimentary comparison between the gij expressions given in Table 2 is displayed in Figure 1.
In this figure, gij is plotted as a function of solids fraction in the monodisperse limit, and a ϕmax = 0.65
is assumed. For this case, the gij proposed by Boublik (1970)/Mansoori et al. (1971) and Santos et al.
(2002) are indistinguishable on the scale shown. As alluded to above, the first three expressions listed
in Table 1 give rise to finite values of gij at the packing limit, whereas the latter three approach infinity.
With regard to quantitative comparisons against simulation data, the Iddir & Arastoopour gij was shown
to match well with monodisperse data over a large range of solids fraction, namely ϕ ~ 0 – 0.65 (Iddir
& Arastoopour, 2005). Also, over a range of ϕ ~ 0 – 0.6, the gij expression proposed by Santos et al. was
shown to provide an excellent match with binary data over a range of diameter ratios and compositions
(Santos et al., 2002).
Finally, also related to gij, it is worthwhile to mention that several of the polydisperse kinetic theories
listed in Table 1 include a species chemical potential in their closures of the constitutive relations. This
quantity is derived directly from gij, so a change in gij requires a corresponding change in the chemical
potential. The interested reader is referred to Chapter 9 of Reed & Gubbins (1973) for further details on
the derivation of the chemical potential.
Standard Enskog Theory (SET) vs. Revised Enskog Theory (RET). For those theories in which gij
appears (those using the Enskog equation), two different treatments of this quantity are possible during
the derivation process. In SET, gij is treated as a function of solids concentration– i.e., it depends only on
the local value of concentration. In RET, gij is treated as a functional – i.e., it depends on the local value
and its gradient. SET has been shown to be inconsistent with irreversible thermodynamics, regardless of
what point is chosen to evaluate the local concentration (e.g., the midpoint), whereas RET is consistent.
For polydisperse systems, a choice of SET vs. RET will impact the closures governing segregation. For
example, in theories utilizing U as a hydrodynamic variable, the form of the diffusive mass flux (j0i)
will depend on whether SET or RET is followed. For theories utilizing Ui as a hydrodynamic variable,
Arnarson & Jenkins (Arnarson & Jenkins, 2000) showed that the choice of SET vs. RET impacts the
closures appearing in the diffusion velocity equation (this equation, and its relation to segregation, is
113
described in more detail below). Quantitative approximations of the differences resulting from SET and
RET, however, are unavailable. For those theories using the Enskog equation as their starting kinetic
equation, JM87, M99, R03, and IA05 use SET, whereas JM89, Z95, H01, and GHD07 use RET.
Solution Method. The Chapman Enskog (CE) perturbative expansion is the solution method followed
in all of the polydisperse models contained in Table 1, except Z95 in which Grad’s moment method
was utilized. As mentioned above, theories derived using a CE expansion are restricted to the small
parameter(s) around which they are perturbed, namely Kn for all theories listed (and 1-eij for a subset
of those theories: JM87, JM89, M99, H01, R03, IA05, and S06). Recall Kn is the ratio of the mean free
path to the characteristic length of the mean-flow gradients, and thus low-Kn (“small-gradient”) theories
are not strictly applicable to systems in which free-molecular effects (Knudsen layers, shocks, etc.) play
a role [see, for example, (Brey, Ruiz-Montero, & Moreno, 2001; Forterre & Pouliquen, 2001; Galvin,
Hrenya, & Wildman, 2007; Goldhirsch, Noskowicz, & Bar-Lev, 2004; Hrenya, Galvin, & Wildman,
2008; Khain, Meerson, & Sasorov, 2008; Martin, Huntley, & Wildman, 2006; Rericha, Bizon, Shattuck,
& Swinney, 2002; Wassgren, Cordova, Zenit, & Karion, 2003)]. Nonetheless, theories based on a first-
order expansion in spatial gradients (i.e., Navier Stokes order, as detailed below) often have a wider
range of applicability than is expected in principle, as discussed in the previous chapter on monodisperse
systems [see, for example (Xu, Louge, & Reeves, 2003)]. On the other hand, the Grad moment method
does not contain similar restrictions, though the more complex nature of the corresponding calculations
has not been performed without making other simplifying assumptions (e.g., equipartition of energy).
Chapman Enskog Expansion Order. The CE expansion order refers to how many terms are used in
the perturbation expansion. All of the theories listed in Table 1 that utilize the CE expansion are based
on expansions through the first order in gradients, or Navier-Stokes order. A second order expansion is
referred to as Burnett order, whereas a third-order expansion is known as super-Burnett. Higher-order
expansions such as Burnett or super Burnett have not been carried out for polydisperse flows and indeed
are limited for monodisperse flows due to the added complexity of the calculations. In particular, Sela
& Goldhirsch (1998) derived a Burnett-order theory for dilute, nearly elastic granular flows, the results
from which provide qualitative improvement to Navier-Stokes-order predictions when compared to MD
simulations (Hrenya et al., 2008). Nonetheless, it is worth mentioning that Burnett equations also require
higher-order boundary conditions that cannot be obtained solely from physical principles, and such condi-
tions remain an active area of research. Moreover, solutions at the Burnett level are inherently unstable.
Sonine Expansion Order. For those models derived using the, an analytical evaluation of the collision
integrals is achieved via a truncated expansion based on Sonine polynomials. For all the relevant models
listed in Table 1, only the leading term of the expansion (i.e., 1st order) is used, with the exception of S06,
in which a third-order expansion is utilized. Generally speaking, previous studies for both monodisperse
and polydisperse systems indicate that terms beyond the lowest-order truncation may provide noticeable
corrections for eij ∼<0.7, but are negligible for less dissipative systems (Brey & Ruiz-Montero, 2004;
Brey, Ruiz-Montero, Maynar, & Garcia de Soria, 2005; Garzó & Montanero, 2002, 2003, 2004; Garzó,
Reyes, & Montanero, submitted; Montanero & Garzó, 2003b; Montanero, Santos, & Garzó, 2007;
Noskowicz, Bar-Lev, Serero, & Goldhirsch, 2007; Serero et al., 2006).
Additional Comments on Models

Monodisperse Limit. For purposes of consistency, polydisperse kinetic-theory-based models are expected
to reduce to a monodisperse theory if the number of species is chosen equal to one (s=1) or if multiple
(s>1), identical (σ1=σ2, m1=m2, etc.) species are used and compared to a monodisperse system of same
114
total solids fraction (ϕ1+ϕ2+…=ϕ). As described by Benyahia (2008), the theories R03 and IA05 do not
satisfy this monodisperse limit since the sum of the (kinetic) stresses of two identical components do
not add to that of the equivalent monodisperse system. Accordingly, Benyahia (2008) proposed an ad
hoc modification to the IA05 theory based on the earlier work of Mathiesen et al. (2000) to ensure that
the monodisperse limit is achieved.
Explicit vs. Implicit Constitutive Relations. For each of the polydisperse theories listed in Table 1,
the constitutive relations take an explicit form. In other words, constitutive quantities (y) like the stress
tensor, heat flux, mass flux, etc. are expressed in terms of the hydrodynamic variables (x1, x2,...) accord-
ing to the form y = f (x1, x2,...). The one exception is GHD07, in which the majority of closures take
an implicit form. More specifically, for given values of the hydrodynamic variables, a set of algebraic
equations must be solved to find the constitutive quantity. Note that only one of these constitutive quanti-
ties (dissipation rate of granular energy, γ) involves a nonlinear set of algebraic equations, whereas the
remaining quantities are characterized by linear algebraic equations.
Non-equipartition of Energy when T is a Hydrodynamic Variable. It is worthwhile to note that two
of the theories noted above, namely GD02 and GHD07, incorporate a non-equipartition of energy (Ti ≠
Tj) while also using T as a hydrodynamic variable, which may appear contradictory at first glance. In
fact, the solution to the set of nonlinear algebraic equations needed for determination of the dissipation
rate of granular energy (γ) as noted above also leads to values for Ti, Tj, etc. In other words, species tem-
peratures are available from the GD02 and GHD07 theories via the solution of algebraic equations. As
mentioned above, in other theories that utilize Ti is as a hydrodynamic variable instead of T, differential
conservation balances (species energy balances) must instead be solved to find Ti, Tj, etc.
Continuous Distribution of Particle Sizes. To date, all of the polydisperse, kinetic-theory-based
models are targeted at a discrete (s) number of particles species, rather than a continuous size distribu-
tion (Gaussian, lognormal, etc.). However, because continuous distributions are prevalent in nature and
industry alike, the application of existing theories to such systems is needed. To do so, the continuous
distribution must be approximated as a set of discrete species. Accordingly, two questions arise: (i) how
should the particles sizes (σi) and corresponding concentrations (ni) of the discrete approximation be
chosen, and (ii) how many species (s) are needed to achieve the desired level of accuracy for the approxi-
mation? One approach to the first question is to determine σi and ni by matching the first 2s moments of
the discrete approximation to the known 2s moments of the continuous distribution (Fan & Fox, 2008;
Fan, Marchisio, & Fox, 2004). With regard to the second question, two different approaches have been
followed. Fan & Fox (2008) applied a polydisperse kinetic-theory-based model to a gas-fluidized bed,
and found that s=3 for Gaussian and lognormal distributions provided sufficient accuracy. Rather than
considering a specific system, Hrenya & coworkers (ongoing work) are examining the dependency of
all transport coefficients contained in the GHD07 theory to the value of s used to approximate a variety
of continuous distributions (Gaussian, lognormal, and bidisperse). An examination of the transport
coefficients over a wide range of parameters (ϕ and eij) will indicate the appropriate value of s – i.e., as
s increases, the values of the transport coefficients eventually collapse on one another. In this manner,
the results will be system independent, since some of the transport coefficients may be important in one
system but not another (e.g., the conductivity is immaterial in a simple shear flow system which displays
constant temperature, but is influential in systems with a nonzero temperature gradient).
Availability in CFD Software Tools. Of the polydisperse kinetic theories listed in Table 1, IA05 and
GHD07 have been incorporated into MFIX, an open-source CFD (computational fluid dynamics) code
which can be downloaded cost-free from https://mfix.netl.doe.gov/. The IA05 theory was implemented
115
with some corrections to the original work, as detailed by Galvin (2007). Similarly, GHD07 was imple-
mented with corrections to typographical errors in the original paper, as detailed by Garzó and Hrenya
(in preparation). To the best of the author’s knowledge, none of the kinetic theories contained in Table
1 are available off-the-shelf in commercial CFD codes.
Species Segregation
As mentioned in the introduction, a ubiquitous phenomenon associated with polydisperse flows is the
tendency of particles which differ in size and/or density to segregate or de-mix. Based on the kinetic-
theory models proposed above, considerable insight into the driving forces for this phenomenon is pos-
sible. Similar to mass transport in a single-phase fluid, the driving forces are revealed via the constitutive
equation for the species (diffusive) mass flux [ j0i = nimi (Ui - U) ] or equivalently the “diffusion velocity”
(Ui - Uj). The former follows directly from theories which utilize ni, U, and T as hydrodynamic variables,
whereas the latter can be derived from theories using ni, Ui, and Ti as hydrodynamic variables (and the
two can be related via the appropriate definitions).
For purposes of illustration, consider first a theory which uses ni, U, and T as hydrodynamic variables.
For example, in GHD07, the constitutive relation for the mass flux of species i takes the form
s nj s
j0i = −∑ mi m j Dij ∇ ln n j − pDiT ∇ ln T − ∑ DijF Fj, ext (11)
j =1 ρ j =1
where Dij is the ordinary diffusion coefficient, DiT is the thermal diffusion coefficient, and DijF is the
mass mobility coefficient. Expressions for these coefficients in terms of the hydrodynamic variables for
ni, U, and T and material properties (σi, mi, eij, etc.) are given in the original paper (Garzó, Hrenya et al.,
2007) and are not repeated here for sake of brevity.
An examination of the right-hand-side of the above equation reveals that the driving forces for
mixing/segregation mimic those of molecular gases: gradients in species concentration (ordinary diffu-
sion), gradients in granular temperature (thermal diffusion), and external forces (forced diffusion). This
similarity gives rise to the question: why is species segregation upon agitation (shearing, shaking, etc.)
so prevalent in granular materials but not so in molecular gases? The answer lies in the large tempera-
ture gradients present in such granular systems relative to their molecular counterparts. Because grains
(unlike molecules) are dissipative, areas of high concentration lead to increased collision frequencies,
which thereby lead to lower granular temperatures. Correspondingly, the gradients of concentration
and temperature are often in opposite directions: concentration gradients drive the mixing process and
temperature gradients drive the segregation process. Indeed, for systems without external forces pres-
ent, numerous theoretical, simulation, and experimental works (Dahl & Hrenya, 2004; Galvin et al.,
2005; Hsiau & Hunt, 1996; Liu et al., 2007; Luding, Strauss, & McNamara, 2000; Xu et al., 2003) have
shown that overall concentration and temperature are inversely related, and that more massive particles
segregate preferentially toward low-temperature regions.
For purposes of comparison, now consider a theory which uses ni, Ui, and Ti as hydrodynamic variables.
As evident from the conservation equations (5) – (7), such theories do not involve a species mass flux
(j0i), and hence the segregation mechanisms cannot be gleaned in the manner described above. Instead,
an equation for an analogous quantity, namely the diffusion velocity Ui - Uj, can instead be derived from
116
the species momentum balances (Galvin et al., 2005; Jenkins, 1998; Jenkins & Mancini, 1987) or calcu-
lated from the distribution function obtained from the CE expansion (Arnarson & Jenkins, 2000; Hsiau
& Hunt, 1996; Jenkins & Mancini, 1989). Because this derivation has been detailed in several previous
publications, only the basic steps of the former method are outlined here. In particular, the diffusion
velocity equation is found by subtracting the species momentum balance of j from that associated with i,
coupled with a few common assumptions (i.e., inertia associated with diffusion and nonlinear terms are
negligible). The constitutive relation for the species collisional momentum source (solids-solids drag)
is then substituted into the equation. Since this constitutive relation contains a term proportional to the
diffusion velocity, the diffusion velocity itself can then be solved for. The final result is shown here for
the JM87 theory [see (Galvin, 2007) for detailed derivation]:
n2
Ui − U j = − D (a ∇P − a2∇Pj − a3 Fi, ext + a4 F j, ext − a5∇T − a6∇ni + a7∇n j ) (12)
ni n j ij 1 i
where the species pressures Pi and Pj, the diffusion coefficient Dij and coefficients a1-a7 are functions of
the hydrodynamic variables and physical properties as specified by JM87. Note that the right-hand-side
of this equation reveals similar driving forces to those identified from Equation (11) – species concentra-
tion gradients, temperature gradient, and external forces – in addition to a gradient in species pressure
(pressure diffusion). It is important to note, however, that because the constitutive relation for species
pressure is a function of species concentration and temperature, this pressure diffusion can be replaced
by terms containing gradients in species concentration and temperature. Accordingly, the driving forces
for segregation revealed by Equation (11) are analogous to those of Equation (12), though the transport
coefficients have different bases, and thus cannot be directly compared. (Note that Equation (12) was
derived simply to illustrate the segregation mechanisms and does not provide an additional equation for
the JM87 theory. Because it was derived using both of the species momentum balances, however, it can
be used in substitution of one of those balances when solving the system of equations. It is worthwhile
to point out that its derivation does involve additional assumptions, as described in (Galvin et al., 2005).)
Based on the segregation mechanisms inherent in kinetic-theory models, a wide range of segregation
behavior has been predicted from such models. In addition to the force-free system mentioned above (in
which massive particles segregate preferentially toward lower-temperature regions), Jenkins and Yoon
(2002) consider a constant-temperature system, and shown that the presence of gravity alone can give
rise to one species either rising or falling, depending on the size and density ratio. The incorporation
of all three mechanisms – ordinary, thermal, and forced diffusion – has been considered in a variety of
segregating systems and was found to give rise to a number of counter-intuitive segregation patterns
(Brey, Ruiz-Montero, & Moreno, 2005; Garzó, 2006; Khakhar, McCarthy, & Ottino, 1999; Liu et al.,
2007; Serero et al., 2006; Wildman, Jenkins, Krouskop, & Talbot, 2006b). In addition, the influence of
non-equipartition on qualitative and quantitive nature of segregation has been documented in several
systems (Brey, Ruiz-Montero, & Moreno, 2005; Galvin et al., 2005; Garzó, 2006; Liu et al., 2007; Yoon
& Jenkins, 2006). Finally, although such species segregation is typically linked to size and/or density
differences associated with the thermal and/or forced diffusion, Serero et al. (2006) have shown that
differences in the restitution coefficients among otherwise identical species will also lead to species
segregation via thermal diffusion.
117
gas-solids flows
Up to this point, all of the polydisperse theories discussed have been restricted to granular flows, or flows
in which the role of the interstitial fluid phase is negligible. To formally account for the effects of a fluid
phase, the instantaneous force on a particle from the fluid phase must be incorporated into the starting
kinetic equation, Equation (1). Note that this force, which is typically dominated by a drag force, is
non-conservative in nature – i.e., it depends on fluctuations in particle velocity and fluid velocity, rather
than just mean velocities. In all the polydisperse theories reported to date, the Fi,ext term in Equation (1)
has been treated as a conservative force (Fi,ext,cons) in ensuing calculations, such that only the effects of
mean velocities are rigorously incorporated. Nonetheless, the additional effects of instantaneous velocites
have been reported for monodisperse systems, as described below. Such efforts are illustrative of the
remaining work to be done for polydisperse systems.
In monodisperse systems, previous work has indicated that the addition of a non-conservative fluid
force in the starting kinetic equation leads to additional terms in the resulting conservation equations,
as well as to modifications to the closures for existing constitutive quantities (see, for example, Balzer,
Boelle, & Simonin, 1995; Gidaspow, 1994; Koch, 1990; Koch & Sangani, 1999; Ma & Ahmadi, 1988;
Tsao & Koch, 1995). More specifically, with regard to the first point, the presence of a fluid phase leads
to (i) an additional term in the momentum balance that accounts for mean drag acting on the particles,
and (ii) two additional terms in the granular energy balance stemming from velocity fluctuations, namely
a viscous sink associated with particle velocity fluctuations, and a viscous source associated with fluc-
tuations in the fluid velocity. With regard to the second point, the closures for existing quantities like
stress will also depend on the parameters of the instantaneous force (i.e., the acceleration model) which
is incorporated into the starting kinetic equation (Equation (1)).
As alluded to above, a prerequisite for the inclusion of the fluid phase into the starting kinetic equation
is an expression for the instantaneous force on the particle from the fluid. Several of the monodisperse
studies mentioned above used Stokes drag as a first approximation (in conjunction with the slip velocity
v-Ug, where v is the instantaneous particle velocity and Ug is the mean gas velocity); this approxima-
tion is restricted to low Reynolds (Re) flow, does not account for effect of neighboring particles (i.e.,
dilute limit), and does not account for fluid-phase fluctuations. Extensions to this first approximation
are possible via empirical correlations for the drag force (Gidaspow, 1994; Leboreiro, Joseph, & Hre-
nya, 2008), which account for higher Re and higher concentrations (though do not explicitly depend
on fluid-phase fluctuations). Analogous empirical correlations for polydisperse systems, however, are
essentially non-existent due to experimental difficulties (e.g., lack of clear interface between the vari-
ous species in settling experiments). Accordingly, direct numerical simulations (DNS) of polydisperse
flows are required to obtain corresponding drag laws. Recent work in this area has been performed by
van der Hoef and co-workers (Beetstra, van der Hoef, & Kuipers, 2007a, 2007b; van der Hoef, Beetstra,
& Kuipers, 2005), who developed a drag law in which the drag force on a species depends on the mean
relative velocity between that species and the fluid phase. Yin and Sundaresan (2009) also incorporated
the effect of off-diagonal terms; in their expression, the drag force on a species also depends on the
mean velocity of other species. It is worthwhile to mention that both of these efforts considered static
arrays of particles, and hence the effect of velocity fluctuations in the solid phase (due to collisions and/
or transport across a velocity field) and gas phase (due to configurational and velocity changes of sur-
rounding particles with time) were not considered.
118
The polydisperse drag laws described above are an important first step in the rigorous extension of
polydisperse kinetic-theory models to gas-solids flows. The formal incorporation of an instantaneous
fluid-force model into the starting kinetic equation will lead to further modifications of the polydisperse
granular theory (i.e., in addition to the mean drag term), which have yet to be identified and evaluated.
That being said, it is worthwhile to mention that papers involving kinetic-theory-based modeling of
gas-fluidized beds, based on a less rigorous treatment of the polydisperse drag law, have appeared in
the literature. For example, prior to the availability of DNS-based, polydisperse drag laws, monodis-
perse drag laws were modified in an ad hoc fashion for polydispersity; such treatments have since been
shown to have a non-negligible influence on species segregation predictions when compared to their
DNS counterparts (Beetstra, Hoef, & Kuipers, 2007; Leboreiro, 2008). On a broader scale, it is also
worthwhile to point out that many earlier examples of using ad hoc, polydisperse drag laws to model
bubbling fluidized beds, circulating fluidized beds, etc. (via kinetic theory and/or discrete element mod-
els) can be found in the engineering literature; unfortunately, a coverage of these contributions is well
beyond the scope of this chapter.
cOncluding reMarKs
As detailed above, significant recent progress has been made in advancing the state-of-the-art for polydis-
perse, kinetic-theory models for granular flows in the recent past. In particular, many of the assumptions
associated with earlier models (e.g., equipartition of energy) are now treated in a more rigorous fashion.
Accordingly, a concerted validation effort is needed to critically assess both the physical accuracy and
computational efficiency associated with such models. A combination of MD simulations and experi-
mental data will be useful for this task, beginning with relatively simple non-segregating [homogeneous
cooling system (Dahl, Hrenya et al., 2002) and simple shear (Clelland & Hrenya, 2002; Dahl, Clelland,
& Hrenya, 2003)] and segregating [Couette flows (Karion & Hunt, 2000; Liu et al., 2007; Xu et al.,
2003) and bounded conduction (Dahl & Hrenya, 2004; Galvin et al., 2005)] systems in which specific
quantities and mechanisms can be isolated, and proceeding to more complex systems with a complex
interaction of driving forces and boundaries, such as vibrating fluidized beds (Feitosa & Menon, 2002;
Wildman, Jenkins, Krouskop, & Talbot, 2006a; Wildman & Parker, 2002).
Compared to their granular counterparts, the incorporation of gas effects into polydisperse models
is in its infancy. The first step in this process is the development of a model for the instantaneous force
of a fluid on a particle in a polydisperse suspension, for incorporation into the starting kinetic equation.
Following the corresponding kinetic-theory-based description, a stepwise validation effort is needed
again ranging from simple to more complex systems, including both MD simulations and experiments
(see, for example, Bokkers, van Sint Annaland, & Kuipers, 2004; Brito, Enriquez, Godoy, & Soto, 2008;
Dahl & Hrenya, 2005; Joseph, Leboreiro, Hrenya, & Stevens, 2007; Mathiesen et al., 1999b).
Lastly, it is worthwhile to point out that the topic of this review has been limited to polydisperse
models in fairly idealized systems. Non-idealities such as friction (Benyahia, 2008), non-spherical
shapes, and impact-velocity-dependent restitution coefficients will also influence species segregation.
Knudsen effects may play a role in relatively simple systems and experiments [see, for example, (Brey
et al., 2001; Forterre & Pouliquen, 2001; Galvin et al., 2007; Goldhirsch et al., 2004; Hrenya et al., 2008;
Khain et al., 2008; Martin et al., 2006; Rericha et al., 2002; Wassgren et al., 2003)]. For polydisperse
119
systems, boundary conditions will have added complexity, and instabilities such as clusters may behave
differently (Alam & Luding, 2005; Rice, 2009).
acKnOwledgMent
The author is indebted to the Department of Energy (Award DE-FC26-07NT43098) and the National
Science Foundation (Award CBET-0650893) for providing the funding support for this work.
references
Alam, M., & Luding, S. (2003). Rheology of bidisperse granular mixtures via event-driven simulations.
Journal of Fluid Mechanics, 476, 69–103. doi:10.1017/S002211200200263X
Alam, M., & Luding, S. (2005). Energy nonequipartition, rheology, and microstructure in sheared bi-
disperse granular mixtures. Physics of Fluids, 17, 063303. doi:10.1063/1.1938567
Alam, M., Willits, J. T., Arnarson, B. O., & Luding, S. (2002). Kinetic theory of a binary mixture of nearly
elastic disks with size and mass disparity. Physics of Fluids, 14(11), 4085–4087. doi:10.1063/1.1509066
Arnarson, B. O., & Jenkins, J. T. (2000). Particle segregation in the context of species momentum bal-
ances. In Helging, D., Herrmann, H. J., Schreckenberg, M., & Wolf, D. E. (Eds.), Traffic and Granular
Flow ‘99: Social, Traffic, and Granular Dynamics. Berlin: Springer.
Arnarson, B. O., & Willits, J. T. (1998). Thermal diffusion in binary mixtures of smooth, nearly elastic
spheres with and without gravity. Physics of Fluids, 10, 1324–1328. doi:10.1063/1.869658
Bagnold, R. A. (1954). Experiments on a gravity-free dispersion of large solid spheres in a Newtonian
fluid under shear. Proc. Roy. Soc., A225, 49.
Balzer, G., Boelle, A., & Simonin, O. (1995). Eulerian gas-solid flow modelling of dense fluidized beds.
Paper presented at the Fluidization VIII.
Barrat, A., & Trizac, E. (2002). Lack of energy equipartition in homogeneous heated binary granular
mixtures. Granular Matter, 4(2), 57–63. doi:10.1007/s10035-002-0108-4
Beetstra, R., Hoef, M. A. d., & Kuipers, J. A. M. (2007). Numerical study of segregation using a new
drag force correlation for polydisperse systems derived from lattice-Boltzmann simulations. Chemical
Engineering Science, 62, 246–255. doi:10.1016/j.ces.2006.08.054
Beetstra, R., van der Hoef, M. A., & Kuipers, J. A. M. (2007a). Drag force of intermediate Reynolds
number flow past mono- and bidisperse arrays of spheres. AIChE Journal. American Institute of Chemi-
cal Engineers, 53, 489–501. doi:10.1002/aic.11065
Beetstra, R., van der Hoef, M. A., & Kuipers, J. A. M. (2007b). Erratum. AIChE Journal. American
Institute of Chemical Engineers, 53, 3020.
120
Benyahia, S. (2008). Verification and validation study of some polydisperse kinetic theories. Chemical
Bokkers, G. A., van Sint Annaland, M., & Kuipers, J. M. (2004). Mixing and segregation in a bidis-
perse gas-solid fluidized bed: a numerical and experimental study. Powder Technology, 140, 176–186.
doi:10.1016/j.powtec.2004.01.018
Boublik, T. (1970). Hard-sphere equation of state. The Journal of Chemical Physics, 53, 471.
doi:10.1063/1.1673824
Brey, J. J., & Ruiz-Montero, M. J. (2004). Simulation study of the Green-Kubo relations for dilute granu-
lar gases. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 70, 051301. doi:10.1103/
PhysRevE.70.051301
Brey, J. J., Ruiz-Montero, M. J., Maynar, P., & Garcia de Soria, M. I. (2005). Hydrodynamic modes,
Green-Kubo relations, and velocity correlations in dilute granular gases. Journal of Physics Condensed
Matter, 17, S2489–S2502. doi:10.1088/0953-8984/17/24/008
Brey, J. J., Ruiz-Montero, M. J., & Moreno, F. (2001). Hydrodynamics of an open vibrated system. Physical
Brey, J. J., Ruiz-Montero, M. J., & Moreno, F. (2005). Energy partition and segregation for an intruder
in a vibrated granular system under gravity. Physical Review Letters, 95, 098001. doi:10.1103/Phys-
RevLett.95.098001
Brito, R., Enriquez, H., Godoy, S., & Soto, R. (2008). Segregation induced by inelasticity in a vibrofluid-
ized granular mixture. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 77, 061301.
doi:10.1103/PhysRevE.77.061301
Chapman, S., & Cowling, T. G. (1970). The Mathematical Theory of Non-Uniform Gases (3rd ed.).
Cambridge, UK: Cambridge University Press.
Clelland, R., & Hrenya, C. M. (2002). Simulations of a binary-sized mixture of inelastic grains in rapid
shear flow. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 65, 031301. doi:10.1103/
PhysRevE.65.031301
Curtis, J. S., & van Wachem, B. (2004). Modeling particle-laden flows: a research outlook. AIChE
Dahl, S. R., Clelland, R., & Hrenya, C. M. (2002). The Effects of Continuous Size Distributions on the
Rapid Flow of Inelastic Particles. Physics of Fluids, 14(6), 1972–1984. doi:10.1063/1.1476917
Dahl, S. R., Clelland, R., & Hrenya, C. M. (2003). Three-dimensional, rapid shear flow of particles
with continuous size distributions. Powder Technology, 138, 7–12. doi:10.1016/j.powtec.2003.08.036
Dahl, S. R., & Hrenya, C. M. (2004). Size segregation in rapid, granular flows with continuous size
distributions. Physics of Fluids, 16(1), 1–13. doi:10.1063/1.1626682
Dahl, S. R., & Hrenya, C. M. (2005). Size segregation in gas-solid fluidized beds with continuous par-
ticle size distributions. Chemical Engineering Science, 60, 6658–6673. doi:10.1016/j.ces.2005.05.057
121
Dahl, S. R., Hrenya, C. M., Garzó, V., & Dufty, J. W. (2002). Kinetic temperatures for a granular
mixture. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 66, 041301. doi:10.1103/
PhysRevE.66.041301
Fan, R., & Fox, R. O. (2008). Segregation in polydisperse fluidized beds: Validation of a multi-fluid
model. Chemical Engineering Science, 63, 272–285. doi:10.1016/j.ces.2007.09.038
Fan, R., Marchisio, D. L., & Fox, R. O. (2004). Application of the direct quadrature method of moments
to polydisperse gas-solid fludized beds. Powder Technology, 139, 7–20. doi:10.1016/j.powtec.2003.10.005
Feitosa, K., & Menon, N. (2002). Breakdown of energy equipartition in a 2D binary vibrated granular
gas. Physical Review Letters, 88(19), 198301. doi:10.1103/PhysRevLett.88.198301
Ferziger, J. H., & Kaper, H. G. (1972). Mathematical theory of transport processes in gases. New York:
Elsevier.
Forterre, Y., & Pouliquen, O. (2001). Longitudinal vortices in granular flows. Physical Review Letters,
86(26), 5886–5889. doi:10.1103/PhysRevLett.86.5886
Galvin, J. E. (2007). On the hydrodynamic description of binary mixtures of rapid granular flows and
gas-fluidized beds. PhD Thesis, University of Colorado, Boulder, CO.
Galvin, J. E., Dahl, S. R., & Hrenya, C. M. (2005). On the role of non-equipartition in the dynam-
ics of rapidly-flowing, granular mixtures. Journal of Fluid Mechanics, 528, 207–232. doi:10.1017/
S002211200400326X
Galvin, J. E., Hrenya, C. M., & Wildman, R. D. (2007). On the role of the Knudsen layer in rapid granular
flows. Journal of Fluid Mechanics, 585, 73–92. doi:10.1017/S0022112007006489
Garzó, V. (2006). Segregation in granular binary mixtures: Thermal diffusion. Europhysics Letters, 75,
521–527. doi:10.1209/epl/i2006-10143-4
Garzó, V., & Dufty, J. (1999). Homogeneous cooling state for a granular mixture. Physical Review E:
Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics, 60(5), 5706–5713. doi:10.1103/
PhysRevE.60.5706
Garzó, V., & Dufty, J. W. (2002). Hydrodynamics for a granular binary mixture at low density. Physics
of Fluids, 14(4), 1476–1490. doi:10.1063/1.1458007
Garzó, V., Dufty, J. W., & Hrenya, C. M. (2007). Enskog theory for polydisperse granular mixtures. I.
Navier-Stokes order transport. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 76,
031303. doi:10.1103/PhysRevE.76.031303
Garzó, V., Hrenya, C. M., & Dufty, J. W. (2007). Enskog theory for polydisperse granular mixtures. II.
76, 031304. doi:10.1103/PhysRevE.76.031304
Garzó, V., & Montanero, J. M. (2002). Transport coefficients of a heated granular gas. Physica A, 313,
336–356. doi:10.1016/S0378-4371(02)00994-9
122
Garzó, V., & Montanero, J. M. (2003). Shear viscosity for a moderately dense granular mixture. Physical
Garzó, V., & Montanero, J. M. (2004). Diffusion of impurities in a granular gas. Physical Review E:
Garzó, V., Reyes, F. V., & Montanero, J. M. (Manuscript submitted for publication). Modified Sonine
approximation for granular binary mixture. Journal of Fluid Mechanics.
Gidaspow, D. (1994). Multiphase Flow and Fluidization. San Diego, CA: Academic Press.
Goldhirsch, I., Noskowicz, S. H., & Bar-Lev, O. (2004). Theory of granular gases: some recent results
and some open problems. Journal of Physics Condensed Matter, 17, 2591–2608. doi:10.1088/0953-
8984/17/24/015
Hrenya, C. M., Galvin, J. E., & Wildman, R. D. (2008). Evidence of higher-order effects in thermally-
driven, rapid granular flows. Journal of Fluid Mechanics, 598, 429–450. doi:10.1017/S0022112007000079
Hsiau, S. S., & Hunt, M. L. (1996). Granular thermal diffusion in flows of binary-sized mixtures. Acta
Mechanica, 114, 121–137. doi:10.1007/BF01170399
Huilin, L., Gidaspow, D., & Manger, E. (2001). Kinetic theory of fluidized binary granular mixtures.
RevE.64.061301
Iddir, H. (2004). Modeling of the multiphase mixture of particles using the kinetic theory approach.
Ph.D. Thesis, Illinois Institute of Technology, Chicago.
Iddir, H., & Arastoopour, H. (2005). Modeling of Multitype Particle Flow Using the Kinetic Theory
Approach. AIChE Journal. American Institute of Chemical Engineers, 51(6), 1620–1632. doi:10.1002/
aic.10429
Jenkins, J. T. (1998). Particle segregation in collisional flows of inelastic spheres. In Hermann, H. J.,
Hovi, J. P., & Luding, S. (Eds.), Physics of Dry Granular Media (pp. 645–658). Amsterdam: Kluwer.
Jenkins, J. T., & Mancini, F. (1987). Balance laws and constitutive relations for plane flows of a dense,
binary mixture of smooth, nearly elastic, circular disks. Journal of Applied Mechanics, 54, 27–34.
doi:10.1115/1.3172990
Jenkins, J. T., & Mancini, F. (1989). Kinetic theory for binary mixtures of smooth, nearly elastic spheres.
Phys. Fluids A, 1(12), 2050–2057. doi:10.1063/1.857479
Jenkins, J. T., & Yoon, D. K. (2002). Segregation in Binary Mixtures under Gravity. Physical Review
Letters, 88(19), 194301. doi:10.1103/PhysRevLett.88.194301
Joseph, G. G., Leboreiro, J., Hrenya, C. M., & Stevens, A. R. (2007). Experimental segregation profiles in
bubbling gas-fluidized beds. AIChE Journal. American Institute of Chemical Engineers, 53, 2804–2813.
doi:10.1002/aic.11282
Karion, A., & Hunt, M. L. (2000). Wall streses in granular couette flows of mono-sized particles and
binary mixtures. Powder Technology, 109, 145–163. doi:10.1016/S0032-5910(99)00233-8
123
Khain, E., Meerson, B., & Sasorov, P. V. (2008). Knudsen temperature jump and the Navier-Stokes
hydrodynamics of granular gases driven by thermal walls. Physical Review E: Statistical, Nonlinear,
and Soft Matter Physics, 78, 041303. doi:10.1103/PhysRevE.78.041303
Khakhar, D. V., McCarthy, J. J., & Ottino, J. M. (1999). Mixing and segregation of granular materials
in chute flows. Chaos (Woodbury, N.Y.), 9(3), 594–610. doi:10.1063/1.166433
Knowlton, T. M., Carson, J. W., Klinzing, G. E., & Yang, W.-C. (1994). The importance of storage,
transfer, and collection. Chemical Engineering Progress, 90(4), 44–54.
Koch, D. L. (1990). Kinetic theory for a monodisperse gas-solid suspension. Phys. Fluids A, 2, 1711–1723.
doi:10.1063/1.857698
Koch, D. L., & Sangani, A. S. (1999). Particle pressure and marginal stability limits for a homogenous
monodisperse gas-fluidized bed: kinetic theory and numerical simulations. Journal of Fluid Mechanics,
400, 229–263. doi:10.1017/S0022112099006485
Leboreiro, J. (2008). Influence of drag laws on segregation and bubbling behavior in gas-fluidized beds.
Ph.D. Thesis, University of Colorado, Boulder.
Leboreiro, J., Joseph, G. G., & Hrenya, C. M. (2008). Revisiting the standard drag law for bubbling,
gas-fluidized beds. Powder Technology, 183, 385–400. doi:10.1016/j.powtec.2008.01.008
Lebowitz, J. L. (1964). Exact solution of generalized Percus-Yevick equation for a mixture of hard
spheres. Physical Review, 133, 895–899. doi:10.1103/PhysRev.133.A895
Lee, Y. Y. (1997). Design considerations for CFB boilers. In Grace, J. R., Avidan, A. A., & Knowlton,
T. M. (Eds.), Circulating Fluidized Beds (pp. 417–440). New York: Blackie Academic & Professional.
Liu, X., Metzger, M., & Glasser, B. J. (2007). Couette flow with a bidisperse particle mixture. Physics
of Fluids, 19, 073301. doi:10.1063/1.2741245
Luding, S., Strauss, O., & McNamara, S. (2000). Segregation of polydisperse granular media in the
presence of a temperature gradient. In A. D. Rosato & D. L. Blackmore (Eds.), IUTAM Symposium on
Segregation in Granular Flows. Boston: Kluwer Academic Publishers.
Lun, C. K. K., Savage, S. B., Jeffrey, D. J., & Chepurniy, N. (1984). Kinetic theories for granular flow:
inelastic particles in couette flow and slightly inelastic particles in a general flowfield. Journal of Fluid
Mechanics, 140, 223–256. doi:10.1017/S0022112084000586
Ma, D., & Ahmadi, G. (1988). A kinetic model for rapid granular flows of nearly elastic particles in-
cluding interstitial fluid effects. Powder Technology, 56, 191–207. doi:10.1016/0032-5910(88)80030-5
Mansoori, G. A., Carnahan, N. F., Starling, K. E., & Leland, T. W. (1523-1525). J. (1971). Equilibrium
Thermodynamic Properties of the Mixture of Hard Spheres. The Journal of Chemical Physics, 54(4).
Martin, T. W., Huntley, J. M., & Wildman, R. D. (2006). Hydrodynamic model for a vibrofluidized
granular bed. Journal of Fluid Mechanics, 535, 325–345. doi:10.1017/S0022112005004866
124
Mathiesen, V., Solberg, T., Arastoopour, H., & Hjertager, B. (1999a). Experimental and Computational
Study of Multiphase Gas/Particle Flow in a CFB Riser. AIChE Journal. American Institute of Chemical
Mathiesen, V., Solberg, T., Arastoopour, H., & Hjertager, B. H. (1999b). Experimental and computational
study of multiphase gas / particle flow in a CFB riser. AIChE Journal. American Institute of Chemical
Mathiesen, V., Solberg, T., & Hjertager, B. H. (2000). An experimental and computational study of
multiphase flow behavior in a circulating fluidized bed. International Journal of Multiphase Flow, 26,
387–419. doi:10.1016/S0301-9322(99)00027-0
Montanero, J. M., & Garzó, V. (2002). Monte Carlo simulation of the homogeneous cooling state for a
granular mixture. Granular Matter, 4, 17–24. doi:10.1007/s10035-001-0097-8
Montanero, J. M., & Garzó, V. (2003a). Energy nonequipartition in a sheared granular mixture. Molecular
Simulation, 29(6-7), 357–362. doi:10.1080/0892702031000117207
Montanero, J. M., & Garzó, V. (2003b). Shear viscosity for a heated granular binary mixture at low
density. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 67, 021308. doi:10.1103/
PhysRevE.67.021308
Montanero, J. M., Santos, A., & Garzó, V. (2007). First-order Chapman-Enskog velocity distribution
function in a granular gas. Physica A, 376, 75–93. doi:10.1016/j.physa.2006.10.080
Muzzio, F. J., Shinbrot, T., & Glasser, B. J. (2002). Powder technology in the pharmaceutical industry:
the need to catch up fast. Powder Technology, 124(1-2), 1–7. doi:10.1016/S0032-5910(01)00482-X
Noskowicz, S. H., Bar-Lev, O., Serero, D., & Goldhirsch, I. (2007). Computer-aided kinetic theory and
granular gases. Europhysics Letters, 79, 60001. doi:10.1209/0295-5075/79/60001
Ottino, J. M., & Khakhar, D. V. (2000). Mixing and segregation of granular materials. Annual Review
of Fluid Mechanics, 32, 55–91. doi:10.1146/annurev.fluid.32.1.55
Paolotti, D., Cattuto, C., Marconi, U. M. B., & Puglisi, A. (2003). Dynamical properties of vibrofluidized
granular mixtures. Granular Matter, 5(2), 75–83. doi:10.1007/s10035-003-0133-y
Pell, M., & Jordan, S. P. (1988). Effects of fines and velocity on fluid bed reactor performance. AIChE
Symp. Ser., 84(262), 68-73.
Rahaman, M. F., Naser, J., & Witt, P. J. (2003). An unequal temperature kinetic theory: description
of granular flow with multiple particle classes. Powder Technology, 138, 82–92. doi:10.1016/j.pow-
tec.2003.08.050
Reed, T. M., & Gubbins, K. E. (1973). Applied Statistical Mechanics. New York: McGraw-Hill.
Rericha, E. C., Bizon, C., Shattuck, M. D., & Swinney, H. L. (2002). Shocks in supersonic sand. Physi-
cal Review Letters, 88(1), 014302. doi:10.1103/PhysRevLett.88.014302
125
Rice, R. B. (2009). (Manuscript submitted for publication). Clustering in rapid granular flows of binary
and continuous particle size distributions. Physical Review E: Statistical, Nonlinear, and Soft Matter
Physics.
Rosato, A., Prinze, F., Standburg, K. J., & Swendsen, R. (1987). Why Brazil nuts are on top: size
segregation of particulate matter by shaking. Physical Review Letters, 58, 1038–1040. doi:10.1103/
PhysRevLett.58.1038
Santos, A., Yuste, S., & López de Haro, M. (2002). Contact values of the radial distribution functions
of additive hard-sphere mixtures in d dimensions: A new proposal. The Journal of Chemical Physics,
117(12), 5785–5793. doi:10.1063/1.1502247
Schröter, M., Ulrich, S., Kreft, J., Swift, J. B., & Swinney, H. L. (2006). Mechanisms in the size segre-
gation of a binary granular mixture. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics,
74(011307).
Sela, N., & Goldhirsch, I. (1998). Hydrodynamic equations for rapid flows of smooth inelastic spheres,
to Burnett order. Journal of Fluid Mechanics, 361, 41–74. doi:10.1017/S0022112098008660
Serero, D., Goldhirsch, I., Noskowicz, S. H., & Tan, M.-L. (2006). Hydrodynamics of granular gases and
granular gas mixtures. Journal of Fluid Mechanics, 554, 237–258. doi:10.1017/S0022112006009281
Sundaresan, S. (2001). Some outstanding questions in handling of cohesionless particles. Powder Tech-
nology, 115, 2–7. doi:10.1016/S0032-5910(00)00423-X
Tsao, H.-K., & Koch, D. L. (1995). Simple shear flows of dilute gas-solid suspensions. Journal of Fluid
Mechanics, 296, 211–245. doi:10.1017/S0022112095002114
van der Hoef, M. A., Beetstra, R., & Kuipers, J. A. M. (2005). Lattice Boltzmann simulations of low
Reynolds number flow past mono- and bi-disperse arrays of spheres: results for the permeability and
drag forces. Journal of Fluid Mechanics, 528, 233–254. doi:10.1017/S0022112004003295
van Wachem, B. G. M., Schouten, J. C., van den Bleek, C. M., Krishna, R., & Sinclair, J. L. (2001). CFD
modeling of gas-fluidized beds with a bimodal particle mixture. AIChE Journal. American Institute of
Chemical Engineers, 47(6), 1292–1302. doi:10.1002/aic.690470607
Wassgren, C. R., Cordova, J. A., Zenit, R., & Karion, A. (2003). Dilute granular flow around an immersed
cylinder. Physics of Fluids, 15(11), 3318–3330. doi:10.1063/1.1608937
Wildman, R. D., Jenkins, J. T., Krouskop, P. E., & Talbot, J. (2006a). A comparison of the predictions of
a simple kinetic theory with experimental and numerical results for a vibrated granular bed consisting
of nearly elastic particles of two sizes. Physics of Fluids, 18, 073301. doi:10.1063/1.2210500
Wildman, R. D., Jenkins, J. T., Krouskop, P. E., & Talbot, J. (2006b). A comparison of the preodictions
of a simple kinetic theory with experimental and numerical results for a vibrated granular bed consisting
of nearly elastic particles of two sizes. Physics of Fluids, 18, 073301. doi:10.1063/1.2210500
Wildman, R. D., & Parker, D. J. (2002). Coexistence of two granular temperatures in binary vibrofluid-
ized beds. Physical Review Letters, 88(6), 064301. doi:10.1103/PhysRevLett.88.064301
126
Willits, J. T., & Arnarson, B. O. (1999). Kinetic theory of a binary mixture of nearly elastic disks. Phys-
ics of Fluids, 11(10), 3116–3122. doi:10.1063/1.870169
Xu, H., Louge, M., & Reeves, A. (2003). Solutions of the kinetic theory for bounded collisional granular
flows. Continuum Mechanics and Thermodynamics, 2003, 321–349. doi:10.1007/s00161-003-0116-6
Yin, X., & Sundaresan, S. (2009). Drag law for bidsiperse gas-solid suspensions containing equally sized
spheres. Industrial & Engineering Chemistry Research, 48, 227–241. doi:10.1021/ie800171p
Yoon, D. K., & Jenkins, J. T. (2006). The influence of different species’ granular temperature on segre-
gation in a binary mixture of dissipative grains. Physics of Fluids, 18, 073303. doi:10.1063/1.2219437
Zamankhan, P. (1995). Kinetic theory of multicomponent mixtures of slightly inelastic spherical particles.
Physical Review E: Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics, 52(5),
4877–4891. doi:10.1103/PhysRevE.52.4877
endnOtes
1
“Dilute” refers to a theory based on the Boltzmann kinetic equation, whereas “dilute & dense”
refers to the Enskog kinetic equation.
2
“No restriction” refers to a theory based on an expansion about the HCS, whereas eij ~ 1 refers to
an expansion about the nearly-elastic state.
3
“No restriction” refers to a theory that includes non-equipartition effects, whereas mi/mj ~ 1 refers
to a theory based on an equipartition assumption.
4
Three of the references defined the granular temperature differently than others, namely (i) M99
utilizes a species temperature (Ti,m) which is independent of mass, (ii) IA05 defines species tem-
perature (Ti, IA) in terms of velocity fluctuations relative to the mean species velocity Ui instead of
the mixture velocity U, and (iii) S06 utilized the mixture temperature Ts=3T instead of T.
127
128
Chapter 4
Interfacial Interactions:
Drag
Wei Ge
Chinese Academy of Sciences, China
Ning Yang
Wei Wang
Jinghai Li
abstract
The drag interaction between gas and solids not only acts as a driving force for solids in gas-solids flows
but also plays as a major role in the dissipation of the energy due to drag losses. This leads to enormous
complexities as these drag terms are highly non-linear and multiscale in nature because of the variations
in solids spatio-temporal distribution. This chapter provides an overview of this important aspect of the
hydrodynamic interactions between the gas and solids and the role of spatio-temporal heterogeneities
on the quantification of this drag force. In particular, a model is presented which introduces a mesoscale
description into two-fluid models for gas-solids flows. This description is formulated in terms of the
stability of gas-solids suspension. The stability condition is, in turn, posed as a minimization problem
where the competing factors are the energy consumption required to suspend and transport the solids
and their gravitational potential energy. However, the lack of scale-separation leads to many uncertain-
ties in quantifying mesoscale structures. The authors have incorporated this model into computational
fluid dynamics (CFD) simulations which have shown improvements over traditional drag models.
Fully resolved simulations, such as those mentioned in this chapter and the subject of a later chapter
on Immersed Boundary Methods, can be used to obtain additional information about these mesoscale
structures. This can be used to formulate better constitutive equations for continuum models.
DOI: 10.4018/978-1-61520-651-3.ch004
Interfacial Interactions
intrOductiOn
To a large extent, hydrodynamic interactions between the gas and solid phases are responsible for the
complexity of gas-solids flows since otherwise the two phases would move independently. Drag force
is almost always the dominant component of these interactions and its quantification is critical for the
predictability of simulations on gas-solids flows. Although dozens of correlations have been proposed
for this force, considerable uncertainties and discrepancies remain. Significant multi-scale heterogene-
ity and its dynamic behavior present a major difficulty for accurate estimation of drag force, most drag
force models are based on the assumption of homogenous two-phase flows and seems to be not accurate
enough for typical heterogeneous gas-solids flows in engineering.
In fact, it is all very often to find that quantification of meso-scale structures is a challenge in many
different areas. For the drag force in gas-solids systems, we have relatively accurate correlation of a
single particle in unbounded flow field and we can measure its average value for a large amount of
particles through pressure drop or solids concentration. However, correlations for the drag force in a
computational cell in continuum models can give predictions that are different in orders. For heat and
mass transfer in gas-solids flows, when correlating the transfer rate to average parameters such as the
particle Reynolds number based on the mean slip velocity, the difference can even reach more orders and
we may expect even larger discrepancies when chemical reactions are involved, due to the meso-scale
heterogeneity prevailing in such systems.
This challenge is tackled from different angles in this chapter. A two-phase two-fluid model is used to
develop a drag force model that has considered the meso-scale (MS) structure in the simulation of gas-
solids flows. A criterion for the stability of the gas-solids suspension is introduced to close this model that
involves more variables than homogeneous models. This stability condition is based on an analysis of the
compromise between the minimization of energy consumption for suspending and transporting solids in
unit space and minimization of its gravitational potential. Along with the drag force, parameters for the
meso-scale structure are also provided by this model, which can be used to improve the characterization
of mass and heat transfer properties in heterogeneous systems. This model has been incorporated into
computational fluid dynamics (CFD) simulations and has contributed substantially to its predictability.
However, partly due to the lack of scale separation in gas-solids systems, many uncertainties remind
in quantifying meso-scale structures. To understand these structures from a more fundamental level,
direct numerical simulations (DNS) of gas-solids suspension, down to scales far below particle size,
has been performed using macro-scale particle methods. Such simulations needs no drag correlations
as model input, and in fact, it can provide the interaction details between the two phases, the flow field
around each particle, the pressure and tangential forces on particle surfaces, from which the drag force
properties on larger scales can be analyzed and, in long term, new correlations can be proposed. The
authors’ preliminary studies have shown pronounced anisotropy of the magnitude of the drag force in a
heterogeneous gas-solids systems.
Sections 1 and 2 of this chapter was written by Ning Yang, Section 3.1-3.3 was written by Wei Wang,
and the rest was rewitten by Wei Ge. The whole chapter is edited by Jinghai Li and Wei Ge.
129
1. HydrOdynaMic interactiOns between Particles and fluid
1.1 forces exerted on a Particle immersed in a fluid
Forces exerted on a particle immersed in a fluid can fall into three categories: volume forces acted on
bulk volume, surface forces on surfaces and line forces on curves (Wörner, 2003). Inertia and buoyancy
forces fall within the scope of volume forces. The former tends to retain the original direction and mag-
nitude of the particle motion unchanged and the latter tends to reduce the apparent weight of an object.
In multiphase flow buoyancy is sometimes redefined as the difference between Archimedes force and
gravity to represent the net action of these two forces. Pressure force and viscous force belong to surface
forces. The former tends to accelerate fluid in the direction of pressure gradient, and the latter tends to
dampen the velocity gradient. Surface tension acts on a line or curve element and tends to minimize the
surface energy of a bubble or droplet. However the above classification is not very unambiguous since
buoyancy force results from the hydrostatic pressure difference, and other hydrodynamic forces like
drag and lift forces represent the macroscopic effects of pressure and shear stress around the particle.
The resultant force exerted by the surrounding fluid on a particle is given by integrating the pressure
and shear stress over the particle surface
F= ∫Ap
(− pII + 2µD)indS 1.1
where
1
D=
2
(
∇v + (∇v)T . ) 1.2
For creeping flows where Reynolds number is less than about 0.1, the pressure distribution at the
surface of a sphere can be analytically derived as (ref., e.g., Bird, et al., 1963)
3 µv∞
p |r = R = p0 − ρ f gR cos θ − cos θ 1.3
2 R
where p0 denotes the pressure far away from the sphere and v∞ the approaching velocity of fluid. When
the three terms on the right hand side of Eq. 1.3 are substituted into Eq. 1.1 and the integration performed,
the first term vanishes and the second term gives buoyancy force Fb and the third term gives the so-called
form drag Ff, as shown below:
4 3
Fn = Fb + F f = πR ρ f g + 2πµRV∞ 1.4
3
The shear stress distribution on the sphere surface is
130
3 µv∞
τ rθ |r = R = sin θ 1.5
2 R
Substitution of Eq. 1.5 into the shear stress part of integral in Eq. 1.1 gives the “friction drag”
Ft = 4πµRv∞ 1.6
Hence the resultant force can be given by the sum of Eq. 1.4 and Eq. 1.6
4 3
F= πR ρ f g + 6πµRV∞ 1.7
3
From Eq. 1.7 the resultant force can be physically classified into a buoyancy force and a hydrody-
namic force.
For unsteady creeping flows, Basset derived the hydrodynamic force as
 1 dV p t dV (τ) dτ 

dτ(−e r )
p
Fh = 6πµRV p + ρ f V p + 6 πµρ f R 2 ∫ 1.8
 2 dt 0
t−τ 
where Vp(τ) represents the instantaneous velocity of the sphere and er the unit vector in the direction of
relative velocity (Wörner, 2003). For steady flows, the second and third terms vanish and the first term
represents the Stokes drag force. The second term represents the force required to accelerate the fluid
surrounding the particle (added or virtual mass force) and it seems that particles resist accelerations
stronger than is apparent from their actual mass. The third term represents the force due to the lagging
boundary layer development with changing velocity (Basset force).
The hydrodynamic force can be decomposed into a component with direction opposite to the rela-
tive velocity, and a component normal to the relative velocity. There are no normal components in Eq.
1.8 for creeping flows. For more general cases in gas-solids flow, the normal component can lead to
transversal motion and involves Magnus force and Saffman force. The Magnus force is related to the
rotation of particles placed in a uniform flow field. The rotation results in an increase of velocity on
one side and a decrease on the other, thereby causing an asymmetrical pressure distribution around the
particle and a transversal force. Saffman force is caused by the pressure distribution around the particle
immersed in a non-uniform flow field without necessary particle rotation. The above two forces are
generally denoted as lift force.
1.2 forces independent of slip velocity: buoyancy
According to Archimedes’ principle, buoyancy equals to the weight of the fluid displaced by the object
wholly or partly immersed in a fluid, that is,
Fb = ρ f Vg 1.9
131
where ρf denotes the density of fluid and V denotes the volume of the displaced fluid. Buoyancy reduces
the apparent weight of that object and therefore it is generally easier to lift the object up through the
water than it is to pull it out of the water. For hydrostatic conditions at which a particle suspends in fluid
at rest, buoyancy can also be defined as the pressure gradient in the fluid around the object resulting
from the gravity field on the fluid, and momentum conservation for fluid gives
V ∇p = V ρ f g 1.10
While the left hand side of Eq. 1.10 represents the resultant force on the body of volume V due entirely
to the presence of the fluid, the right side coincides with the definition of Archimedes’ principle. Both
the left and right hand sides of Eq. 1.10 can be used to define buoyancy when the pressure gradient is
completely driven by gravity field.
For systems where a large number of particles are supported by their surrounding fluid in a state of
rest or uniform motion, the total pressure drop supports the total weight of the bed,
−∆p = ρ m gh = α f ρ f + (1 − α f )ρ p  gh 1.11

 
where ρm denotes the mixture density of particles and fluid, and αf represents the volume fraction of
fluid. This can be obtained either from experimental verification or mechanical energy balance analysis.
Foscolo et al. (1983) and Foscolo & Gibilaro (1984) argued that the buoyancy force experienced by an
individual particle in this case arises from the total pressure drop and can be given by
Fb = ρ mVg 1.12
It looks as if the particle would suspend in a pseudo-fluid at rest with a density ρm, and the pressure
gradient influenced by the presence of other particles becomes ρmg.
Epstein (1984), Clift et al. (1987), Fan et al. (1987) and Jean & Fan (1992) argued that the buoyancy
force should still maintain its original definition of Archimedes’ principle, that is, ρfVg which represents
the pressure gradient in a static fluid in the absence of fluid, and the calculation of the buoyancy based
on the bulk density should be restricted to the case when fluidized particles are much smaller than the
immersed object itself. Since the pressure gradient in fluid can also be induced by the relative motion
between particles and fluid in addition to gravity, Clift et al. (1987) attributed the total pressure gradient
to conservative buoyancy and dynamic drag:
−∆p = ρ f gh + gh(1 − α f )(ρ p − ρ f ) 1.13
The former term at the right hand side represents the conservation buoyancy reflecting the contri-
bution due to gravity, whereas the latter stands for the form drag exerted on the particle by a flowing
fluid and this term gives rise to mechanical energy dissipation. Fan et al. (1987) pointed out that only
the static contribution in the pressure distribution is pertinent to the buoyancy and suggested that the
132
expression given by Eq. 1.12 can be referred as apparent or effective buoyancy in order to distinguish
from the classical gravitational buoyancy defined by Eq. 1.9.
The definition of buoyancy is actually pertinent to the expression of drag force and momentum equation.
Jackson (2000) gives a systematic analysis on the different ways of decomposition of the resultant force
and the intrinsic relation of buoyancy with momentum equation and other forces due to relative motion
of particles and fluid. If the system is very dilute with few particles, the fluid-particle interaction can be
treated with one-way coupling, i.e., neglecting the action of particles on fluid. In this case, the pressure
gradient would be independent of the relative motion, and therefore the definition of buoyancy either
from pressure gradient or Archimedes’s principle makes little difference on the additional hydrodynamic
part in the resultant force. But for the dense cases like fluidized bed systems, the relative motion can
also contribute a part to the pressure gradient in addition to gravity and inertial terms. Therefore differ-
ent definition of buoyancy would lead to different ways for partitioning the resultant force exerted on
particles into buoyancy and interaction forces, and in turn leads to different appearances of momentum
equations and forms of interaction forces. Khan & Richardson (1990) discussed the various forms of
drag relationships depending on how one considers buoyancy and relative velocity.
1.3 forces related to velocity gradient and acceleration:

added Mass, lift and basset forces
(1) Added Mass Force
Particles or bubbles moving in a non-uniform velocity field can accelerate some of the surrounding fluid,
and the corresponding reacting force of displaced fluid on this acceleration due to its inertia is called
added or virtual mass force. It should be emphasized that particles or bubbles moving with a constant
terminal velocity can also accelerate surrounding fluid, but this acceleration of fluid is driven by fluid
viscosity rather than added mass force. For a particle moving in a static inviscid fluid, the added mass
force can be analytically derived as
4 3 d(u f − u p )
Fam = πR ρ f 1.14
3 dt
In some Eulerian simulation of particle-fluid flow, the averaged added mass force is usually expressed as
d(u f − u p )
Fam = −α f α p ρ mCam 1.15
dt
where Cam stands for the added mass coefficient and is often assumed to be 0.5, and ρm denotes the
volume-weighted mixture density of particles and fluid.
The role of added mass force in multiphase flow modeling is not very clear yet. For gas-solids flow,
Zhang & VanderHeyden (2002) reported that the added mass of particles can be negligible because of
the larger density difference between gas and solids, but the added mass force of meso-scale structures
derived from a two-average approach for conservation equations is in contrast quite important. The
133
two-average approach is usually introduced as a so-called filtered model, and the maximum generalized
added mass coefficient was reported to increase with decreasing the filtered frequency defined as the
frequency up to which phenomena can be accurately resolved and spatially or temporally determined
(De Wilde, 2005, 2007). For gas-liquid flow, Mudde & Simonin (1999) reported that their 3D simulation
of bubbly flow with only drag force resulted in an under-prediction of the amplitude and frequency of
the bubble plume oscillations, while inclusion of added mass force can give satisfactorily comparable
results with experiments. However, Oey et al. (2003) found that right magnitude of the oscillation of
meandering plume can be obtained without using added mass force.
(2) Lift Forces
Lift forces are generally normal to the relative motion of fluid and particles and involves Magnus force,
Saffman force and the force caused by a slanted wake behind a distorted bubble in a shear field (Tomi-
yama et al., 1995; Chen, 2004). The Magnus lift force is related to the particle or bubble rotation in a
uniform flow field which gives rise to the velocity difference on the two sides of the particle and an
asymmetric pressure distribution around the particle. Taneda (1957) reported that a lift force opposite
to the Magnus force was experienced by the spheres with certain rotation speed in a range of Reynolds
number, and can be explained by a transition of the boundary layer from laminar on one side to turbu-
lent on the other side of the bubbles or particles. A second lift force, the Saffman force, is caused by the
pressure difference on a non-rotating particle under a shear velocity field. It is reported that the lateral
lift force is an order of magnitude larger than Magnus force at low Reynolds number unless the rotating
speed of the particle is much larger than the shear rate.
The mean lift force is usually expressed in Eulerian simulation of particle-fluid flow as
FL = α f α p ρ f CL (u p − u f ) × ∇ × u f 1.16
where CL denotes the lift coefficient and is often assumed to be 0.5. The lift forces are also often neglected
both in gas-solids and gas-liquid flow simulation because they are less important in magnitude compared
to the drag force. But Lucas et al. (2005) reported that the lift force may be one important reason for the
transition from homogeneous to heterogeneous flow in gas-liquid bubble columns.
(3) Basset Force
The Basset force or history force is caused by the lagging response of boundary layer to particles accel-
eration. Particles therefore experiences a unsteady force due to the instability of boundary layer. Because
it depends on the acceleration history up to the present time, Basset force is also called history force and
defined as the integral of all past particle accelerations. The Basset force constitutes an instantaneous
flow resistance. It arises from the generation of vorticity at the surface of the particle to diffuse into sur-
rounding fluid to describe the initial motion of particle (Seville et al., 1997), and becomes substantial
when the particle is accelerated at a higher rate and the observed drag force becomes many times that
of the steady state (Thomas, 1992). A brief review of Basset force was given by Mabrouk, et al. (2007)
and the force can be calculated by
134
3d p2µ π t p du / dt
FB = CB ∫ r
dτ 1.17
2 γ tp0
tp − τ
where tp0 refers to the time when particles start to be accelerated, ur the interstitial slip velocity, γ kine-
matic viscosity and
0.52 An 3
CB = 0.48 + 1.18
(1 + An)3
and
dur / dt
An = dp 1.19
ur2
1.4 forces related to slip velocity: drag
The drag force acts in the direction opposite to the relative motion of particle and fluid. As discussed in
the preceding sections, integration of pressure and viscous stress around the surface of the particle can
give form drag and friction drag, respectively. Drag force can be defined as
1
FD = ρ f ACD Ur2 1.20
2
where A is the frontal area presented to the flow by the particle, Ur the relative velocity and CD is called
the drag coefficient. In particular, the drag coefficient for a single particle in a infinite flow field in the
absence of other particles is a function of Reynolds number, particle shape and orientation, and is called
the standard drag coefficient (CD0).
For a smooth sphere immersed in a creeping flow at Re<0.2, the friction drag is two times the form
drag and CD can be given by stokes law neglecting inertial effect
24
CD0 = 1.21
Re
The Oseen’s formula taking account of inertial effect is suitable for Re<5:
24  3 
CD0 = 1 + Re 1.22
Re  16 
With further increase of Reynolds number, the linearity of drag curve is broken and the contribution
of form drag becomes more important and finally dominates over the friction drag. For 0.2<Re<1000,
the Schiller & Naumann(1933) correlation is well consistent with experiments:
135
24
CD0 = (1 + 0.15 Re0.687 ) 1.23
Re
The drag coefficient approaches a constant in the range of 750<Re<3×105 known as inertial or
Newton’s regime
CD0 = 0.44 1.24
Transition from laminar to turbulent flow in the boundary layers on the forward portion of the sphere
causes a dramatic decrease in drag coefficient at a critical Reynolds number about 2-3×105. The boundary
layer is better able to resist the adverse pressure gradient on the rear of the sphere, and hence separation
is delayed and the wake is smaller, causing less form drag.
There have been numerous correlations for standard drag coefficient in literature, as summarized by
Lapple and Shepherd (1940), Clift et al. (1978), Flemmer and Banks (1986), and recently by Almedeij
(2008) and Chen (2009). Clift et al. (1978) recommended drag correlations for 10 subintervals of the
whole range of Re. Flemmer and Banks (1986) proposed a drag coefficient correlation suitable for a
wide range of Reynolds number (Re<3×105).
24 E
CD 0 = 10 1.25
Re
where
0.124
E = 0.261 Re0.369 − 0.105 Re0.431− 1.26
1 + (log10 Re)2
A comprehensive summary of various correlations for standard drag coefficient can also be found in
Perry’s Chemical Engineer’s Handbook (Perry and Green, 2007).
The aforementioned correlations are generally obtained from experimental or theoretical analysis
for spherical particles. For irregularly shaped particles, it is observed that the departure from spherical
shape causes a decrease in terminal velocity and a higher drag coefficient. Several shape factors such as
volume equivalent sphere diameter, surface equivalent sphere diameter, particle aspect ratio and particle
sphericity can be used to describe the effect of the shape on drag coefficient. For example, Tran-Cong
et al. (2004) proposed the expression for non-spherical particles as
2
d 
 0.42  A 
 
0.687
24 d A  0.15  d A  d n 
CDA = 1 +  Re  + 1.27
Re d n  c  d n     
−1.16 
   4  dA 
c 1 + 4.25 × 10  Re  
  d n 
 
 
136
where dA/dn represents the flatness ratio and c stands for the particle circularity. The volume equivalent
sphere diameter dn is defined by
d n = 3 6V π 1.28
and the surface equivalent sphere diameter is defined by
d A = 4 Ap π 1.29
where V and Ap stands for particle volume and projected area of the sphere respectively. The particle
circularity c is also called surface sphericity is defined by
c = πd A Pp 1.30
where PP is the projected perimeter of the particle in its direction of motion. Review on drag coefficients
for non-spherical particles can be found in Tran-Cong et al. (2004), Loth (2008), and Holzer and Som-
merfeld (2008).
2. drag fOrce MOdels: aPPlicatiOns and liMitatiOns
2.1 Many-body effects
Due to the influence of its surrounding particles, the effective drag coefficient CD for a particle is generally
different from the standard drag coefficient CD0. The surrounding particles may either hinder or accelerate
the motion of the particle, depending on local flow structures in gas-solids systems. The standard drag
coefficient CD0 is related to terminal velocity UT with
4d p (ρ p − ρ f )
CD 0 = , 2.1
3ρ f U T2
and in a similar manner the effective drag coefficient CD can be related to the slip velocity US with
4d p (ρ p − ρ f )
CD = 2.2
3ρ f U S2
if the slip velocity Us remains constant and therefore a force balance between buoyancy and drag can be
established. Note that the form of CD is also pertinent to the definition of slip velocity US and buoyancy.
In literature one can denote US as either the superficial slip velocity based on the overall cross-section
137
of the bed or the interstitial slip velocity based on local voidage. As discussed in the preceding sections,
both mixture density and fluid density were used to calculate the buoyancy. The effect of buoyancy
definition and slip velocity on the form of drag coefficient was discussed by Khan & Richardson (1990).
It is common to relate the two drag coefficients by a correction factor:
CD = f ⋅ CD 0 2.3
Numerous correlations were presented to relate these two drag coefficients, mostly via the voidage
e One of the well known correlation is proposed by Wen & Yu (1966) from the bed expansion experi-
ment of particulate fluidized bed,
−4.7
f =ε 2.4
for which the superficial slip velocity
U Up 
  .
U S = ε ⋅  g −  2.5
 ε 1− ε 
should be used. For many other gas-solids systems with typical heterogeneous structure like cluster
formation in circulating fluidized beds, the applicability of this correction factor is still an open question
(Yang et al., 2003). Though different approaches are developed, accurate formulation for the effective
drag coefficient taking into account of corresponding many-body effects is a challenging issue for com-
putational modeling of gas-solids flow.
Actually the effective drag coefficient CD is extremely important in triggering the heterogeneous flow
structure formation and evolution in gas-solids flow modeling. Previous studies, however, generally
focused on the closure of particle-particle interaction. It is only in recent decade that much attention is
paid in this respect. By using the Energy-Minimization Multi-scale model, Li et al. (1993) calculated the
slip velocities and drag coefficients for the cluster phase, dilute broth and the interphase between them,
showing great difference among different phases. For simplified configurations of dilute and dense phases
with the same average voidage, as shown in Figure 1, the calculation indicated significant dependence
of effective drag coefficient CD on structural changes and the difference may reach several orders (Li et
al., 1998). Another example of many-body effects involving meso-scale structure is illustrated in Figure
2 by using the so-called pseudo particle simulation (Li & Kwauk, 2003). The occurrence of meso-scale
structure can reduce the effective drag coefficient, and the non-uniform distribution of these structures
leads to further decrease. This demonstrates that the effective drag coefficient is essentially not only a
function of Reynolds number, voidage, and physical properties of particles and fluid, but is germane to
some parameters which can reflect the topology or structural changes.
The difference between particulate and aggregative fluidization has been identified very early (Wil-
helm & Kwauk, 1948; Lewis et al., 1949). The roles of meso-scale structure and effective drag coefficient
in computational modeling of gas-solids flow have been gradually recognized in recent decade. Agraw-
al et al.(2001) reported that the coarse-grid simulation commonly used in the Eulerian simulation of
138
Figure 1. Effects of simplified flow structure on effective drag coefficient in a unit with the same amount
of particles and the same gas flow rate. (Li et al., 2001 ©2001, American Chemical Society. Used with
permission)
Figure 2. Effects of meso-scale structure (cluster) on effective drag coefficient in a unit with the same
amount of particles and the same gas flow rate. (Li & Kwauk, 2003 ©2003, Elsevier. Used with permission.)
gas-solids flow completely ignored the sub-grid structure and would overestimate the effective drag
force. Zhang & VanderHeyden (2002) proposed a two-averaging approach for conservative equations
to resolve the added-mass force and drag force due to meso-scale structure from those at the particle
length scale, showing their essential roles on the macroscopic momentum equations. Grace & Tuot
(1979) extended the linear stability analysis of Jackson (1963) to more dilute systems, suggesting that
all systems are unstable to the growth of voidage waves. Helland et al. (2000) studied the influence of
porosity functions of correction factor for drag coefficient with a Eulerian-Lagrangin approach, indicat-
ing that a homogeneous fluidization was unstable to small porosity perturbations due to the non-linear
drag inducing cluster formation, and the collision parameters had important influence on the shape and
voidage of clusters. Li & Kuipers (2003) reported that the non-linear drag plays an essential role in pat-
tern formation of heterogeneous flow structure in dense fluidized beds. Here the non-linearity in drag
force arises from the group effect of particles in a cloud. Their discrete particle simulation demon-
strated the large drag acting on dense clusters and small drag acting on dilute phase, leading to the ac-
139
celeration of particles in dense clusters and deceleration in dilute gas. Zhang & Reese (2003) reported
that the averaged drag force needed not only to incorporate the influence of solid volume fraction but
also to address the effect of the random fluctuational motion of particles.
Yang et al. (2003, 2004) utilized the energy-minimization multi-scale model proposed by Li & Kwauk
(1994) to construct a structure-dependent drag model, and the CFD simulation clearly demonstrated the
difference between the Wen & Yu correlation and the structure-dependent drag model in the numerical
simulation of entrained solid flux and meso-scale flow structure in a circulating fluidized-bed riser.
Schemes to solve this issue were classified into four categories (Yang et al., 2004). The first category
includes the so-called direct numerical simulation (DNS) (e.g., implemented in the Lattice-Boltzmann
method (LBM)). By using these models, the interface between each particle and the surrounding fluid
can be tracked and the hydrodynamic interactions between interfaces need not to be modeled explicitly.
The ensemble averaging in the derivation of conservation equations for Eulerian-Eulerian approach
generates the interphase coupling terms including drag which needs closure (Drew, 1983). Therefore
analogous to the definition DNS in turbulence, the DNS here refers to the direct simulation of particle-
fluid interaction without resorting to closure equations. The second category implements the two-fluid
model with a higher grid resolution in an attempt to capture meso-scale structure. Conventional drag
coefficient correlations or standard drag coefficient can still be used. This is based on the assumption
that heterogeneity is weakened when reducing the grid size, but its physical validity needs further proofs.
For example, the assumption of solid as a continuum phase may be weakened when decreasing grid
size. The third category utilized the advantage of the first two categories of resolving fine structure, and
thus theoretically cascade description, i.e., extracting closure correlations for the effective drag coef-
ficient from microscopic or direct numerical simulations, may suggest a way to explore the multi-scale
heterogeneity. The fourth category is to establish a model capable of describing the main characteristics
of heterogeneous multi-scale structure and the effective drag coefficient can then be calculated based
on structure parameters.
2.2 experimental/empirical correlations
Numerous experimental or empirical correlations have been put forward over several decades for fluid-
ized bed systems and two different approaches emerged. Garside & Al-Dlbouni (1977) and Cox & Clark
(1991) reviewed some effective particle drag relationships. The first one is to modify the standard drag
coefficient CD0 to include the particle concentration or voidage effect. The Wen & Yu correlation, i.e., Eq.
2.4, is one of this category, showing that the effective drag coefficient CD increases as particles become
more concentrated in the fluid and hence the superficial slip velocity US for suspending particles drops
with the increase of particle concentration. Richardson & Zaki (1954) proposed that
n
U S = Utε 2.6
where
140
Table 1. Correlations of effective drag coefficient
Authors Formulation
C D = C D 0 f (ε )
53
Happel (1958) 3 + 2(1 − ε )
f (ε ) = 13 53
3 − 4.5(1 − ε ) + 4.5(1 − ε ) − 3(1 − ε )2
−4.7
Wen and Yu (1966) CD = CD 0ε
Barnea and Mizrahi (1973) CD = CD 0[1 + (1 − ε )1/3 ] / ε 3
CD = CD 0ε −3.8
Gibilaro et al.(1985) 17.3  ρu 2 L
∆P =  + 0.336 (1 − ε )ε −4.8
 Re  d
−4.8
Khan and Richardzon (1990) CD = (24 Re + 0.44)ε
FD = FD 0ε −β
Di Felice (1994)
β = 3.7 − 0.65 exp[−(1.5 − log Re)2 / 2]
CD = ε 1−n (24 Re−1 + 5Re−0.25 )

Xie (1997) 339 ( Ret ≥ 500)
n=
1 + 4.45Ret−0.1 (1 ≤ Ret < 500)

−m
CD = CD 0ε
Mostoufi and Chaouki (1999) m = 3.02 Ar 0.22 Re−t 0.33(d p d s )−0.40
1 1 1
CD (πd p2 ) (u 2ρ) = (πd p3 g )(ρ p − ρ) × (a ε 4.78 + (1 − a)ε 2.78 )
Yang and Renken (2003) 4 2 6
a = 0.7418 + 0.9670 Ar−0.5 1 < Ret < 50, 24 < Ar < 3000
4.65 Ret < 0.2

 −0.03
4.4 Ret 0.2 < Ret < 1
n= 2.7
4.4 Re−t 0.1 1 < Ret < 500

2.4 Ret > 500
The second approach derived the velocity-voidage relationship from the correlation of pressure drop
across a fixed or packed bed such as the Ergun equation. Table 1 lists some effective drag efficient cor-
relations for a particle in multi-particle systems.
Application of these effective drag coefficient correlations into each control volume of computa-
tional fluid dynamic modeling, one can obtain the total drag force for particles per unit control volume
and hence the effective drag coefficient for a control volume β. For example, the Ergun equation can be
extended to derive the correlation for β as below (Gidaspow, 1993)
141
(1 − ε )2 µ g (1 − ε )ρ g | u g − u p |
β = 150 + 1.75 2.8
ε d p2 dp
Since the Ergun equation is based upon data for a voidage range of 0.4-0.6 with accuracy dropping
off at larger voidages, Gidaspow (1993) proposed to use Wen & Yu equation for ε>0.8:
3 (1 − ε )ε
β= ρ g | u g − u p | CD 0ε−2.7 2.9
4 dp
Almost all of the above correlations are based on the experimental data for liquid-solid fluidized bed
or fixed-bed systems in which the flow structure is generally more homogeneous than that in gas-solids
systems. Their direct application to other gas-solids systems, particularly, the so-called fast fluidization
in circulating fluidized systems, is questionable as the multi-body effects on the effective drag coef-
ficient is quite different than that in liquid-solid or fixed-bed systems. This can be demonstrated from
the historical development of the hydrodynamic studies on fast fluidization and circulating fluidization
technology. It took several decades before the circulating fluidized bed was recognized as unique, differ-
ing from the bubbling fluidized bed and pneumatic conveying (Reh, 1971; Yerushalmi, et al. 1976). Li &
Kwauk (1980) presented the evidence of the existence of this flow regime by their experimental data of
an S-shaped axial voidage distribution with a gradual transition between a top dilute and a bottom dense
region. The effective slip velocity between gas and particle in this state is usually 20-30 times terminal
velocity. To explain this phenomenon, Reh (1971), Yerushalmi et al. (1976) and Li & Kwauk (1980) sup-
posed that most particles did not remain single, but tended to aggregate in the form of so-called clusters
or strands, which was then confirmed by experimental measurements through high-speed photography
(Yerushalmi & Squires, 1977), video camera and optical fiber (Li et al., 1991) and capacitance probes
(Brereton & Grace, 1993) later. The clusters are found to be irregular in shape with variable size, formed
and then degenerated dynamically due to wake effect, slowing down particles motion and hence causing
the larger slip velocity between fluid and particles. The effective drag coefficient is decreased with the
increase of particle concentration in a certain range. The cluster phenomenon and larger slip velocity
cannot be described by the correlations obtained from experimental data for liquid-solid fluidized bed
or fixed-bed systems such as that of Wen & Yu (1966) or Richardson & Zaki (1954).
Some researchers measured the effective drag coefficient considering the cluster effect and the
characteristics of fast fluidization. Gunn & Malik (1967) found that, if particles were grouped in ag-
gregates for a given rate of fluid flow and overall voidage of the system, the measured drag coefficient
decreased because of the decrease of the gas flow penetrating the aggregates and the increase of the
gas flow passing around the aggregates. Muller & Reh (1993) reported that the formation of particle
strands resulted in considerable drag reduction, leading to a longer acceleration region in risers. Based
on the experimental data for fast fluidization, Li (1994) proposed a correlation showing a sharp drop of
effective drag coefficient with increasing voidage, as shown below:
−13.49
CD / CD 0 = 0.0232ε 2.10
For pneumatic transport, the correlation becomes
142
−35.01
CD / CD 0 = 0.0633ε 2.11
Bai et al. (1991) proposed another correlation which leads to a deep valley in the curve of correction
factor vs. voidage
CD CD 0 = 1.68ε−0.253(Re Ret )−1.213(d p / D)0.105 2.12
where D denotes the diameter of fluidized bed and Ret the Reynolds number based on terminal velocity.
Cruz et al. (2006) obtained a correlation for Re<4.53×10-4 through regression analysis and the average
drag coefficient for a control volume in their CFD simulation was
3 (1 − ε )ε
β= ρ g | u g − u p | CD ε−2.7 2.13
4 dp
where e s is the concentration of the solids and
 2 × 10−3 
CD = 16.9 + ε s  − 2.5 × 106 Re− 44.4 2.14
 Re 
ρgd p | ug − u p |
Re = 2.15
µ lm
  u p − ug  
   − 31.2ε 2 
µ lm = 0.806 0.47 + ε s 19.3 − 3.7   2.16
 
 Ug 
s

 
Considering the effects of particle acceleration in the acceleration zone and Basset force, Mabrouk
et al. (2007) proposed a correlation from experimental data obtained in a internal CFB lab-scale using
sand and alumina particles for several superficial gas velocities:
CD = 4330 Re−p 2 2.17
where
ρ g d p ε (ug − u p )
Re p = 2.18
µ
Their numerical study with a one-dimensional model and this drag correlation is in good agreement
with experimental data. O’Brien & Syamlal (1993) presented an empirical correlation for two operat-
ing conditions based on pressure drop data and the Dallavalle correlation of standard drag coefficient:
143
3 (1 − ε )ε ρ g | u g − u p |  
β= 0.63 + 4.8 Vr, cor  2.19
 
4 Vr2, cor d p Re 
 
where
Vr, cor = (1 + mRe(1 − ε )e n )Vr , 2.20
n = −0.005( Re − 5)2 − 90(ε − 0.92)2 , 2.21
and
250 (Gs = 98kg ⋅ m−2 ⋅ s−1 )

m =  2.22
1500 (Gs = 147kg ⋅ m−2 ⋅ s−1 )

2.3 Microscopic simulation results
Microscopic simulation is capable of revealing the mechanism of gas-solids interaction at each particle
level and hence more accurate than empirical correlations. The weakness of this method lies in the enor-
mous computational cost and the complicated mesh-generation treatment for some methods. With the
quick development of computer hardware and parallel computational techniques, various microscopic
simulation can be found in literature such as direct numerical simulations (DNS), Lattice-Boltzmann
simulation (LBS) and other methods to investigate the drag coefficient and drag force in multi-particles
systems in recent decades.
Hu et al. (1996, 2001) developed a direct simulation of particle-fluid systems using the arbitrary
Lagrangin-Eulerian (ALE) technique. The moving body-fitted unstructured grids were frequently generated
with the motion of particles. The influence of particles on fluid was reflected by the non-slip boundary
condition of fluid flow field which was then computed by solving the fluid momentum equation, whereas
the influence of fluid on particles was reflected by the integration of the stress on each element around
the surface of particles. In this way, the drag forces are directly computed from the fluid flow and the
motion of fluid and particles are fully coupled.
Pan et al. (2002) studied the fluidization of 1204 spheres at Reynolds number in the thousands using
the distributed Lagrange-multiplier-based fictitious domain method (DLM) developed by Glowinski
et al. (1997). The basic idea of this method is to extend a problem on a time-dependent geometrically
complex domain to a simple stationary domain, and constraints of rigid-body motion of the particles is
enforced using DLM to represent the additional body force required to maintain the rigid-body motion.
This allows a simple mesh treatment and eliminates the need for dynamic reconstruction of mesh. A dis-
tributed Lagrange multiplier was employed to impose the velocity constraint on each particle boundary.
The log-log plots of fluidization velocity vs. voidage curve with simulation were found to have similar
trends with Richardson-Zaki correlation. However, these studies were limited to liquid-solid systems.
Unlike the above direct simulations based on numerical discretization of macroscopic continuum
equations, LBM is based on microscopic models and mesoscopic kinetic equations. Hill et al. (2001) and
144
Kandhai et al. (2003) performed LBM studies concerning the dependence of the drag force on Re and
εs and compared simulation with Wen & Yu and Ergun equations. Van der Hoef et al. (2005) measured
the drag force with LBM simulation on the spheres for a range of diameter ratios, mass fractions and
packing fractions and formulated a drag force relations for both monodisperse and polydisperse systems.
More detail about LBM study on drag force can be found in the second section of this book.
It should be noticed that the so-called discrete particle simulation based on Eulerian-Lagrangian ap-
proach also tracks the motion of each single particle. It has been extensively investigated for gas-solids
flow modeling (Tsuji, et al., 1993; Hoomans, et al., 1996; Xu and Yu, 1997), and some recent reviews
can be found in Deen et al. (2007) and Zhu et al. (2007, 2008). But unlike the microscopic or direct
numerical simulation aforementioned, the fluid phase equations are formulated and solved on the scale
of control volumes and therefore fluid-solid coupling resorts to some averaging treatment or empirical
correlations as used in Eulerian-Eulerian simulation, which may to some extent counteract its advantage
in particle tracking. The influences of effective drag coefficient on simulation have been studied by Kafui
et al. (2002, 2004), Feng et al. (2004) and Helland et al. (2007). In contrast to linear interpolations for
calculating the local fluid velocity at the position of each particle and the effective drag force experienced
by the particle, Xu et al. (2007) proposed an method for calculating the local fluid velocity by distributing
fluid flow within each cell according to weighted local porosities from pressure balance considerations.
So far, fully resolved simulation of dynamic gas-solids systems with high density ratio which will be
very helpful to understand the drag in such heterogeneous systems, is still very few. Further discussion
on this topic can be found in section 4.
2.4 Multi-scale Models
Application of microscopic simulation is constrained by the enormous computational cost to track the
motion of each particle. Since it is suitable for parallel computation, one can imagine that this approach
is promising in future either to directly compute the real gas-solids flow or extract useful correlations
for continuum-based two-fluid models. At present, a practical way is to establish a structure model to
take the effect of heterogeneous structure on effective drag force into consideration. But such struc-
ture models are usually based on some averaged approach with specified flow structure obtained from
experimental findings, and one need to pre-identify the main characteristics of flow structure in each
local control volume. Exact structure information cannot be obtained unless microscopic simulation is
utilized, and therefore how to retrieve the lost structure information is an challenging problem. Li et al.
(1999) pointed out that if one could not reach such a small scale at which there is only one particle in
the control volume, the so-called variational criterion or something equivalent must be established to
identify the prevailing state from all possible arrangements of particles in the control volume.
In the Energy-Minimization Multi-Scale (EMMS) model of Li & Kwauk (1994), a multi-scale analysis
leads to the resolution of the structure into a gas-rich dilute phase and a particle-rich dense phase. The
gas-solids interaction was resolved into what occurs inside dilute phase and dense phase at particle-scale
respectively, as well as that between dense clusters and dilute broth at the interface of clusters. The drag
force for the dense cluster phase includes a bypassing drag (ki) and a permeating drag (kc), and their
relative magnitude would be adjusted automatically depending on the structure heterogeneity dominated
by the stability condition. The two phases are characterized by the fluid and solids flow velocities in
each phase (Uc, Udc, Uf, Udf), and the voidage (εc, εf), while the heterogeneity is specified by the cluster
diameter (dcl) and dense phase fraction (f). Only six dynamic constraints are found for the eight struc-
145
ture parameters, that is, the momentum balance equations for both phases, the interphase pressure drop
balance equations, the continuity equations for the fluid and the solids, and finally the cluster diameter
correlation. To close the model, a stability condition is proposed to serve as the variational criterion.
Calculation of Li et al. (1993) by using the EMMS model demonstrates the dramatic difference among
the drag coefficients for particles in dense phase CDC and in dilute phase CDf, and that for dense clusters
in dilute broth CDi.
Yang et al. (2003, 2004) utilized the EMMS model to establish the relationship between effective
drag force and structure parameters, and CFD simulation was performed to compare the role of different
drag models. Since the detail of the EMMS model and its recent development will be presented in the
next section of this chapter, only the main idea is briefly introduced here.
When retrieving the structure information, the transfer parameters between phases such as the aver-
age acceleration and the average drag coefficient for the particles of a cell can be calculated. Yang et al.
(2003, 2004) adapted the steady EMMS model so that these transfer parameters were implicitly correlated
with structure parameters through the coupling of non-linear equations and optimization process related
to stability condition. In contrast to the empirical correlations employed in current two-fluid models,
the transfer parameters obtained in this way embody the information of heterogeneous structure, and is
therefore important for the simulation of computational fluid dynamics.
To incorporate this approach into the two-fluid model, a simplified treatment is proposed by Yang et
al. (2003, 2004) by assuming the dense phase voidage as a constant. The relationship of effective drag
coefficient and voidage obtained from the operating parameters for the whole bed was then extended to
each cell in computation. This transforms the previous implicit coupling into some explicit correlations
without losing the physical essence of the model, and hence circumvents the problem of computational
cost resulting from the iterative process and certain limitations of the model stemming from its original
derivation from global fluidized bed systems. For the operating conditions in their simulation, the drag
coefficient can be correlated as
 3 (1 − ε)ε
 ρ g | ug − up | C D 0 ⋅ ω(ε) (ε > ε * )
 4 d
β = 
p
2.23
 (1 − ε)2 µ g (1 − ε)ρ g | ug − up |
150 + 1 .75 (ε ≤ ε * )
 εd p2 dp
where ε* is equal to 0.74 and the correction factor is correlated for Ug=1.52 m/s and Gs=14.3 kg/m2 as
below
 0.0214
-0.5760 + (ε* < ε ≤ 0.82)
 4(ε − 0.7463)2 + 0.0044


 0.0038
ω(ε) = -0.0101 + (0.82 < ε ≤ 0.97) 2.24
 4(ε − 0.7789)2 + 0.0040


-31.8295 + 32.8295ε (ε > 0.97)


146
Figure 3. Comparison between the calculated outlet solid flux and the experimental data. (Yang et al.,
2003 ©2003, Elsevier. Used with permission.)
It should be mentioned that correlations different from Eq. 2.24 could be theoretically obtained for
other operating parameters following the above approach.
Then CFD simulation is implemented with the new drag model to simulate a air/FCC riser, indicating
that the EMMS-based drag model produces more reasonable results than the Wen & Yu/Ergun correlations.
The output solid flux predicted from the combination of the EMMS-based drag model (drag model B in
Figure 3) with the two-fluid model is close to the experimental measurement, whereas the predication
from the Wen & Yu/Ergun correlations (drag model A in Figure 3) is far beyond the experimental data.
Correspondingly, the meso-scale structure taking the form of particle strands or clusters can be clearly
observed with the former approach (Figure 4), reflecting the unique feature of gas-solids fast fluidiza-
tion. The latter hybrid correlations generated only the homogeneous structure due to the overestimation
of the drag coefficient. Since the same mesh size is used for both cases, the difference in the figure is
only caused by drag models in which model B takes the cluster effect into account while model A not.
More details can be found in Yang et al. (2003, 2004). It has been demonstrated that drag model B is at
least suitable for Geldart A-type particles (Jiradilok et al., 2006).
This new drag model can also be used to offer some physical understanding of the so-called “chok-
ing” in gas-solids fluidization and necessary parameters for specifying the hydrodynamics of CFB risers.
Choking which represents the regime transition between dilute transport and dense fluidization is char-
acterized by a dilute top on the transition and bottom dense regions of the axial voidage distribution,
reflecting the maximum solid carrying capability under given superficial gas velocity. Yang (2003) and
Yang et al. (2004) first captured this important phenomena with CFD simulation integrated with the
EMMS-based drag model, as illustrated in Figure 5. The superficial gas velocity Ug was kept constant
(1.52 m/s) at the bottom inlet and the two different solid inventories (15kg and 20kg) were set by impos-
ing two initial bed heights evaluated from the experimental solid hold-ups in the riser. The solid output
flux Gs was dynamically measured at the top outlets of the bed and fed back to the bottom-inlet instan-
147
Figure 4. Snapshot of voidage distribution. (Left) Drag model A; (Right) Drag model B. (Air/FCC sys-
tem, Ug=1.52m/s, Gs=14.3kg/m2s, d=54μm) (Yang et al., 2003 ©2003, Elsevier. Used with permission.)
taneously. It can be seen from Figures 7 and 8 that when the solid inventory is increased from 15 kg to
20 kg, the output flux Gs is almost invariable and the cross-sectional voidages in both the dilute top
region and the dense bottom region are almost coincident. The only difference between the two axial
voidage distribution curves is the inflection point which moves up when increasing solid inventory.
These simulations indicate that the system reaches a state of saturation for entraining particles despite
the increase of solid inventory from 15 to 20 kg, and is in agreement with the experiments of Li et al.
(1998) and Li & Kwauk (1994) in which the relationship of Ug, Gs and solid inventory I is depicted
with a saddle-shaped curve and the horizontal segment wedged between the curve slopes. The two points
at the horizontal segment, corresponding to the two inventories in the simulation, indicate the same
solid output flux. But when solid inventory is further increased from 20 to 35 kg, the output flux is in-
creased once again and the inflection point moves to the top and the dilute region almost disappears.
The whole saddle-shaped area was then simulated by Wang et al. (2007a) with the model proposed by
Yang et al. (2003, 2004), and Lu et al. (2007) simulated the hydrodynamics of an industrial FCC reactor.
Further discussion on choking simulation with this model is presented in Yang et al. (2005, 2006). Wang
and Li (2007) proposed another version of the EMMS-based drag model which will be described in the
next section.
3. Multi-scale MOdel fOr drag fOrce
3.1 characterizing of Heterogeneity in gas-solids flows
Gas-solids flows are normally heterogeneous, with various forms of structures excited over a wide spec-
trum of spatial and temporal scales. If the micro-scale is defined as the scale with respect to the smallest
space being observed, e.g. single particles or one grid in computation, and the macro-scale is defined as
the scale with respect to cross-section averaging over the whole domain, then, the wide span of scales
148
Figure 5. Simulation of choking and the relationship between Ug, Gs and solid inventory I. 1: I=15kg;
2: I=20kg; 3: I=35kg (Adapted from Yang 2003, Yang et al., 2003, 2004)
in between can be named the meso-scale, which actually represents the spatio-temporal characteristics
of the multi-scale structures. Consequently, in what follows we may refer to the “structure” tantamount
to the “meso-scale structure” in most discussions.
On the meso-scale, structures can be observed in various forms such as bubbles, clusters and stream-
ers, which deform, break up and merge constantly. These structures can be associated with the inertial
instability due to gas-solids slip and the instability due to inelastic collisions between particles (Agrawal
et al., 2001), as well as the global constraint with respect to operating and boundary conditions (Gi-
daspow & Mostofi, 2003; Wang & Li, 2007). In most situations, these structures are shapeless, their
interfaces being hardly clear-cut, and in fact, big controversy still remains on their definitions (Manyele
et al., 2002) and measures (Reh & Li, 1991). To account for their effect on the interphase momentum
transfer, nonetheless, one may assume the multi-scale structure can be relaxed on two scales, in terms
of, e.g. for a circulating fluidized bed (CFB), cluster scale and single-particle scale, respectively. Such a
division can be supported by experiments that two kinds of phase structures exist in bubbling, turbulent,
circulating fluidized beds and even in pneumatic transport, in which the dilute phase structure (void)
is characterized by a continuous gaseous phase with a log-normal solids fraction distribution and the
dense phase structure (cluster) is characterized by a continuous solids phase with a normal solids frac-
tion distribution (Lin et al., 2001).
The term of meso-scale is not so distinct in that the meso-scale structure in gas-solids flows may
overlap the micro-scale for continuum-based pseudo-fluid approaches. In fact, one of the basic charac-
teristics of the multi-scale structure in two-phase flow and granular flow is the lack of scale separation
between the micro- and meso- scales (Goldhirsch, 2003; Wang & Li, 2007), whose influence depends
on, however, the operating conditions and physical properties. As shown in the scale analysis for fluid-
ized beds (Wang & Li, 2007) and certain experimental data (Kunii & Levenspiel, 1991), the meso-scale
structure in form of bubbles or voids is normally larger than the typical grid size in TFM simulation,
which implies weak demand of sub-grid modeling of structures for bubbling fluidized beds, though
this advantage may be violated near the wall region by strong interference of boundary constraints. For
149
dilute two-phase flows such as those encountered in risers, however, the meso-scale structure in forms
of clusters spans a wide range of scales. It can even be identified in a small space as little as the size of
a few particle diameters. In practice, even two particles that align to or cross over the streamwise direc-
tion enable significant difference on the interphase momentum transfer (Mueller & Reh, 1993), while
this size or scale is smaller than the micro-scale required by the continuum assumption. In this regard,
unlike the case of Navier-Stokes equations for single phase flow where sub-grid contribution goes to
zero as the grid size approaches the Kolmogorov scale, the sub-grid contribution never vanishes in the
framework of two-fluid modeling (Wang et al., 2009). The structure-dependent drag force, therefore,
is inevitable for simulations of the riser and the other flow systems without distinct scale separations.
In what follows we will focus on an example of multiscale approaches to present structure-dependent
drag force modeling.
3.2 energy Minimization Multi-scale Model (eMMs) and its solution
To account for the multiscale structure or heterogeneity in a CFB, the energy-minimization multi-scale
(EMMS) model was proposed by Li and Kwauk (1994, 1988) for the time-mean global behavior. EMMS
model characterizes the multiscale structure in riser flows with a bi-scale resolution, i.e. the cluster scale
in terms of particle-rich dense phase (marked by the subscript “c”) and the dispersed particle scale in
terms of fluid-rich dilute phase (marked by the subscript “f”). Assuming overall balance between ef-
fective gravity and drag force, the hydrodynamic equation set of EMMS model, Φ(X), is as follows:
Force balance for particles in the dense phase:
Φ1 ( X ) = nc Fc f + ni Fi − f (1 − ε c )(ρ p − ρ g ) g = 0 , 3.1
Force balance for particles in the dilute phase:
Φ2 ( X ) = n f F f − (1 − ε f )(ρ p − ρ g ) g = 0 , 3.2
Pressure drop balance between the dense and the dilute phases:
Φ3 ( X ) = n f F f + ni Fi (1 − f ) − nc Fc = 0 , 3.3
Mass balance of particles:
Φ4 ( X ) = U p − U pf (1 − f ) − U pc f = 0 , 3.4
Mass balance of gas:
Φ5 ( X ) = U g − U gf (1 − f ) − U gc f = 0 , 3.5
Correlation for cluster diameter:
150
Table 2. Summary of relevant parameters for EMMS model(Adapted from Li & Kwauk, 1994)
Interaction between the dilute and the

Dense phase Dilute phase
dense phases
U pc ε c U pf ε f U pc ε f
Superficial slip velocity U sc = U c − U sf = U f − U si = (U f − )(1 − f )
1 − εc 1− ε f 1 − εc
Characteristic
Re c = ρ g d pU sc / µ g Re f = ρ g d pU sf / µ g Rei = ρ g d clU si / µ g
Reynolds number
Number density of particles or

clusters
(
nc = 6 (1 − ε c ) πd p3 ) (
n f = 6 (1 − ε f ) πd p3 ) ni = 6 f ( πd ) 3
cl
Standard drag coefficient for single 24 3.6 24 3.6 24 3.6

CD 0, c = + CD 0, f = + CD 0,i = +
particles Re c Re0c.313 Re f Re0f.313 Rei Rei0.313
Effective drag coefficient for par-

CD, c = CD 0, c ε−c 4.7 CD, f = CD 0, f ε−f 4.7 CD,i = CD 0,i (1 − f )−4.7
ticles or clusters
Drag force acting on particles or πd p2 ρ g πd p2 ρ g πd cl2 ρ g

Fc = CD, c U sc2 F f = C D, f U sf2 Fi = CD,i U si2
clusters 4 2 4 2 4 2
d cl U p (1 − ε max ) − U mf + U p ε mf (1 − ε mf )

Φ6 ( X ) = −   = 0, 3.6
dp N st g ⋅ρ p (ρ p − ρ g ) − U mf + U p ε mf (1 − ε mf )

 
where εmf, Umf are the voidage and gas velocity at the minimum fluidization state, respectively, εmax is the
maximum voidage that allows the existence of clusters. For a first approximation, it takes on value of
0.9997 for fine particles fluidized by ambient air (Matsen, 1982). The relevant parameters in Equations
(1~6) are summarized in Table 2.
The set of independent variables, X, in these six equations encompasses eight unknowns, i.e.. {Uc,
Upc, Uf, Upf, εf, εc, f, dcl}. The mismatch between equations and unknown variables means infinite solu-
tions in the sense of hydrodynamics. To close these equations, Li and Kwauk (1994) argued that certain
stability condition should be obeyed to sustain the dissipative structure. A variational criterion expressed
as Nst→min was hence proposed as follows:
ρ − ρg  
N st =
Wst
= p U − ε f − ε f (1 − f )U  ⋅ g = min , 3.7
(1 − ε )ρ p ρp  g 1− ε gf 
 
which can be explained by the “compromise”, or in other words, mutual optimization, between gas
dominance (Wst→min) and particle dominance (ε→min) (Li, 2000; Ge & Li, 2002). Here, the particle
dominance is to minimize the potential energy of particles and then to reach minimal voidage, while
the gas dominance is to minimize the energy consumption for suspending and transporting particles in
unit volume. Later exploration on the underlying physics verified this critical presumption through a
pseudo-particle method simulation (Zhang et al., 2005). Thus, the total variable set was closed by a set
151
Figure 6. The EMMS solution and explanation of the choking phenomenon. (a): Variation of energy
consumption for suspending and transporting particles Nst with voidage in clusters εc for different solids
flux Gs (Air-FCC system, Ug=1.5m/s; dp=54 μm, ρp=930 kg/m3; εmax=0.9997, εmf=0.5). Solutions
are indicated by the lowest points on the curves. At choking point (Gs≈10.5kg/m2s), two solutions could
coexist (Adapted from Ge & Li, 2002 and Wang et al., 2007b). (b): The intrinsic flow regime diagram
for the air-FCC system calculated by using the EMMS model. The intrinsic flow regime diagram is
independent of the riser height (Wang et al., 2008b ©2008, Wiley. Used with permission.)
of conservation equations (Eqs 3.1-3.6) and a stability condition that is particular to circulating fluidized
beds. Based on the idea that if two variables, e.g. εc and εf, are provided, the six non-linear equations
can be solved by the remaining six variables (Meynard, 1997), a rigorous full-search scheme has been
developed to analyze the solutions of the model (Ge and Li, 2002), as follows:
1. Calculate Usf from Eq. 3.2;

2. Select a trial value for f;
3. Calculate Usc from Eq. 3.1;
4. With the definitions of Usf and Usc, calculate Uf, Uc, Upf and Upc from Eqs. 3.4 and 3.5;
5. Calculate dcl from Eq. 3.6;
6. Calculate Usi,1, Usi,2 from the definition and Eq. 3.3, respectively, denote their difference ΔUsi=|Usi,1-
Usi,2| as the convergence criterion;
7. If not converged, go back to step 3 with new trial value of f.
Finally, the solution of the model can be achieved only when the minimum of Nst is satisfied for a sweep
over all pairs of εc and εf within the range of [εmf, εmax]. Calculations show that Nst→min always requires
the maximization of εf. On the other hand, the variation of Nst with εc is not always monotonous in that
Nst may reach its minimum at εc=εmf and certain higher value of εc. Especially, it has given fresh insight
to the long-lasting dispute on the “choking” phenomenon in CFB risers (Ge & Li, 2002). As shown in
Figure 6 (a), the choking, which occurs under the operating conditions of (Ug=1.5m/s, Gs=10.5kg/m2s),
has two solutions with the same value of minimum Nst, corresponding to, respectively, the coexistence
of the dilute transport denoted by the point (B) and the dense transport denoted by the point (A). Under
the same gas velocity, higher solids flux, e.g. Gs=20.0kg/m2s, will result in the complete dense transport
152
in the riser which is denoted by the point (D), while lower solids flux, e.g. Gs=5.0kg/m2s, will result in
the complete dilute transport denoted by the point (C). Thus, the “choking” transition actually represents
a jump between the dilute and the dense transports with bi-stable states, both stable branches satisfying
the minimum requirement of Nst. The relevant calculation is now available on the Internet (http://pevrc.
ipe.ac.cn/emms/emmsmodel.php3).
To reveal the “choking” physics without geometric limitation, EMMS model can be used to calculate
the intrinsic flow regime diagram (Wang et al., 2007a; Wang et al., 2008b). Here the flow regime diagram
is capped “intrinsic” because it is determined purely by hydrodynamics without taking riser height into
account. As shown in Figure 6 (b), the intrinsic flow regime diagram is illustrated with a set of iso-
aeration which relates the solid flux Gs with the averaged solids fraction of the riser εs0 at constant Ug.
The choking transition occupies a bell-shaped area, in which the dense transport coexists with the dilute
transport and the iso-aeration levels off with constant Gs. The summit of the choking area or the critical
point separates the choking transition from the continuous transition. Owing to the freedom from the
riser-height limitation, the critical point rises to a very high position, i.e., Ug≈10m/s and Gs≈1000kg/
m2s. It should be noted that, however, the intrinsic flow regime is hard to reproduce in experiments as
much longer tube than those reported in literature was needed to eliminate the geometrical effects.
3.3 extended eMMs Models and the drag force expression
EMMS model was proposed for time-averaged global behavior under force balance. To account for the
time-dependent inertial effects in local space, EMMS was extended to the sub-grid level by taking into
account the inertial terms (ac, af, ai) (Wang & Li, 2007). This extended model features a two-step scheme.
The first step is to determine the cluster parameters in terms of its diameter and voidage (dcl and εc) with
the global stability condition Nst→min under given Ug and Gs. The reason of isolating the determination
of dcl and εc from that of velocities and inertial terms lies in that these meso-scale structure parameters
were assumed dependent on the macro-scale conditions. This presumption can be verified in part by the
good correlation between the cross-section averaged solids fraction and (dcl, εc) in experiments (Harris
et al., 2002). The hydrodynamic equations involved in the first step are as follows:
nc Fc − (1 − ε ) (ρ p
− ρ g ) ( ac + g ) = 0 , 3.8
ni Fi − f (ρ p − ρ g )(ε − ε c )(ai + g ) = 0 , 3.9
n f F f − (1 − ε f )(ρ p − ρ g )(a f + g ) = 0 , 3.10
where the pressure drop balance assumption relates ai with the other two inertial terms by
(1 − f ) (1 − ε )(a + g ) − (1 − ε f )(a f + g )

ai =
c −g, 3.11
f (ε − ε c )
The mass balance equations of the gas and particles as well as the definitions of dcl and Nst remain
the same with Eqs. 3.4-3.7. In addition, the average voidage relates εf and εc by
153
ε = f ε c + (1 − f ) ε f , 3.12
For given Ug, Gs and a sweep of ε, the relevant scheme is as follows:
1. Traverse trial values for εc and εf within the range of [εmf, εmax];
2. Calculate f from Eq. 3.12;
3. Traverse trial values for ac and af within the range of [-g, amax], here amax is the upper limit of the
search region;
4. Calculate Usc and Usf from Eqs. 3.8 and 3.10;
5. With the definition of Usc and Usf, calculate Ugc, Usc, Ugf, Usf from Eqs. 3.4 and 3.5;
6. Calculate dcl from Eq. 3.6, and then Usi,1 from Eq. 3.9;
7. Calculate Usi,2 from its definition and then the difference ΔUsi=|Usi,1-Usi,2| as the convergence criterion;
8. If not converged, jump back to step 3 with new trial values of ac and af; if converged, store the
solution set and Nst for later comparison;
9. Continue the inner loop of ac and af as well as the outer loop of εc and εf, find the optimal root with
respect to the minimization of Nst.
The variation of dcl and εc as functions of ε can be determined with this scheme. Calculations show
that Nst→min requires the maximization of εf and minimization of af, that is, εf→εmax and af→-g. To
reduce the computational cost, these two criteria are used directly in later calculation.
It should be noted that the solution for dcl and εc can be obtained only within the range of (εmf, ε*),
where ε*=Ug/(Ug+Gs/ρp), since the global slip velocity cannot be negative for a concurrent-up gas-solids
riser. Then, to extrapolate the solution to the full range of voidage, we may assume when ε=εmax, particles
are discretely distributed in the fluid and the cluster disappears, i.e. dcl=dp, beyond which εc=εf=ε. One
may expect to adopt alternative models or correlations to calculate the cluster parameters dcl and εc. One
of these efforts can be referred to the work of Wang et al. (2008a).
In the second step, with the input from CFD results, viz. ug, up and ε, the remaining variables, i.e.,
(Uc, Upc, f) for the dense phase and (Uf, Upf) for the dilute phase as well as inertial terms associated to
each phase, are resolved by deterministic solution of the set of conservation equations. According to
Galilean relativity, the second step can be simplified by organizing the conservation equations as func-
tions of slip velocities (Wang et al., 2008b; Lu et al., 2009). The three equations concerning the closure
of three unknown variables (Usc, Usi, ac) read
3 (1 − ε c )
CD, c ρ g Usc Usc = (ρ p − ρ g )(1 − ε ) (a c
− g) , 3.13
4d p
3
C ρ U U = (ρ p − ρ g ) ( ε − ε c ) (ai − g ) , 3.14
4d cl D,i g si si
ε f (1 − ε ),
Usi = (Us − fUsc ) ⋅ 3.15
εf − ε
154
The EMMS-based drag coefficient read
ε2 ε2
β=
Us
FD =
Us
( fn Fc c
+ ni Fi + (1 − f ) n f F f )
(ρ p − ρ g ) ε 2  f 1− ε − g ) + f (ε − ε c ) (ai − g ) + (1 − f ) (1 − ε f ) (a f − g )
=
Us  ( )(a c

(ρp − ρg ) ε2
=
Us
(1 − ε )(a c
− g)
3.16
Here the vector division means that vectors on both numerator and denominator hold the same ori-
entation and the division result is a scalar. Finally, for given Ug, Gs, εc and dcl have been calculated as
functions of ε and stored for later interpolation. The relevant algorithm EMMS/matrix for remaining
variables (Usc, Usi, ac) is as follows:
1. Sweep over the possible pairs of the local slip velocity Us and voidage ε;
2. Initiate trial values for ac within the range of [g, amax];
3. Calculate Usc from Eq. 3.13;
4. Calculate first trial slip velocity Usi,1 from Eq. 3.14.
5. Calculate second trial slip velocity Usi,2 from Eq. 3.15, and then their difference ΔUsi=│Usi,1-Usi,2│;
6. Compare the value of ΔUsi with the convergence criterion, if converged then calculate the drag coef-
ficient from Eq. 3.16 and store it in a drag coefficient matrix as a function of Us and ε, otherwise,
directly jump back to the step 2.
Taking the drag coefficient from Wen and Yu (1966) as the standard,
3 ε (1 − ε )ρ C
β0 = g D0
u s ε−2.7 , 3.17
4 dp
we can define the heterogeneity index as follows:
β
HD ≡ , 3.18
β0
Figure 7 shows the results of EMMS/matrix in forms of the heterogeneity index as a function of local
Reynolds number Rep (Rep=εdp│us│ρg/μg) and ε for an air-FCC system. For the sake of visualization, only
the vertical direction of slip velocity is used here. It is obvious that the drag coefficient is reduced due to
structures and the voidage is the main factor affecting this reduction in most of the range investigated.
The heterogeneity index approaches unity near the two ends of voidage span, which correspond to the
packed bed state and the dilute flow of isolated single particles, respectively. Reynolds number seems to
takes the secondary role in reducing the drag coefficient in most of the range. Higher Reynolds number
155
Figure 7. Surface plot of the heterogeneity index as a function of voidage and Reynolds number (Rep=ε(ug-
up)dpρg/μg) for air-FCC particle system (ρg=1.225kg/m3, ρp=1400kg/m3, dp=60μm, Ug=3.5m/s,
Gs=86.1kg/(m2·s), εmf=0.448) (Zhang et al., 2008 ©2008, Elsevier. Used with permission.)
corresponds to higher value of HD as the formation of aggregates is suppressed by higher slip velocity. For
the sake of use, the surface as shown in Figure 7 can be fitted by piecewise functions (Lu et al., 2009).
Recent simulation reveals that this EMMS-based drag coefficient seems to allow a mesh-independent
solution of the effect of sub-grid structures on the interphase momentum transfer (Lu et al., 2009; Wang
et al., 2009). To testify the mesh dependency of the model, following the scheme proposed by Agrawal
et al. (2001), we restricted our simulation to a periodic 2-D domain of 15×60 mm2, which is comparable
to the coarse grid size commonly used in TFM simulations. The domain was resolved into uniformly
spaced, square grids. As no global acceleration exists in a periodic domain, drag force exerted on par-
ticles can be related to the effective gravity as follows: β=ε(1-ε)(ρp-ρg)g/us. Figure 8 shows the variation
of time-average slip velocity against grid size together with snapshots of the solids distribution. The
coordinates are scaled with the terminal velocity of single particles (uT≈21.84 cm/s) and particle diam-
eter dp, respectively. It is obvious that the structure resolution improves with the mesh refining, and the
EMMS-based drag coefficient remains almost unchanged in the sense that the predicted slip velocity
maintains around five times the terminal velocity over the dimensionless grid size spanning from 1 to
50. This mesh-independency means that the EMMS-based drag coefficient correctly captures the effects
of sub-grid structure, and in the sense of practical use, it allows use of coarse grid scheme when simulat-
ing an industrial reactor and reduces the demand of computing capacity.
It should be emphasized that EMMS model and its current extensions are proposed only for gas-
solids CFB simulations, as its stability condition, Nst→min, which is the key difference from the other
approaches, is satisfied only for gas-solids risers. It should not be used directly for the other kinds of
flow systems. To our knowledge, for nonlinear, dissipative systems normally encountered in multi-phase
flows, a universal stability condition may not exist, though certain general methodology can be ex-
pected to explore their respective stability conditions. Actually in recent years, the concept of the EMMS
156
Figure 8. Effect of grid resolution on time-averaged axial slip velocity scaled with terminal velocity of
single particles (us/uT). EMMS/matrix drag coefficient was used. Error bar denotes the square mean root
of slip velocity. Physical properties used read (dp=75μm, ρp=1500kg/m3, ρg=1.3kg/m3, μg=1.8×10-5Pa·s)
(Wang et al., 2009 ©2009, Elsevier. Used with permission.)
model has been generalized to the so-called analytical multi-scale methodology (Li & Kwauk, 2003),
featuring different variational criteria for different systems.
In practice, the EMMS-based drag coefficients have found successful applications in simulating
circulating fluidized bed risers. The advantage of these structure-dependent drag coefficients can be
reflected by the correct prediction of the solids flux. For a circulating simulation mode, the solids are
carried out of the riser by the air. For the conventional drag coefficient such as the hybrid form of Wen
and Yu (1966) and Ergun (1952), the solids flux will be overestimated to a large extent. By comparison,
the EMMS-based drag coefficient predicts a flux in good agreement with the experimental data (Yang
et al., 2003; Wang & Li, 2007).
3.4 alternative cluster description
In the previous sub-section, the cluster size in the EMMS model is estimated by a semi-empirical ex-
pression and its value depends on the maximum voidage possible for clustering, which has not been
well determined. To avoid this uncertainty, we are making efforts to give an alternative and implicit
description of cluster diameter. Wang and Ge (2005) and Wang et al. (2008a) used a stochastic geom-
etry approach named doubly stochastic Poisson processes (Grandell, 1976) to analyze the fluctuation
characteristics of solid volume fraction in risers, together with the meso-scale added mass effect of the
clusters and the gas-solids mixture of the dilute phase. They then arrived at a correlation between the
accelerations between the dense and dilute phases, which can replace the cluster diameter correlation of
Eq. 3.6 for the closure of the EMMS model, that is, the EMMS model can now be solved which gives dcl
as a result. Coupling of this revised model to CFD framework has been reported by Wang et al. (2008a)
157
Figure 9. Evolution of the flow pattern in a CFB riser as simulated by discrete particle model (DPM),
left: standard DPM, right: DPM with meso-scale model (Xu et al., 2007 ©2007, Elsevier. Used with
permission.)
also, which seems to be equally effective in improving the TFM description of gas-solids fluidization
as original EMMS model.
3.5 sub-grid drag for discrete Particle Model
With the success of introducing sub-grid drag model, in our case, the EMMS model, into two-fluid
model, it is tempting to extend this approach to discrete particle model (DPM). This has been tried by
Xu and Ge (2007), and a brief introduction to this work is presented here.
In traditional DPM, the governing equation for the fluid flow possesses a momentum exchange source
term, which is the total drag on the particles with unit space. Since the position and velocity of each par-
ticle is known in DPM, the drag force on each particle can be calculated specifically based on the local
slip velocity, which traditionally use bi-linearly interpolated four grid velocities (e.g., Hoomans et al.,
2000). However, when heterogeneity presents at sub-grid scales, this interpolation can lead to apparent
inaccuracy. The fluid flow will actually adapt to the local configuration of the particles to suffer least
resistance, which is also an ingredient of the EMMS model. To account for this effect, Xu et al. (2007)
developed a meso-scale model for DPM.
The basic idea of Xu et al. (2007) is to consider, in addition, the gradients of voidage at different
points in the cell. Restricting themselves to a relatively simple case where the gradients are similar and
hence can be describe a single vector ∇ε, they decompose the flow velocity in the cell into the compo-
nents parallel and perpendicular to ∇ε. The perpendicular component is distributed among particle with
different local voidage on the basis of equal pressure drop, whereas the parallel component is distributed
uniformly. The local gas velocities are then reconstructed from these two components and return to the
traditional DPM approach.
158
Table 3.
Particle left right Gas left right

density ρp (kg·m )
-3
810 1170 density ρg (kg·m ) -3
1.29 1.29
number Np 4800 6000 velocity ug (m·s ) -1
1.6 5.6
restitution coefficient 0.8 0.9 viscosity μg(kg·m-1·s-1) 1.8×10-5 1.8×10-5
friction coefficient 0.4 0.2 incipient fluidization velocity 0.8 2.7
Umf (m·s-1)
diameter dp (mm) 0.8 2
Geometry left right Numerical left right
height H (m) 1.0 1.2 grids 50×10 40×15
width W (m) 0.04 0.2 time step Δt (s) 1×10 -4
1×10-4
Smoothing length h 1.5 dp 3 dp simulation time t (s) 3 20
porosity interval of a layer 0.02 0.02
To demonstrate the effectiveness of this improvement, we simulated a CFB riser with both standard
DPM and that with meso-scale model (Xu et al., 2007). Typical “core-annulus” structure (Nakamure,
1973) and more distinguished clustering phenomena can be observed in the latter case (see the right
inset) where the circulating solids flux is also much lower. As analyzed in Xu et al. (2007), the latter
case is considered more reasonable under the conditions simulated. (see Table 3)
4. cHaracteristics Of drag fOrce distributiOn in dynaMic flOws
As discussed above, although different approaches, either theoretical, experimental or numerical, has been
practiced to quantify the drag force in gas-solids flows, and many correlations has been proposed, little
is known about the drag force distribution in real gas-solids systems. It is possible to track the motion of
few individual particles in a three-dimensional gas-solids system using, for example, Positron Emission
Particle Tracking (PEPT, [Seville et al., 1994, 1995]), or Magnetic Resonance Imaging (MRI, [Gladen
et al., 2003, 2007]); and it is possible to measure in quasi-two-dimensional systems the motion of many
particles using high speed photograpy [Rhodes et al., 1992]. However, simultaneous measurement of
multiple moving particles together with the dynamic flow field around them is still very difficult, if not
impossible. By image analysis, Martin et al. (2005) tracked the motion of 8 and 16 particles in gas-solids
fluidized bed, and using Particle Image Velocimetry (PIV), they managed to visualize the fluid flow in
between, but its resolution in the wake area were not so satisfactory.
For larger systems of practical significance or, at least, statistical meanings, direct numerical simulation
(DNS) may prove to be a feasible alternative. For liquid-solid systems, as mention in previous sections,
quite a lot work (Hu, 1996; Pan et al., 2002; Ladd, 1994a, 1994b, 2001) has been reported on DNS of
suspensions of thousands of particles that are numerous enough to give some constitutive correlations.
But as a matter of fact, very little work has been done on gas-solids system which has density ratios typi-
cally of the order of 1000. A practical difficulty is that, for most approaches used, finer step time have
to be used for higher density ratio so as to keep enough accuracy or convergence (van de Hoef, 2008).
In this section, we will introduce some preliminary results on our DNS of particle-fluid suspensions
with density ratios of the order of 100, which, though not very typical, can be justified as gas-solids
159
Table 4. Simulation and physical parameters
System H (m) W (m) ρ (kg/m3) μ(m2.s1) d(m) h(m) ΔP (Pa) F (m/s2)

A 6.14e-2 0.39e-2 1.225 1.46e-05 6.54e-6 7.27e-6 18.586 2.471e2
B 24.96e-2 6.24e-2 1.225 1.46e-05 6.54e-6 7.27e-6 90.28 3.617e2
m(kg) R(m) gs (m/s2) ms (kg) ρs (kg/m3) Δt (s) Vrmy(m/s) Re
A 1.142e-15 1.09e-4 -9.8 1.306e-10 160.49 1.627e-08 0.1340 1.84
B 1.142e-15 1.09e-4 -9.8 1.306e-10 160.49 1.627e-08 0.1173 1.61
systems. Macro-scale pseudo-particle modeling is the numerical method used, interested readers can
find detailed discussions on the method in Ge & Li (2001; 2003) and Ma et al. (2006; 2009).
4.1 setup of the simulations
The gas-solids systems studied are enclosed in rectangular domains periodic in both width (W) and height
(H) directions. The major parameters are listed in Table 4, where R, ρs and ms are the radius, density
and mass of a solid particle. For systems A and B, 1024 and 99856 solid particles are initially arranged
evenly on an orthogonal matrix of 8×128 and 158×632, and then start to fall under gravity. The gas flow
is driven by a pressure drop that exactly balances the total weight of the solids. The listed Re is based
on the resultant average slip velocity in the height direction.
4.2 local flow structure
Figure 10 is a snapshot of the flow field of system B in the steady state, with part of it enlarged. The
clustering of the particles is apparent, though the shape and size of cluster is hard to define, partly due
to the limited particle number. We have found that, in general, higher drag forces are exerted on the
leading particles in a cluster or on individual particle, and the direction of the drag force is under quick
and frequent change. In fact, from the animation simulation results, we also find that the magnitudes and
directions of the drag forces are under significant and frequent changes, in response to the dynamic local
flow patterns, for example, the interaction between the wakes of the particles. It seems very difficult
to apply any drag correlation on any single particle based on a local slip velocity. In fact, the concept
of local slip velocity also becomes ambiguous on this scale. It provides a kind of support to the remark
that the effective resolution of discrete particle model (DPM) should be much lower than the particle
level (ref. Sect. 3.5)
4.3 velocity and drag force distribution
Magnitude and characteristics of the velocity fluctuations of the solids is the basic properties of the solid
phase for TFM, which determines granular temperature and pressure, as well as many other properties
of the system. Figure 11 illustrates the distributions of particle fluctuation velocities in the width and
height directions for systems A and B, respectively. The data are sampled from all the particles at every
50000th step during 2.4 million steps under the steady state, and the fluctuating velocities are calculated
based on the average velocity of all these samples.
160
Figure 10. Detailed flow field around solids in unbounded suspension. Left: the whole flow field; Right:
partial enlargement. The magnitudes of the flow velocities are presented by the color spectrum, where
white represents velocities higher than the red one (Adapted from Chen et al., 2009)
It can be found in Figure 11 that both components of the velocity fluctuation is nearly Gaussian which
takes the form,
fi = 1 2πu 'i 2 exp(u 'i 2 u 'i 2 ) 4.15
where i denotes the width or height components, respectively. However, anisotropy is clearly seen, the
fluctuation in the height direction ( u '2 2 ≈4.67186×10-3m2/s2 for system A and 9.13754×10-2 m2/s2 for
system B) is considerably stronger than that in the width direction ( u '12 =2.29283×10-3 m2/s2 for system
A and 3.84168 ×10-2 m2/s2 for System B). This anisotropy may be related to the fact that gravity is ex-
erted on the height direction. To counterbalance this gravity with drag force, and to adapt to the cluster-
ing of the particles, the particles must be accelerated and de-accelerated constantly. Goldschmidt et al.
161
Figure 11. Distribution of radial (left) and axial (right) components of fluctuation velocities (a and b for
system A and c and d for system B), the solid lines correspond to Gaussian distribution (adapted from
Ma et al., 2006; Xiong et al., 2009.©2006 Elsevier. Used with permission.)
(2002) reported similar anisotropy in DEM simulation of fluidization, and for liquid-solid systems, such
evidence was found by Carlos and Richardson (1968) in experimental measurements.
As to a quantitative analysis of the drag distribution in the system, we have to admit that, it is still
very difficult to distinguish the various contributions in the overall interactions between the gas and
solids, for example, how important is added mass force and lift forces. We assume that, due to their
large density ratio, drag force is the predominant ingredient, and in this sense, we may take Figure 12
as an indication that drag distribution is also anisotropic and even not Gaussian for its component in the
height direction. Large upward drag forces occur rather more frequently than the downward ones. We
suggest that it is a reflection of the nearly periodic motion of the solids which are slowly accelerated in
free falling in dilute phase followed by a hard break as they are sucked into the wakes of the clusters
and hit the bulk of the cluster.
Figure 12 also suggests that the dispersion of instant and individual drag force on each particle is
fairly large as compared with their average which equals gravity. In the height direction, the root mean
square deviation of this fluctuation calculated from Figure 12b for system A is about 74% of particle
gravity, and 71% for the vector value analyzed in Figure 12c. Similar results are obtained for system B
which are demonstrated in Figure 12 d, e, and f.
162
Figure 12. Distribution of the width (a and d) and height (b and e) components and the module (c and
f) of drag forces on the particles, the solid lines are Gaussian. (adapted from Ma et al. 2006 and Xiong
et al., 2009 ©2006, Elsevier. Used with permission.)
Ma et al. (2009) also measured the evolution of slip velocity in another simulation of gas-solids
suspension. The arrangements are similar to the previous system described in Sec. 4.1, so the total
weight of the solids is constant and, as seen in Figure 13, the balance between gravity and drag force is
established very soon. The average slip velocity, however, increases in a much longer period in a violent
fluctuating mode, in response to the development of heterogeneity in the system. This is another example
that the drag force is not only dependent on the voidage and average slip velocity, but also local flow
163
Figure 13. Temporal evolution of average inter-phase force and slip velocity in the height direction for a
system of 389 particles (the dimensional simulation parameters are: domain width W=7.72mm, domain
height H=7.68mm, gas density ρg=1.225kg/m3, solid density ρs=160.49kg/m3, gas dynamical viscosity
υ=1.46e-5m2/s, solid particle diameter Ds=2.181e-4m, gravitational acceleration g=9.8m/s2; reduced
values are used in the figure with unit time being 1.627e-8s, unit mass being 1.142e-15kg and unit length
being 2.181e-5m). (Ma et al., 2009 ©2009, Elsevier. Used with permission.)
structure and solids distribution. New parameters should be introduced to reflect the characteristics of
the sub-grid scale structures.
5. suMMary
In this chapter, we revisited different forms of the interactions between gas flow and solid particles,
however, drag force is in many cases the predominant ingredient and effectively the only nontrivial
interaction considered in the mainstream simulation methods for gas-solids flow. Though various cor-
relations have been proposed and introduced in this chapter for the drag on single particles and particles
in some simple configurations, we have demonstrated that the drag on a swarm of particles can be a very
difficult problem, especially when the particles are in dynamic motion and interactions. Characterising
the meso-scale structure of particle and flow distribution is important to establish reasonable correlations
for the drag force under these conditions, and the Energy Minimization Multi-Scale (EMMS) model we
have introduced seems to be one of the practical approaches to this problem. Of course, plenty of future
work is called for, such as proposing more general and reliable cluster size correlations based on more
realistic and detailed physical pictures obtained from the DNS simulations discussed in Sec. 4.
Although this chapter is devoted to the problem of drag force in gas-solids systems, it also demon-
strates, in a more general sense, how meso-scale structures can be quantified in complex systems. The
compromise of different mechanisms in the system is key to the formation of these structures, and the
164
stability condition expressing this compromise may provide a physically meaningful way to close the
model for such systems, which is, perhaps, the most profound implication of the EMMS model. Interest-
ing problems may be, for example, whether similar models can be developed to describe the amazing
diversity of material structures on meso-scale based on our relatively better understanding in their atom-
istic/molecular structure and bulk properties, though it is completely beyond the scope of this chapter.
references
Almedeij, J. (2008). Drag coefficient of flow around a sphere: matching asymptotically the wide trend.
Powder Technology, 186, 218–223. doi:10.1016/j.powtec.2007.12.006
Baeyens, J., & Geldart, D. (1974). An investigation into slugging fluidized beds. Chemical Engineering
Science, 29, 255–259. doi:10.1016/0009-2509(74)85051-7
Bai, D., Shibuya, E., & Nakagawa, N. (1997). Fractal characteristics of gas-solids flow in a circulating
fluidized bed. Powder Technology, 90, 205. doi:10.1016/S0032-5910(96)03212-3
Bai, D. R., Jin, Y., & Yu, Z. (1991). Momentum exchange between gas and solids in fast fluidized bed.
Journal of Chemical Industry and Engineering (China), 5, 548–553.
Barnea, E., & Mizrahi, J. (1973). Generalized approach to the fluid dynamics of particulate systems.
Part I: general correlation for fluidization and sedimentation in solid multiparticle systems. Chemical
Engineering Journal, 5(2), 171–189. doi:10.1016/0300-9467(73)80008-5
Bi, H. T., & Grace, J. R. (1995). Effect of measurement method on the velocities used to demarcate the
onset of turbulent fluidization. Chemical Engineering Journal, 57, 261.
Bird, R. B., Stewart, W. E., & Lightfoot, E. N. (1963). Transport phenomena. Chichester, UK: John
Wiley & Sons.
Brereton, C. M. H., & Grace, J. R. (1993). Microstructural aspects of the behavior of circulating fluidized
beds. Chemical Engineering Science, 48, 2565–2568. doi:10.1016/0009-2509(93)80267-T
Carlos, C. R., & Richardson, J. F. (1968). Solids movement in liquid fluidised beds-I Particle velocity
distribution. Chemical Engineering Science, 23, 813–824. doi:10.1016/0009-2509(68)80016-8
Chen, N. S. (2009). Comparison of formulas for drag coefficient and settling velocity of spherical par-
ticles. Powder Technology, 189, 395–398. doi:10.1016/j.powtec.2008.07.006
Chen, P. (2004). Modeling the fluid dynamics of bubble column flows. Doctoral dissertation, Washington
University, Saint Louis, LA.
Chen, Y. M., Jang, C. S., & Cai, P. (1991). On the formation and disintegration of particle clusters in a
liquid--solid transport bed. Chemical Engineering Science, 46, 2253. doi:10.1016/0009-2509(91)85124-G
165
Chou, C.S. (2001). Collisional source of the second moment and pressure tensor for inhomogeneous
rapid granular flows of highly inelastic spheres. Physica A: Statistical Mechanics and its Applications,
290, 341-359.
Chou, C. S., & Richman, M. W. (1998). Constitutive theory for homogeneous granular shear flows of
highly inelastic spheres. Physica A. Statistical and Theoretical Physics, 259, 430–448. doi:10.1016/
S0378-4371(98)00265-9
Clift, R., Grace, J. R., & Weber, M. E. (1978). Bubbles, drops and particles. New York: Academic Press.
Clift, R., Seville, J. P. K., Moore, S. C., & Chavarie, C. (1987). Comments on buoyancy in fluidized
beds. Chemical Engineering Science, 42(1), 191–194. doi:10.1016/0009-2509(87)80228-2
Cox, J. D., & Clark, N. N. (1991). The effect of particle drag relationships on prediction of kinematic
wave velocity in fluidized beds. Powder Technology, 66, 177–189. doi:10.1016/0032-5910(91)80099-5
Cruz, E., Steward, F. R., & Pugsley, T. (2006). New closure models for CFD modeling of high-density
circulating fluidized beds. Powder Technology, 169, 115–122. doi:10.1016/j.powtec.2006.08.005
De Wilde, J. (2005). Reformulating and quantifying the generalized added mass in filtered gas-solid
flows models. Physics of Fluids, 17, 113304. doi:10.1063/1.2131925
De Wilde, J. (2007). Filtered gas-solid momentum transfer models and their applications to 3D steady-
state riser simulations. Chemical Engineering Science, 62, 5451–5457. doi:10.1016/j.ces.2006.12.028
Deen, N. G., Van Sint Annaland, M., Van der Hoef, M. A., & Kuipers, J. A. M. (2007). Review of
discrete particle modeling of fluidized beds. Chemical Engineering Science, 62, 28–44. doi:10.1016/j.
ces.2006.08.014
Di Felice, R. (1994). The void function for fluid-particle interaction systems. International Journal of
Multiphase Flow, 20, 153–159. doi:10.1016/0301-9322(94)90011-6
Ding, J., & Gidaspow, D. (1990). A bubbling fluidization model using kinetic theory of granular flow.
AIChE Journal. American Institute of Chemical Engineers, 36, 523. doi:10.1002/aic.690360404
Drew, D. A. (1983). Mathematical modeling of two-phase flow. Annual Review of Fluid Mechanics, 15,
261. doi:10.1146/annurev.fl.15.010183.001401
Eaton, J. K., & Fessler, J. R. (1994). Preferential concentration of particles by turbulence. International
Journal of Multiphase Flow, 20, 169. doi:10.1016/0301-9322(94)90072-8
Epstein, N. (1984). Comments on a unified model for particulate expansion of fluidized beds and flow in
fixed porous media. Chemical Engineering Science, 39, 1533–1534. doi:10.1016/0009-2509(84)80020-2
Ergun, S. (1952). Fluid Flow through Packed Columns. Chemical Engineering Progress, 48, 89–94.
Español, P. (1998). Fluid particle model. Physical Review E: Statistical Physics, Plasmas, Fluids, and
Related Interdisciplinary Topics, 57, 2930–2948. doi:10.1103/PhysRevE.57.2930
Español, P., & Revenga, M. (2003). Smoothed dissipative particle dynamics. Physical Review E: Statisti-
cal, Nonlinear, and Soft Matter Physics, 67, 026705. doi:10.1103/PhysRevE.67.026705
166
Fan, L. S., Han, L. S., & Brodkey, R. S. (1987). Comments on the buoyancy force on a particle in a fluid-
ized suspension. Chemical Engineering Science, 42(5), 1269–1271. doi:10.1016/0009-2509(87)80087-8
Feng, Y. Q., & Yu, A. (2004). Comments “Discrete particle-continuum fluid modelling of gas-solid flu-
idsized beds” by Kafui et al. Chemical Engineering Science, 59(3), 719. doi:10.1016/j.ces.2003.11.003
Fessler, J. R. (1994). Preferential concentration of heavy particles in a turbulent channel flow. Physics
of Fluids, 6, 3742. doi:10.1063/1.868445
Flemmer, R. L. C., & Banks, C. L. (1986). On the drag coefficient of a sphere. Powder Technology, 48,
217–221. doi:10.1016/0032-5910(86)80044-4
Fortes, A. F., Joseph, D. D., & Lundgren, T. S. (1987). Nonlinear mechanics of fluidization of beds of
spherical particles. Journal of Fluid Mechanics, 177, 467–483. doi:10.1017/S0022112087001046
Foscolo, P. U., & Gibilaro, L. G. (1984). A fully predictive criterion for the transition between particu-
late and aggregate fluidization. Chemical Engineering Science, 39(12), 1251–1260. doi:10.1016/0009-
2509(84)80100-1
Foscolo, P. U., Gibilaro, L. G., & Waldram, S. P. (1983). A unified model for particulate expansion
of fluidized-beds and flow in fixed porous-media. Chemical Engineering Science, 38(8), 1251–1260.
doi:10.1016/0009-2509(83)80045-1
Garside, J., & Al-Dlbouni, M. R. (1977). Velocity-voidage relationships for fluidization and sedimen-
tation in solid-liquid systems. Industrial & Engineering Chemistry Process Design and Development,
16(2), 206–214. doi:10.1021/i260062a008
Ge, W. (2005). Particle methods for multiscale simulation of complex flows. Chinese Science Bulletin,
50, 1057–1069. doi:10.1360/04wb0108
Ge, W., & Li, J. (2001). Macro-scale pseudo-particle modeling for particle-fluid systems. Chinese Sci-
ence Bulletin, 46, 1503–1507. doi:10.1007/BF02900568
Ge, W., & Li, J. (2002). Physical mapping of fluidization regimes-the EMMS approach. Chemical En-
gineering Science, 57, 3993–4004. doi:10.1016/S0009-2509(02)00234-8
Ge, W., & Li, J. (2003). Macro-scale phenomena reproduced in microscopic systems-pseudo-particle model-
ing of fluidization. Chemical Engineering Science, 58, 1565–1585. doi:10.1016/S0009-2509(02)00673-5
Ge, W., Wang, W., Dong, W., et al. (2008) Meso-scale structure-a challenge of computational fluid dy-
namics for circulating fluidized bed risers. 9th International Conference of Circulating Fluidized Beds.
May 13-16, Hamburg, Germany.
Geldart, D. (1973). Types of gas fluidization. Powder Technology, 7, 285. doi:10.1016/0032-
5910(73)80037-3
Gibilaro, L. G., Di Felice, R., Waldram, S. P., & Foscolo, P. U. (1985). Generalized friction factor and
drag coefficient correlations for fluid-particle interactions. Chemical Engineering Science, 40(10),
1817–1823. doi:10.1016/0009-2509(85)80116-0
167
Gidaspow, D. (1993). Multiphase flow and fluidization: continuum and kinetic theory descriptions. San
Diego: Academic Press.
Gidaspow, D., & Mostofi, R. (2003). Maximum Carrying Capacity and Granular Temperature of A, B
and C Particles. AIChE Journal. American Institute of Chemical Engineers, 49(4), 831–843. doi:10.1002/
aic.690490404
Gingold, R. A., & Monaghan, J. J. (1977). Smoothed particle hydrodynamics: theory and application to
non-spherical stars. Monthly Notification of the Royal Astronamy Society, 181, 375.
Gladden, L. F. (2003). Magnetic resonance: Ongoing and future role in chemical engineering research.
AIChE Journal. American Institute of Chemical Engineers, 49, 2–9. doi:10.1002/aic.690490102
Gladden, L. F., Anadon, L. D., Dunckley, C. P., Mantle, M. D., & Sederman, A. J. (2007). Insights into
gas-liquid-solid reactors obtained by magnetic resonance imaging. Chemical Engineering Science, 62,
6969–6977. doi:10.1016/j.ces.2007.08.084
Glowinski, R., Hestla, T., Joseph, D. D., Pan, T. W., & Periaux, J. (1997). Distributed lagrange multiplier
methods for particulate flows. In Bristeau, M. O., Etgen, G., Fitzgibbon, W., Lions, J. L., Periaux, J., &
Wheeler, M. F. (Eds.), Computational Science for the 21st Century (pp. 270–279). New York: Wiley.
Glowinski, R., Pan, T. W., & Hesla, T. I. (1999). A distributed Lagrange multiplier/fictitious domain
method for particulate flows. International Journal of Multiphase Flow, 25, 755–794. doi:10.1016/
S0301-9322(98)00048-2
Goldhirsch, I. (2003). Rapid granular flows. Annual Review of Fluid Mechanics, 35, 267–293. doi:10.1146/
Goldschmidt, M. J. V. (2002). Hydrodynamic modelling of dense gasfluidised beds: comparison of the
kinetic theory of granular flow with 3D hard-sphere discrete particle simulations. Chemical Engineering
Science, 57, 2059. doi:10.1016/S0009-2509(02)00082-9
Grace, J. R., & Tuot, J. (1979). A theory for cluster formation in vertically conveyed suspensions of
intermediate density. Transactions of the Institution of Chemical Engineers, 57, 49–54.
Grandell, J. (1976). Doubly stochastic Poisson processes. Berlin: Springer-Verlag.
Gu, W. K., & Chen, J. C. (1998). A model for solid concentration in circulating fluidized beds. In Fan,
L. S., & Knowlton, T. M. (Eds.), Fluidization (p. 501). Durago, CO: Engineering Foundation.
Gunn, D. J., & Malik, A. A. (1967). The structure of fluidized beds in particulate fluidization. In A. A.
Dringkenbrug (Ed.), Proceedings of the International Symposium on Fluidization (pp. 52-65). Eindhoven,
The Netherlands: Netherlands University Press.
Happel, J. (1958). Viscous flow in multi-particle systems: Slow motion of fluids relative to beds of
spherical particles. AIChE Journal. American Institute of Chemical Engineers, 4, 197. doi:10.1002/
aic.690040214
Harris, A. T., Davidson, J. F., & Thorpe, R. B. (2002). The prediction of particle cluster properties in the
near wall region of a vertical riser. Powder Technology, 127, 128–143. doi:10.1016/S0032-5910(02)00114-6
168
Helland, E., Bournot, H., Occelli, R., & Tadrist, L. (2007). Drag reduction and cluster formation in a
circulating fluidised bed. Chemical Engineering Science, 62, 148–158. doi:10.1016/j.ces.2006.08.012
Helland, E., Occelli, R., & Tadrist, L. (2000). Numerical study of cluster formation in gas-particle flow
in a circulating fluidized bed. Powder Technology, 110, 210–221. doi:10.1016/S0032-5910(99)00260-0
Hill, R. J., Koch, D. L., & Ladd, A. J. C. (2001). Moderate-Reynolds-number flows in ordered and ran-
dom arrays of spheres. Journal of Fluid Mechanics, 448, 243–278. doi:10.1017/S0022112001005936
Hirt, C. W., & Nichols, B. D. (1981). Volume of fluid (VOF) method for the dynamics of free boundar-
ies. Journal of Computational Physics, 39, 201–226. doi:10.1016/0021-9991(81)90145-5
Holzer, A., & Sommerfeld, M. (2008). New simple correlation formula for the drag coefficient of non-
spherical particles. Powder Technology, 184(3), 361. doi:10.1016/j.powtec.2007.08.021
Hoogerbrugge, P. J., & Koelman, J. M. V. A. (1992). Simulating microscopic hydrodynamic phenomena
with dissipative particle dynamics. Europhysics Letters, 19(3), 155–160. doi:10.1209/0295-5075/19/3/001
Hoomans, B. P. B. (2000). Granular dynamics of gas-solid two-phase flows. Ph.D. dissertation, Twente
University, The Netherlands.
Hoomans, B. P. B., Kuipers, J. A. M., Briels, W. J., & Van Swaaij, W. P. M. (1996). Discrete particle
simulation of bubble and slug formation in a two-dimensional gas-fluidised bed: a hard-sphere approach.
Chemical Engineering Science, 51(1), 99–118. doi:10.1016/0009-2509(95)00271-5
Horio, M., & Kuroki, H. (1994). Three-dimensional flow visualization of dilutely dispersed solids in
bubbling and circulating fluidized beds. Chemical Engineering Science, 49, 2413. doi:10.1016/0009-
2509(94)E0071-W
Hu, H. H. (1996). Direct simulation of flows of solid-fluid mixtures. International Journal of Multiphase
Flow, 22, 335–352. doi:10.1016/0301-9322(95)00068-2
Hu, H. H., Patankar, N. A., & Zhu, M. Y. (2001). Direct numerical simulations of fluid-solid systems
using the arbitrary Lagrangian-Eulerian technique. Journal of Computational Physics, 169, 427–462.
doi:10.1006/jcph.2000.6592
Issangya, A. S. (1998). Flow dynamics in high density circulating fluidized beds. Ph.D thesis. The Uni-
versity of British Columbia, Canada.
Issangya, A. S., Grace, J. R., & Bai, D. R. (2000). Further measurements of flow dynamics in a high-
density circulating fluidized bed riser. Powder Technology, 111, 104. doi:10.1016/S0032-5910(00)00246-1
Jackson, R. (1963). The mechanics of fluidized beds. I: the stability of the state of uniform fluidization.
Transactions of the Institution of Chemical Engineers, 41, 13–21.
Jean, R. H., & Fan, L. S. (1992). On the model equations of Gibilaro and Foscolo with corrected buoy-
ancy force. Powder Technology, 72, 201–205. doi:10.1016/0032-5910(92)80038-X
169
Jiradilok, V., Gidaspow, D., Damronglerd, S., Koves, W. J., & Mostofi, R. (2006)... Chemical Engineer-
ing Science, 61, 5544. doi:10.1016/j.ces.2006.04.006
Kafui, K. D., Thornton, C., & Adams, M. J. (2002). Discrete particle-continuum fluid modelling of gas-
solid fluidised beds. Chemical Engineering Science, 57, 2395–2410. doi:10.1016/S0009-2509(02)00140-9
Kafui, K. D., Thornton, C., & Adams, M. J. (2004). Reply to comments by Feng and Yu on “Discrete
particle-continuum fluid modelling of gas-solid fluidsized beds” by Kafui et al. Chemical Engineering
Science, 59(3), 723. doi:10.1016/j.ces.2003.11.004
Kajishima, T. (2004). Influence of particle rotation on the interaction between particle clusters and
particle-induced turbulence. International Journal of Heat and Fluid Flow, 25, 721–728. doi:10.1016/j.
ijheatfluidflow.2004.05.007
Kandhai, D., Derksen, J. J., & Van den Akker, H. E. A. (2003). Interphase drag coefficients in gas-solid
flows. AIChE Journal. American Institute of Chemical Engineers, 49(4), 1060–1065. doi:10.1002/
aic.690490423
Karlis, D., & Xekalaki, E. (2005). Mixed Poisson distributions. International Statistical Review, 73, 35.
doi:10.1111/j.1751-5823.2005.tb00250.x
Khan, A. R., & Richardson, J. F. (1990). Pressure gradient and friction factor for sedimentation and flui-
disation of uniform spheres in liquids. Chemical Engineering Science, 45(1), 255–265. doi:10.1016/0009-
2509(90)87097-C
Koshizuka, S., & Oka, Y. (1996). Moving-particle semi-implicit method for fragmentation of incom-
pressible fluid. Nuclear Science and Engineering, 123, 421.
Koshizuka, S., Tamako, Y., & Oka, Y. (1995). A paticle method for incompressible viscous flow with
fluid fragmentation. Journal of Computational Fluid Dynamics, 4, 29.
Kostinski, A. B., & Jameson, A. R. (1997). Fluctuation properties of precipitation. Part 1: On the devia-
tions of single-size drop counts from the Poisson distribution. Journal of the Atmospheric Sciences, 54,
2174. doi:10.1175/1520-0469(1997)054<2174:FPOPPI>2.0.CO;2
Kunii, D., & Levenspiel, O. (1991). Fluidization Engineering (2nd ed.). Boston: Butterworth- Heinemann.
Kurose, R., & Komori, S. (1999). Drag and lift forces on a rotating sphere in a linear shear flow. Journal
of Fluid Mechanics, 384, 183–206. doi:10.1017/S0022112099004164
Lackermeier, U., Rudnick, C., & Werther, J. (2001). Visualization of flow structures inside a circulating
fluidized bed by means of laser sheet an image processing. Powder Technology, 114, 71. doi:10.1016/
S0032-5910(00)00265-5
Ladd, A. J. C. (1994a). Numerical Simulations of Particulate Suspensions Via a Discretized Boltzmann-
Equation. 1. Theoretical Foundation. Journal of Fluid Mechanics, 271, 285–309. doi:10.1017/
S0022112094001771
170
Ladd, A. J. C. (1994b). Numerical Simulations of Particulate Suspensions Via a Discretized Boltzmann

Equation Part Ii. Numerical Results. Journal of Fluid Mechanics, 271, 311–339. doi:10.1017/
S0022112094001783
Ladd, A. J. C. (2001). Verberg R.Lattice-Boltzmann Simulations of Particle-Fluid Suspensions. Journal
of Statistical Physics, 104(5-6), 1191–1251. doi:10.1023/A:1010414013942
Lapple, C. E., & Shepherd, C. B. (1940). Calculation of particle trajectories. Industrial & Engineering
Chemistry, 32(5), 605. doi:10.1021/ie50365a007
Lewis, W. K., Gilliland, E. R., & Bauer, W. C. (1949). Characteristics of Fluidized Particles. Industrial
and Engineering Progress, 41(6), 1104–1117. doi:10.1021/ie50474a004
Li, H., Xia, Y., Tung, Y., & Kwauk, M. (1991). Micro-visualization of clusters in a fast fluidized bed.
Powder Technology, 66, 231–235. doi:10.1016/0032-5910(91)80035-H
Li, J. (2000). Compromise and Resolution — exploring the multi-scale nature of gas-solid fluidization.
Powder Technology, 111, 50–59. doi:10.1016/S0032-5910(00)00238-2
Li, J., Chen, A., Yan, Z., Xu, G., & Zhang, X. (1993). Particle-fluid contacting in circulating fluidized
beds. In A. A. Avidan (Ed.), Preprint Volume for Circulating Fluidized Beds IV (pp. 49-54). Somerset:
AIChE.
Li, J., Cheng, C., Zhang, Z., Yuan, J., Nemet, A., & Fett, F. N. (1999). The EMMS model—its applica-
tion, development and updated concepts. Chemical Engineering Science, 54, 5409–5425. doi:10.1016/
S0009-2509(99)00274-2
Li, J., & Kwauk, M. (1994). Particle-fluid two-phase flow: the energy-minization multi-scale method.
Beijing: Metallurgical Industry Press.
Li, J., & Kwauk, M. (2001). Multiscale nature of complex fluid-particle systems. Industrial & Engineer-
ing Chemistry Research, 40, 4227–4237. doi:10.1021/ie0011021
Li, J., & Kwauk, M. (2003). Exploring complex systems in chemical engineering—the multi-scale
methodology. Chemical Engineering Science, 58, 521–535. doi:10.1016/S0009-2509(02)00577-8
Li, J., Tung, Y., & Kwauk, M. (1988a). Axial voidage profiles of fast fluidized beds in different operating
regions. In P. Basu, & J.F. Large (Eds.), 2nd international conference on circulating fluidized beds(pp.
193). Oxford, UK: Pergamon Press.
Li, J., Wen, L., Ge, W., Cui, H., & Ren, J. (1998). Dissipative structure in concurrent-up gas-solid flow.
Chemical Engineering Science, 53(19), 3367–3379. doi:10.1016/S0009-2509(98)00130-4
Li, Y. (1994). Hydrodynamics. In Kwauk, M. (Ed.), Advances in Chemical Engineering (pp. 85–196).
San Diego: Academic Press.
Li, Y., & Kwauk, M. (1980). The dynamics of fast fluidization. In Grace, J. R., & Matsen, J. M. (Eds.),
Fludization (pp. 537–544). New York: Plenum Press.
171
Lin, Q., Wei, F., & Jin, Y. (2001). Transient density signal analysis and two-phase micro-structure
flow in gas-solids fluidization. Chemical Engineering Science, 56(6), 2179–2189. doi:10.1016/S0009-
2509(00)00499-1
Liu, X. (2005). Aggregation and lateral transfer behaviour of particles in gas-solid fluidized beds. Ph.D
thesis. Institute of Process Engineering, Chinese Academy of Science, Beijing, P. R. China.
Loth, E. (2008). Drag of non-spherical solid particles of regular and irregular shape. Powder Technology,
182(3), 342. doi:10.1016/j.powtec.2007.06.001
Lu, B., Wang, W., & Li, J. (2009). Searching for a mesh-independent sub-grid model for gas-solids riser
flows. Chemical Engineering Science, 64, 3437–3447. doi:10.1016/j.ces.2009.04.024
Lu, B., Wang, W., Li, J., Wang, X., Gao, S., & Lu, W. (2007). Multi-scale CFD simulation of gas-
solid flow in MIP reactors with a structure-dependent drag model. Chemical Engineering Science, 62,
5487–5494. doi:10.1016/j.ces.2006.12.071
Lucas, D., Prasser, H. M., & Manera, A. (2005). Influence of the lift force on the stability of a bubble
column. Chemical Engineering Science, 60, 3609–3619. doi:10.1016/j.ces.2005.02.032
Lucy, L. B. (1977). A numerical approach to the testing of the fission hypothesis. The Astronomical
Journal, 83, 1013–1024. doi:10.1086/112164
Ma, J., & Ge, W. (2008). Is standard symmetric formulation always better for smoothed particle hydrody-
namics? Computers & Mathematics with Applications (Oxford, England), 55, 1503–1513. doi:10.1016/j.
camwa.2007.08.010
Ma, J., Ge, W., Wang, X., Wang, J., & Li, J. (2006). High-resolution simulation of gas-solid suspension
using macro-scale particle methods. Chemical Engineering Science, 61, 7096–7106. doi:10.1016/j.
ces.2006.07.042
Ma, J., Ge, W., Xiong, Q., Wang, J., & Li, J. (2009). Direct numerical simulation of particle cluster-
ing in gas-solid flow with macro-scale particle method. Chemical Engineering Science, 64(1), 43–51.
doi:10.1016/j.ces.2008.09.005
Mabrouk, R., Chaouki, J., & Guy, C. (2007). Effective drag coefficient investigation in the accelera-
tion zone of an upward gas-solid flow. Chemical Engineering Science, 62, 318–327. doi:10.1016/j.
ces.2006.08.055
Manyele, S. V., Pärssinen, J. H., & Zhu, J.-X. (2002). Characterizing particle aggregates in a high-density and
high-flux CFB riser. Chemical Engineering Journal, 88, 151–161. doi:10.1016/S1385-8947(01)00299-6
Martin, T. W. (2005). Capturing gas and particle motion in an idealised gas-granular flow. Powder
Technology, 155, 175–180. doi:10.1016/j.powtec.2005.05.043
Marzocchella, A., Zijerveld, R. C., & Schouten, J. C. (1997). Chaotic behavior of gas-solids flow in
the riser of a laboratory-scale circulating fluidized bed. AIChE Journal. American Institute of Chemical
Engineers, 43, 1458. doi:10.1002/aic.690430609
172
Matsen, J. M. (1982). Mechanisms of choking and entrainment. Powder Technology, 32, 21–33.
doi:10.1016/0032-5910(82)85003-1
Meynard, F. (1997). An analysis of the energy minimization multi-scalemodel for circulating fluidized
beds. Unpublished.
Monaghan, J. J. (1992). Smoothed particle hydrodynamics. Annual Review of Astronomy and Astrophys-
ics, 30, 543–574. doi:10.1146/annurev.aa.30.090192.002551
Monazam, E. R., Shadle, L. J., & Lawson, L. O. (2001). A transient method for determination of satura-
tion carrying capacity. Powder Technology, 121, 205. doi:10.1016/S0032-5910(01)00354-0
Morris, J. P., Fox, P. J., & Zhu, Y. (1997). Modeling low Reynolds number incompressible flows using
SPH. Journal of Computational Physics, 136, 214–226. doi:10.1006/jcph.1997.5776
Mostoufi, N., & Chaouki, J. (1999). Prediction of effective drag coefficient in fluidized beds. Chemical
Engineering Science, 54, 851–858. doi:10.1016/S0009-2509(98)00290-5
Mudde, R. F., & Simonin, O. (1999). Two- and three-dimensional simulations of a bubble plume using a
two-fluid model. Chemical Engineering Science, 54(21), 5061–5069. doi:10.1016/S0009-2509(99)00234-1
Mueller, P., & Reh, L. (1993). Particle drag and pressure drop in accelerated gas-solid flow. In A. A.
Avidan (Ed.), Preprint Volume for Circulating Fluidized Beds IV (pp. 193-198). Somerset: AIChE.
Nakamure, K., & Capes, C. E. (1973). Vertical pneumatic conveying: a theoretical study of uniform
and annular particle flow models. Canadian Journal of Chemical Engineering, 51, 39–46. doi:10.1002/
cjce.5450510107
Neri, A., & Gidaspow, D. (2000). Riser hydrodynamics: simulation using kinetic theory. AIChE Journal.
American Institute of Chemical Engineers, 46, 52. doi:10.1002/aic.690460108
O’Brien, T. J., & Syamlal, M. (1993). Particle cluster effects in the numerical simulation of a circulaiting
fluidized bed. In A. A. Avidan (Ed.), Preprint Volume for Circulating Fluidized Beds IV (pp. 430-435).
Somerset: AIChE.
Oey, R. S., Mudde, R. F., & Van den Akker, H. E. A. (2003). Numerical simulations of an oscillating
internal-loop airlift reactor. Canadian Journal of Chemical Engineering, 81(3-4), 684–691. doi:10.1002/
cjce.5450810347
Osher, S., & Fedkiw, R. P. (2001). Level set methods: an overview and some recent results. Journal of
Computational Physics, 169, 463. doi:10.1006/jcph.2000.6636
Pan, T. W., Joseph, D. D., Bai, R., Glowinski, R., & Sarin, V. (2002). Fluidization of 1204 spheres: simu-
lation and experiment. Journal of Fluid Mechanics, 451, 169–191. doi:10.1017/S0022112001006474
Perry, R. H., & Green, D. W. (2007). Perry’s Chemical Engineers’ Handbook. New York: Mcgraw-Hill.
Qi, X., Tao, Z., & Huang, W. (2005). Experimental study of solids holdups inside particle clusters in
CFB risers. Journal of Sichuan University, 37(5), 46.
Reh, L. (1971). Fluidized bed processing. Chemical Engineering Progress, 67, 58–63.
173
Reh, L., & Li, J. (1991). Measurement of voidage in fluidized beds by optical probe. In Basu, P., Horio,
M., & Hasatani, M. (Eds.), Circulating Fluidized Beds Technology III (p. 105). Oxford, UK: Pergamon
Press.
Rhodes, M., Mineo, H., & Hirama, T. (1992). Particle motion at the wall of a circulating fluidized bed.
Powder Technology, 70, 207–214. doi:10.1016/0032-5910(92)80055-2
Rhodes, M. J., Sollaart, M., & Wang, X. S. (1998). Structure of the dense-dilute interface in fast flu-
idization. Presented at Proceedings of the Ninth Engineering Foundation Conference on Fluidization,
Engineering Foundation, (pp.141). Durango, Colorado.
Richardson, J. F., & Zaki, W. N. (1954). Sedimentation and fluidisation: Part I. Transactions of the
Institution of Chemical Engineers, 32, 35–53.
Schiller, L., & Naumann, A. (1933). Fundamental calculations in gravitational processing. Zeitschrift
Des Vereines Deutscher Ingenieure, 77, 318–320.
Segrè, P. N., Liu, F., & Umbanhowar, P. (2001). An effective gravitational temperature for sedimenta-
tion. Nature, 409, 594. doi:10.1038/35054518
Seville, J. P. K., Simons, S. R. J., Broadbent, C. J., Martin, T. W., Parker, D. J., & Beynon, T. D. (1995).
Particle velocities in gas-fluidized beds. In Fluidization VIII (p. 319). New York: Engineering Foundation.
Seville, J. P. K., Simons, S. R. J., Broadbent, C. J., Parker, D. J., & Beynon, T. D. (1994). Particle ve-
locities in gas fluidized beds. In 1st International Particle Technical Forum of AIChE (p. 493). Denver.
Seville, J. P. K., Tüzün, U., & Clift, R. (1997). Processing of particulate solids. London: Blackie Aca-
demic & Professional.
Sharma, A. K., Matsen, J., Tuzla, K., et al. (2001). A correlation for solid fraction in clusters in fast flu-
idized beds. In M. Kwauk, J. Li, & W.C. Yang (Eds.), Proceedings of the 10th Engineering Foundation
Conference on Fluidization, (pp. 301). New York: Engineering Foundation.
Sharma, A. K., Tuzla, K., & Matsen, J. (2000). Parametric effects of particle size and gas velocity on cluster
characteristics in fast fluidized beds. Powder Technology, 111, 114. doi:10.1016/S0032-5910(00)00247-3
Shi, D., Nicolai, R., & Reh, L. (1998). Wall-to-bed heat transfer in circulating fluidized beds. Chemical
Engineering and Processing, 37, 287. doi:10.1016/S0255-2701(98)00039-7
Soong, C. H., Tuzla, K., & Chen, J. C. (1994). Identification of particle clusters in circulating fluidized
bed. In AA Avidan (Ed.), Proceeding of the Fourth International Conference on Circulating fluidized
beds, (pp. 615). New York: Pergamon Press.
Soong, C. H., Tuzla, K., & Chen, J. C. (1995). Experimental determination of cluster size and velocity
in circulating fluidized bed. In Large, J. F., & Laguerie, C. (Eds.), Fluidization (p. 219). New York:
Engineering Foundation.
Squires, K. D., & Eaton, J. K. (1991). Preferential concentration of particle by turbulence. Physics of
Fluids, 3, 1169. doi:10.1063/1.858045
174
Stewart, P. S. B., & Davidson, J. F. (1967). Slug Flow in Fluidized Beds. Powder Technology, 1, 61–66.
doi:10.1016/0032-5910(67)80014-7
Strumendo, M., & Canu, P. (2002). Method of moments for the dilute granular flow of inelastic spheres
- art. no. 041304. Physical Review E: Statistical, Nonlinear, and Soft Matter Physics, 6604, 1304.
Takeda, H., Miyama, S. M., & Sekiya, M. (1994). Numerical simulation of viscous flow by smoothed
particle hydrodynamics. Progress of Theoretical Physics, 92, 939–960. doi:10.1143/PTP.92.939
Taneda, S. (1957). Negative magnus effect. Republic Research Institution of Mechanics, 5, 123–128.
Thomas, P. J. (1992). On the influence of the Basset history force on the motion of a particle through a
fluid. Physics of Fluids A, 4(9), 2090–2093. doi:10.1063/1.858379
Tomiyama, A., Sou, A., Zun, I., Kanami, N., & Sakaguchi, T. (1995). Effects of Eötvös number and
dimensionless liquid volumetric flux on lateral motion of a bubble in a laminar duct flow. In Serizawa,
A., Fukano, T., & Bataille, J. (Eds.), Advances in multiphase flow (pp. 3–15). Amsterdam: Elsevier.
Tran-Cong, S., Gay, M., & Michaelides, E. E. (2004). Drag coefficients of irregularly shaped particles.
Powder Technology, 139(1), 21. doi:10.1016/j.powtec.2003.10.002
Tryggvason, G., Bunner, B., & Esameeli, A. (2001). A front-tracking method for the computations of
multiphase flow. Journal of Computational Physics, 169, 708–759. doi:10.1006/jcph.2001.6726
Tsuji, Y., Kawaguchi, T., & Tanaka, T. (1993). Discrete particle simulation of two-dimensional fluidized
bed. Powder Technology, 77(1), 79–87. doi:10.1016/0032-5910(93)85010-7
Tuzla, K., Sharma, A. K., & Chen, J. C. (1998). Transient dynamics of solid concentration in downer
fluidized bed. Powder Technology, 100, 166. doi:10.1016/S0032-5910(98)00137-5
Unverdi, S. O., & Tryggvason, G. A. (1992). Front-tracking method for viscous, incompressible multi-
fluid flows. Journal of Computational Physics, 100, 25–37. doi:10.1016/0021-9991(92)90307-K
Van der Hoef, M. A., Beetstra, R., & Kuipers, J. A. M. (2005). Lattice-Boltzmann simulations of low-
drag force. Journal of Fluid Mechanics, 528, 233–254. doi:10.1017/S0022112004003295
Van der Hoef, M. A., van Sint Annaland, M., Deen, N. G., & Kuipers, J. A. M. (2008). Numerical
simulation of dense gas-solid fluidized beds: a multi-scale modeling strategy. Annual Review of Fluid
Mechanics, 40, 47–70. doi:10.1146/annurev.fluid.40.111406.102130
Wang, J., & Ge, W. (2005). Collisional particle-phase pressure in particle-fluid flows at high particle
inertia. Physics of Fluids, 17, 128103. doi:10.1063/1.2145757
Wang, J., Ge, W., & Li, J. (2008a). Eulerian simulation of heterogeneous gas-solid flows in CFB risers:
EMMS-based sub-grid scale model with a revised cluster description. Chemical Engineering Science,
63, 1553–1571. doi:10.1016/j.ces.2007.11.023
Wang, W., Ge, W., Yang, N., & Xu, G. (2007b). Response to “Evaluating EMMS model for simulating
high solid flux risers”. Chemical Engineering Research & Design, 85(A9), 1338–1339. doi:10.1205/
cherd.le.0709
175
Wang, W., & Li, J. (2007). Simulation of gas-solid two-phase flow by a multi-scale CFD approach-
extension of the EMMS model to the sub-grid level. Chemical Engineering Science, 62, 208–231.
doi:10.1016/j.ces.2006.08.017
Wang, W., Lu, B., Dong, W., & Li, J. (2008b). Multiscale CFD simulation of operating diagram for gas-
solid risers. Canadian Journal of Chemical Engineering, 86, 448–457. doi:10.1002/cjce.20067
Wang, W., Lu, B., & Li, J. (2007a). Choking and flow regime transitions: simulation by a multi-scale
CFD approach. Chemical Engineering Science, 62, 814–819. doi:10.1016/j.ces.2006.10.010
Wang, W., Lu, B., Zhang, N., Shi, Z., & Li, J. (2010). A review of multiscale CFD for gas-solid CFB
modeling. International Journal of Multiphase Flow..doi:10.1016/j.ijmultiphaseflow.2009.01.008
Wilhelm, R. H., & Kwauk, M. (1948). Fluidization of solid particles. Chemical Engineering Progress,
44(3), 201–218.
Wood, A. M. (2005). Preferential concentration of particles in homogeneous and isotropic turbulence.
International Journal of Multiphase Flow, 31, 1220. doi:10.1016/j.ijmultiphaseflow.2005.07.001
Wylie, J. J., & Koch, D. L. (2000). Particle clustering due to hydrodynamic interactions. Physics of
Fluids, 12, 964. doi:10.1063/1.870351
Xie, H. Y. (1997). Drag coefficient of fluidized particles at high Reynolds numbers. Chemical Engineer-
ing Science, 52(17), 3051–3052. doi:10.1016/S0009-2509(97)00058-4
Xu, B. H., & Yu, A. B. (1997). Numerical simulation of the gas-solid flow in a fluidized bed by combin-
ing discrete particle method with computational fluid dynamics. Chemical Engineering Science, 52(16),
2785–2809. doi:10.1016/S0009-2509(97)00081-X
Xu, M., Ge, W., & Li, J. (2007). A discrete particle model for particle-fluid flow with considerations of
sub-grid structures. Chemical Engineering Science, 62(8), 2302–2308. doi:10.1016/j.ces.2006.12.008
Yan, A., Manyele, S. V., Parssinen, J. H., et al. (2002). The interdependence of micro and macro flow
structures under a high-flux flow. In Proceedings of 7th International Circulating Fluidized Beds Con-
ference, (pp. 357). Canadian Society for Chemical Engineering.
Yang, J. Z., & Renken, A. A. (2003). Generalized correlation for equilibrium of forces in liquid-solid
fluidized beds. Chemical Engineering Journal, 92, 7–14. doi:10.1016/S1385-8947(02)00084-0
Yang, N. (2003). Computer simulation of heterogeneous gas-solid two-phase flow—integration of two-fluid
model with the multi-scale methodology. Doctoral dissertation, Chinese Academy of Sciences, Beijing.
Yang, N., Ge, W., Niu, G., Yang, C., & Li, J. (2005). Simulation of gas/solid flow behaviors and choking
for a CFB riser: the EMMS/CFD approach. In Cen, K. (Ed.), Circulating Fluidized Bed Technology VIII
(pp. 291–298). Hangzhou, China: World Publishing Corporation.
Yang, N., Wang, W., Ge, W., & Li, J. (2003). CFD simulation of concurrent-up gas-solid flow in cir-
culating fluidized beds with structure-dependent drag coefficient. Chemical Engineering Journal, 96,
71–80. doi:10.1016/j.cej.2003.08.006
176
Yang, N., Wang, W., Ge, W., Wang, L., & Li, J. (2004). Simulation of heterogeneous structure in a cir-
culating fluidized-bed riser by combining the two-fluid model with the EMMS approach. Industrial &
Engineering Chemistry Research, 43, 5548–5561. doi:10.1021/ie049773c
Yang, N., Wang, W., Zhao, H., Ge, W., & Li, J. (2006). Structure-oriented multi-scale simulation of two-
phase flows—methodology and application. In Fifth International Conference on CFD in the Process
Industries (CDROM). Melbourne, Australia: CSIRO.
Yerushalmi, J., & Squires, A. M. (1977). The phenomenon of fast fluidization. AIChE Symposium Series,
72, 44-47.
Yerushalmi, J., Tuner, D. H., & Squires, A. M. (1976). The fast fluidized bed. Industrial & Engineering
Chemistry Process Design and Development, 15, 47–53. doi:10.1021/i260057a010
Zhang, D. Z., & VanderHeyden, W. B. (2002). The effects of mesoscale structures on the macroscopic
momentum equations for two-phase flows. International Journal of Multiphase Flow, 28, 805–822.
doi:10.1016/S0301-9322(02)00005-8
Zhang, J., Ge, W., & Li, J. (2005). Simulation of heterogeneous structures and analysis of energy con-
sumption in particle-fluid systems with pseudo-particle modeling. Chemical Engineering Science, 60,
3091–3099. doi:10.1016/j.ces.2004.11.057
Zhang, N., Lu, B., Wang, W., & Li, J. (2008). Virtual experimentation through 3-D full-loop CFD simu-
lation of a CFB. Particuology, 6, 529–539. doi:10.1016/j.partic.2008.07.013
Zhang, Y., & Reese, J. M. (2003). The drag force in two-fluid models of gas-solid flows. Chemical
Engineering Science, 58, 1641–1644. doi:10.1016/S0009-2509(02)00659-0
Zhu, H. P., Zhou, Z. Y., Yang, R. Y., & Yu, A. B. (2007). Discrete particle simulation of particulate
systems: theoretical developments. Chemical Engineering Science, 62(13), 3378–3396. doi:10.1016/j.
ces.2006.12.089
Zhu, H. P., Zhou, Z. Y., Yang, R. Y., & Yu, A. B. (2008). Discrete particle simulation of particulate sys-
tems: a review of major applications and findings. Chemical Engineering Science, 63(23), 5728–5570.
doi:10.1016/j.ces.2008.08.006
Zou, B., Li, H., & Xia, Y. (1994). Cluster structure in a circulating fluidized bed. Powder Technology,
78, 173. doi:10.1016/0032-5910(93)02786-A
Zuber, N. (1964). On the dispersed two-phase flow in the laminar flow regime. Chemical Engineering
Science, 19, 897. doi:10.1016/0009-2509(64)85067-3
177
178
Chapter 5
Mass and Heat
Transfer Modeling
Ronald W. Breault
abstract
Mass and heat transfer are important to reactor modeling using CFD. Reactants (mass) move within
the flow structure in order to react with other species and thereby form the products, both desirable and
undesirable. This movement in mass occurs either by diffusion or by turbulent dispersion. In a similar
fashion, heat is transferred from one point in the flow structure to another point by convective transfer
between the phases and with the translational effect that occurs with the turbulent dispersion of the
mass. In addition, heat can be transferred across the system boundary. In all these cases, fundamental
mechanistic models are put forward that can be incorporated in the CFD code to calculate these trans-
fer properties based upon the local hydrodynamic conditions. The chapter is organized such that dense
phase systems are covered first and then dilute phase systems. Within each of these areas mass transfer
is covered first and followed by heat transfer. The topics are covered in the following order: Diffusional
mass transfer, Turbulent dispersion, Convective heat transfer between phases and Convective heat
transfer at the system boundary.
intrOductiOn
Prior to jumping into the discussion of the diffusional mass transfer, heat transfer and the turbulent disper-
sion, a short overview of the balances contained within MFIX is shown below such that these properties
can be identified. Interphase mass transfer occurs by two mechanisms in heterogeneous gas-solids flows.
The first of these is through diffusion across the boundary layer and the other is from turbulent transport
of the species in gas eddies or solids clusters. Before to getting into the details of these two mechanisms
DOI: 10.4018/978-1-61520-651-3.ch005
and the methods which will be used to calculate these quantities, it is important to understand how and
where these properties are used in reacting computational multiphase flow fluid dynamics.
For isothermal conditions, the continuity, momentum, and species balance equations are given below.
Gas-Phase Continuity
N
∂ g
(ε g ρ g ) + ∇ (ε g ρ g v g ) = ∑ Rgn (1)
∂t n =1
Solids-Phase Continuity
N
∂ s
(ε s ρ s ) + ∇ (ε s ρ s vs ) = ∑ Rsn (2)
∂t n =1
Gas-Phase Momentum
∂
(ε g ρ g v g ) + ∇ (ε g ρ g v g v g ) = −ε g ∇Pg + ∇ τ g + Fgs (vs − v g ) + ε g ρ g g − R0 (ξ 0v g + ξ 0vs ) (3)
∂t
Solids-Phase Momentum
∂
(ε s ρ s vs ) + ∇ (ε s ρ s vs vs ) = −ε s∇Pg + ∇ Sm − Fgs (vs − v g ) + ε s ρ s g − R0 (ξ 0v g + ξ 0vs ) (4)
∂t
Species Balance
∂
(ε m ρ m X mn ) + ∇ (ε m ρ m X mn vm ) = ∇ Dmn∇X mn + Rmn (5)
∂t
where m = g or s for the gas or solids phase, Dmn is the turbulent diffusive term, Rmn is the reaction rate
which is dependent upon the diffusional mass transfer, and ξ̅0 = 1 − ξ0, ξ0 = 1 if R0< 0 else ξ0 = 0. The
eight dependent hydrodynamic variables in three dimensions are void fraction εg (the solids fraction
εs = 1 − εg), pressure Pg, and six velocity components. These are found by using MFIX to numerically
solve the coupled non-linear partial differential equations. The number of species mass fractions (Xmn)
tracked are given in the chemistry model. Constitutive relations needed to close the system, and the gas/
solids energy balance equations can be found in the following references (Syamlal, M., Rogers, W., &
O’Brien, T. (1993); Syamlal, M.(1998)). A discussion on the solution procedure and further numerical
references was prepared by Guenther & Syamlal (2001).
In general, the reaction rate or kinetics follow a conventional resistance model pathway, where the
rate is is proportional to the sum of the resistance elements such as transfer from the bulk to the particle
surface plus the internal resistance which may include diffusion of the gas reactant in through an ash/
179
product layer, the heterogeneous intrinsic gas solid reaction rate and diffusion of the product gas out
through the ash/product layer plus the transfer of the gaseous product from the surface to the bulk. Note
that for some reactions, not all of the resistances exist or are of significant magnitude to be considered.
As an example, the default reaction mechanism in MFIX for carbon reacting with oxygen to form
carbon monoxide is discussed to show where and how these terms appear within MFIX. Therefore, for
the heterogeneous reaction 2C + O2 → 2CO the reaction rate is given by
−6ε s PO
R= 2
(mol / cm 3 s) (6)
1 1 1
d p( + + )
kf ka kr
where the film resistance is given by the mass transfer coefficient, kf. Presently, kf is evaluated from
DO Sh
kf = 2
(7)
d p RO T f
2
where Sh is the Sherwood number which is given by Gunn (1978)
Sh = (7 − 10ε g + 5ε 2g )(1 + .7 Re0.2 Sc1/ 3 ) +

(8)
(1.33 − 2.4ε g + 1.2ε 2g )Re0.7 Sc1/ 3
and the Schmidt number is given by
µg
Sc = (9)
ρ g DO
2
and for turbulent cases in MFIX is taken to be 0.7μg.

The ash layer resistance is given by
2 rd De
ka = (10)
(1 − rd )d p RO Ts
2
and the ratio of core diameter to particle diameter is
1/ 3
X0 X 

rd =  s0, ash s, carbon  . (11)
 X X x, ash 
s, carbon
180
The surface reaction is given by Desai and Wen (1978) for a bituminous coal.
 27000 
k r = 8710 exp −  r 2 . (12)
 1.987Ts  d
This chapter does not address the resistances associated with diffusion through the ash layer or the
surface reaction, only the interphase mass transfer from the bulk to the solid particles.
dense PHase
Mass and Heat transfer in dense systems have been extensively discussed in numerous journal articles
and book chapters. As a result, only a very brief summary of these topics is included in this chapter and
the readers are referred to these excellent sources for further information.
diffusional Mass transfer
Mass transfer in dense fluidized systems has been studied since Resnick and White published their
work in 1949, (Resnick, W. & White, R. R., (1949)). They presented their data graphically in terms of
the J-factor. Breault (2006) calculated Sherwood numbers, which were all less than 1, for Resnick and
White’s data thereby presenting the difficulty one would have in trying to model mass transfer in these
systems using a modified single sphere model
Sh = 2 + a Reb Sc c (13)
Many researchers since 1949 have calculated Sherwood numbers less than 1 from their experimental
tests indicating that a more complex model of the phenomena was required. Two classical models for
mass transfer in dense fluidized systems were developed in the 1960’s and 1970’s. Both of these models
break the fluidized system into a dense part and a lean part and include an exchange factor between
these two parts. The first of these models was presented by Beek, W. J. in Fluidization (Ed. Davidson,
J. F. and Harrison, D., 1971). The second of these is the Bubbling Bed Model presented in Fluidization
Engineering by Kunii, D and Levenspiel, O. (1977). The readers can review either of these two sources
for more extensive discussions on their respective models, although as they are presented neither lends
itself to direct application in CFD codes.
Both of these sources, however, provide a good source of mass transfer data in dense fluidized beds.
Both identify correlations for mass transfer coefficients that can have values of the Sherwood number
less than 2. This contrasts with the expression for the Sherwood number by Gunn (1978). Gunn’s model
has a minimum value for the Sherwood number of 2 and increases with solids fraction and Reynolds
number. It cannot explain the data of Resnick and White and other presented in the summaries by Beek
or Kunii and Levenspiel.
A mechanistic model is needed for dense fluidized beds as is presented later for dilute or clustering
fluidized beds. This would allow the mass transfer coefficient to be calculated directly from the local
181
hydrodynamics rather than rely on correlations which may or may not represent the hydrodynamic
conditions within the unit being modeled. There is some very recent data by R. Hays, S.B. Reddy Karri,
R. Cocco and T.M. Knowlton (2008) which indicates that clusters might be formed within the dense
fluidized bed. If this is the case and these clusters dominate the solids volume fraction, a method similar
to that presented later for use in dilute clustering flows might also be applicable for these conditions.
turbulent dispersion
Turbulent dispersion coefficients have been measured by tracer studies to provide both axial and lateral
values for both the gas and the solids. These values tend to depend on the type and size of solids as well
as the gas velocity. Two excellent summaries for dispersion coefficient data in dense fluidized systems
are by Potter, O. e. in Fluidization (Ed. Davidson, J. F. and Harrison, D., 1971) and in Fluidization
Engineering by Kunii, D and Levenspiel, O. (1977).
The in situ prediction of turbulent dispersion coefficients as a function of the local hydrodynamics
has recently been incorporated in CFD calculation in work published by Jiradilok, V., Gidaspow, D.,
and Breault, R. W. (2007). In this work, turbulent dispersion coefficients were calculated using the auto
correlation technique and compared to literature values. Readers are referred to this paper for detailed
discussions on this topic.
convective Heat transfer between Phases
Interphase heat transfer in dense fluidized systems has been studied much more than mass transfer. This
fact is likely due to the relative ease in making heat transfer measurements compared to making mass
transfer measurements. Researchers for the most part have measured heat transfer coefficients in the
form of Nusselt numbers less than 2. As a result, many of them have presented their data graphically in
terms of the J-factor. Two classical excellent summaries for heat transfer data in dense fluidized systems
are by Gelperin, N. I. and Einstein, V. G. in Fluidization (Ed. Davidson, J. F. and Harrison, D., 1971)
and in Fluidization Engineering by Kunii, D and Levenspiel, O. (1977).
Both of these sources identify correlations for heat transfer coefficients that can have values of the
Nusselt number less than 2. This contrasts with the expression for the Nusselt number by Gunn (1978).
Gunn’s model has a minimum value for the Nusselt number of 2 and increases with solids fraction and
Reynolds number. It cannot explain the data in the summaries by Gelperin and Einstein or Kunii and
Levenspiel.
A mechanistic model for heat transfer as noted above for mass transfer is needed for dense fluidized
beds as is presented later for dilute or clustering fluidized beds. This would allow the heat transfer coef-
ficient to be calculated directly from the local hydrodynamics rather than rely on correlations which may
or may not represent the hydrodynamic conditions within the unit being modeled. There is some very
recent data by R. Hays, S.B. Reddy Karri, R. Cocco and T.M. Knowlton (2008) which indicates that
clusters might be formed within the dense fluidized bed. If this is the case and these clusters dominate
the solids volume fraction, a method similar to that presented later for use in dilute clustering flows
might also be applicable for these conditions.
182
convective Heat transfer at the system boundary
`Heat transfer between the fluidized bed and surfaces – walls and internals – is one of the most widely
studied fields in fluidization. Again as with other dense bed transport properties, two excellent sum-
maries for heat transfer data between surfaces and the bed in dense fluidized systems are by Gelperin,
N. I. and Einstein, V. G. in Fluidization (Ed. Davidson, J. F. and Harrison, D., 1971) and in Fluidization
Engineering by Kunii, D and Levenspiel, O. (1977).
Both of the above references present summaries of the “packet theory” for predicting heat transfer
coefficients from the bed hydrodynamics as opposed to using correlations that are widely available in the
literature, but might not apply well to a specific application due to slight differences in the bed proper-
ties – especially – at the local level. Again, as noted above, the work of R. Hays, S.B. Reddy Karri, R.
Cocco and T.M. Knowlton (2008) suggests that clusters might be formed within the dense bed and with
a similar approach that is taken later on in this chapter for dilute-clustering flows, a local theory might
be developed to allow these properties to be calculated based upon the local hydrodynamics.
dilute PHase: clustering flOw
Prior to getting into the specifics for each of these transport processes, it is important to briefly discuss
the hydrodynamic flow structure of the solids. The reason for this will become apparent in the discussions
on the heat and mass transfer processes. It is postulated that large macroscale clusters that might be as
large as 1 m or more in characteristic length are made up of microscale clusters that have a characteristic
length ranging from say about 6 particle diameters to maybe as many as 50 particle diameters or more.
The distribution of these cluster sizes are skewed with a relatively long tail towards the larger sizes. The
data from three different experimental tests are used to prove the postulate:
• LDV data,
• Wavelet analysis of fiber optic data and
• High speed, high resolution photographic data.
The LDV data have been analyzed and reported on in great detail (Breault, R.W., Ludlow & J.C., Yue,
P.C., (2005); Breault, R. W., Shadle, L. J. & Pandey, P. (2005);Breault, R. W., Guenther, C. & Shadle,
L.J., (2008)). The clusters shown in Figure 1 were used in the analysis by Breault, Ludlow & Yue (2005)
using statistical arguments to define clusters as noted in their paper. The three clusters had characteristic
lengths of 6, 7 and 13 particle diameters, respectively and thus categorizing each as a microcluster. The
three microclusters shown are contained in a large macrocluster spanning a time greater than 0.5 seconds
resulting in a characteristic length of a little more than 0.5 m using the average velocity of 1m/s.
Microclusters and macroclusters can also be observed with fiber optic data through wavelet analysis
of the amplitude time series data. Figure 2 shows approximately one second of fiber optic data for 200μm
glass beads. The microclusters are first found using the wavelet technique of Guenther & Breault (2007).
These are shown in the figure with the pink lines as can be seen clearly for the two microclusters at a
count position of approximately 5000. There are three macroclusters identified in the time series shown.
The first of these is observed to have four distinct microclusters in it. These have a characteristic length
of 12, 64, 6 and 108 particle diameters, respectively. The overall macrocluster characteristic length is
183
Figure 1. Microclusters from LDV data
435 particle diameters which is approximately a 20 cm length macrocluster. The third macrocluster
shown is about twice this size and the second somewhat smaller.
A macrocluster can be seen directly in the picture insert of Figure 4. In this figure, the light inten-
sity as measured in normalized grey level versus distance as traveled along the line inscribed on the
inserted picture. Traveling along the line from bottom to top, the grey level intensity is low, on the order
of 15% on average, prior to the intersection of the line with the leading edge of the macrocluster, where
the grey level jumps to a high level of about 65% on average. Looking at the macrocluster data, the
intensity rises and falls as clumps of particles are more tightly compacted and than those that immedi-
ately surround them. Identifying these higher intensity areas as microclusters gives sizes that agree with
both the LDV data and the fiber optic data. The result being that this particular macrocluster was
roughly four microclusters long, at least in the location along the length of the inserted line. These mi-
croclusters were 11, 15, 6 and 7 particle diameters with the overall being 43 particles.
Mass transfer
As a result of the review of mass transfer and dispersion conducted by Breault (2006), simple computa-
tional experiment were conducted by Guenther and Breault (2005) and Breault & Guenther (2005) with
MFIX to assess the relative importance in the choice of the mass transfer coefficient and the turbulent
dispersion coefficient within MFIX. These simple tests showed that both of these parameters have a
large effect on the predicted conversion, generating the conclusion that better information on both of
these parameters is needed. Therefore, this chapter provides methods to have the code directly compute
these parameters rather than rely on correlations that may or may not be applicable to the flow regime
being predicted by the code.
184
Figure 2. Macrocluster with LDV data
Figure 3. Microclusters and macroclusters with fiber optic data
185
Figure 4. Microclusters in macroscopic cluster as determined with analysis
Diffusional Mass Transfer
Summary of Mass Transfer Literature in Circulating Fluidized Beds

As noted above, the correlation by Gunn (1978) is the default correlation used in MFIX to provide the
mass transfer coefficient. A review of the literature by Breault(2006) revealed six experimental investi-
gations (Li & Wang (2002); Subbarao, D. & Gambhir, S. (2002); Bolland, O., (1998); Kalil, J., Chu, J.
C., & Wetteroth, W. A., (1953); Resnick, W. & White, R. R., (1949); Venderbosch, R. H., Prins, W., van
Swaaij, W. P. M., (1999)) from which correlations for mass transfer coefficients can be obtained. These
studies were generally conducted in small diameter, bench-top facilities; the exception being the work of
Bolland (1998). Also, the bulk of the work was conducted at ambient conditions or near ambient condi-
tions with the only exception being the work of Vanderbosch et al. (1999), which was conducted at 750K.
The correlations developed from the six experimental investigations above along with the correlation
by Gunn (1978) are presented graphically in Figure 5. There is little agreement between the reported
results with values of the Sherwood number ranging from a low of 10-5 reported by Bolland (1998) to a
high of 200 as reported by Kalil et al. (1953). Details of this comparison can be found in Breault (2006).
Looking at the range in values of the Sherwood number in Figure 5, leads to questioning the valid-
ity of using a modified single sphere expression of the form
Sh or Nu = 2 + a Reb Sc c (14)
especially, since a large quantity of the data fall below the value of 2.
186
Figure 5. Summary of experimental mass transfer investigations (Breault, 2006)
In addition to the literature data presented and discussed above, limited experimentation for both heat
and mass transfer in the core-annular flow regime has provided transfer results that are less than single
phase flow past a single sphere at the same gas velocity. This peculiar behavior combined with the need
to model and simulate the performance of reactors has motivated the author to reexamine the transport
phenomena from a more fundamental approach.
Theoretical Basis for Mass Transfer in Thin Films
For mass transfer into a liquid film, which moves down along a wall, from a gas containing a soluble
species a number theories have been put forward to explain the phenomena on a fundamental basis. These
are – in increasing complexity – penetration theory (Higbie, R., (1935)), penetration theory with random
surface renewal (Danckwerts, P.V., (1951)), film-penetration theory (Toor, H. L. and Machello, J. M.,
(1958)) and transfer to falling turbulent wavy films (Banerjee, S., Scott, D. S., & Rhodes, E. (1968)).
In all of these theories, the mass transfer coefficient is proportional to the square root of the diffusivity
and inversely proportional to the square root of a time parameter which represents the contact time.
4D 4 Du
k≈ ≈ sD ≈ (15)
π te πδ
where
187
Figure 6. Conceptual Picture of Riser for Mass Transfer
1 u
= ≅s (16)
te δ
where te is the exposure time, u is the eddy velocity, δ is the eddy size and s is the fractional rate of
surface renewal.
Development of Mass Transfer Coefficient for a Cluster in a CFB

In modeling the riser, it is important have a clear understanding of the phenomena driving the mass
transfer. In this case, the riser contains large streamers or macroclusters as shown in Figure 6. Each of
these macroclusters is made up of microclusters which in turn are made up of particles as shown.
In analyzing the phenomena of mass transfer in a riser operating under core-annular conditions or
fast fluidization, we will first introduce the concept that clusters are small packages of particles that
move about in a random manner not significantly different than the movement of eddies in a turbulent
single phase system, i.e. the analogy to Banerjee et al’s work on turbulent wavy films (Banerjee, S.,
Scott, D. S., & Rhodes, E., (1968)). Visualize the system to consist of clusters moving along a plane as
shown in Figure 7. In the model, clusters are made up of particles moving together as a relatively dense
suspension, possibly approaching a packed bed density which have very little internal mass transport.
That is to say that the clusters move with the fluctuating velocity, uc, collide with one another to gener-
ate new surface area at a time scale shorter than that for diffusion through the quasi-static interial region
of the clusters. These clusters have an average cluster size of δ, the measurement of the length width
and height.
188
Figure 7. Micro scale structure driving mass transfer
Given this model, the governing differential equation for mass transfer from the bulk to the cluster
is
∂C ∂2 C
uc =D 2 (17)
∂y ∂x
where
C − Ccs
C= (18)
C∞ − Ccs
and the boundary values are for Ccs being zero at the cluster surface:
y = 0, C = 0 (19)
x = 0, C = 1 (20)
x = δ, C = 0 (21)
The solution of this equation can be found in any number of texts, for example see Bird, R. B.,
Stewart, W. E. & Lightfoot, E. N., (2002), to be
189
x
C = 1 − erf (22)
4 Dδ / uc
From this the mass transfer coefficient can be found to be
4 Duc
kf = (23)
πδ
which is identical with the relationship developed by Banerjee, et al. (1968) except that the velocity term
is the cluster velocity and the length term, δ, now represents the cluster size. As mentioned above, the
surface of the cluster is renewed through the interaction of adjacent clusters. Recalling that the cluster
velocity term, uc, is the fluctuating velocity and the square of this velocity is the turbulent kinetic energy.
Therefore, the mass transfer coefficient can be expressed as
2
4D (u )c
kf = (24)
πδ
Techniques for determining cluster size and velocity have been recently been reported by Breault,
R.W., Ludlow, J.C., Yue, P.C., (2005) and Guenther, C. and Breault, R. W., (2007). Breault et al (2005)
using a statistical approach for determining the existence of a cluster from LDV data found that clusters
consist of a minimum of six particles. Therefore, the minimum cluster characteristic length can be taken
to be 6 particle diameters. Guenther, C. and Breault, R. W., (2007) using wavelet analysis techniques
of fiber optic time series data found that cluster frequency and size were functions of the solids and gas
flow rates. Summarizing those results at constant gas velocity, the cluster frequency increases 25% when
the solids flow rate increases by a factor of 5 while the cluster size increases by a factor of 3 for the
same increase in the solids flow rate. Also, at constant solids flow rate, the cluster frequency is relatively
independent of the gas velocity for nearly doubling the gas velocity while the cluster size decreases by
about 30% for this same change in the gas velocity.
Application of Mass Transfer Theory to CFD Simulation

In the CFD system, the cell solids fraction will be determined. The solids are evenly dispersed over the
entire cell volume at the solids fraction determined from the model. In reality, the solids exist in a dense
cluster phase and a dilute dispersed phase (Breault et al. (2005). It is assumed that particles move into
and out of clusters at a rate significantly fast such that all solids can be assumed to exist within clusters
for this analysis. There is a difference between clusters at the wall and clusters in the center as was found
by Guenther and Breault (2007). Clusters in the center are less dense than those at the wall and have
higher turbulent kinetic energy values.
Therefore, as noted, all solids will be represented as exiting in clusters of various sizes and various
fluctuating energies. Typical examples of these values can be seen in Figures 8 and 9 - shown for low
(<3%) and high (>3%) solids flux values. Both Figures 8 and 9 were obtained by fitting the cluster size
data to the Weibull distribution function. The cluster size data for low solids fraction produced an “L”
190
Figure 8. PDF of cluster size for 60μm particles under lean conditions
or exponential distribution as shown in Figure 8. The cluster size data for the dense condition resulted
in a distribution which is unimodal, highly skewed bell-shaped distribution as is shown in Figure 9.
Taking the solids fraction for a particular cell as determined from the CFD, the total number of aver-
age particles that this represents can be determined.
(1 − ε)Vc
N = (25)
d p3
π
6
Now, given this number, a series of random number pairs (δ, PDF) can be assumed and compared
to either Figure 8 for lean flow conditions or Figure 9 for dense flow conditions to generate a series of
clusters with random particle counts. This procedure would continue until the total particles in the series
of clusters equals the total number of particles in the cell. Calculate the mean cluster size.
For example, assume that the solids fraction is 4%, the particle size is 60μm and the cell volume is
1cm3. The number of particles, N, could be calculated to be 353,678. Selecting random valid pairs of
Cluster Size and PDF values such that the pairs fall below the curve presented in Figure 9, as shown in
Figure 10 having a total of 353,675 particles. From this, the cluster size can be obtained to be 0.00068m
– a little more than 10 particle diameters.
In a similar fashion, a second series, of length equal to the number of clusters just determined, of
random number pairs (ETK, PDF) can be assumed. These values are compared with Figure 11a for lean
conditions and Figure 11b for dense condition to ensure that the pairs identify a point below the curve.
Again, the energy data for both Figures 11a and 11b were fitted to Weibull distributions. Once a set of
energies is obtained, the mean can be determined. For this example, valid random pairs of the cluster
191
Figure 9. PDF of cluster size for 60μm particles under dense conditions
Figure 10. Cluster size PDF and random pairs
kinetic energy and the PDF were obtained by comparison to Figure 11b which resulted in a cluster ki-
netic energy value of 1.79m2/s2.
192
Figure 11. (a) PDF of cluster kinetic energy for 60μm particles under lean conditions and (b) PDF of
cluster kinetic energy for 60μm particles under dense conditions
Finally, the local interphase mass transfer coefficient for that cell can be determined from Equation
14 which resulted in a mass transfer coefficient of 0.21m2/s for an oxygen diffusion coefficient of
0.18x10-4 m2/s.
turbulent disPersiOn
The underlying theory for turbulent dispersion can be found in a number of texts that discuss the kinetic
theory of turbulence, for example see Brodkey, R. S., (1967). This theory basically relates the turbulent
dispersion to the time constant for the auto correlation function of the velocity. The theory as it appears
in most texts is applied only to single phase flow of gas and liquids. The approach was utilized first by
Chaouki, J., Godfrey, L. & Larachi, F., (1999) and later by Breault, R. W., Guenther, C. and Shadle,
L.J., (2008) to determine the turbulent dispersion of the solids phase in a multiphase flow system. The
dispersion, whether for fluid of solids, due to turbulence can be calculated from the autocorrelation of
the velocity fluctuations obtained from the velocity time series data. It should be noted here that these
micro/meso scale values are not the same as the macroscale dispersion values obtained from tracer stud-
ies and incorporated into 1-D models to account for non-ideal flow behavior.
Summarizing the above noted work, the solids turbulent dispersion coefficient can be calculated from
the fluctuating particle velocity. The first step in this process is to determine the Lagrangian autocor-
relation function in the axial direction of concern. For the calculation of the axial solids dispersion, the
axial solids fluctuating velocity will be used to determine the Lagrangian auto correlation function, fzz.
The function is evaluated from
Vz′(t ) • Vz′(t + τ
f zz = (26)
V ′ 2 (t )
193
where 〈 〉 is the operator for taking ensample average and V′z(t) is the z-component of the instantaneous
fluctuation velocity. The fluctuating velocities are calculated by subtracting the mean velocity, 〈 Vz(t)〉,
from the instantaneous velocity, Vz(t). That is
Vz′(t ) = Vz (t ) − Vz (t ) (27)
Assuming that the there is moderate to strong correlation of the data, the autocorrelation function will
decrease and eventually reach zero for increasing time lags, τ. That is to say, V′z(t) and V′z(t+τ) become
less correlated at τ increases The Lagrangian time scale, TL, is found by integration of fzz for all t, from
zero to infinity and is presented mathematically below:
TL = ∫ f zz (τ)d τ (28)
0
The axial solids dispersion coefficient can be approximated from the product of the Lagrangian time
scale, TL, and the square of the cross-sectionally averaged axial velocity, Vz.
2
Dz @ TLV z (29)
The axial dispersion coefficients at the wall were obtained from the fluctuating particle velocity data
by the autocorrelation technique discussed earlier. The Lagrangian autocorrelation function, fzz, for a
typical data set is presented in Figure 12, where fzz is plotted against the time lag, τ. The area under this
curve is the Lagrangian time scale, TL, and for this example is 0.285.
The product of TL and the cross-sectionally averaged velocity, Vz, squared is the axial dispersion
coefficient. The value of Vz is obtained from the solids flux according to equation
Gs
Vz = (30)
ρ s (1 − ε)a
Where Gs is the solids flux, ρs is the solids density and (1-ε)a is the apparent volume fraction.
aPPlicatiOn Of turbulent disPersiOn tHeOry tO cfd siMulatiOn
The application of this procedure to obtain the turbulent dispersion id the same, whether for the gas or
the solids. To begin, the velocity in the center of the cell is known at time zero. After one iteration, the
velocity is updated. At this point, the turbulent dispersion coefficient can be calculated. This is done as
follows. Calculate the quantity fii according to Equation 27, where i = x, y, or z for the time zero condi-
−t / T
tion and at time τ after 1 iteration. Fit the results to the function f ii = e L to determine TL. Calculate
2
the turbulent dispersion coefficient for that cell and direction from Di = TLV i where Vi is the average
194
Figure 12. Lagrangian autocorrelation function for a typical data set
velocity in the i direction for the cell and its neighbors. Repeat this process for the next 1 or 2 iterations,
selecting the best value for TL from the curve fit. From this iteration onward, drop the first time step
when adding the next. That is to say for iteration number 6, the velocity data for iterations 3, 4, 5 and 6
will be used. These functions decay quite rapidly such that fitting only a few points will provide a good
value for the turbulent dispersion coefficient.
Heat transfer
Heat transfer can be divided into two subcategories; heat transfer between the gas and the solids phase
and heat transfer from the process media across the system boundary. The first of these categories is
analogous to mass transfer and the following discussion on the heat transfer process between the gas and
the solids will parallel that discussion. Analysis of the heat transfer for the other category, heat transfer
between the medium and the system boundary, results in the development of a different mechanistic
model of heat transfer process. Thus, since the mechanisms differ for these heat transfer processes, they
are discussed separately below.
Heat Transfer between the Gas and the Solids Phase
As briefly alluded to above, the mechanism for heat transfer between the gas and the solid phases can
be arrived at by analogy to the discussion of mass transfer. Like mass transfer, numerous researchers
195
have investigated this process and have presented their findings in various journals. The results of their
work can be seen in Figure 13, as presented by Kunii & Levenspiel (1969). Like the mass transfer ex-
perimental investigations, the results which these investigators reported are real, only some critical piece
of information is missing. It is contended that the missing information is on the clustering phenomena
and that a mechanistic model that includes the clustering phenomena will be able to accurately predict
the heat transfer coefficient as has been shown for mass transfer.
The differential equation for heat transfer is
∂T ∂2 T
uc =α 2 (31)
∂y ∂x
which will result in a sulution of the same form. The heat transfer coefficient can be obtained as
2
4α (uc )
h= (32)
πδ
The example for mass transfer can be followed, substituting the thermal diffusivity, α, for the mass
transfer diffusion coefficient, D.
Heat transfer acrOss tHe systeM bOundary
Numerous researchers have analyzed heat transfer from the fluidized media to the system boundaries,
for example Dutta & Basu (2004), Glicksman (1997), Noymer & Glicksman (1998), Wu, Grace & Lim
(1990), Mickley & Fairbanks (1955) and Vijay & Reddy (2005). All of these investigations have mod-
eled the heat transfer coefficient from a maroscale reference relating the heat transfer coefficient to the
large, pseudo-steady clusters which are observed in the wall region. For these cases, the heat transfer
coefficient is most often presented in the following form
4kc ρ cCc 0.5

hc = ( ) (33)
πt c
In most of the recent publications noted above, the time term is obtained from estimations for the
cluster decent length, Lc, and the cluster decent velocity, Uc, such that
Lc
tc = (34)
Uc
The original work by Mickley & Fairbanks did not intend this value to be so generally described.
They specifically state that they provided the functionality for the “heat transfer coefficient
196
Figure 13. Heat transfer coefficients in fluidized systems by a number of researchers as presented by
Kunii and Levenspiel (1969) pp. 215
197
kc ρ cCc − 12
hc = (τ) (35)
π
wherein the effects of the bed thermal properties are separated from the effects of the stirring factor,
which accounts for bed motion and geometry”. In their work, the stirring factor relates the solids clusters
or packets age distribution and motion. They also state that “At present it is impossible to evaluate the
stirring term from the dynamic factors that determine it”. This statement must now be qualified given
the 50 plus years that have gone into the research and modeling of multiphase flow processes such as
the clustering that occurs in riser flow.
Gross mixing of the bed is a function of the large macroclusters rather than the microclusters, the
latter do not have enough momentum or energy to drive the large scale motion that Mickley & Fairbanks
refer to as “stirring”. Referring to Figures 2 and 3, the time in which a macrocluster is in contact with a
system boundary can be obtained. In the case of Figure 2, the time is directly obtained by knowing the
count of particles in a macrocluster (approximately 3700 for the large macrocluster on the right) and
dividing that by the frequency in which the data were recorded. For this particular cluster, this value is
0.3 s. This is one particular macrocluster, some are larger and some are smaller. A distribution of the
size of these macroclusters is needed. This distribution can be obtained by analysis of fiber optic data as
has been done here or by running an isothermal, periodic inlet/outlet simulation to generate the cluster
size and velocity. Once the distribution is at hand, the mean contact time can be obtained and the heat
transfer coefficient calculated.
references
Banerjee, S., Scott, D. S., & Rhodes, E. (1968). Mass Transfer To Falling Wavy Liquid Films In Turbu-
lent Flow. Industrial & Engineering Chemistry Fundamentals, 7(1), 22–27. doi:10.1021/i160025a004
Bird, R. B., Stewart, W. E., & Lightfoot, E. N. (2002). Transport Phenomena. New York: John Wiley
and Sons.
Bolland, O. (1998). Describing Mass transfer in Circulating Fluidized Beds by Ozone Decomposition.
Trondheim, Norway: Norwegian University of Science and Technology, Department of Thermal Energy
and Hydropower.
Breault, R. W. (2006). A review of gas–solid dispersion and mass transfer coefficient correlations in
circulating fluidized beds. Powder Technology, 163(1-2), 9–17. doi:10.1016/j.powtec.2006.01.009
Breault, R. W., & Guenther, C. (2005). Sensitivity of Gas-Solids Dispersion and Mass Transfer Coef-
ficient in an Eulerian-Eulerian CFD Modeling. Paper presented at the Spring Meeting of the American
Institute of Chemical Engineers, Atlanta, GA.
Breault, R. W., Guenther, C., & Shadle, L. J. (2008). Velocity fluctuation interpretation in the near wall
region of a dense riser. Powder Technology, 182(2), 137–145. doi:10.1016/j.powtec.2007.08.018
198
Breault, R. W., Ludlow, J. C., & Yue, P. C. (2005). Cluster Particle Number and Granular Temperature
of Cork Particles at the Wall in the Riser of a CFB. Powder Technology, 149(2-3), 68–77. doi:10.1016/j.
powtec.2004.11.003
Breault, R. W., Shadle, L. J., & Pandey, P. (2005). Granular Temperature, Turbulent Kinetic Energy and
Solids Fraction of Cork Particles at the wall in the Riser of a CFB. In Cen, K. (Ed.), Circulating Fluid-
ized Bed Technology VIII (pp. 755–761). Hangzhou, China: Int. Academic Publishers.
Brodkey, R. S. (1967). The Phenomena of Fluid Motions. Reading, MA: Addison-Wesley.
Chaouki, J., Godfrey, L., & Larachi, F. (1999). Position and Velocity of a Large Particle in a Gas/Solid
Riser using the Radioactive Particle Tracking Technique. Canadian Journal of Chemical Engineering,
77, 253. doi:10.1002/cjce.5450770210
Danckwerts, P. V. (1951). Significance of liquid-film coefficients in gas absorption. Industrial & Engi-
neering Chemistry, 43, 1460–1467. doi:10.1021/ie50498a055
Davidson, J. F., & Harrison, D. (1971). Fluidization. London: Academic Press.
Desai, P. & Wen, C. (1978). Computer modeling of the MERC fixed bed gasifier, MERC/CR-78/3.
Dutta, A., & Basu, P. (2004). An improved cluster-renewal model for the estimation of heat transfer
coefficients on the furnace walls of commercial circulating fluidized bed boilers. Transactions of the
ASME, 126, 1040–1043. doi:10.1115/1.1833360
Glicksman, L. R. (1997). Heat transfer in circulating fluidized beds. In Grace, J. R., Avidan, A. A.,
& Knowlton, T. M. (Eds.), Circulating Fluidized Beds (pp. 261–307). London: Blackie Academic &
Professional.
Guenther, C., & Breault, R. W. (2005). Two and three dimensional simulations investigating dispersion
and mass transfer coefficients in an Eulerian-Eulerian CFD model, In S. T. Johansen & I. Gamst Page,
(Ed.), Proceedings of the Fourth International Conference on Computational Fluid Dynamics in the Oil
and Gas, Metallurgical & Process Industries, Paper 53, Trondheim, Norway.
Guenther, C., & Breault, R. W. (2007). Wavelet analysis to characterize cluster dynamics in a circulating
fluidized bed. Powder Technology, 173(3), 163–173. doi:10.1016/j.powtec.2006.12.016
Guenther, C., & Syamlal, M. (2001). The effect of numerical diffusion on isolated bubbles in a gas-solids
fluidized bed. Powder Technology, 116, 142–154. doi:10.1016/S0032-5910(00)00386-7
Gunn, D. J. (1978). Transfer of Heat or Mass to Particles in Fixed and Fluidized Beds. International
Journal of Heat and Mass Transfer, 21, 467–476. doi:10.1016/0017-9310(78)90080-7
Hays, R., & Karri, S. B. Reddy, Cocco, R. & Knowlton, T.M. (2008) Small Particle Cluster Formation
in Fluidized Beds and its Effect on Entrainment. In J. Werther, W. Nowak, K-E. Wirth & E-U. Hartge,
(Ed.), Proceedings of the 9th International Conference on Circulating Fluidized Bed, (pp. 93), Hamburg,
Germany, May 13-16, 2008.
Higbie, R. (1935). The rate of absorption of a pure gas into a still liquid during short periods of exposure.
Transactions of the A.I. Ch.E., 31, 365–387.
199
Jiradilok, V., Gidaspow, D., & Breault, R. W. (2007). Computation of gas and solid dispersion coefficients
in turbulent risers and bubbling beds. Chemical Engineering Science, 62, 3397–3409. doi:10.1016/j.
ces.2007.01.084
Kalil, J., Chu, J. C., & Wetteroth, W. A. (1953). Mass Transfer in a Fluidized Bed. Chemical Engineer-
ing Progress, 49(3), 141–149.
Kunii, D., & Lenenspiel, O. (1977). Fluidization Engineering. Huntington, New York: Krieger Publishing.
Li, J., & Wang, L. (2002). Concentration Distributions During Mass Transfer in Circulating Fluidized
Beds. In 7th International Conference on Circulating Fluidized Beds, Niagra Falls, Ontario, Canada.
Mickley, H. S., & Fairbanks, D. F. (1955). Mechanism of heat transfer to fluidized beds. A.I. Ch.E.
Journal, 1(3), 374–384.
Noymer, P. D., & Glicksman, L. R. (1998). Cluster and particle-convective heat transfer at the wall of a
circulating fluidized bed. International Journal of Heat and Mass Transfer, 41(1), 147–158. doi:10.1016/
S0017-9310(97)00065-3
Resnick, W., & White, R. R. (1949). Mass Transfer in Systems of Gas and Fluidized Solids. Chemical
Engineering Progress, 45(6), 377–389.
Subbarao, D., & Gambhir, S. (2002). Gas Particle Mass Transfer in Risers. 7th International Conference
on Circulating Fluidized Beds, Niagra Falls, Ontario, Canada.
Syamlal, M. (1998). MFIX documentation, numerical technique. EG&G technical report, DE-AC21-
95MC31346.
Syamlal, M., Rogers, W., & O’Brien, T. (1993). MFIX documentation: theory guide. Technical Note,
DOE/METC-95/1013.
Toor, H. L., & Machello, J. M. (1958). Film-Penetration Model For Mass And Heat Transfer. AIChE
Venderbosch, R. H., Prins, W., & van Swaaij, W. P. M. (1999). Mass Transfer and Influence of the local
catalyst activity on the conversion in a Riser reactor. Canadian Journal of Chemical Engineering, 77,
262–274. doi:10.1002/cjce.5450770211
Vijay, G. N., & Reddy, B. V. (2005). Effect of dilute and dense phase operating conditions on bed-to-
wall heat transfer mechanism in a circulating fluidized bed combustor. International Journal of Heat
and Mass Transfer, 48, 3276–3283. doi:10.1016/j.ijheatmasstransfer.2005.03.013
Wu, L. R., Grace, J. R., & Lim, C. J. (1990). A model for heat transfer in circulating fluidized beds.
Chemical Engineering Science, 45(12), 3389–3398. doi:10.1016/0009-2509(90)87144-H
200
nOMenclature
C Concentration at location (x,y) (mole/m3)

Cc Heat capacity for suspension (kJ/kgC)
Ccs Concentration at cluster surface (mole/m3)
C0 Concentration at source far from cluster (mole/m3)
C Dimensionless concentration
dp particle diameter (microns)
D Diffusion coefficient (m2/s)
DO oxygen diffusivity (m2/s)
2
De ash layer diffusivity (m2/s)

Dmn turbulent diffusive or dispersion term (m2/s)
fii Autocorrelation function for direction i
fzz Autocorrelation function for direction z

g Force due to gravity (m/s2)
Gs Solids flux (kg/m2s)
h Heat transfer coefficient between gas and solids
hc Heat transfer coefficient to surface
ka Ash layer mass transfer conductance (m/s)
kc Thermal conductivity of suspension (kJ/mK)
kf Mass transfer coefficient (m/s)
kr Reaction rate constant (m/s)
Lc Macrocluster length (m)
N Number of particles
Nu Nusselt number
Pg pressure in the gas phase (Pa)
R rate of production of the nth gas/solids species (g/cm3s)
R0 total rate of production of the sum of all gas species (g/cm3s)
Re Reynolds number
S s solids phase stress tensor (Pa)
Sh Sherwood number
Sc Schmidt number
tc Cluster contact time as defined by Equation 33 (s)
te Effective time (s)
Tg gas temperature(K)
Ts solids temperature (K)
TL Lagrangian time scale (s)
Uc Macrocluster velocity (m/s)

vm gas/solids velocity vector (cm/s)
Xmn nth gas/solids species mass fraction
X0 initial value of Xmn
201
greeK syMbOls
α thermal diffusivity (cm2/s)

εm gas/solids void fraction
μg gas viscosity (g/cms)
ρm gas/solids density (g/cm3)
t g gas phase stress tensor (Pa)
τ cluster contact time with wall (s)
202
Section 2
Numerical Methods
204
Chapter 6
Coupled Solvers for
Gas-Solids Flows
Berend van Wachem
Imperial College, UK
abstract
In recent years, the application of coupled solver techniques to solve the Navier-Stokes equations has
become increasingly popular. The main reason for this, is the increased robustness originating from the
implicit and global treatment of the pressure-velocity coupling. The drawback of a coupled solver are the
increase in memory requirement and the increased complexity of implementation. However, fully coupled
methods are reported to have an overall favorable computational cost when a suitable pre-conditioner
and algorithm for solving the resulting set of linear equations are employed. In solving multi-phase
flow problems, the coupled solver approach is even more advantageous than in single-phase, due to the
presence of large source terms arising from the coupling of the phases. In this chapter, various strategies
for the fully coupled approach are discussed. These strategies include employing artificial compress-
ibility, applying physically consistent cell face interpolation, and applying momentum weighted cell
face interpolation. The idea behind the strategies is outlined and their advantages and disadvantages
are discussed. The treatment of source terms and volume fraction in coupled methods is also shown.
Finally, a number of examples of implementations and calculations are presented.
1 intrOductiOn
There is no dispute that on Cartesian grids, computation of incompressible gas-solids flows is best
performed with the staggered scheme proposed by Harlow and Welch (1965). However when the grid
is arbitrary, the use of collocated schemes is prevalent and more natural. One of the reasons for us-
ing collocated grids, is that the generalization from Cartesian to a general case grid with a collocated
DOI: 10.4018/978-1-61520-651-3.ch006
variable arrangement is, at least on paper, quite straightforward. The application of staggered grids to
non-Cartesian meshes is much more complicated, as the nodes containing the scalars no longer lie on
the line connecting the velocity neighbours. Staggered grids applied to non-Cartesian geometries often
suffer from accuracy problems on non-uniform grids due to the use of Christoffel symbols (Aris, 1962)
which are difficult to compute accurately on general grids (e.g. Wesseling et al., 1998). Although some
researchers have proposed remedies to this, e.g.Perot (2000); Wenneker et al. (2003), the approach is
still less natural than employing a collocated discretisation.
Although a collocated variable arrangement is more prevalent for non-Cartesian geometries, there are
a number of drawbacks and potential pitfalls when employing collocated grids as well. A straightforward
discretisation on a collocated grid results in severe pressure oscillations (Patankar, 1980), also referred
to as pressure-velocity decoupling. To remedy this, the discretisation for the momentum convective ve-
locity needs to differ from the continuity convection. Currently, there are two main approaches to solve
this on a collocated grid: the pressure-weighted interpolation (Rhie and Chow, 1983; Zwart, 1999) and
the physical consistent interpolation (Schneider and Raw, 1987; Deng et al., 1994). Moreover, apply-
ing source terms and body forces in the equation also introduces problems, as they cannot be precisely
matched by the discretisation stencil of the pressure gradient stencil.
In the physical consistent interpolation, first proposed by Scheider and Raw (1987), the complete
momentum equation is averaged to the cell faces. The resulting expression at the cell faces is then used
to close the continuity equation. Although the method is very elegant, the major problem of the method
is the interpolation of the viscous terms. Due to the size of the discretisation stencil of the viscous terms,
their interpolated stencil becomes rather large (Deng et al., 1994) and may not perform well in near-wall
regions. Moreover, the method is rather complex to code for a multiphase case, where care has to be
taken with volume fractions and various kinds of source terms.
In pressure-weighted interpolation (Rhie and Chow, 1983; Zwart, 1999), the face velocity for the con-
vective term in the continuity equation is determined by subtracting the difference between the pressure
gradient and the interpolated pressure gradient from the linearly determined face fluxes. Mathematically,
this can be seen as a filter for the velocity and pressure fields, forcing the local pressure profile to be
linear, without affecting the accuracy of the continuity equation. The drawback of this approach becomes
clear in the presence of source terms and gradients of density and volume fraction, as the physical pres-
sure profile is no longer locally linear. This is one of the subjects of the current article.
Another benefit of collocated grids is that its discretisation enables to fully couple the velocity and
pressure equations, as will be shown in this chapter. Although coupling the velocity and pressure equa-
tions lead to an increase in memory usage, the large advantage is increased robustness due to the implicit
and global treatment of the pressure-velocity coupling. Moreover, fully coupled methods are reported to
have an overal favourable computational cost when a suitable pre-conditioner and solver are employed
(Ammara and Masson, 2004).
Almost all multiphase CFD solvers today are based upon standard decoupled approaches (e.g.
SIMPLE, SIMPLER, PISO, fractional step, and other pressure projection methods, Ferziger and Peric,
2002) and most often employ a staggered variable arrangement. In these frameworks, the momentum
equations are solved with a guessed pressure, and thereafter the pressure is corrected, so the velocity
field satisfies the continuity equation. This procedure is repeated until the velocity field nearly satisfies
the momentum equation as well. In multiphase flows, these approaches require an additional iteration,
often called an outer iteration, updating the volume fraction of the phases.
205
Because of increasing computer speed and memory, the numerical solution of the gas-solids flow
equations by a fully coupled approach is an attractive trend in computational fluid dynamics (CFD) cal-
culations. This can be done with a penalty formulation (Vanka, 1985), or using a density based framework
and, for instance, applying artificial compressibility (Wesseling, 2000). A fully incompressible approach
can also be employed, in which the face velocities required for discretising the continuity equation are
interpolated from the discretised momentum equations at cell centers (Rhie and Chow, 1983; Schneider
and Raw, 1987; Deng et al., 2001).
The main advantage of this approach is an increased robustness due to the implicit treatment of
the pressure velocity coupling. Although the equations describing multiphase flows appear similar to
single-phase flow equations, their nature is often much more difficult due to the presence of volume
fractions, large source terms, and gradients of these as well as of density. This makes the requirement
for a robust solving approach even more desirable. A number of approaches for a fully coupled solving
approach are discussed.
2 artificial incOMPressibility
For illustration, the method is explained for the single phase incompressible flow equations, and the
extension to multiphase flow, where appropriate, will be treated later. Considering the single-phase
continuity equation
∂ i
u =0 (1)
∂x i
and the momentum equation
∂u j ∂ ∂p ∂τ
ρ
∂t ∂x
(
∂x ∂x
)
+ ρ i u i u j = − j + iji − Ru j − S j (2)
which are locally linearized by discretisation. In this equation, Cartesian tensor notation is applied, i.e. ui
represents the velocity in the i direction, where i can be chosen freely (1,2, or 3). The term Sj represents
a general source term in the j direction, where the term R represents the linearized source term.
The most straightforward way to solve the unknowns, i.e. the velocity field and the pressure, is to
write the linearized equations as
... ... ... ... u   RH 

   1   u1 
... ... ... ... u   RH 
   2  =  u2  (3)
... ... ... ... u   RH 
   3   u3 
... ... ... 0   p   RH p 
206
where the dots in the left matrix are determined by the discretisation scheme – they represent the linear-
ized flow equations. The “obvious” discretisation scheme does not provide a diagonal for the pressure
in the above equation; this term is zero as in Equation 3 above. This is because there is no equation of
state for pressure in incompressible flow calculations relating it to density or velocity. Although it is
possible to solve the problem by accepting the zero on the diagonal, as is beautifully shown by Kunz et
al. (1999), it requires a lot of additional effort and standard solving strategies, which typically require
the discretisation matrix to be diagonally dominant, are not applicable.
In compressible flow calculations, the coupled matrix might have a non-zero diagonal for the pressure
variables, which is determined by the equation of state used for the pressure. For compressible flows,
the continuity equation becomes
∂p ∂
β + i ui = 0 (4)
∂t ∂x
For real compressible flow problems, β is determined by the equation of state. For artificial incom-
pressibility, β is a parameter to be chosen (Wesseling, 2000). The choice of this parameter is important: it
should be chosen so the diagonal on the matrix is substantial, but it should not interfere with the physics
described in the equation. Basically, artificial compressibility introduces an fictitious speed of sound in
the fluid. This fictitious speed of sound should be far from the velocity of the flow, not to hinder with the
incompressible nature of the fluid, but on the other hand should allow some compressibility to enable
significant coupling between the velocity and the pressure drop (Peyret and Taylor, 1985). Decoupling
of the equation can also be treated by filtering the expression for density (Rauwoens et al, 2007). Ar-
tificial compressibility methods are straightforward for approximating a steady-state solution, but the
time dependence of a solution may be incorrect, as the erroneous speed of sound may interfere with the
solution. This is the main reason why the method has not been used frequently in transient problems.
3 interPOlatiOn MetHOds
Discretising the incompressible continuity equation, Equation 1, in a finite volume framework leads to
the equation
∑M ∑u
i
f
= f
s if = 0 (5)
f = faces f = faces
Where s if represents the normal vector to the cell face, with a magnitude corresponding to the cell face
area. To solve this equation, an expression for the face velocity, u if , needs to be obtained. The objective
of the interpolation methods is to obtain an expression for the face velocity from the momentum equa-
tions. The momentum weighted interpolation and the physically consistent interpolation have a different
way to obtain this.
207
4 PHysically cOnsistent interPOlatiOn
In physically consistent interpolation, introduced by Schneider and Raw (1987), the face velocities, u if ,
are found by interpolation from the complete momentum equation. The method is illustrated on a one
dimensional steady equation, as the extension to a more general equation and more dimensions is triv-
ial. The nodal points are the integer values of i, i + 1, and so forth, whereas the face value are the halves:
i +1/2 and i − 1/2. The equation considered is
du dp d 2u
ρu 0 =− +µ 2 (6)
dx dx dx
The first term is the advective term and is typically discretized by an upwind type discretisation, e.g.
u 1
− ui
du
0
i+
ρu ≈ ρu 0 2
(7)
dx ∆x
2
The other two terms can be satisfactorily approximated with central discretisation schemes,
dp p − pi
≈ i +1 (8)
dx ∆x
ui +1 − 2u 1
+ ui
d 2u i+
µ 2 ≈ 2
2
(9)
dx  ∆x 
 
 2 
Employing the above discretisation, an analytical expression for the face velocity can be found
 Pe + 2   2  ∆x  P − P 
u = ui   + ui +1    i i +1 
(10)
i+
1
 Pe + 4   Pe + 4  +   
4   ∆x 

2
2ρu 0 1 + 
 Pe 
where the so-called mesh Peclet number is given by
ρu 0 ∆ x
Pe = (11)
µ
208
The above analytical equation for the face velocity does not contain any other variable to be evalu-
ated at the cell faces; only variables evaluated at the nodes are present. Therefore, this equation can
directly close the discretized continuity equation. It provides an expression for the fluid velocity at the
cell face based upon the nodal values of velocity, pressure and possibly other variables in the problem
(such as source terms).
The method has not been used widely due to its complexity in implementation for general flow
equations. Deng et al. (1994) show the method works well, but note the large stencil for the viscous
terms. From the above 1 dimensional example it can be seen that the method has already a 5 node wide
molecule. This may lead to stability problems in some cases.
5 MOMentuM weigHterd interPOlatiOn
The momentum equations of an incompressible single phase flow including source terms are
∂u j ∂ ∂p ∂τ
ρ
∂t ∂x
( )
+ ρ i u i u j = − j + iji − Ru j − S j
∂x ∂x
(12)
where the first term indicates the transient contribution and the second the convective contribution. The
source terms are given by Sj, which is a general source term (not linear in velocity) and a linearized
source term, R, representing the part of the source term which is linear in velocity. The cell centered
finite volume discretisation at P of the velocity terms for one direction of the equation 12, is of the form
 ρV       
 P + a( u j ) + V R( u j )  u j =  a( u j )u j  − V  b( j ) P + S u j  +  ρVP  u j,O
 ∆t

P P P 

P  ∑ nb nb  P  ∑ nb nb   ∆t  P
(13)
 nb P  nb    P
where a and b are coefficients obtained from the discretisation stencils, and VP is the volume of the
 ∂p j 
computational cell surrounding point P. The net force driving the flow is given by  j + S u  , and is
 ∂x 
 
defined as
∂p  ∂p 
uj 
= + S (14)
∂x j  ∂x j 

It is important to treat the pressure gradient and the sources similar, as their sum is the contribution
to accelerating the fluid. With this, Equation 13 becomes

  (u j ) j  
unb  

1 + ρ V P
 

uj =  ∑ nb
anb 
P 
  
∆t a( u ) + V R( u )  P  (u ) 
j j j j
(u )
   a + V R 
P P P 


P P P 


209
 ∂p 
 
 ∂x j 
  ρ VP
−VP j   P j +   uPj,O
 
 ∆t  P aP + VP RP
(u )
aP + VP RP (u ) ( u j
) ( u j
) (15)
or, with the abbreviations,
ρ
c= (16)
∆t
j VP
d (u ) = j j
(17)
a(Pu ) + VP RP( u )

 a( u )u j  
j

  ∑ nb nb nb  P 
j   (18)
u = j 


(u )
a P + VP RP  (u j ) 


 


For cell Pequation 15 becomes

1 + c d (u j )  u j = uj P − d (u j )  ∂p  + c d (u j )u j,O
 p P  P P  j (19)
  ∂x  P
P P P
A similar equation can be written for all cell centers, taking boundary conditions into account. Figure
1 shows two cell center nodes (P and E) with a line connecting these two points. The point e’ is defined
in such a way that it lies in the center of this line connecting the nodes P and E. An analogous equation
can be written down for the velocity at point e’,

1 + c d (u j )  u j = uj e' − d (u j )  ∂p  + c d (u j )u j,O
 e' e'  e' e'  j (20)
  ∂x  e'
e' e' e'
Halfway the line connecting points P and E is point e’. The center of the face between P and E is
shown by point e, and the local normal at this point is given by ni. and an expression for the velocity at
point e’ is obtained by using the velocity expressions for points E and P.
d e('u )   ∂p  1  ∂p  

1  ∂p  
j
j uPj + uEj
u = − − −  j  
e'
2 1 + c d (u j )    ∂x j  2  ∂x j  2  ∂x  
 e'        E
 e' P
210
Figure 1. A two-dimensional boundary fitted cell (P) with its east neighbour (E)
j
ce' d e('u  j,O 1 j,O 1 j,O 
)
+ u − u − u  (21)
1 + c d ( u )   e' 2 P
j
2 E 
 e' e' 
It is important to notice that the coefficients of the pressure gradient terms are 1/2 and are not inter-
polated to point e, as is sometimes proposed (e.g. Choi et al., 2003), as the discretised pressure terms
3
2 ∂ p
needs to be : ∆ , hence acting as a filter on locally non-linearity. Moreover, the total pressure term
∂x 3
is of second order accuracy. The expression for the face velocity in point e’, Equation 21, is used to
obtain the cell face velocity at point e by
−j −j
∂u j i
ue = ue' + ( x − xe ) (22)
∂ x i e' e'
Equation 22 can be directly used to close the discretised continuity equation. Now, the continuity
equation is dependent upon the pressure field, and a coupled formulation can be given as,
... ... ... ... u   RH 

   1   u1 
... ... ... ... u   RH 
   2  =  u2  (23)
... ... ... ... u   RH 
   3   u3 
... ... ... ...  p   RH p 
where the matrix is given as
211
  ρV  
  p u
… aUu … aWu  ∆t + a u
P
+ V R
P P
 au
 E … aSu … bNu 
   
… 
bWu bPu bEu … bDu 
 
 
 
  ρV  
 p v v 
… aUv … aWv  ∆t + a v
+ V R
P P
 av
 E … aSv … bN 
 
P
  (24)
 
… bv
bv
bv
… bDv 
 W P E 
 


… hUc … hWc hPc hEc … hDc … g Nc 
 
gWc g Pc g Ec g Sc 
… … 
 
 
and the coefficients are given by
nf
aFi + = M f if M f <0
| nf |
nf
aPi + = M f if M f ≥0
| nf |
bPi + = l f s if
bFi + = (1 − l f )s if
hPj + = l f s fj
hFj + = (1 − l f )s fj
i
− | sf |1 ˆ sf
g F + = dˆ f + d ∑sl
i
| rP − rF | 2 f VF q = Ffaces
q q
i
| sf |
1 ˆ sf
g P + = dˆ f + d ∑sl
i
| rP − rF | 2 f VP q = Pfaces
q q
212
i
1 ˆ sf i
gQ + = df s (1 − lq )
2 VP q
i
1 ˆ sf i
gQ + = df s (1 − lq )
2 VF q
where F and Q represent the neighbouring nodes, f the surrounding faces, and q denotes the faces of
neighbouring node Q. l are the linear interpolation factors from the nodes to the cell faces.
A single phase problem consisting of n1n2n3 computational cells, leads to a matrix size of 4n1n2n3 ×
4n1n2n3.
6 treatMent Of sOurce terMs
One of the most important features of multiphase flows, is the presence of source terms originating from
closure models, such as drag models or surface tension models. These source terms are relatively large
and have large gradients over the domain, due to the presence of interfaces (gas-liquid), drag forces
(droplets, particles), and a varying gravity force due to density changes in the domain.
Although constant body forces, such as gravity for a constant density flow, can be fairly easily dealt
with (Choi et al, 2003), this is generally not the case when the sources are varying, as in most multi-
phase flows. In these cases, it is important that the disretisation of the pressure gradient matches the
representation of the local source terms. When the source terms are a function of a gradient, for instance
the surface tension force, this is fairly natural, as the discretised gradient operator for the source terms
should match the discretised gradient operator for the pressure. However, in a general case, the source
terms should closely match the discretisation of the gradient operator applied for pressure. This operator
should be applied on an integral form of the local source term,
∫ S dx
i i
Q≈ (25)
V
so that
∂Q
Si ≈ i (26)
∂x
Although the formulation of Q is probably not unique, I have chosen to base Q upon the co-variant
cell face vectors, employing weighting factors of 1/2 to determine the cell face source terms. With Q
defined at the cell face center, an approximation of the gradient of Q can be made at the cell center,
1
SiP = ∑Q f
s if (27)
∆Ω f = faces
213
which is similar to the discretisation of the pressure gradient. This representation is employed in the
discretised cell face velocity equation. Note that in some cases, the actual direction of the source terms
is important, as in gravity, and the direction should be maintained in the procedure outlined above.
7 treatMent Of vOluMe fractiOn
Considering the governing equations for a multiphase flow problem, volume fractions are introduced
in the continuity and momentum equations (Jackson, 1997),
∂α1ρ1 ∂
∂t
+ i α1ρ1u1i = 0
∂x
( ) (28)
∂α1ρ1u1j ∂ ∂ ∂
∂t ∂x ∂x
( ∂x
)
+ i α1ρ1u1j u1i = i α1τ1ij − α1 j p + R1u1j + T1 j α1 + S1j (29)
where the volume fraction of phase 1 is represented by α1, and the source term linear in the volume
fraction is denoted by T1. The procedure as outlined in the above section can be followed, the local
constants becoming
ρ
c= (30)
∆t
j α1VP
d (u ) = (31)
(u j ) j
a P
+ VP RP( u )
With these constants, the face velocity expression of equation 15 is valid for the above momentum
equations, which can be directly inserted into the discretised continuity equation,
ρVP
∆t
(
α P − α oP + ) ∑ ρα f U if s if = 0 (32)
f = faces
The source and body force terms, R, T, and S need to be treated separately, as outlined in the previous
section. For the buoyancy term, the direction of the variable T is maintained by employing equation 27
on the directionless T.
8 eulerian-lagrangian fluid-Particle MetHOd
Due to increasing computer power, discrete element models (DEM), or Lagrangian models, have be-
come a very useful and versatile tool to study the hydrodynamic behavior of particulate flows. Next to
214
the continuous phase equations, the Newtonian equations of motion are solved for each particle, and an
interaction model is applied to handle particle encounters. In the numerical framework, both the behaviour
of the particles, which details are not treated in this article, as well as the behaviour of the interstitial
fluid are considered. The motion of the gas-phase is calculated from the volume averaged gas-phase
governing equations as put forward by Jackson (1997), with the continuity equation
∂αρ ∂
∂t ∂x
(
+ i αρu i = 0 ) (33)
where α represents the volume fraction of the fluid, ρ the intrinsic density of the fluid, and ui the ve-
locity field of the fluid. The volume fraction is given by the location and number of particles from the
Lagrangian calculations. The momentum equations are given by
∂αρu j ∂ ∂ ∂
∂t ∂x
( ∂x ∂x
)
+ i αρu j u i = i ατij − α j p + Su j + T j (34)
where tija represents the stress tensor of the fluid, p represents the pressure, and S and T represent the
source terms due to interphase momentum transfer from the fluid phase to the particle phase. The dis-
cretisation of these equations is done in a similar fashion as for single-phase Navier-Stokes equations.
9 euler-euler equatiOns
The equations represented by the matrix are, respectively,
∂α1ρ1U1j ∂ ∂ ∂
∂t ∂x
( ∂x
)
+ i α1ρ1U1jU1i = i α1τ1ij − α1 j p + R1U1j + T1 j α1 + S1j + β(U1j − U 2j )
∂x
(35)
∂α 2ρ 2U 2j ∂ ∂ ∂
∂t ∂x
( ∂x
)
+ i α 2ρ 2U 2jU 2i = i α 2τij2 − α 2
∂x j
p + R2U 2j + T2j α 2 + S2j + β(U 2j − U1j )
(36)
∑∇ ⋅ α U
i
k k
=0 (37)
k =1, 2
∂α1ρ1 ∂
∂t ∂x
(
+ i α1ρ1U1i = 0 ) (38)
The coefficients for the volume fraction of the second phase are eliminated with the equation
215
α1 + α 2 = 1 (39)
The discretised momentum equation for phase k in direction j can be written as
ρ V j   j   ∂P 
 k P α + a(U k ) + V R  U j =  a(U k )U j  − V α j
 ∑ nb k , nb 
j
 ∆t k , P P k,P 
− S − α T +
P k,P P  k ∂x j k k k 
   nb P  P
ρ V 
 k P  α o U j,o (40)
 ∆t  k , P k , P
  P
From Equation 40 the net force on phase k is given by
 ∂P 
  = α ∂P − α › j − α T j (41)
 j k k k k
 ∂x  k ∂x j
j j i i
where Λ = Sk / α k . the required mass flow in the k-phase continuity equation 32, M k , f = ρ kU k , f s f , can be de-
termined after computing the normal velocity at face f
 ∂P
U ki , f sif = l f U ki , P + (1 − l f )U ki , F  s if − dˆk , f s if  fi − Ski , f − Tki, f α k , f
   ∂x

1   ∂ P   ∂ P 
+ −  Pi − Ski , P − Tki, Pα k , P  +  Fi − Ski , F − Tki, F α k , F  (42)
2  ∂x   ∂x  
 
ˆ
{
+ck d k , f U k , f s f − l f U k , P + (1 − l f )U k , F  s f
i,o i 

i,o
}
i,o  i

The final equations are inserted into a fully coupled linear system,
   
⋅
 ⋅ ⋅ ⋅ ⋅U 1   RHU 1 
   
⋅ ⋅ ⋅ ⋅ ⋅    RH  
 U 2  =  U 2  (43)
⋅ ⋅ ⋅ ⋅ ⋅   RH 
  P   P 
⋅ ⋅ ⋅ ⋅ ⋅   RH 
α1   α1  
where the coefficients in the matrix are obtained by discretising the governing equations of both phases.
A two-fluid mode problem consisting of n1n2n3 computational cells, leads to a matrix size of 8n1n2n3 ×
8n1n2n3.
216
10 bOundary cOnditiOns
Special care needs to be taken regarding boundary conditions for the pressure. The pressure variable
appears in two places, in the continuity equation and in the momentum equations. At boundaries where
the velocity is specified (Dirichlet), no additional pressure terms are required in the cell face velocity
interpolation, hence no boundary condition for pressure is required. The remaining pressure terms, in
the momentum equations, can be determined by extrapolation. Extrapolation for a variable ϕ can be
described as
∂φ i
φ f = φP + x (44)
∂x i P (W − f )
where ϕf denotes the value of variable ϕ at the wall. The above equation can be directly employed to
close the equation for the face velocity.
At boundaries where the pressure is directly specified, velocity is typically extrapolated. In this case,
the boundary pressure can be directly specified in both the continuity equation as well as the momentum
equation. The velocity at such a boundary can be obtained from equation 15 by inserting the value of
pressure at the wall. An excellent discussion of the boundary conditions can be found in Zwart (1999).
11 sOlving strategy
The coefficients in the discretized equations are determined from the discretisation stencils from the
individual terms. The convective terms are discretized using a higher order upwind scheme, the stress
and pressure terms are discretized centrally (applying Gauss law), and the transient term is discretized
with a second order backward scheme (trapezium rule).
The resulting system to be solved, Equation 23, results in a block-banded linear matrix with slightly
less favourable coefficients than the systems in the segregated counterpart. For instance, if the whole
pressure term discretisation, from Equation 21 is solved for implicitly, the resulting one dimensional
1 3 1 
coefficients for the continuity equation on a uniform Cartesian mesh are  −1 −1  which
4 2 4 
is less favourable than the regular Poisson stencil. Although applying deferred correction to the two
central pressure gradients in Equation 21 improves the convergence, it does require additional outer
iterations to be taken and might lead to a loss in robustness. The main numerical bottleneck consists in
limiting the effort devoted to the formation of a reliable preconditioner.
12 validatiOn: lid driven cavity flOw
To validate the procedure and its implementation, a lid-driven square cavity flow is considered, which
is extensively used as a benchmark for CFD code validation. This case was previously studied by Ghia
et al. (1982), whom provide a highly accurate solution. The problem comprises a Newtonian fluid in a
square box with three fixed walls, the fluid flow is generated by the movement of the upper wall.
217
Figure 2. Velocity profiles at centerline of the cavity. (a) Re = 100; (b) Re = 400
Figure 3. Lid-drive cavity flow, streamline countors. (a) Re = 100; (b) Re = 400
In the current work, simulations at two different Reynolds numbers are performed. The simulations
are carried out in two uniform Cartesian meshes with 70 × 70 and 140 × 140 grid points for Reynolds
numbers 100 and 400, respectively. Figures 2a and 2b show a comparison of the velocity distributions
at horizontal (y = 0.5) and vertical (x = 0.5) center-lines of the cavity at Reynolds numbers Re = 100
and Re = 400, respectively. The results computed using the procedure outlined in this work agree very
well with the data from Ghia et al (1982). The coarse mesh results shows a peak deviation of 10% in
the velocity profiles at Reynolds number Re = 400.
The cavity flow is characterized by a main vortex in the centre and two secondary vortices at the
lower corners for cases with Reynolds number less than 1000. Figure 3 shows the streamline contours
218
for Reynolds 100 and 400, the characteristic flow behavior can be easily observed in both simulations.
The secondary vortices increase the size with the increase of the Reynolds number.
13 cOnclusiOn
Several strategies for a fully coupled discretisation and solver have been proposed and employed for
solving gas-solids flows, including the Eulerian-Lagrangian particle method and the Eulerian-Eulerian
or two-fluid model. The physical interpolation and momentum weighted interpolation are worked out
in detail and the discretisation matrix for the latter is presented. Also, a strategy for dealing with source
terms and volume fraction gradients is presented. The global and implicit treatment of the pressure-
velocity coupling and the special treatment of source terms and volume fraction and density gradients in
the main reason for the increased robustness. The application of the fully coupled approach as described
in this chapter requires a computationally efficient and robust preconditioner and solver – more so than
the pressure correction type of approaches.
references
Ammara, I., & Masson, C. (2004). Development of a fully coupled control-volume finite element method
for the incompressible navier-stokes equations. International Journal for Numerical Methods in Fluids,
44, 621–644. doi:10.1002/fld.662
Aris, R. (1962). Vectors, Tensors, and the Basic Equations of Fluid Mechanics. London: Dover.
Choi, S.-K., Kim, S.-O., Lee, C.-H., & Choi, H.-K. (2003). Use of the momentum interpolation method
for flows with a large body force. Numerical Heat Transfer Part B, 43, 267–287. doi:10.1080/713836204
Deng, G., Piquet, J., Queutey, P., & Visonneau, M. (1994). Incompressible flow calculations with
a consistent physical interpolation finite volume approach. Computers & Fluids, 23, 1029–1047.
doi:10.1016/0045-7930(94)90003-5
Deng, G., Piquet, J., Vasseur, X., & Visonneau, M. (2001). A new fully coupled method for computing
turbulent flows. Computers & Fluids, 30, 445–472. doi:10.1016/S0045-7930(00)00025-6
Ferziger, J., & Peric, M. (2002). Computational Methods for Fluid Dynamics. Berlin: Springer.
Ghia, U., Ghia, K., & Shin, C. (1982). High-re solutions for incompressible flow using the Navier-Stokes
equations and a multigrid method. Journal of Computational Physics, 48, 387–411. doi:10.1016/0021-
9991(82)90058-4
Harlow, F., & Welch, J. (1965). Numerical calculation of time-dependent viscous incompressible flow
of fluid with a free surface. Physics of Fluids, 8, 2182–2189. doi:10.1063/1.1761178
Jackson, R. (1997). Locally averaged equations of motion for a mixture of identical spherical particles and
a newtonian fluid. Chemical Engineering Science, 52, 2457–2469. doi:10.1016/S0009-2509(97)00065-1
219
Kunz, R., Cope, W., & Venkateswaran, S. (1999). Development of an implicit method for multi-fluid
flow simulations. Journal of Computational Physics, 152, 78–101. doi:10.1006/jcph.1999.6235
Patankar, S. (1980). Numerical Heat Transfer and Fluid Flow. New York: Hemisphere Publising Cor-
poration.
Perot, B. (2000). Conservation properties of unstructured staggered mesh schemes. Journal of Compu-
tational Physics, 159, 58–89. doi:10.1006/jcph.2000.6424
Peyret, R., & Taylor, T. (1985). Computational Method for Fluid Flows. Berlin: Springer.
Rauwoens, P., Vieerndeels, J., & Merci, B. (2007). A solution for the odd-even decoupling problem
in pressure-correction algorithms for variable density flows. Journal of Computational Physics, 227,
79–99. doi:10.1016/j.jcp.2007.07.010
Rhie, C., & Chow, W. (1983). Numerical study of the turbulent flow past an airfoil with trailing edge
separation. AIAA Journal, 1(21), 1527–1532.
Schneider, G., & Raw, M. (1987). Control volume finitie-element method for heat transfer and fluid flow
using collocated variables - 1. Computational procedure. Numerical Heat Transfer Part A, 11, 363–390.
doi:10.1080/10407788708913560
Vanka, S. (1985). Block-implicit calculation of steady turbulent recirculating flows. International Journal
of Heat and Mass Transfer, 28, 2093–2103. doi:10.1016/0017-9310(85)90103-6
Wenneker, I., Segal, A., & Wesseling, P. (2003). Conservation properties of a new unstructured staggered
scheme. Computers & Fluids, 32, 139–147. doi:10.1016/S0045-7930(01)00094-9
Wesseling, P. (2000). Principles of Computational Fluid Dynamics. Berlin: Springer.
Wesseling, P., Segal, A., Kassels, C., & Bijl, H. (1998). Computing flows on general two-dimensional non-
smooth staggered grids. Journal of Engineering Mathematics, 34, 21–44. doi:10.1023/A:1004341115180
Zwart, P. (1999). The integrated space-time finite volume method. PhD thesis, University of Waterloo,
Canada.
220
221
Chapter 7
Quadrature-Based Moment
Methods for Polydisperse
Gas-Solids Flows
Alberto Passalacqua
Iowa State University, USA
Prakash Vedula
University of Oklahoma, USA
Rodney O. Fox
abstract
Classical Euler-Euler two-fluid models based on the kinetic theory of granular flows assume the particle
phase to be dominated by collisions, even when the particle volume fraction is low and hence collisions
are negligible. This leads to erroneous predictions of the particle-phase flow patterns and to the inability
of such models to capture phenomena like particle trajectory crossing for finite Stokes numbers. To cor-
rectly predict the behavior of dilute gas-particle flows a more fundamental approach based on solving
the Boltzmann kinetic equation is necessary to treat non-zero Knudsen-number and finite Stokes-number
conditions. In this chapter an Eulerian quadrature-based moment method for the direct solution of the
Boltzmann equation is adopted to describe the particle phase, and it is fully coupled with an Eulerian
fluid solver to account for the two-way coupling between the phases. The solution algorithm for the mo-
ment transport equations derived in the quadrature-based moment method and the coupling procedure
with a fluid solver are illustrated. The predictive capabilities of the method are shown considering a
lid-driven cavity flow with particles at finite Stokes and Knudsen numbers, and comparing the results
with both Euler-Euler two-fluid model predictions and with Euler-Lagrange simulations.
DOI: 10.4018/978-1-61520-651-3.ch007
intrOductiOn
The quadrature method of moments (QMOM) was proposed for the first time to solve population bal-
ance equations by McGraw (1997) for studying size-dependent growth in aerosols, and then applied to
aggregation problems by Barrett and Webb (1998). The method was extended by Wright et al. (2001)
to treat bi-variate aerosol distributions. Marchisio et al. (2003a) tested the QMOM approach for solving
the length-based population-balance equation for molecular growth and aggregation. Breakage processes
were considered in Marchisio et al. (2003b), while the QMOM implementation in CFD codes was dis-
cussed in Marchisio et al. (2003c)
In the work of Marchisio (2003c), QMOM is applied to the solution of a population balance equation
(PBE) in the size-based number density function, to treat aggregation and breakage phenomena. Fan et
al. (2004) applied the direct quadrature method of moments (DQMOM) to study aggregation and break-
age processes in gas-solids fluidized beds showing that the approach is very effective in modeling solid
segregation and elutriation, and in tracking the evolution of the particle size distribution. Recently Stru-
mendo & Arastoopour (2008) developed the Finite size domain Complete set of trial functions Method
Of Moments (FC-MOM), where the size distribution function is represented as a series expansion by a
complete system of orthonormal functions. The latter is useful when the functional shape of the number
density function is needed. In comparison, QMOM provides only the moments but is computationally
very efficient given its overall accuracy.
A particularly interesting use of QMOM is the solution of the Boltzmann kinetic equation, which
represents the foundation of the kinetic theory of granular gases. Classical Euler-Euler multi-fluid
models are based on conservation equations for the lower-order moments of the velocity-based number
density function, the unknown in the Boltzmann equation. Closures for these equations are obtained
under restrictive assumptions such as isotropy of the particle velocity fluctuations, and energy equipar-
tition, assuming a Maxwellian velocity distribution. The latter hypothesis corresponds to considering
collision-dominated flows, at the equilibrium condition, and imposes strong restrictions on the range of
the Knudsen numbers, defined by the ratio of the mean free path and the characteristic length scale of
the system (Jenkins & Savage, 1983; Gidaspow, 1994).
The behavior of a granular gas and its mathematical description are a function of its Knudsen number
(Goldhirsch, 2005). In particular, in the limit of zero Knudsen number, an elastic granular gas can be
described using the Euler equation, neglecting the viscous stresses in the Navier-Stokes equation. For
Knudsen numbers between zero and 0.01, the Navier-Stokes equation with no-slip boundary conditions
at walls provides a satisfactory description of the flow, while for Knudsen numbers between 0.01 and
0.1, partial slip boundary conditions are required to deal with the presence of a non-negligible Knudsen
layer at the walls, where the rarefaction effects start to appear. For Knudsen numbers higher than 0.1, the
Knudsen layer extends into the bulk of the granular gas, and the validity of the Navier-Stokes equation
is questionable in the whole domain. In these conditions, higher-order terms in the Chapman-Enskog
expansion or direct solutions of the Boltzmann equation are necessary to correctly describe the flow
behavior (Bird, 1994, Goldhirsch et al., 2005, Galvin et al., 2007).
Dilute gas-particle flows present the same non-equilibrium effects in the particle phase as in rarefied
granular gases, which cannot be described correctly by the multi-fluid model, because such models rely
on the assumption of an equilibrium velocity distribution (Simonin, 1991). This means it is impossible
for such models to properly predict peculiar phenomena of these flows like particle trajectory crossing
(Desjardins et al., 2008), in which the velocity distribution strongly deviates from the equilibrium, and
222
discontinuities in the particle mean velocity field are present. These intrinsic limitations of the two-fluid
model will be further discussed and clarified in this work through a comparison with the QMOM results
from a quadrature-based moment model presented here.
A better approach to treat these kinds of flows would be to solve the kinetic equation directly, in
order to avoid simplifying assumptions. However, an accurate solution of the Boltzmann equation with
a direct solver that discretizes the velocity phase space, or with a Lagrangian method is prohibitively
expensive from a computational point of view. In this context, QMOM provides a framework to find an
approximate solution of the Boltzmann equation, with an accuracy that can be increased in a rational
manner, by increasing the order of the controlled moments (Fox, 2008). QMOM represents a viable al-
ternative to direct solvers of the Boltzmann equation in various applications of interest for the aerospace
and the multiphase community, because it enables to obtain a higher-order solution of the Boltzmann
equation with a relatively low computational cost, even in large scale systems, where the application of
techniques like DSMC is not possible.
The basic idea of quadrature-based moment methods is to obtain a set of transport equations for
the moments of the velocity-based number density function, which is the unknown in the Boltzmann
equation, and to provide closures for the source terms of these transport equations and for the moments
spatial fluxes by approximating them using Gaussian quadrature formulas. The number density func-
tion is approximated as a sum of Dirac delta function in a number equal to the number of the quadrature
weights, and centered on the abscissas. The quadrature-based method does not make any further as-
sumption on the shape of the distribution function than assuming that it can be represented by a finite
set of delta functions, enabling the method to deal with highly non-equilibrium flows. This contrasts
with Grad’s (1949, 1958) method, where the moment transport equations are closed assuming that the
number density function can be approximated using an expansion into Hermite polynomials of the
Maxwellian equilibrium distribution. Moreover, Grad’s approach leads to difficulties in prescribing
boundary conditions for the higher-order moments and their spatial fluxes (Struchtrup, 2005), while
this difficulty is absent in QMOM, where all the boundary conditions used in Lagrangian solvers can
be easily accommodated by expressing them in terms of the quadrature weights and abscissas.Recently
Desjardin at al. (2008) developed a two-node quadrature-based algorithm to solve the kinetic equation
used to describe the behavior of dispersed flows. In their work, they show how the procedure is capable
of dealing with highly non-equilibrium flows like particle jets with particle trajectory crossing and finite
Stokes number flows. Fox (2008) proposed a third-order quadrature-based moment method to deal with
dilute gas-particle flows, accounting for the one-way coupling with the fluid phase and for inter-particle
collisions using the Bhatnagar et al. (1954) linear collision operator. Fox & Vedula (2010), Passalacqua
et al. (2010) derived a quadrature-based closure for the complete form of the Boltzmann collision inte-
gral for the single species case, and extended it to arbitrary numbers of species in Fox & Vedula (2010).
In the first part of this chapter the foundations of the quadrature-based moment methods are provided
to set the background of the second part, where the solution of the kinetic equation for gas-particle flows
and the quadrature closures of the collision kernels for mono and polydisperse flows are discussed. The
third part of the chapter deals with the strategies to couple the quadrature-based algorithms with a fluid
solver to describe the behavior of gas-solids flows. Both the dilute case, where only the momentum
coupling is accounted for, and the fully coupled moderately dense case are examined. Finally, a set of
applications and results are presented to show the predictive capabilities of the quadrature-based mo-
ment method.
223
bacKgrOund
Gas-solids flows can be described in analogy with granular gases (Goldhirsch, 2005), by considering the
velocity-based number density function f(v;x,t), defined so that f(v;x,t)dvdx is the average number of
particles of the considered species with velocity between v and v+dv, at position between x and x+dx,
and at the time t. In practical applications it is not important or not possible to know the exact form of
the number density function, but it is enough to know a set of its lower-order moments.
The velocity moment of order k of f is defined as
+∞
mk = ∫ v k f ( v; x ,t ) dv, (1)
−∞
where the integral is taken over all of the three-dimensional velocity phase space. Particularly important
moments are the zero-order moment, which represents the number of particles per unit volume, and
the first-order moment, which, divided by the first-order moment, provides the mean particle velocity.
The evolution of the velocity-based number density function is represented by a kinetic equation,
which, in the case of mono-dispersed particles, has the form
∂ t f + v ⋅ ∂ x f + ∂ v (Ff ) = C, (2)
where v represents the particle velocity, F the force acting on each particle (e.g. drag, gravity), and C is
the rate of change in the number density function due to binary collisions between particles. For dilute
gas-particle flows, the force acting on each particle is given by
3m p rg
F= Cd (U g − v( (U g − v ), (3)
4d p rp
where ρg, ρp, dp are the gas density, the particle density and the particle diameter, respectively, mp is the
particle mass, Cd is the drag coefficient and Ug is the gas velocity. The drag coefficient can be calculated
according to the Schiller & Naumann (1935) correlation:
24
Cd =
Re p
(
1 + 0.15Re0.687
p
, ) (4)
where the particle Reynolds number is Rep = dpρg|Ug – v|/μg, and μg is the gas kinematic viscosity.
The particle-particle collision term C can be computed, for collision-dominated flows, using the
linear collision operator proposed by Bhatnagar et al. (1954):
1
C= ( f − f ),
τ c eq
(5)
224
where τc is the local collision time and feq is the Maxwellian equilibrium distribution function. The BGK
collision operator satisfies the essential properties of the collision operator of mass, momentum and
energy conservation, it respects the H-theorem, providing a change in entropy that is always positive,
and returns Maxwellian statistics at the equilibrium. However, it leads to a Prandtl number equal to 1,
instead of the expected value of 2/3. To correct this inconsistency, Holway (1966) proposed to replace
the Maxwellian distribution in (5) with an anisotropic Gaussian distribution, originating the ellipsoi-
dal statistical BGK (ES-BGK) (Struchtrup, 2005). Andries et al. (2002) proposed a consistent BGK
model for gas mixtures. Zheng & Struchtrup (2005) extended the ES-BGK model to deal with velocity-
dependent collision frequencies, and compared different BGK models with discrete simulation Monte
Carlo results, showing good agreement with them in the case of shock waves with Mach number up to
6, and in the case of planar coquette flow with Knudsen number up to 0.5. Recently Xu (2008) devel-
oped a generalized BGK model for non-equilibrium flows, to account for the relaxation to equilibrium
through multiple particle collision processes. Ramanathan and Koch (2009) developed a Monte Carlo
method based on the BGK model, and applied it to both a pressure-driven flow through a nanochannel
and to a flow around an isolates sphere, showing results in excellent agreement with traditional DSMC.
BGK kinetic models represent a practical and computationally inexpensive approach, suitable in many
engineering applications, however a more complete description of the binary collisions between point
particles is provided by the Boltzmann collision integral (Cercignani, 1998), valid for dilute flows, where
the finite-size effects can be neglected:
C= ∫∫ e
−2
( ) ( )
f v1' ; x ,t f v'2 ; x ,t − f ( v1 ; x ,t ) f ( v 2 ; x ,t )Bdσdv 2 ,

(6)
where v1 and v2 the pre-collisional velocities, v1’ and v2’ the post-collisional ones, e is the restitution
coefficient, and B is related to the collision cross-section.
The case of a poly-disperse particle phase can be treated considering the multi-species form of the
kinetic equation (Cercignani, 1998):
∂ t f i + v i ⋅ ∂ x f i + Fi ⋅ ∂ v f i = Ci, j , (7)
i
where the collision term is evaluated by computing a Boltzmann collision integral for each of the pos-
sible pair combinations of particle-particle interactions:
Ci, j = ∑ ∫∫ e
j
−2
i, j ( ) ( )
f v'1, i ; x ,t f v'2,j ; x ,t − f ( v1, i ; x ,t ) f ( v 2 , j ; x ,t ) Bi, j dσdv 2, j .

(8)
In principle it is possible to find a solution to the kinetic equation discussed in this section by dis-
cretizing the velocity phase space, or by using Lagrangian solvers; however the computational cost of
both these approaches is too high for practical engineering applications (Fox, 2008), where, in most of
the cases, only approximate solutions are required. In this perspective, a third-order quadrature-based
moment method was developed by Fox (2008) and will be discussed in this chapter. In particular we show
how the moment method can be coupled to a fluid solver to describe gas-particle flows, and illustrate
225
its predictive capabilities through the comparison of the results with those of Euler-Euler and Euler-
Lagrange simulations in a lid-driven cavity flow with particles at finite Stokes and Knudsen numbers.
qMOM fOr tHe sOlutiOn Of tHe bOltzMann equatiOn
The quadrature-based moment methods are based on the idea of tracking the evolution in space and time
of a defined set of moments of the velocity-based number density function f. Fox (2008) considered
moments of the velocity-based number density function f up to the third order, defined as
M0 = ò f dv M i1 = ò v fdv
i
(9)
M 2
ij
= ò v v fdv
i j
M 3
ijk
= ò vv vi j k
fdv
and tracked the evolution of the following set W3 of twenty moments:
M0
M 11 ,M 21 ,M 31
(10)
M 112 ,M 122 ,M 132 ,M 22
2 2
,M 23 ,M 332 ,M 12
3 3 3 3 3 3 3 3 3 3
M 111 ,M 112 , M 113 ,M 122 ,M 123 ,M 133 ,M 222 ,M 223 ,M 233 ,M 333
Transport equations for these moments can be derived from the kinetic equation by applying the
definitions of the moments given in equation (9), where the forces acting on the particles have been
decomposed into the gravity and drag contributions. This leads to a set of twenty transport equations,
one for each of the considered moments:
1
¶M 0 ¶M i
+ =0
¶t ¶xi
2
¶M i1 ¶M ij
+ = gi M 0 + Di1
¶t ¶x j
(11)
¶M ij2 ¶M ijk3 1 1 2 2
+ = gi M + g j M + C + D
j i ij ij
¶t ¶x k
¶M ijk3 4
¶M ijkl
+ = gi M 2jk + g j M ik2 + g k M i2j + Cijk3 + Dijk3
¶t ¶xl
where repeated indices indicate summation, gi, gj, gk are the components of the gravity vector, Di, Dij,
Dijk are the source terms due to the drag force, and Cij, Cijk are those due to collisions, while the second
term on the right-hand side of equation (11) represents the moment spatial flux, showing that the role of
the moments of order n+1 is to transport the information to the moments of order n.
Once equations (11) are solved, it is possible to recover the macroscopic quantities of interest in
gas-particle flow simulations. In particular, the volume fraction is given by
226
as = Vp M 0 (12)
where Vp is the volume of a single particle. The mean velocity can be computed from the first-order
moment:
1
Us = M1 (13)
M0
in which M1 is the set of the three first-order moments of f.

The components of the granular temperature are given by
2 2 2
 M 1 
M 112  M 1 
M 222 M 332 M 1 
  2   3 
Qx = 0 −  10  Qy = −   Qz = −  , (14)
M  M  M 0  M 0  M 0  M 0 
while the granular temperature is given by
1
Q= (Q + Qy + Qz ).
3 x
(15)
Note that no assumptions, like isotropy of the velocity fluctuations and energy equipartition, typical
of the classical kinetic theory closures adopted in two-fluid models (Gidaspow, 1994), are made on the
values of the components, nor they are required to be smoothly varying in space and time (e.g., they
may be discontinuous, as it happens, for example, when particle trajectory crossing occurs).
quadrature-based Moment closures
The moment transport equations (11) can be solved only if closures are provided for the spatial fluxes
and the source terms they contain. These closures are obtained in terms of a finite set of N weights nα
and velocity abscissas Uα. In particular, following the work of Fox (2008), a set of eight weights and
eight abscissas (N = 8) is considered here. This allows the integrals in the moment definitions (9) to be
replaced by summations:
N N
M0 = ån a
M i1 = ån U a ai
a=1 a=1
N N
M ij2 = ån U a
Ua j
ai
M ijk3 = ån U a ai
U a jU ak (16)
a=1 a=1
Key to the quadrature-based moment method is the definition of a moment-inversion algorithm to

find the set of eight weights and abscissas V8 from the moments in W3. The algorithm is treated in detail
in Fox (2008), and briefly summed up here. The first step to define the inversion algorithm is to consider
the mean particle velocity vector (13), and the velocity covariance matrix
227
 M 2' − U 2 M 122' − U p1U p2 M 132' − U p1U p3 

 11 p1
sU =  M 212' − U p1U p2 M 222' − U p22 M 232' − U p2U p3 

 2' 
 M 31 − U p1U p3 M 322' − U p2U p3 M 332' − U p32 

where Mij2’ = Mij2/M0 is the normalized second-order moment. The inversion is performed by introduc-
ing the vector
X = A−1 ( v − U p ).
where the linear transformation A is defined by the lower Cholesky decomposition of the velocity cova-
riance matrix. The set of moments W3 is then normalized with respect to X, defining a set of moments
for each X component with the form
{
Wi 3* = mi0 = 1, mi1 = 0, mi2 = 1, mi3 . }
Weights and abscissas in each direction are computed by applying the two-node quadrature formulas
to the three set of pure normalized moments Wi3*:
1 − 2γi
ni1 = 0.5 + γ i , X i1 = −
1 + 2γi
,
1 + 2γi
ni 2 = 0.5 − γ i , X i2 =
1 − 2γi
where
0.5mi3
γi = .
2
4+ m ( )
3
i
With this operation, the sets of univariate weights and abscissas are found:
V1* = {n11, X 11; n12 , X 12 } ,

V2* = {n21, X 21; n22 , X 22 } ,
V3* = {n31, X 31; n32 , X 32 } .
The three dimensional quadrature approximation is then defined using the tensor product of the
univariate abscissas, defining V8*:
228
V8* = (
 n*, X , X , X , n*, X , X , X ,
 1 11 21 31 2 12 )( 21 31 )
( )(
n3 , X 11, X 22 , X 31 , n4 , X 12 , X 22 , X 31 ,
* *
)
(n , X *
4 11
, X 21, X 32 ), (n , X
*
6 12
, X 21, X 32 ),
(n , X *
7
, X 22 , X 32
11 ), (n , X
*
8 12
, X 22 , X 32 ) .
Finally, the weights ni* are determined by imposing the constraints given by the univariate nodes
and solving the corresponding system of linear equations:
n1* + n3* + n5* + n7* = n11,

n2* + n4* + n6* + n8* = n12 ,
n1* + n2* + n5* + n6* = n21,
n3* + n4* + n7* + n8* = n22 ,
n1* + n2* + n3* + n4* = n31,
n5* + n6* + n7* + n8* = n32 .
The rank of the linear system reported above is four, as a consequence four additional equations are
required, three of them are obtained by observing that the three second-order cross moments are null due
to the linear transformation applied to the original set of moments, and the fourth equation is obtained
by rewriting the third-order moment X1X2X3, whose value is known, in terms of weights and abscissas.
Once the quadrature weights and abscissas are computed, it is possible to evaluate the moment
spatial fluxes and the source terms in (11) as follows. The spatial fluxes, whose role is to transport the
information from the higher-order to the lower-order moments, are represented according to their kinetic
formulation (Perthame, 1992) by considering the definition of the moments and decomposing them in
two components, one over the negative velocity values and the other over the positive ones. For example,
to compute the spatial flux of the zero-order moment, we can decompose the first-order moment, whose
spatial derivative provides the desired spatial flux, in two components
(∫ ) (∫ )
0 +∞
M i1 = ∫
−∞
vi f dv j dvk dvi + ∫
0
vi fdv j dvk dvi (17)
and evaluate the two integrals independently. Once this is achieved, the first-order moment can be
reconstructed, and its spatial derivative can be calculated to obtain the spatial flux. With a similar de-
composition procedure it is possible to define the spatial fluxes for all the moments transport equations.
This decomposition is essential to obtain a stable solution of the moment equations when the mo-
ment fields are not continuous, as in the case of shock wave propagation, or when particle trajectory
crossing occurs in gas-particle flows. In these situations, the conventional evaluation of the spatial flux
derivatives based on finite differences would fail because of the discontinuity in the properties fields. If
the number density function is approximated with a delta-function representation:
229
N
f ( v) = ∑ n δ ( v − U ),
α α (18)
α=1
expression (17) for the first-order moment can be rewritten in closed form as
N N
M i1 = ∑ na min (0,U ia ) + ∑ na max (0,U ia ). (19)
a=1 a=1
Analogous expressions are found for the higher-order moments.

Closures for the collision source terms can be easily found if the BGK collision operator is used, by
applying the moment definitions to (5):
M0
Cij2 =
τ
(σeq − σij )
(20)
1
(
C3ijk = ∆ijk − M ijk3
τ
)
where σeq and σ are the variance of the Maxwellian distribution and of f, respectively, while Δijk is the set
of third-order moments of the Maxwellian distribution. It is worth to notice that the solution procedure
presented here for the Boltzmann equation, and valid for dilute flows, can be easily extended to solve
the Boltzmann-Enskog equation, accounting for finite-size particles. The core of the procedure remains
unaffected, because the difference is all contained in the collision integral, which would simple lead to
a different formulation of the corresponding source terms in the moments transport equations. In the
case of a gas constituted by point particles, the moment transport equations provide results consistent
with the hydrodynamic limit for small Knudsen numbers, and correctly degenerate in the Euler equation
for Kn equal to zero.
Quadrature-based closures for the Boltzmann collision integral were obtained by Fox & Vedula
(2010), Passalacqua et al. (2010). Only the closure equations are reported in this work, and the reader
is redirected to the original works for their derivation.
The source term for the i-th order moment for the species k provided by equation (8) can be written
in terms of quadrature weights and abscissas as
1 N N
Ckl l l =
12 3
∑ ∑ n n mk β g I
2 α =1 β =1 α β α αβ αβ l1l2l3
where mαk is the mass of a particle of specie k, gαβ = |Uα-Uβ| is the relative velocity of the two particles
involved in the collision, l1, l2 l3 are the indices of the moment (i.e., 000 for the zero order moment, 100,
010 and 001 for the first-order moments, and so on), and
βαβ = π ( rα + rβ )
230
Table 1.
I 000 = 0 ωg1 ωg 2 ωg 3
I100 = − I 010 = − I 001 = −
2 2 2
2 2
ω2 g 2 ω g1 ω2 g1 g 2 ωg 2 ωg
I 200 = + − ωg1U α1 I110 = − U α1 − 2 U α 2
12 4 4 2 2
2
ω2 g 2 ω g 2
2
ω2 g1 g 3 ωg 3 ωg
I 020 = + − ωg 2U α 2 I101 = − U − 1 Uα3
12 4 4 2 α1 2
2 2
ω2 g 2 ω g3 ω2 g 2 g 3 ωg 3 ωg
I 002 = + − ωg 3U α 3 I 011 = − U − 2 Uα3
12 4 4 2 α2 2
ω3 g 2 g1 ω3 g13  ω2 g 2 3ω2 g12  3ωg1 2

I 300 = − − +  + U − U α11
8 8  4 4  α1 2
ω3 g 2 g 2 ω3 g 23  ω2 g 2 3ω2 g 22  3ωg 2 2
I 030 = − − +  + U − Uα 2
8 8  4 4  α 2 2
ω3 g 2 g 3 ω3 g 33  ω2 g 2 3ω2 g 32  3ωg 3 2
I 003 = − − +  + U − Uα3
8 8  4 4  α 3 2
ω3 g 2 g 2 ω3 g12 g 2 ω2 g1 g 2  ω2 g 2 ω2 g  ωg 2 2
I 210 = − − + U α1 +  + 1
U − U − ωg1U α1U α 2
24 8 2  12 4  α 2 2 α1
ω3 g 2 g 3 ω3 g12 g 3 ω2 g1 g 3  ω2 g 2 ω2 g  ωg 3 2
I 201 = − − + U α1 +  + 1
U − U − ωg1U α1U α 3
24 8 2  12 4  α 3 2 α1
ω3 g 2 g 3 ω3 g12 g3 ω2 g2 g3  ω2 g 2 ω2 g  ωg 3 2
I 021 = − − + U α 2 +  + 2
U − U − ωg 2U α 2U α 3
24 8 2  12 4  α 3 2 α2
ω3 g 2 g1 ω3 g12 g 2  ω2 g 2 ω2 g 22  ω2 g1 g 2 ωg
I120 = − − +  + U + U α 2 − ωg 2U α1U α 2 − 1 U α2 2
24 8  12 4   α1
2 2
ω3 g 2 g1 ω3 g12 g 3  ω2 g 2 ω2 g 32  ω2 g1 g 3 ωg
I102 = − − +  + U α1 + U α 2 − ωg 3U α1U α 3 − 1 U α2 3
24 8  12 4  2 2
ω3 g 2 g 2 ω3 g 2 g 32  ω2 g 2 ω2 g 32  ω2 g 2 g3 ωg
I 012 = − − +  + U α 2 + U α 2 − ωg 3U α 2U α 3 − 2 U α2 3
24 8  12 4  2 2
ω3 g1 g 2 g 3 ω2 g 2 g 3 ω2 g1 g 3 ω2 g1 g 2 ωg
I111 = − + U α1 + Uα 2 + U α 3 − 3 U α1U α 2
8 4 4 4 2
ωg ωg
− 2 U α1U α 3 − 1 U α 2U α 3
2 2
is the collision cross-section. The expressions for I(i) are found by integrating over the collision cross-
section and have the following forms for the 20 moments up to third order (see Table 1)where g=|Uα-Uβ|
is the relative velocity magnitude and g1, g2, g3 are the relative velocity components, while ω is given
by
231
m*(1 + e)
ω= ,
m + m*
in which m is the mass of the particle the collision term is computed for, m* is the mass of the particle
of the other species and e is the restitution coefficient.
The source terms due to the drag force are computed by writing the force exerted by the fluid on each
particle as a function of the characteristic drag time τα,:
mp
Fα =
τα
(U g
− U α ). (21)
Introducing this notation, it is possible to write the drag source terms as a function of the weights
and abscissas:
N
na
Di1 = ∑m Fia
a=1 p
N
na
D = 2
ij ∑ m (U ia
Fja + U ja Fia )
a=1 p
N
na
Dijk3 = ∑ m (U ia
U ja Fka + U jaU ka Fia + U iaU ka Fja ) (22)
a=1 p
It is worth noting that the drag time is computed for each node α, using the corresponding abscissa
in the evaluation of the relative velocity. This detail will be important when the coupling of the QMOM
algorithm with a fluid solver is discussed.
The moment transport equations (11) can be solved using the finite-volume method for the spatial
integration, and Runge-Kutta methods for integration in time. The solution sequence proposed in Fox
(2008) is the following:
• Specify the initial conditions in terms of weights and abscissas, which are the only information
required to start the solution process, and to eventually restart a calculation. In particular, it is not
necessary to explicitly specify the initial values of the moments, because this is done implicitly
when weight and abscissas are initialized, since it possible to compute the moments from the
weights and abscissas using (16).
• Compute the time step as the minimum of the time steps imposed by the local Courant number,
evaluated in terms of the maximum quadrature velocity abscissa in each computational cell, the
drag time and the particle collision time.
• Account for the effect of the spatial fluxes by solving the expression
dW
= G (W ) (23)
dt
232
where W represents the set of twenty moments, and G(W) is the change in the moments due to the
spatial fluxes, which can be computed for all the moments using equations with the form shown
in (19).
• Account for the effect of collisions by solving
dW
= C (W ) (24)
dt
where C(W) is the rate of change in the moments due to collisions.

• Account for the effects of gravity and drag. This step differs from the previous two, because it is
possible to operate directly on the quadrature weights and abscissas. Drag and gravity do not af-
fect the quadrature weights (we assume that particles do not break or aggregate, hence the local
number density changes only due to convective phenomena), as a consequence, it is possible to
write
dna (25)
=0
dt
while the gravity and drag influence the velocity abscissas:
dU ia Fia
= - gi (26)
dt mp
where gi is the gravity component in the i-th direction. It is worth noting that the evaluation of the
drag contribution is moved here at the end of the solution procedure of the moment transport equa-
tions, so that the drag source term that has to be introduced in the fluid-phase momentum equation
in gas-solids flow simulations is not altered by other phenomena.
• Every time an operation is performed on the space of the moments, it is necessary to apply the
inversion algorithm to compute the updated weights and abscissas, and to perform the projection
step, re-computing the moments using the weight and abscissas, to ensure that the moments are
realizable.
• At the end of each time step, boundary conditions are applied. In this work, perfectly reflective
boundary conditions are used, which can be represented directly in terms of the weights and ab-
scissas, as done in Fox (2008). If i=0 represents the wall, oriented along the first axis of the refer-
ence frame, and i=1 the neighboring computational cell, the boundary condition can be written as
 n   n / e 
 a   a w 
U    
 1,a  =  U1,a 
U  −e U  (27)
 2,a   w 2,a 
U 3,a   U 3,a 
i= 0 i=1
233
where ew is the particle-wall restitution coefficient.
cOuPling Of qMOM witH a fluid sOlver
The quadrature-based algorithm for the solution of the particle kinetic equation presented in the previ-
ous section has to be coupled to a fluid-flow solver to obtain a complete description of a gas-particle
flow. Two situations are possible, according to the volume fraction of the particulate phase present in the
system and the intensity of the interaction between the phases. If particles do not affect the continuous
phase, or their influence is negligible, the phase coupling can be assumed to be uni-directional, in the
sense that only the influence of the gas phase on the solid phase has to be taken into account. Otherwise,
when the presence of particles cannot be neglected, it is necessary to account for the reciprocal interac-
tion of one phase on the other.
One-way coupling
The case of very dilute gas-particle flows, with limited mass loading, is straightforward because the ef-
fect of the presence of the particles on the fluid flow can be neglected, and the coupling is realized only
accounting for the momentum transfer from the fluid to the particle phase. In this situation, the fluid
flow equations have the usual form adopted for single-phase flows:
∂
(r ) + ∇ ⋅ (rg Ug ) = 0
∂t g
∂
(r U ) + ∇ ⋅ (rg Ug Ug ) = ∇ ⋅ tg + rg g
∂t g g
(28)
where τg is the gas-phase stress tensor, usually modeled assuming the fluid to be Newtonian (Bird et al.,
2002). The equations for the gas phase can be solved using the traditional SIMPLE (Patankar, 1980)
algorithm, until a steady solution for the fluid phase is obtained. At that point, the integration of the
moment transport equations can be performed as explained in the previous section, until a steady state
is reached by the dispersed phase.
two-way coupling
The one-way coupling discussed in the previous section is not suitable to deal with denser flows or with
situations where the mass loading, defined by
εs ρs
ξ= , (29)
εg ρg
is not small, because in these cases the hypothesis of uni-directional action of the continuous phase
on the dispersed phase cannot be considered valid, and the effect of the dispersed phase on the fluid is
not negligible. Under these conditions, the coupling happens at two levels, because it is necessary to
234
account for the momentum exchange between the fluid phase and each particle phase (i.e., the velocity
abscissas), and for the effect of the presence of the particle phase (i.e., the solids volume fraction). Inter-
phase momentum exchange is accounted for through the introduction of a source term in the gas-phase
momentum equation, while the volume occupied by the particle phase is considered by introducing the
fluid-phase volume fraction in the continuity and momentum equations of the fluid phase.
The gas-phase continuity (30) and momentum equations (31) have the usual form adopted in two-
fluid models:
∂
(ε ρ ) + ∇ ⋅ (εg ρg Ug ) = 0
∂t g g
(30)
∂
(ε ρ U ) + ∇ ⋅ (εg ρg Ug Ug ) = ∇ ⋅ tg + εg ρg g + M gs
∂t g g g
(31)
where mass transfer, heat transfer and chemical reactions are not considered, and where εg is the gas-
phase volume fraction, τg is the gas-phase stress tensor, and Mgs is the momentum exchange term that
accounts for the the interaction among the gas phase, and the dispersed phases. The way this source term
is computed in QMOM is peculiar to the algorithm, and depends on how the drag term is evaluated as a
function of the velocity abscissas instead of the mean solids velocity. In the general case of M dispersed
phases (e.g., multiple particle sizes), Mgs is then computed as
( )
M N
M gs = ∑∑ ² gs
m
,a U sm ,a − U g (32)
m=1 a=1
where βgsm,α is the gas-particle drag coefficient evaluated as a function of the quadrature velocity abscissa
at node α for dispersed phase m.
It is well known (Oliveira & Issa, 1994; Karema & Lo, 1999) that, when the drag coefficient be-
comes large, iterative solvers can fail to converge or the convergence rate is very slow if the drag term
is managed explicitly. A robust treatment of the momentum transfer term can be obtained by applying
the partial elimination algorithm (Spalding, 1980). The procedure needs to be adapted when QMOM
is used to describe the particle phase, because no mean momentum equation is solved. For simplicity,
let us consider the case of a gas phase and a single particle phase. Using the usual finite-volume nota-
tion (Ferziger & Peric, 2001), it is possible to write a pair of semi-discretized momentum equations for
the gas and for the virtual particle phase represented by a single abscissa. In other words, to obtain the
partially eliminated equations for the fluid phase, we imagine that each abscissa represents a different
particle phase, and we formally write a semi-discretized momentum equation for this fictitious phase:
( )
N
AgPU gP + ∑ Agf U gf = Sg + ∑ βgs ,α U sP ,α
− U gP
f α=1 m
(33)
(
AsP,αU s,Pα + ∑ Asf,αU sf,α = Ss ,α + βgs ,α U gP − U sP,α )
f
235
where the first term represents the central term evaluated at the cell centroid, the second is the contri-
bution due to the neighboring cells, S contains all of the source terms, and the last element represents
the momentum exchange. It is worth to notice, for clarity, that the momentum equation for the virtual
phases represented by the abscissas are not solved during an actual simulation, because all the required
information to compute the coefficients that appear in the partially eliminated fluid-phase momentum
equation can be obtained from the QMOM algorithm.
Grouping the second term on the left-hand side and the first on the right-hand side of (33):
H g = Sg − ∑ Agf U gf
f
(34)
H s ,a = Ss ,a − ∑ Asf,aU sf,a
f
leads to
N
AgPU gP = H g + ∑ βgs ,α U sP,α − U gP ( ) (35)
α=1
P
A U
s ,α
P
s ,α
= H s ,α + βgs ,α (U P
g
−U P
s ,α )
Rearranging the second expression in (35), an expression for the particle velocity abscissa, to elimi-
nate it in the gas-phase semi-discretized momentum equation, is obtained:
1 H + β U  = η H + β U 
U s,Pα = (36)
A + βgs ,α 
P s ,α gs ,α g  s ,α  s ,α gs ,α g 
s ,α
By substituting (36) into the first expression in (35), and collecting Ug, the partially eliminated equa-
tion for the gas-phase momentum is obtained:
 N 
U g = ηg  H g + ∑ βgs ,α ηs ,α H s ,α  (37)
 
 α=1 
where ηg is given by
1
ηg = N
. (38)
A + ∑ ηs,α βgs ,α A
P
g
P
α
α=1
The general case of an arbitrary number of dispersed phases M is obtained from the previous one in
a straightforward manner. The semi-discretized form of the gas-phase momentum equation is
( )
M N
AgPU gP = H g + ∑ ∑ βgs ,α U sP ,α − U gP . (39)
m m
m=1 α=1
236
For each dispersed phase, it is possible to consider a semi-discretized equation in the form
m m m m
(
AsP ,αU sP ,α = H s ,α + βgs ,α U gP − U sP ,α
m
) (40)
where only the momentum exchange between a given dispersed-phase velocity abscissa and the gas phase
are considered. The drag between different particle phases, due to collisions, is accounted for directly
in the collision term in QMOM, and its effect is already accounted for in the particle velocity abscissa
that the QMOM algorithm provides to the gas flow solver.
Applying the same procedure shown in the single dispersed-phase case, obtaining the particle velocity
from (40), and substituting it in (39), the final equation for the gas-phase velocity is obtained:
1
U gP = .
M N (41)
A + ∑ ∑ ηs ,α βgs ,α A
P
g
P
α
m m
m=1 α=1
The gas-phase volume fraction can be computed directly from the particle-phase volume fractions using
M
eg = 1 − ∑ es ,m . (42)
m=1
If the quadrature weights are properly scaled so that the sum of the weights provides directly the volume
fraction corresponding to each dispersed phase, which is convenient also to limit the truncation error in
the moment-inversion algorithm when the number density is high, (42) leads to
M N
εg = 1 − ∑ ∑ nm,α . (43)
m=1 α=1
The model equations can be solved in sequence, using the SIMPLE algorithm to deal with the gas-
phase equations (Patankar, 1980; Syamlal et al., 1993), calling the QMOM routine to solve the moment
transport equations at each time step, to update the quantity fields of the particle phases. It is worth noting
that the solution algorithm used for the moment transport equations (Fox, 2008) uses explicit methods.
To reduce the computational cost of the solution process, it is possible to implement the algorithm so
that a certain number of time steps are performed internally to the QMOM algorithm per each fluid flow
time step, avoiding the expensive operations required to solve the momentum and the pressure Poisson
equation for the fluid, when not required, because the characteristic time scale of the fluid is, in such
conditions, significantly larger than the particle collision time (Boyalakuntla & Pannala, 2006). The
quadrature-based moment method described above has been implemented in the gas-solids flow solver
MFIX and is referred to as MFIX-QMOM.
237
Figure 1. Schematic representation of the two-dimensional lid-driven cavity
Table 2. Fluid and particles properties
ρgU lid Llid

Re = 100
µg
2
rp  d p 
St =   Re 1
18rg  Llid 
dp
Kn = 1
6 2as Llid
L 1
aPPlicatiOn tO gas-Particle flOws
In this section an application of the method presented in this chapter is illustrated. A two-dimensional
lid-driven cavity, with mono-dispersed particles is considered, and results provided by the MFIX-QMOM
algorithm are compared with Euler-Lagrange and Euler-Euler two-fluid model predictions, to show the
predictive capabilities of the quadrature-based method (Garg et al., 2008). The system examined is rep-
resented in Figure 1, while the properties of the fluid and the particles are summed up in table 1 where
Re, St, and Kn are the Reynolds, Stokes, and Knudsen numbers, respectively, and L is the length scale
of the cavity. Note that gravity effects are neglected in this example.
The cavity is initially filled with the fluid and with uniformly distributed mono-dispersed particles
with properties chosen to respect the constraints given by the Knudsen, Reynolds and Stokes numbers
of Table 2. Both phases have the initial mean velocity set to zero. At t = 0 the top wall of the cavity starts
to move as show in figure 1, with a constant velocity Ulid, inducing the fluid and the particles to move.
In all of the simulations, particle-particle and particle-wall collisions are assumed to be perfectly elastic.
In addition, particle-wall collisions are assumed to be perfectly specular.
The Lagrangian simulation was performed using an in-house Euler-Lagrange code (Garg et al., 2008),
while the two-fluid model simulations were performed using MFIX (Syamlal et al., 1993) with the ki-
netic theory closures for the particulate phase of Gidaspow (1994), and the Johnson & Jackson (1987)
238
Figure 2. Volume fraction contour plots of the particle phase at t=20s predicted by the Euler-Lagrange,
Euler-Euler two-fluid, and MFIX-QMOM models
boundary conditions at the walls for both the particle velocity and the granular energy, with specularity
coefficient equal to zero. Collisions were described in MFIX-QMOM using the BGK linear collision
operator.
Representative volume fraction contours of the particle phase are shown in Figure 2, comparing the
predictions of the three solvers in the case of Stokes and Knudsen numbers equal to one. The Euler-
Lagrange and MFIX-QMOM results agree well in the prediction of the peculiar structure of the flow
that forms in the top corners of the cavity. Particles are transported by the vortex induced in the fluid by
the moving lid and segregate at its borders, with a clear tendency to accumulate in the top-right corner
of the cavity. It is worth considering what happens in the top corners of the cavity with more attention:
particles hit the wall because they are transported by the fluid, and then are reflected specularly by the
walls. At the top-left corner, they are removed towards the right by the movement of the lid, while in
the top-right corner they are reflected in the opposite direction with respect to the flow. This leads to the
formation of the structure that causes the spread of the particles in the centre of the cavity.
At the top-right corner of the cavity the peculiar phenomenon of particle trajectory crossing, sche-
matically represented in Figure 3, takes place. The particles reflected by the wall, and the particles going
towards the wall have intersecting trajectories, but the particles do not collide with one another due the
relatively low volume fraction of particles. As is clear from the comparison in Figure 2, this character-
istic of the flow is not captured by the two-fluid model simulation.
In the dilute limit, the two-fluid model can be considered as analogous to a quadrature-based moment
method, where only one velocity abscissa is used. Thus, only the local mean solids velocity, which is
proportional to the first-order moment, is considered. As a consequence, in each computational cell, the
number density function is represented by a single Dirac delta function, located in velocity phase space
at the local mean solids velocity. When the trajectories of two particles cross each other at a given point,
two different velocity vectors are defined at that point. By definition, such behavior cannot be repre-
sented by the mean velocity alone. In other words, the local velocity distribution strongly deviates from
a Maxwellian distribution, which is assumed to be the particle velocity distribution in the derivation of
the two-fluid model. Hence, because it cannot predict multiple velocities at a single point, the two-
fluid model is unable to properly predict trajectory crossing phenomena.
It is evident from the previous considerations that where particle trajectory crossing takes place, the
mean particle velocity field shows discontinuities due to the presence, locally, of particles with different
velocities, as it is possible to notice in figure 4a. Similar considerations can be applied to what happens
in the top-left corner, and along the length of the top wall of the cavity, where particles are respectively
239
Figure 3. Schematic representation of the particle trajectory crossing at the top-right corner of the lid-
driven cavity
reflected and then transported towards the right side of the cavity, until they meet the front of reflected
particles coming back after having collided with the wall on the right side of the cavity. This phenomenon
produces a shock-like structure where reflected particles accumulate at the point where the oncoming
fluid stops their advancement towards the left-hand wall. Due to the high local solids volume fraction,
the interaction with the gas phase is large at the accumulation point, and the gas is diverted downward
carrying particles with it. The resulting complex flow structure is present in the Lagrange-Euler and
MFIX-QMOM simulations, but cannot be predicted by the standard two-fluid model.
When particle trajectory crossing happens, a significant increase in the granular temperature is pres-
ent, as shown in Figure 4b. For example, at the top-right corner of the cavity, the particles hitting the
wall and those reflected by the wall give origin to a zero net flux of particles, which corresponds to a
zero mean local solids velocity. However the particle abscissas have non-zero velocity, which means
that the velocity variance is high, leading to locally high values of the granular temperature.
future researcH directiOns
The quadrature-based algorithm discussed in this chapter showed the capabilities of the approach to deal
with highly non-equilibrium gas-particle flows, being able to capture behavior such as particle trajectory
crossing due to rebounds at walls. A mono-disperse case, in a lid-driven cavity was considered, with per-
fectly elastic collisions between particles. Future work will be in the direction of extending the procedure
to track the evolution of higher-order moments, in order to improve the accuracy of the predictions. In
particular, it would be interesting to compare the results obtained controlling the entire set of third-order
moments, versus the ones provided by the current algorithm, where only fourteen of the twenty moments
are controlled; because of the role these moments play in the computation of the heat flux.
240
Figure 4. Contour plots of the mean particle velocity (a) and granular temperature (b) after 20s of
simulation
A further step is to include the possibility of accounting for particle breakage and aggregation, to be
able to describe phenomena like the formation of fines in a riser, or the sticking of particles together.
Incorporation of heat transfer, mass transfer and chemical reaction with the QMOM algorithm presented
here is straightforward if the chemical reactions do not affect the particle phase, in the sense of chang-
ing their size or their velocity, because it would imply no changes to the moment transport equations.
In-depth validation of the model both in the case of mono-dispersed and poly-dispersed systems is then
necessary to explore the range of applicability and the accuracy in different practical applications like
fluidized beds, riser flows, and particle jets.
Finally, we should note that the finite-volume algorithms used in the current implementation of MFIX-
QMOM are first order in space and time. The extension of the algorithms to higher order to increase
the accuracy for a given spatial and temporal grid is a logical next step. In principle, this can be done
using existing numerical methods for compressible flows. However, care must be taken to ensure that
the higher-order expressions for the spatial fluxes generate realizable moments (i.e., moments that can
be generated by a non-negative velocity distribution function). The latter is guaranteed for first-order
kinetic methods, but must be built into higher-order methods.
cOnclusiOn
A quadrature-based moment method for the solution of the kinetic equation describing the evolution of a
granular phase was coupled with a fluid solver to simulate the behavior of fully coupled dilute gas-particle
flows. The applicability and the capabilities of the method were discussed in comparison to traditional
two-fluid models, pointing out the limitations of the latter due to the simplifying hypothesis their deri-
vation relies on. The procedure to couple the QMOM algorithm with a fluid solver was explained both
in the case of one-way and two-way coupled flows. An example of application in a lid-driven cavity
with particles was provided, showing the good agreement between the predictions of the QMOM-based
algorithm and the Eulerian-Lagrangian simulations. In particular, MFIX-QMOM was able to properly
capture the peculiar structures of the flow due to particle trajectory crossing, which were not predicted
by the two-fluid model.
241
references
Andries, P., Aoki, K., & Perthame, B. (2002). A consistent BGK-type model for gas mixtures. Journal
of Statistical Physics, 106, 993–1018. doi:10.1023/A:1014033703134
Barrett, J. C., & Webb, N. A. (1998). A comparison of some approximate methods for solving the
aerosol general dynamic equation. Journal of Aerosol Science, 29(1-2), 31–39. doi:10.1016/S0021-
8502(97)00455-2
Bhatnagar, P. L., Gross E. P., & Krook M. (1954). A model for collisional processes in gases. I. Small
amplitude processes in charged and neutral one-component systems. Physical Reviews 94, 4(3), 511 – 525.
Bird, R. B., Stewart, W. E., & Lightfoot, E. N. (2002). Transport phenomena. New York: John Wiley
& Sons, Inc.
Boyalakuntla, D. S., & Pannala, S. (2006). Summary of Discrete Element Model (DEM) implementation
in MFIX. Retrieved from http://www.mfix.org/documents/MFIXDEM2006-4-1.pdf
Cercignani, C. (1998). The Boltzmann equation and its applications. New York: Springer.
Desjardin, O., Fox, R. O., & Villedieu, P. (2008). A quadrature-based moment method for dilute fluid-
particle flows. Journal of Computational Physics, 227(4), 2514–2539. doi:10.1016/j.jcp.2007.10.026
Fan, R., Marchisio, D. L., & Fox, R. O. (2004). Application of the direct quadrature method of moments to
polydisperse gas-solid fluidized beds. Powder Technology, 139(1), 7–20. doi:10.1016/j.powtec.2003.10.005
Ferziger, J. H., & Peric, M. (2001). Computational methods for fluid dynamics. New York: Springer.
Fox, R. O. (2008). A quadrature-based third-order moment method for dilute gas-particle flows. Journal
of Computational Physics, 227(12), 6313–6350. doi:10.1016/j.jcp.2008.03.014
Fox, R. O., & Vedula, P. (2010). Quadrature-based moment model for moderately dense polydisperse
gas-particle flows. Industrial & Engineering Chemistry Research, 49(11), 5175–5187. doi:10.1021/
ie9013138
Galvin, J. E., Dahl, S. R., & Hrenya, C. M. (2005). On the role of the non-equipartition in the dynam-
ics of rapidly flowing granular mixtures. Journal of Fluid Mechanics, 528, 207–232. doi:10.1017/
S002211200400326X
Galvin, J. E., Hrenya, C. M., & Wildman, R. D. (2007). On the role of the Knudsen layer in rapid granular
flows. Journal of Fluid Mechanics, 585, 73–92. doi:10.1017/S0022112007006489
Garg, R., Passalacqua, A., Subramaniam, S., & Fox, R. O. (2008). Comparison of Euler-Euler and Euler-
Lagrange simulations of finite-Stokes-numbers gas-particle flows in a lid-driven cavity. AIChE Annual
Meeting, November 16th – 21st, Philadelphia.
Gidaspow, D. (1994). Multiphase flow and fluidization. New York: Academic Press.
Goldhirsch, I. (2003). Rapid granular flows. Annual Review of Fluid Mechanics, 35, 267–293. doi:10.1146/
242
Goldhirsch, I., Noskowicz, S. H., & Bar-Lev, O. (2005). Theory of granular gases: some recent results
and some open problems. Journal of Physics Condensed Matter, 17, S2591–S2608. doi:10.1088/0953-
8984/17/24/015
Grad, H. (1949). On the kinetic theory of rarified gases. Communications on Pure and Applied Math-
ematics, 2, 329. doi:10.1002/cpa.3160020403
Grad, H. (1958). Principles of the kinetic theory of gases. In Flügge, S. (Ed.), Handbuch der Physik XII:
Thermodynamic der Gase. Berlin, Germany: Springer.
Holway, L. H. (1966). New statistical models for kinetic theory: methods of construction. Physics of
Fluids, 9(9), 1658–1673. doi:10.1063/1.1761920
Issa, R. I. (1986). Solution of the implicitly discretised fluid flow equations by operator splitting. Journal
of Computational Physics, 61(1), 40–65. doi:10.1016/0021-9991(86)90099-9
Johnson, P. C., & Jackson, R. (1987). Frictional-collisional constitutive relations for granular materials, with
applications to plane shearing. Journal of Fluid Mechanics, 176, 67–93. doi:10.1017/S0022112087000570
Karema, H., & Lo, S. (1999). Efficiency of interphase coupling algorithms in fluidized bed conditions.
Computers & Fluids, 28, 323–360. doi:10.1016/S0045-7930(98)00028-0
Marchisio, D. L., Pikturna, J. T., Fox, R. O., Vigil, R. D., & Barresi, A. (2003a). Quadrature method of
moments for population-balance equations. AIChE Journal. American Institute of Chemical Engineers,
49(5), 1266–1276. doi:10.1002/aic.690490517
Marchisio, D. L., Vigil, R. D., & Fox, R. O. (2003b). Quadrature method of moments for aggregation-
breakage processes. Journal of Colloid and Interface Science, 258(2), 322–334. doi:10.1016/S0021-
9797(02)00054-1
Marchisio, D. L., Vigil, R. D., & Fox, R. O. (2003c). Implementation of the quadrature method of
moments in CFD codes for aggregation-breakage problems. Chemical Engineering Science, 58(15),
3337–3351. doi:10.1016/S0009-2509(03)00211-2
McGraw, R. (1997). Description of aerosol dynamics by the quadrature method of moments. Aerosol
Science and Technology, 27(2), 255–265. doi:10.1080/02786829708965471
Oliveira, P. J., & Issa, R. I. (1994). On the numerical treatment of interphase forces in two-phase flow.
Numerical Methods in Multiphase Flows, 185, 131–140.
Passalacqua, A., Galvin, J. E., Vedula, P., Hrenya, C. M., & Fox, R. O. (2010). A quadrature-based kinetic
model for dilute non-isothermal granular flows. Communication in Computational Physics.
Patankar, S. V. (1980). Numerical Heat Transfer and Fluid Flow. New York: Hemisphere Publishing
Corporation.
Perthame, B. (1992). Second-order Boltzmann schemes for compressible Euler equations in one and two
space dimensions. SIAM Journal on Numerical Analysis, 29(1), 1–19. doi:10.1137/0729001
243
Ramanathan, S., & Koch, D. L. (2009). An efficient direct simulation Monte Carlo method for low Mach
number noncontinuum gas flows based on the Bhatnagar-Gross-Krook model. Physics of Fluids, 21,
033103. doi:10.1063/1.3081562
Schiller, L., & Naumann, A. (1935). A drag coefficient correlation. V.D.I. Zeitung, 77, 318–320.
Simonin, O. (1991). Prediction of the dispersed phase turbulence in particle-laden jets. In 4th Interna-
tional Symposium on Gas-Solid Flows (pp. 197 – 206, Vol. 121), ASME-FED.
Spalding, D. B. (1980). Numerical computation of multi-phase fluid flow and heat transfer. In Taylor,
C., & Morgan, K. (Eds.), Recent advances in numerical methods in fluids (pp. 139–168). Swansea, UK:
Pineridge Press Limited.
Struchtrup, H. (2005). Macroscopic Transport Equations for Rarified Gas Flows. Berlin: Springer-Verlag.
Strumendo, M., & Arastoopour, H. (2008). Solution of PBE by MOM in finite size domains. Chemical
Syamlal, M., Rogers, W., & O’Brien, T. J. (1993), MFIX documentation: theory guide. (Tech. Rep. DOE/
METC-94/1004, DE9400087). Morgantown, WV: US Department of Energy, Office of Fossil Energy,
Morgantown Energy Technology Center.
Wright, D. L., McGraw, R., & Rosner, D. E. (2001). Bivariate extension of the quadrature method of
moments for modeling simultaneous coagulation and sintering particle populations. Journal of Colloid
and Interface Science, 236(2), 242–251. doi:10.1006/jcis.2000.7409
Xu, K. (2008). A generalized Bhatnagar-Gross-Krook model for nonequilibrium flows. Physics of Fluids,
20, 026101. doi:10.1063/1.2837174
Zheng, Y., & Struchtrup, H. (2005). Ellipsoidal statistical Bhatnagar-Gross-Krook model with velocity-
dependent collision frequency. Physics of Fluids, 17, 127103. doi:10.1063/1.2140710
244
245
Chapter 8
Direct Numerical Simulation of
Gas-Solids Flow Based on the
Immersed Boundary Method
Rahul Garg
Sudheer Tenneti
Jamaludin Mohd. Yusof

Los Alamos National Laboratory, USA
Shankar Subramaniam
abstract
boundary method (IBM) is described where the governing Navier-Stokes equations are modeled with exact
boundary conditions imposed at each particle surface using IBM and the resulting three dimensional
time-dependent velocity and pressure fields are solved. Since this model has complete description of
the gas-solids hydrodynamic behavior, one could extract all the Eulerian and Lagrangian statistics for
validation and development of more accurate closures which could be used at coarse-grained simula-
tions described in other chapters.
MOMentuM transfer in gas-sOlids flOw
Accurate representation of the momentum transfer between particles and fluid is necessary for predic-
tive simulation of gas-solids flow in industrial applications. Such device-level simulations are typically
based on averaged equations of mass and momentum conservation corresponding to the fluid and particle
DOI: 10.4018/978-1-61520-651-3.ch008
Direct Numerical Simulation of Gas-Solids Flow Based on the Immersed Boundary Method
phase(s) in gas-solids flow (Syamlal, Rogers, & O’Brien, 1993), and these constitute the multi-fluid
theory. The momentum conservation equation in this theory contains a term representing the average
interphase momentum transfer between particles and fluid. The dependence of this term on flow quanti-
ties such as the Reynolds number based on mean slip velocity, solid volume fraction, and particle size
distribution must be modeled in order to solve the set of averaged equations, and is simply referred to
as a drag law. If higher levels of statistical representation are adopted—such as the second moment of
particle velocity, or the particle distribution function—then the corresponding terms (such as the inter-
phase transfer of kinetic energy in the second velocity moment equations) appearing in those equations
also need to be modeled.
Direct numerical simulation of flow past particles is a first-principles approach to developing accurate
models for interphase momentum transfer in gas-solids flow at all levels of statistical closure. Since DNS
solves the governing Navier-Stokes equations with exact boundary conditions imposed at each particle
surface, it produces a model free solution with complete three dimensional time-dependent velocity and
pressure fields. In principle, all Eulerian and Lagrangian flow statistics can be extracted from the DNS
data making it a powerful tool for model validation and development (Pope, 2000; Rai, Gatski, & Erle-
bacher, 1995). While there are different numerical approaches available to perform DNS of gas-solids
flow—such as the lattice Boltzmann method (LBM)—here we describe a DNS approach that is based
on the immersed boundary method (IBM). The outline of this chapter is as follows. We first describe the
context in which models for interphase momentum transfer arise. We begin with the transport equation
for the one-particle distribution function that is the starting point for the kinetic theory of granular and
multiphase flows. This is appropriate because all moment-based theories (averaged equations, second
and higher moments) can be derived from this distribution function. Thus, by developing closure models
at the level of the one-particle distribution function, we effectively model all moment equations. The
appropriate physical problem that needs to be set up to approximate statistically homogeneous gas-solids
suspension flow is then described. The expression for the mean interphase momentum transfer term in
steady, homogeneous, gas-solids flow that arises from the averaged conservation equations in the two-
fluid theory is then derived, and related to the equivalent term in the one-particle distribution function
approach. Then the immersed boundary method and its implementation are described. Numerical error
associated with forming statistical estimates of the interphase momentum transfer term is analyzed
and decomposed into spatial, temporal and statistical contributions. This results in the identification
of relevant numerical parameters (grid resolution, size of computational domain, number of particle
configurations) corresponding to each of the error contributions. Numerical convergence of the IBM
DNS code is established, and results from standard tests are presented that validate the simulation ap-
proach. Drag laws obtained from IBM simulations are discussed and compared with those obtained
from other simulation methods. The IBM approach is compared with other simulation approaches, and
relative advantages and disadvantages are discussed. Directions for further research in the formulation
of models of gas-solids flow using DNS based on IBM are outlined. Finally, the contributions of this
chapter are summarized along with concluding remarks regarding the use of IBM for direct numerical
simulation of gas-solids flow.
246
transport of the Particle distribution function
The transport equation for the one-particle distribution function in gas-solids flow for monodisperse
particles is (Chapman & Cowling, 1952; Garzo, Hrenya, & Dufty, 2007; Jenkins, 1998; Koch, 1990;
Liboff, 1990; Subramaniam, 2001)
∂f ∂ ∂
∂t
+
∂xk
(vk f ) +
∂ vk
( )
Ak | x, v, t f = f coll (1.1)
where v is the velocity of the particles, 〈A | x, v, t〉 is the conditional expectation of the acceleration and
fcoll is the term arising due to collisions between particles.
The principal difference between this equation for solid particles and its counterpart in molecular
gases is the appearance of the conditional expectation of the acceleration 〈A | x, v, t〉 inside the veloc-
ity derivative corresponding to transport of the distribution function in velocity space. The conditional
expectation of acceleration cannot be expressed purely in terms of the distribution function, and is hence
denoted an unclosed term in the above equation. It can depend on higher-order distribution functions
(e.g., the two-particle distribution function) in the hierarchy resulting from a description of the particle
system in terms of the Liouville density. It also depends on statistics of the carrier flow. Since analytical
models are difficult to propose for this term beyond dilute particle flow in the Stokes flow regime, it
must be inferred from direct numerical simulation data. Drag laws for steady flow through homogeneous
suspensions are obtained by integrating the conditional expectation of the acceleration over velocity
space to obtain the average force 〈Fd〉 exerted on the particles by the fluid
m
n ∫ i
Fd ,i = A | v fdv , (1.2)
where m is the mass of a particle, and n is the particle number density.
Homogeneous suspension flow
In order to calculate 〈Fd〉 from DNS, it is natural to simulate a statistically homogeneous suspension
flow with freely moving particles, and to then compute volume-averaged estimates of 〈Fd〉 from particle
acceleration data. Imposing a mean pressure gradient to balance the weight of the particles leads to a
steady mean momentum balance. In this setup the particle positions and velocities sample a trajectory
in phase space that corresponds to the specified nonequilibrium steady state, and time averaging can be
used to improve the estimate for 〈Fd〉. However, such freely moving suspensions are computationally
prohibitive especially because in order to propose drag laws these simulations need to be performed
over a range of solid volume fractions and mean flow Reynolds numbers (based on mean slip velocity).
Furthermore, over a wide range of volume fraction and particle Stokes number, the particle configuration
in individual realizations develops spatial structures due to flow instabilities. Wylie and Koch (Wylie
& Koch, 2000) performed simulations of a suspension with particles moving along ballistic trajectories
between elastic hard sphere collisions, but this assumption that the fluid does not affect the particle mo-
tion is valid only in the limit of high Stokes number.
247
Koch and Hill (Koch & Hill, 2001) discuss the relevant non-dimensional parameters that arise in
the context of gas-solids suspensions. As noted in their work, direct numerical simulations are useful
in developing drag laws for suspension flows where the effects of fluid inertia and the particle inertia
cannot be neglected. In the simulations described in this work we neglect gravity, so the relevant nondi-
mensional parameters are the Reynolds number (characterizing the importance of fluid inertia) and the
particle Stokes number (characterizing the importance of particle inertia). While the Stokes flow regime
(negligible fluid inertia) is amenable to analytical treatment, direct simulation is the only approach for
gas-solids suspensions at finite Reynolds number.
Steady Flow Past Homogeneous Assemblies of Fixed Particles
A convenient simplification for high Stokes number suspensions is to replace the ensemble of particle
positions and velocities sampled by the system in its nonequilibrium steady state, by a set of particle
configurations and velocities that would result from a granular gas simulation. Steady flow past fixed
assemblies of particles in configurations (and with velocities) sampled from this set is simulated, and
drag laws are obtained by averaging over this ensemble. The idea of extracting drag laws from steady
flow past random and ordered arrays of particles through particle assemblies has been successfully
exploited by several researchers using the LBM simulation methodology developed by Ladd (Ladd,
1994a, 1994b) for particulate suspensions. For example, Koch and co-workers (Hill, Koch, & Ladd,
2001a) and (Hill, Koch, & Ladd, 2001b), referred to collectively as HKL, studied the steady flow past
both ordered and random arrays. Kuipers and co-workers (van der Hoef, Beetstra, & Kuipers, 2005)
and (Beetstra, van der Hoef, & Kuipers, 2007), collectively referred to as BVK, extended HKL’s LBM
simulations to higher Reynolds numbers.
In the simplest case of a monodisperse suspension, the drag law is extracted by computing steady
nonturbulent flow at a specified mean slip Reynolds number past a set of random particle configurations
(microstates) that correspond to a particular value of the solid volume fraction. The pair-correlation and
higher-order statistics of the particle field are determined by the configurations resulting from the granular
gas simulation. The particle velocity distribution can be initialized either from the granular gas simula-
tion at finite granular temperature or it is often assumed that all particles move with the same velocity.
gOverning equatiOns
The schematic in corresponds to the physical problem of flow past a single particle. Volumes occupied
by the fluid and solid phases are denoted by ςf and ςs respectively, such that the total domain volume ς
= ςf + ςs. The bounding surface of the physical domain is denoted ∂ς, and the bounding surfaces of the
solid phase and fluid phase are denoted by ∂ςs and ∂ςf, respectively. For incompressible flows, the mass
and momentum conservation equations for the fluid-phase are
∂ui
=0 (1.3)
∂xi
248
Figure 1. Schematic of the physical domain with only one particle. Hatched lines represent the volume ςf
occupied by the fluid-phase and solid fill represents the volume ςs of the solid-phase such that the total
volume of physical domain ς = ςf + ςs. The bounding surfaces of the physical domain, solid-phase, and
fluid-phase are denoted by ∂ς, ∂ςs, and ∂ςf, respectively
and
∂ui ∂ui u j ∂ 2ui ∂τ ji

ρf + ρf = −g i + µ f = (1.4)
∂t ∂x j ∂x j ∂x j ∂x j
respectively. In the above equation g = ∇p is the gradient of modified pressure (Mohd-Yusof, 1996),
and ρf and μf are the thermodynamic density and dynamic viscosity of the fluid-phase, respectively. At
the particle-fluid interface, in order to ensure zero slip and zero penetration (for impermeable surfaces)
boundary conditions, the relative velocity should be zero. If the solid particles are held stationary, then
the above boundary conditions translate to
u = 0 on ∂ςs. (1.5)
The averaged equations corresponding to these mass and momentum conservation balances are use-
ful in simulations of practical gas-solids flow applications. In the previous section we described one
statistical approach based on the one-particle distribution function. Here we first describe an alternative
approach called the Eulerian two-fluid theory because it is more natural to derive the averaged equations
corresponding to Equation (1.4) using this approach. The conditional expectation of acceleration ap-
pearing in the one-particle distribution function approach is then related to the mean interphase momen-
tum transfer term in the Eulerian two-fluid theory.
In the Eulerian two-fluid theory phasic averages are defined as follows. If Q(x, t) is any field, then
its phasic average 〈Q(f)〉(x, t) over the fluid volume ςf, referred to as fluid-phase mean, is defined as:
I f (x, t )Q(x, t )
Q( f ) (x, t ) = , (1.6)
I f (x, t )
249
where If is the fluid-phase indicator function which is equal to one if the point x lies in the fluid phase,
and zero otherwise.
The solid-phase mean 〈Q(s)〉(x, t) is similarly defined. The (unconditional) mixture mean 〈Q〉(x, t) is
related to the phasic mean by:
Q = ε f Qf + εs Qs (1.7)
where εf = 〈If(x, t)〉 and εs = 〈Is(x, t)〉 are the volume fractions of the fluid and solid phases, respectively. If
the flow is statistically homogeneous, there is no dependence on x and spatial derivatives are zero. Simi-
larly, if the flow is statistically stationary there is no dependence on t and temporal derivatives are zero.
The mean momentum conservation equation (Drew & Passman, 1999; Pai & Subramaniam, 2009) in
the fluid phase is obtained by multiplying the momentum conservation equation (1.4) by If resulting in
∂ρf ε f ui( f ) ∂ ∂ ∂τ ji
+ ρ ε u( f ) u(j f ) = I f u "(i f ) u "(j f ) + I f , (1.8)
∂t ∂x j f f i ∂x j ∂x j
where u "(i f ) = ui − ui( f ) is the fluctuating component of the fluid velocity field. For steady flow with
an imposed mean pressure gradient in the fluid phase, it is convenient to decompose the pressure gradi-
ent term that appears in the divergence of the fluid-phase stress tensor as g = 〈g〉 + g’, such that remain-
ing part of the stress tensor t ' ji is defined by the expression:
∂τ ji ∂ 2ui ∂τ ' ji
= − g i − g 'i + µ f = − gi + . (1.9)
∂x j ∂x j ∂x j ∂x j
For a statistically homogeneous suspension at steady state (statistically stationary flow), the average
velocity does not depend on x or t, and the unsteady and convective terms on the left hand side of Equation
(1.8) do not contribute. Writing the remaining terms in an integral form, shows that the mean pressure
gradient term εf〈g〉 balances the sum of fluctuating pressure and viscous stress on the solid particles:
ε f gi = − τ ' ji n(js )δ(x − x(1) . (1.10)
In the above equation n(js ) is the normal vector pointing outward from the particle surface into the
fluid, and the stress tensor is evaluated on the fluid side of the interface. The term − τ ' ji n(js )δ(x − x(1)
appears as the drag contribution Fgm(vsm − vg) to the fluid-solids interaction force Igm in the two-fluid
equations derived from a volume-averaging approach (Syamlal, et al., 1993). For statistically homoge-
neous flows, the relationships between the one-particle distribution function approach and the Eulerian
two-fluid theory are established in the context of a comprehensive probability density function approach
to multiphase flows (Pai & Subramaniam, 2009). Using the relationships in Pai & Subramaniam (2009),
it is easy to show that the term on the right hand side of Equation (1.10) is related to the average force
exerted by the fluid on the particles (see Equation (1.2)) as follows:
250
1
Fd ,i = m Ai =
n
{ }
−ε s gi + τ ' ji n(js )δ( x − x(i ) ) . (1.11)
tHe iMMersed bOundary MetHOd
The basic notion of the immersed boundary method is to apply a set of forces on the computational
grid to mimic the presence of an interface. This has several advantages over conventional boundary or
body-fitted grids, especially for problems involving moving interfaces. First, there is no overhead for
grid generation, which can add considerable computational expense even for non-deforming geometries.
Second, the convergence of the solvers is generally better for Cartesian meshes than for unstructured
meshes. Third, IBM is intended to be implemented on regular Cartesian meshes that require much less
storage overhead than general unstructured or curvilinear meshes. The primary disadvantage of IBM
is the reduced resolution near the interface, but this is remedied by adopting adaptive mesh techniques.
There are two basic facets of the IBM, namely the choice of flow field (i.e. what velocity field do we
wish to achieve) and calculation of the force itself (i.e. once we decide on the field we wish to achieve,
how do we specify the force at each time-step). For clarity we will separate these two aspects, dealing
with the force specification first.
The immersed boundary method was originally developed by Peskin (Peskin, 1982) as a way to
incorporate the effect of flexible interfaces into fluid simulations. In that version, the local force is ob-
tained from some constitutive relation commensurate with the nature of the interface (e.g surface tension
in the case of a bubble, Young’s modulus for an elastic membrane) and is, by necessity, iterative over
a timestep since the location of the interface is not known a priori. This method has been applied to a
variety of flows, such as bubbles, blood cells and swimming fish. The issue with this implementation is
that it is not efficient for rigid bodies, since this requires driving the stiffness of the interface membrane
(and effectively the stiffness of the equations to be solved) to infinity. The same is true for the Immersed
Interface method (IIM) which is well suited to the solution to the flow past deformable bodies (Lee &
Leveque, 2003)..
Goldstein (Goldstein, Handler, & Sirovich, 1993) proposed what is essentially proportional-integral
feedback on the force term to produce boundary conditions on a rigid body. The problem with this method
is the lack of efficiency; due to the need to numerically integrate the force in (pseudo-continuous) time
over a single time-step, the effective CFL limit was extremely small, (O(10-3)). Coincident with Goldstein’s
work, Mohd-Yusof (Yusof, 1996) developed what is now termed the Discrete-Time Immersed Boundary
Method (DTIBM). The essential aspect of this formulation is the recognition that examination of the
discretized-in-time equations leads to a straightforward definition of the force at a given point, once we
have decided on the required velocity field (and hence the velocity required at the point in question).
We now turn our attention to the choice of flow field. The implementations to date can be broadly
divided into two classes; ghost fluid and numerical boundary layers. In the former, the flow field in the
region of interest is extrapolated across the interface in such a way as to impose the desired boundary
condition at the interface. This is the method used in the original implementations of Goldstein and Mohd-
Yusof, as well as in this chapter. Such an implementation is natural in situations where the fictitious flow
produced within the rigid body does not affect the solution and is easily accounted for. This choice has
the advantage that the force applied in the fluid region can be zero; that is, the governing equations are
251
unmodified in this region. Additionally, the use of the ghost fluid region allows the effect of, for example,
implicit diffusion operators, to be minimized by forcing linear velocity gradients across the interface.
In the latter method, the immersed boundary force applied at the interface is numerically smoothed
over several grid-points, for numerical stability reasons. As used by Peskin, this is a natural implemen-
tation, since the flow on both sides of the interface is required for the solution. It is possible to use the
numerical boundary layer formulation for rigid body problems, as was done by Verzicco et. al. (Verzicco,
Mohd-Yusof, Orlandi, & Haworth, 2000) where the discrete-time formulation of Mohd-Yusof was applied
with numerical boundary layers in the fluid side, and with exact rigid body fields imposed in the solid.
solution approach
In the immersed boundary method, the mass and momentum equations are solved in the entire domain
that includes the interior regions of the solid particles as well. The mass and momentum conservation
equations solved in IBM are
∂ui
=0 (1.12)
∂xi
and
∂ui 1 ∂2
+ Si = − g IBM,i + vf u + f u, i (1.13)
∂t ρf ∂x j ∂x j i
µf
respectively, vf = is the kinematic viscosity, gIBM is the pressure gradient, S = ∇·(uu) is the convec-
ρf
tive term in conservative form, and u is the instantaneous velocity field. In the above momentum con-
servation equation, fu is the additional immersed boundary force term that accounts for the presence of
solid particles in the fluid-phase by ensuring zero slip and zero penetration boundary conditions (Equa-
tion (1.5)) at the particle-fluid interface.
In Figure 2, a schematic describing the computation of the immersed boundary forcing is shown.
The surface of the solid particle is represented by a discrete number of points called boundary points,
by discretizing the sphere in spherical coordinates. Another set of points called exterior points are gen-
erated by projecting these boundary points onto a sphere of radius r + Δr, where r is the radius of the
particle. Similarly, the boundary points are projected onto a smaller sphere of radius r − Δr and these
points are called interior points. In our simulations, Δr is taken to be same as the grid spacing. The im-
mersed boundary force is computed only at the interior points. At these points, the fluid velocity field
is forced in a manner similar to the ghost cell approach used in standard finite-difference/finite-volume
based methods. Or more specifically for the case of zero solid particle velocity, the velocity field inside
the solid particle at grid points close to the interface is forced to be exact opposite of the fluid veloc-
ity field outside the particle (see Figure 2). The details of this forcing approach are discussed in Yusof
(Yusof, 1996). In Yusof’s original implementation, the IB forcing was also computed on the boundary
points in addition to the interior points. The IB forcing at the boundary points was then interpolated to
252
Figure 2. A schematic showing the computation of the immersed boundary forcing for a stationary
particle. The solid circle represents the surface of the particle at r. Open dot shows the location of one
exterior point at r+Δr (only one exterior point is shown for clarity, although there is one exterior point
for each interior point) and filled dots show the location of interior points at r-Δr where the immersed
boundary forcing is computed. For the special case of a stationary particle, the velocity at the interior
points is forced to be the opposite of the velocity at the corresponding exterior points.
the neighboring grid nodes that could include grid nodes in the fluid phase. This additional forcing leads
to contamination of the fluid velocity and pressure fields by the IB forcing. In the current implementa-
tion of DTIBM, we are able to obtain accurate results even with zero forcing at the boundary points,
avoiding any contamination of the fluid velocity and pressure fields by IB forcing. It is noteworthy that
the discretization of the sphere in spherical coordinates is independent of the grid resolution and hence
to some extent, decouples the grid resolution from the accuracy with which the boundary condition is
imposed. In addition to forcing the velocity field, the IB forcing term also cancels the remaining terms
in the momentum conservation and, at the n + 1 time-step, it is given by
uid − uin ∂2
f un,i+1 = − Sin + gin − vf un (1.14)
∆t ∂x j ∂x j i
where uid is the desired velocity at that location.

Since the immersed boundary force fu is a function of both space and time, its effect on the pressure
field is accounted by solving a modified pressure Poisson equation given by
1 ∂g IBM,i ∂
=− ( S − f u, i ) , (1.15)
ρf ∂xi ∂xi i
253
which is obtained by taking the divergence of the instantaneous momentum conservation equation (1.13)
and using the mass conservation equation (1.12).
For flow past a statistically homogeneous particle assembly, we solve the IBM governing equations
by imposing periodic boundary conditions on fluctuating variables that are now defined. From the defini-
tion of volumetric mean, the velocity field u(x, t) can be decomposed as the sum of a volumetric mean
〈u〉v and a fluctuating component u’(x, t)
ui (x, t ) = ui (t ) + u 'i (x, t ) , (1.16)

v
and similar decompositions are written for the non-linear term S, pressure gradient gIBM, and immersed
boundary forcing fu terms. Substituting the above decompositions in the mass (1.12) and momentum
(1.13) conservation equations, followed by volume averaging, yields the volume-averaged mass and
momentum conservation equations. Since the volumetric means are independent of x, mean mass con-
servation is trivially satisfied. The volume-averaged momentum conservation equation becomes
∂ ui 1
v
=− g + f u, i , (1.17)
∂t ρf IBM,i v v
where it is noted that due to periodic boundary conditions, the volume integrals of convective and dif-
fusive terms are zero.
While mean mass conservation (in the volume-averaged sense) is trivially satisfied, the fluctuating
velocity field needs to be divergence free
∂u ' i
= 0. (1.18)
∂xi
Subtracting the volume-averaged momentum conservation equation from the instantaneous momentum
conservation equation (1.13) yields the following equation for the conservation of fluctuating momentum:
∂u ' i 1 ∂2
+ S ' i = − g ' i + vf u ' + f ' u, i . (1.19)
∂t ρf ∂x j∂x j i
Taking the divergence of the above equation and using (1.18) results in the following modified pres-
sure Poisson equation for the fluctuating pressure gradient:
1 ∂g 'IBM,i ∂
=− ( S 'i − f ' u , i ) . (1.20)
ρf ∂xi ∂xi
The conservation equations (Equations(1.14)-(1.20)) are numerically solved to yield the flow around
immersed bodies.
254
Although the immersed boundary forcing fu ensures zero relative velocity at the particle-fluid inter-
faces, for periodic boundary conditions we need to ensure that the desired fluid-phase mean velocity
will be attained. This is because unlike in inflow/outflow boundary conditions where the flow enters at
a specified mass flow rate, there is no such mechanism for periodic boundary conditions. Therefore, in
order to attain a desired fluid-phase mean velocity 〈u(f)〉d, the mean pressure gradient 〈gIBM〉v is advanced
in pseudo-time such that at the nth time step it is given by
d n
n ui( f ) − ui( f ) 1  ∂uin ( s ) 
 ∂s ∂s ∂x n j dA ,
−∫ ψ ni dA + µ f ∫
n (s)
− g IBM,i = ρf v v
+ (1.21)
v ∆t (1 − εs )  j 
where g’ = ∇ψ, all quantities in the integrand are evaluated on the fluid side of the fluid-particle interface,
and the superscript n implies the relevant quantities at the nth time step. This equation for the volumetrically
averaged pressure gradient is obtained by integrating the IBM momentum conservation equation (1.13)
over the volume occupied by the fluid-phase. A finite difference approximation has been substituted for
the unsteady term on right hand side of the above equation that drives the volume-averaged fluid veloc-
ity to its desired value. Since the immersed boundary force term is zero at grid nodes that lie outside the
solid particles, the fluid-phase volume average of the immersed boundary force term 〈Iffu〉v is zero, thus
resulting in zero contamination of the fluid pressure and velocity fields. The volume-averaged pressure
gradient 〈gIBM〉v given by above equation, and the volume-averaged immersed boundary forcing term
〈fu〉v are used to evolve the volume-averaged velocity 〈u〉v by Equation (1.17). For a statistically station-
ary flow, the equations are evolved in pseudo time until the average quantities reach a steady state, at
which point the first term on the right hand side of Equation (1.21) is negligible, and Equation (1.21)
reduces to the numerical counterpart of Equation (1.11). This establishes that the resulting numerical
solution to the IBM governing equations is a valid numerical solution to steady flow past homogeneous
particle assemblies.
IBM with direct forcing was developed by Mohd-Yusof (Yusof, 1996) for his doctoral dissertation
to solve for turbulent flow past a single particle. This code was subsequently completely rewritten by
the Subramaniam research group at Iowa State University to implement the following improvements:
1. Modification of the forcing to remove the contamination in the fluid

2. Computation of drag for gas-solids suspensions at high volume fraction by establishing the con-
nection with two-fluid theory and one-particle distribution function approaches
siMulatiOn MetHOdOlOgy
We now describe how the physical parameters of the problem—mean flow Reynolds number and solid
volume fraction—are specified in the simulation. For flow past homogeneous particle assemblies, a
Reynolds number based on the magnitude of mean slip velocity between the two phases is defined as
U slip (1 − εs )D
Re = , (1.22)
vf
255
where Uslip = |〈u(f)〉 − 〈u(s)〉| is the magnitude of the mean slip velocity, D is the particle diameter, and
〈u(f)〉 and 〈u(s)〉 are the fluid-phase and solid-phase mean velocities, respectively.
The objective in direct numerical simulations is to solve the instantaneous mass and momentum con-
servation equations (Equations (1.12) and (1.13)) subject to the boundary conditions described earlier, in
such a way that the resulting volumetric mean slip velocity corresponds to a desired Reynolds number.
This system can be solved in three different ways, namely:
1. Specified mean pressure gradient 〈g〉: In this method (Hill, et al., 2001a; Hill, Koch, & Ladd, 2001c),
a mean pressure gradient along with zero particle velocities are specified as inputs. As a result,
the volumetric mean velocity evolves by Equation (1.17) and the steady-state solution implies a
Reynolds number. The drawback of this method is that Reynolds number cannot be specified as
an input.
2a. Specified solid-phase mean velocity 〈u(s)〉: In this method the simulations are carried out in a labo-
ratory frame of reference wherein the mean velocity 〈u〉 is zero. Therefore, from Equation(1.7),
εs
the desired fluid phase mean velocity u( f ) = − u( s ) . Substituting this expression for
(1 − εs
desired fluid-phase mean velocity 〈u(f)〉 in Equation (1.22) results in an expression for |〈u(s)〉| in
terms of the Reynolds number and other physical properties. In these simulations, the desired solid-
phase mean velocity |〈u(s)〉| is attained by specifying equal velocities to all particles.
2b. Specified fluid-phase mean velocity 〈u(f)〉: In this method, particles are assigned zero velocity.
Therefore, from Equation(1.22), the desired fluid-phase mean velocity 〈u(f)〉 is known in terms of
the input Reynolds number and other physical properties.
The advantage of methods 2a and 2b over the first method is that the desired Reynolds number can
be specified as an input to the simulation, whereas it is an output in the first method. However, there is
no relative advantage in choosing between the second and third methods. It is important to note that the
velocity scale (1 − εs)Uslip is the correct scale to use for meaningful comparison of drag laws regardless
of the simulation approach.
The solid volume fraction εs together with the ratio of computational box length to particle diameter
L/D determines the number of solid particles Ns in the simulation:
3
6εs  L 
Ns =   . (1.23)
π  D 
numerical Parameters
The ratio of computational box length to particle diameter L/D, the number of solid particles Ns and the
number of configurations/realizations Μ are numerical parameters of the simulation. Their influence on
the numerical convergence of the IBM simulations is discussed in the following subsections.
The computational box is discretized using M grid cells in each direction, and this introduces a grid
resolution parameter Dm. The number of grid cells is calculated as
256
L L
M = = Dm , (1.24)
∆x D
where L is the length of the computational box, Δx is the size of each grid cell, and Dm is the number
of grid cells across the diameter of a solid particle. The solution algorithm is advanced in pseudo-time
from specified initial conditions to steady state using a time step Δt that is chosen as the minimum of
the convective and diffusive time steps by the criterion
 ∆x ∆x 2 (1 − ε )
∆t = CFL × min  , s 
. (1.25)
 umax v 
 f 
At the beginning of the simulation umax = u( f ) , and as the flow evolves the time step adapts itself
to satisfy the above criterion.
estimation of Mean drag from simulations
Direct numerical simulation of flow through a particle assembly using the immersed boundary method
results in velocity and pressure fields on a regular Cartesian grid. The drag force on the ith particle,
Fd(i)=m(i)A(i), is computed by integrating the viscous and pressure forces exerted by the fluid on the par-
ticle surface. The average drag force on particles in a homogeneous suspension for the μth realization
is computed as
1
Ns
1  ∂u j 
− g nk( s )dA ,
{Fd , j }µv =
Ns
∑ m(i ) A(ji ) = Ns

 IBM , j v  ψn j dA + µ f ∫
s − ∫
∂s
(s)
∂s ∂xk 
(1.26)
i =1
 
which is obtained by integrating the pressure and viscous fields over the surface of each particle. In the
last expression of the above equation, the first term is the force on all particles in the volume due to mean
pressure gradient, the second term is the drag force due to the fluctuating pressure gradient field, and
the third term is the viscous contribution to the drag force. This expression for the drag force is for one
realization, and it is then averaged over Μ independent realizations in order to average over different
particle configurations corresponding to the same solid volume fraction and pair correlation function.
The ensemble-averaged drag is
1 M
{F } { Fd ,i } ,
µ
d ,i V ,M
= ∑
M µ =1 v
(1.27)
which converges to the true expectation of the drag force 〈fd〉 (given by Equations (1.2) and (1.11)) in
the limit NsM → ∞. The ensemble-averaged drag force is later reported as a normalized average drag
force F given by
257
Figure 3. Convergence characteristics of drag force with grid resolution Dm. The drag force contribu-
tion from fluctuating pressure gradient (open symbols) and viscous stresses (filled symbols) for FCC
arrays (with grid resolution Dm) is shown for two CFL values of 0.2 (squares) and 0.05 (triangles). (a)
Re = 40,εs = 0.2; (b) Re=40, εs = 0.4
{Fd }V , M
F= , (1.28)
FStokes
where FStokes = 3πμfDUslip(1 – εs) is the Stokes drag.

Each numerical parameter must be chosen to ensure numerically converged, accurate, and physi-
cally meaningful results. Spatial and temporal discretization contribute to numerical error in the force
on the ith particle that scales as O(Δxp, Δtq), where p and q depend on the order of accuracy of the
method and the interpolation schemes at the particle boundary. For steady flow, the numerical error due
to spatio-temporal discretization is solely determined by the spatial resolution parameter Δx/D = 1/Dm,
which must be sufficiently small to ensure converged results. For the case where the particle positions
are chosen to be randomly distributed, on each realization of the flow the computational domain should
be chosen large enough so that the spatial auto-correlation in the particle force decays to zero. This
guarantees that the periodic boundary condition does not introduce artificial effects due to interaction
between the periodic images. For a given solid volume fraction εs, this determines a minimum value of
Ns = ⌈εsV⌉. The number of multiple independent simulations Μ is determined by the requirement that

the total number of samples ∑N
µ=1
µ
in the estimate for the average force given by Equation(1.26) be
sufficiently large to ensure low statistical error.

Owing to the periodic lattice arrangement of particles in ordered arrays, it is sufficient to solve the
flow for just one unit cell (i.e., one particle for the simple cubic (SC) lattice, and four particles for the
face-centered cubic (FCC) lattice). For the special case of ordered arrays, since the number of particles
is pre-determined, the ratio of computational box length to particle diameter L/D is not an independent
258
numerical parameter. For ordered arrays the only numerical parameter is Dm, which determines the
number of grid cells M required to resolve the flow.
numerical convergence
Here we establish that the IBM simulations result in numerically converged solutions. The test case
chosen is steady flow past an ordered array of particles in a lattice arrangement, because for this case the
only numerical parameter is the grid resolution Dm. Although we consider steady flows, we also verify
that the time step chosen to evolve the flow in pseudo time from a uniform flow initial condition does
not change the steady values of drag that we compute using IBM. For an FCC arrangement of particles
(εs = 0.2, Re = 40), Figure 3a shows the convergence of drag forces due to fluctuating pressure gradient
(open symbols) and viscous stresses (filled symbols) as a function of grid resolution Dm for two differ-
ent values of CFL number (0.2 denoted by squares and 0.05 denoted by triangles). Figure 3b shows the
same convergence characteristics for a denser FCC arrangement with a solid volume fraction of 0.4 and
Re = 40. In both figures it can be seen that the IBM simulation result does not depend on the time step
(CFL). With regard to spatial convergence, the figures show that the resolution requirements increase
with increasing volume fraction. This is because higher local velocities are generated in the interstices
between particles at higher solid volume fraction. While a minimum resolution of Dm = 40 is needed for
converged results at εs = 0.2, the minimum resolution requirement increases to Dm = 60 for εs = 0.4. In
addition to the dependence of grid resolution on volume fraction, increasing the mean flow Reynolds
number also requires progressively higher grid resolution. Therefore, for the higher Reynolds number
cases that are reported later, higher resolutions are used for the volume fractions 0.2 and 0.4, so that
these cases are also adequately resolved.
When studying grid convergence of a numerical scheme it is sometimes useful to calculate the order
of convergence that is implied by the numerical tests. However, the use of a regular Cartesian grid to
solve for flow over spheres necessitates interpolation of pressure and viscous stresses from the grid to
a finite number of particle surface points. For ordered arrays these interpolation errors cause the steady
drag values to exhibit a weak dependence on the location of the particle in the computational box (drag
values can differ up to a maximum of 1%). Even for a fixed particle location in the computational box,
the interpolation error depends on both the number of particle surface points and the grid resolution.
These non-systematic interpolation errors preclude a reliable estimation of the order of convergence of
the numerical scheme, which is formally at least second-order. Although the non-systematic interpola-
tion errors prohibit the reliable quantification of spatial order of convergence, if a relative error is defined
based on the solution at the finest grid, then a spatial order of convergence in the range 1.5-2 is obtained
for the above cases. In other IBM studies (Ikeno & Kajishima, 2007), solution on a highly resolved
unstructured grid is taken as a reference to compare the IBM solutions and convergence rates up to
second order have been reported.
For the random arrays, in addition to errors arising from finite resolution, errors arise due to statistical
fluctuations between different realizations and the box length is also an independent numerical parameter.
Ideally, the effect of each numerical parameter on the numerical error should be investigated by vary-
ing that parameter while holding the other numerical parameters at fixed values. However, the choice
of some numerical parameters must satisfy more than one requirement, and some error contributions
are determined by the choice of more than one numerical parameter. Specifically, the choice of L/D is
determined by more than one requirement (decay of spatial autocorrelation and the need for minimum
259
Figure 4. Convergence characteristics of the normalized drag force with box length to particle diameter
ratio L/D for random arrays at Re=20. Two solid volume fraction values are considered: (a) εs = 0.3, (b)
εs = 0.4. Four different values of Dm are shown: 10 (squares), 20 (upper triangles), 30 (lower triangles),
and 40 (right triangles). Drag values have been averaged over 5 multiple independent simulations. Not
all combinations of Dm and L/D are shown because with a serial code some combinations exceeded
computational memory requirements
number of samples in the average force estimate), and both L/D and the number of multiple independent
simulations Μ determine the number of samples in the force estimate. These considerations as well as
computational limitations did not permit the independent variation of numerical parameters. Therefore,
a limited investigation of numerical parameter variation is presented here. To place this in context, we
note that to our knowledge this is the most comprehensive study of numerical error and convergence
for DNS of gas-solids flow.
While for ordered arrays the box length and number of particles are determined by the volume fraction
and type of lattice arrangement (SC/FCC), in random arrays these parameters have to be carefully chosen.
If L/D is too small, then the spatial autocorrelations that are larger than the box size will not be captured
and the periodic images will interact. For steady flow past random arrays (εs = 0.3, Re = 20), Figure 4a
shows the convergence characteristics of the normalized force with box length to particle diameter ratio
L/D for four different values of Dm equal to 10 (squares), 20 (upper triangles), 30 (lower triangles), and
40 (right triangles). Figure 4b is the same comparison for a denser random arrangement with a volume
fraction equal to 0.4. These results show that the drag value depends on L/D if the simulation is under-
resolved, and the effect of grid resolution Dm is stronger than that of L/D for the cases considered here.
Once the drag values are at their grid-converged values, there is no statistically significant dependence
for L>6D in these cases. The simulations of flow past random arrays that are reported later in this work
use higher resolutions when the Reynolds number exceeds 100, as shown in Table 1.
In summary, these numerical convergence test results show that the IBM simulations yield grid-in-
dependent results, and these results are also independent of the choice of time step used to advance the
solution in pseudo time, provided the stability criterion is met. The tests for random arrays also show
that the grid-converged results do not exhibit a statistically significant dependence on the computa-
tional box length for L>6D. However, these specific values for the numerical parameters should be
260
Table 1. Comparison of the numerical resolutions used for random arrays in IBM simulations with the
past LBM simulations of HKL and BVK. For each entry, first and second rows correspond, respectively,
to the LBM simulations of HKL and BVK, and the third row corresponds to the current IBM simulations.
For the IBM simulations, different numerical parameters are used for Re ≤ 100 and Re > 100. These
are shown in two separate columns
Ns Μ Dm L/D
εs Re < 100 Re > 100 Re < 100 Re > 100 Re < 100 Re > 100
HKL - - -
0.01 BVK - - -
IBM 64 13 5 10 20 15 9
HKL 16 5 9.6 4.38
0.1 BVK 54 20 17.5 6.6
IBM 80 41 5 20 30 7.5 6
HKL 16 5 17.6 3.47
0.2 BVK 54 20 17.5 5.2
IBM 161 34 5 20 40 7.5 4.5
HKL 16 5 17.6 3.06
0.3 BVK 54 20 21.5 3.07
IBM 71 26 5 30 50 5 3.6
HKL 16 5 33.6 2.73
0.4 BVK 54 20 21.5 4.13
IBM 95 20 5 30 60 5 3
treated as tentative because these limited set of tests cannot establish sharp limits on the minimum
resolution required, and further numerical testing could refine these limits. A satisfactory number of
MIS should ideally be determined by the determining the minimum number of samples for a given
level of statistical error in the force estimate. However, this quantity is a strong function of Re and
solid volume fraction. In the plots shown above, we have used 5 MIS for all the cases. While this results
in a statistical error that is on the order of the other numerical error contributions, further testing is
needed to refine this requirement. Clearly, the requirements of minimum L/D, minimum Dm, and mini-
mum Μ, together dictate a trade-off for a fixed level of computational work. Of these parameters, our
tests reveal that the numerical error in IBM exhibits the highest sensitivity to grid resolution Dm. These
numerical convergence tests provide useful guidelines in the choice of these parameters that approxi-
mately balance the error contributions, but further testing is needed for a complete error analysis.
validatiOn tests
isolated sphere
The flow over an isolated sphere in an unbounded medium presents itself as the logical validation test
for any direct numerical simulation approach to gas-solids flow. However, especially for simulations that
261
Figure 5. Normalized drag force F in a simple cubic array (εs = 4.0 × 10−4) as a function of Reynolds
number and angle θ between the mean flow and the x- axis in the (x,y) plane. The symbols are from the
IBM simulations: θ=0 (∆), θ=π/16 (□). The solid line is the drag correlation for an isolated sphere in
unbounded medium (Schiller & Naumann, 1933)
use periodic boundary conditions, this turns out to be a difficult validation test. For simulations using
periodic boundary conditions, flow through a very dilute simple cubic arrangement is taken as a close
approximation to flow over an isolated sphere in an unbounded medium. Since the simple cubic lattice
arrangement is not isotropic, it is known (Hill, et al., 2001b) that the results for drag can depend on
the orientation of the flow with respect to the unit vectors of the lattice for values of Reynolds number
beyond the Stokes flow regime. In contrast, there is of course no preferred direction for flow over an
isolated sphere in an unbounded medium.
Figure 5 shows the comparison of the normalized drag force F in a simple cubic array (εs = 4.0 ×
10−4) as a function of the Reynolds number from IBM simulations with a well- established correlation
for an isolated particle in an unbounded medium (Schiller & Naumann, 1933). The drag computed for
mean flow oriented at two different angles (θ=0 (∆), θ=π/16 (□)) with respect to the lattice unit vector is
shown to illustrate the dependence on flow angle. For Re<1 (in the Stokes regime), the normalized drag
force is independent of the mean flow angle. However, the drag from IBM simulations is about 20%
higher than the established correlation. The drag computed from IBM is within 4% of LBM simulations
of dilute SC arrays using periodic boundary conditions. The interactions between the periodic images
of the spheres result in higher drag values than an isolated sphere. It is expected that as the volume frac-
tion is further reduced, the numerical predictions will get closer to the drag law in the Stokes limit. The
sphere resolution Dm for the simulation shown is equal to 12.8 grid cells. Even more dilute simulations
will require larger computational grids.
For Re>1, the IBM results are in good agreement with the existing drag law only when the mean
flow is directed at an angle of π/16 in the (x, y) plane. This observation is consistent with the earlier
LBM simulations (Hill, et al., 2001b) where the authors argued that for mean flow angles close to 0 or
π/4, the inertial contributions (or pressure gradient contributions) are reduced due to relatively larger
262
Figure 6. Comparison of the normalized drag force as a function of the solid volume fraction εs in in
Stokes flow past simple cubic and FCC arrays from IBM simulations (open symbols) with the results of
Zick & Homsy (filled symbols)
wake interactions than for the case of θ=π/16. The lower inertial contributions result in a lower value
for total drag for those flow angles. For Re<1 the normalized drag force value is independent of the
mean flow angle because momentum transport is dominated by viscous diffusion. Since diffusion is
symmetric about a sphere, the mean flow angle has no effect on the total drag force in the Stokes regime.
stokes flow
Several correlations have been proposed in the literature for the drag force in Stokes flow past ordered
arrays (SC, FCC, BCC) of spheres. Different analytical and numerical techniques, such as analytical
solution to the Stokes equations (Hasimoto, 1959), Galerkin methods (Snyder & Stewart, 1966; Sorensen
& Stewart, 1974), and the boundary-integral method (Zick & Homsy, 1982), have been used to determine
the drag force in Stokes flow past ordered arrays as a function of solid volume fraction. Since Zick and
Homsy’s results are within 6% of all the other studies, and include all three ordered configurations for
the entire range of solid volume fraction, their values are used in as a benchmark to compare with IBM
simulations. Figure 6 shows that the IBM simulations are in excellent agreement with reported values
from dilute to close-packed limits.
The grid resolution in the IBM simulations for the FCC cases is 25.24 grid points per particle diam-
eter for the minimum volume fraction of 0.01, and 104 grid points per particle diameter for the maximum
volume fraction of 0.698. For the simple cubic cases, Dmis equal to 40.08 for the minimum volume
fraction of 0.01, and 149 for the maximum volume fraction of 0.514.
The validation tests described in this section show that the IBM simulations faithfully reproduce
many standard results published in the literature. In cases where there are differences, these are within
acceptable limits, and are mostly due to the higher resolution used in the IBM simulations. Having es-
263
Figure 7. Comparison of the normalized drag force F for SC arrangement obtained from IBM (open
symbols) with the LBM simulations (filled symbols) of HKL. The solid line is the drag law for a single
particle in an unbounded medium. The flow is directed along the x- axis
tablished that the IBM simulations are numerically convergent and having validated them in standard
tests, we now use IBM to study drag in steady flow past ordered and random arrays.
Ordered arrays
Ladd (Ladd, 1994b) and Hill et al. (Hill, et al., 2001b) have studied steady flow past ordered arrays of
particles using LBM simulations. Our purpose in revisiting this problem is two-fold. IBM simulations
of flow past ordered arrays serve to further validate the method for cases where we can compare with
published data of Hill et al. Secondly, we have more comprehensively explored the parameter space
defined by (Re, εs), especially the low volume fraction region, with higher numerical resolution than
reported thus far in the literature. The dilute cases are more computationally demanding, and have
therefore not received as much attention. However, the behavior of the drag force in the dilute limit is
important because it defines a limiting behavior that drag correlations are typically constrained to satisfy.
Our IBM simulations in the dilute regime reveal some new insights into the correct limiting behavior
that should be imposed as a constraint on drag correlations.
Figure 7 shows the behavior of the normalized drag force obtained from IBM simulations (open sym-
bols) for steady flow past a SC arrangement of particles as a function of Reynolds number, for volume
fractions ranging from very dilute to close-packed limits. Also shown in the same figure is the comparison
(wherever the data is available) with the LBM simulations (filled symbols) of HKL. Figure 8 shows the
same comparison for the FCC arrangement. As both figures show, the IBM and LBM simulations are in
excellent agreement. These results serve to further validate the use of IBM for simulation of flow past
homogeneous particle assemblies.
The solid line in Figures 7 and 8 is the drag on a single particle in an unbounded medium from the
Schiller and Naumann correlation. Comparison with the single sphere drag law (solid line) reveals that
for moderate to high Reynolds numbers, the dilute volume fractions in ordered arrays experience lesser
drag than the drag on a single particle. As observed earlier for the dilute SC array (see and its discus-
sion), and also studied comprehensively in HKL, the normalized drag force in ordered arrays is a func-
264
Figure 8. Comparison of the normalized drag force F for FCC arrangement obtained from IBM (open
symbols) and with the LBM simulations (filled symbols) of HKL. The solid line is the drag law for a
single particle in an unbounded medium. Flow is directed along the x- axis
tion of the flow angle. Therefore, in order to avoid the additional parametrization of the problem by flow
angle, all the simulations have been performed for the case where the mean flow is directed along the
x- axis. However, as shown in HKL, a change in the flow angle can result in drag values that differ by
as much as 200-300% from the zero flow angle case. The main conclusion to be drawn from these
simulations is that the single sphere drag law is not the asymptotic limit of the dilute ordered arrays data.
As we shall see in the next section, the same is true for random arrays also, although they do not ex-
hibit the strong dependence on flow angle characteristic of ordered arrays.
random arrays
Fixed assemblies of randomly distributed particles are closer to the freely evolving suspension problem
that we seek to model than ordered arrays. The particle positions are initialized by first allowing them
to evolve to an equilibrium state following elastic hard-sphere collisions.
We have performed IBM simulations with numerical resolutions comparable or higher than those
used in HKL and BVK, again with an emphasis on characterizing the dilute limit, which is used to as a
limiting case constraint to determine drag law coefficients. Later in this section, the numerical parameters
used in the current IBM simulations are compared with those used in the LBM simulations of HKL and
BVK. In the following, the principal IBM results and the underlying physical mechanisms they reveal
are discussed. Implications of the results for drag laws are then summarized.
dilute arrays
Figure 9 shows the dependence of normalized drag force on the Reynolds number for a random con-
figuration at a dilute volume fraction of 0.01. Symbols are the IBM simulations, with square symbols
for the mean flow directed along the x- axis, and triangles for the mean flow directed at an angle of π/16
in the x-y plane. Solid and dashed-dot lines are the monodisperse drag laws from LBM simulations of
HKL and BVK, respectively, and the dashed line is the single sphere drag law of Schiller and Naumann.
265
Figure 9. Normalized drag force F vs. Reynolds number for a random arrangement of particles at solid
volume fraction equal to 0.01. Symbols are the IBM simulations: squares denote the case where the flow
is directed along the x- axis; triangles denote the case where the flow is directed at an angle of π/16 in
the x-y plane
Comparison of the IBM simulations with existing monodisperse drag laws of HKL and BVK shows
an excellent match in the Stokes regime and at low Re, but differences as high as 100-200% in the
moderate and high Re regime. HKL (Hill, et al., 2001a) simulated such dilute volume fractions only for
the Stokes regime, but due to the coarse resolution of less than 2 lattice nodes for particle diameter they
did not simulate higher Reynolds numbers for this volume fraction. BVK did not simulate any case for
εs ≤ 0.1. In HKL it is noted that due to the approximate approach used to obtain the inertial contribution
(denoted by F3in their study) to the total drag, their drag law is a good estimate of the actual drag force
over the entire range of Reynolds number only for relatively high solid volume fractions. This is a plau-
sible explanation for the departure of IBM simulations from the HKL drag law. The departure of IBM
simulations from BVK’s monodisperse drag law is attributed to the incorrect constraint imposed on their
drag law to the single-sphere drag correlation at infinite dilution. The BVK drag law assumes that the
drag in random homogeneous suspensions at infinite dilution (i.e., εs → 0) should tend to the drag on
an isolated particle. From both IBM and LBM simulations, it is clear that this assumption does not hold
true even at the moderately dilute volume fraction of 0.01.
At low Re, viscous terms that are local (short range) dominate. Since the viscous forces are short
ranged, it is reasonable to expect that at infinite dilution and low Re, the normalized drag force will ap-
proach that of single-sphere drag (i.e., F → 1 as εs → 0 and Re → 0). As the Reynolds number increases,
the contribution from inertial terms dominates the viscous effects, and since pressure obeys an elliptic
equation these are long range (nonlocal) interactions. For moderate to high Reynolds numbers flow
past random arrays, even for fairly dilute solid volume fractions the simulation data do not support the
assumption of constraining the drag law to approach that of single-sphere.
Similar to the observations for ordered arrays (Figures 7 and 8), the drag force on dilute suspensions
for moderate to high Reynolds numbers is less than the drag force experienced by an isolated particle
in an unbounded medium. However, unlike in ordered arrays the drag force in random arrays is not
266
dependent on the flow angle due to isotropy of the particle configuration. For ordered arrays, the strong
influence of flow angle on the drag force at moderate to high Reynolds numbers is attributed by HKL
to the different length scales at which the inertial contributions interact. The distribution of neighbor
particles in ordered arrays is anisotropic, and the pair correlation function is sharply peaked at the lat-
tice points. However, in the random particle configurations generated by elastic hard-sphere collisions,
the pair correlation is isotropic. Therefore, the drag force is insensitive to flow angle for all Reynolds
numbers in random arrays.
Moderately dilute to dense arrays
Figure 10 shows the comparison of normalized drag force in random arrays for volume fractions equal
to 0.1 and 0.2 obtained from IBM simulations (open symbols) with the existing monodisperse drag laws
of HKL and BVK. shows the same comparison for volume fractions equal to 0.3 and 0.4. It can be seen
that IBM simulations are in excellent agreement with HKL’s drag law for Re up to 100, which is nearly
the upper limit of Reynolds number simulated by HKL. The extension of their drag law to higher Re
does not agree well with IBM simulations as the solid volume fraction is reduced. This is attributed to
the observation made in HKL that their drag law is a good estimate of the actual drag force over a wide
range of Reynolds number only for relatively high solid volume fractions.
numerical Parameters and resolution
Choosing the numerical resolution for random arrays or a fixed level of computational work should be
based on an optimal combination of box-size and grid resolution. Table 1 compares the numerical resolu-
tions for different volume fractions used in our IBM simulations with those used in LBM simulations of
HKL and BVK. It is noted that not all choices of the numerical parameters for IBM are in the “resolved”
range as determined by our limited set of numerical convergence tests. However, as noted earlier, these
tests are themselves not comprehensive, and so ultimately the choice of numerical parameters reflects
an attempt to balance various contributions to the numerical error. Given the relatively low sensitivity of
the mean drag force to L/D ratio in IBM, we have used values from past LBM simulations as a guideline,
choosing higher grid resolution over larger box size in some of our IBM simulations.
Figure 10. Comparison of the normalized drag force F for random arrays at volume fractions equal to
0.1 and 0.2 from IBM simulations (open symbols) with the monodisperse drag laws of HKL and BVK
267
Figure 11. Comparison of the normalized drag force F for random arrays at volume fractions equal to
0.3 and 0.4 from IBM simulations (open symbols) with the monodisperse drag laws of HKL and BVK
The numerical resolutions for HKL that are reported in Table 1 are those used for the maximum
Reynolds numbers and are taken from Table 1 in (Hill, et al., 2001b). For Stokes flow, HKL have used
similar numerical resolution for εs ≥ 0.1. However, for very dilute volume fractions, very coarse resolu-
tions of less than 2 lattice nodes across a particle diameter have been used. In BVK, a constant resolution
of 17.5 lattice units across a particle diameter was used for εs ≤ 0.2, and for higher volume fractions,
their results were obtained by averaging over two resolutions of 17.5 and 25.5 lattice units. Therefore,
in Table 1, we have used the average value of 21.5 lattice units to report their resolutions for εs ≥ 0.3.
In both the studies, random configurations for volume fraction less than 0.1 were not simulated for the
entire range of Reynolds numbers. So there is no numerical resolution comparison for εs = 0.01. Table
1 shows that the IBM simulations are consistently better resolved in terms of the number of particles,
grid resolution, and the box-size. BVK performed greater number of MIS but the scatter in IBM data
does not point to a need for such high number of MIS. Therefore, normalized drag values averaged over
5 MIS are reported here.
computational cost
The computational cost associated with IBM simulations of flow past fixed assemblies of monodisperse
particles performed using a serial code is discussed in this section. All the simulations are performed
on an AMD Opteron cluster with 148 compute nodes. The compute nodes consist of 2.4 GHz dual core
dual processors. The processor type is AMD 2800 Opteron with a peak performance of 4.8 Gigaflops.
At a given volume fraction and Reynolds number, the simulation time for the mean particle drag to
reach steady state is denoted Tsim and can be written as:
ˆ(ε ) × M (Re, ε ) × N (Re, ε ).

Tsim = T s s T s
In the above equation, T̂ is the computational cost per grid cell per time step, M denotes the number
of grid cells and NT is the number of time steps taken for the mean drag to reach a steady state.
The computational cost per grid cell per time step T̂ is independent of the Reynolds number and
depends only on the volume fraction. In Table 2, the values for T̂ are reported in microseconds for vari-
ous volume fractions. It can be seen from Table 2 that T̂ increases with volume fraction. The number of
268
Table 2. Computational cost per grid cell per time step for IBM simulations of flow past fixed assemblies
of monodisperse particles at various volume fractions
εs T̂(μs)
0.01 4.612
0.1 5.033
0.2 5.693
0.3 5.745
0.4 6.071
grid cells M required for a well resolved simulation also depends on the volume fraction and Reynolds
number as shown in Table 1. From the values of optimal L/D and Dm given in Table 1, the number of
grid cells can be calculated using:
3
L 
M =  Dm  .
 D 
It should be noted that the number of time steps NT that is required for the average drag on the par-
ticles to reach a steady state also depends on the physical parameters of the problem (solid volume
fraction and Reynolds number). Typically, it is observed that at a given Reynolds number, dilute particle
assemblies take more time to attain steady state than the denser ones. And at a given volume fraction,
suspensions at higher Reynolds numbers take longer to attain steady state.
summary
IBM simulations show an excellent match with the drag correlations proposed by HKL and BVK for low
Reynolds number for both dilute and moderately dense random arrays. However, the IBM simulations show
a significant departure from these correlations at higher Re, and for dilute cases. The drag law proposed
by HKL is stated to be more reliable for all Reynolds numbers only at higher volume fraction. The BVK
drag correlation is proposed based on a fit to 5 drag values over a wide range of Reynolds number, and
their simulations appear to be susceptible to numerical resolution errors. For a given volume fraction,
they used a constant resolution of the particle diameter to simulate Reynolds numbers ranging from 21
to 1000. As the volume fraction is increased, the number of grid/lattice nodes in the gaps between the
spheres decrease and a progressively higher grid resolution is required. In the HKL study the particle
resolution was increased from 9.6 lattice units per particle diameter for the lowest volume fraction of 0.1
to 41.6 lattice units for the highest volume fraction of 0.641, which is a four-fold increase. However, in
the BVK study the particle resolution increased by only a fraction for a wide volume fraction range of
0.1-0.6. The IBM simulations suggest that a more complete parametric study at high resolution could
significantly revise these existing drag laws.
269
assessMent Of ibM fOr drag law fOrMulatiOn
Simulations of steady flow past homogeneous particle assemblies using IBM reveal that fundamentally
differing computational approaches to gas-solids flow are in remarkably good agreement for a wide
variety of test cases. Overall this is strong evidence of the consistency between different computational
approaches to the problem of drag law formulation in gas-solids flow, which is difficult to study through
experiment. However, all computational predictions of drag in gas-solids flow are subject to uncertain-
ties arising from numerical error, and should be interpreted as accurate only within 5%. In the following
we compare and contrast the IBM approach with LBM, which is a popular computational approach for
gas-solids flow.
While IBM solves the continuum Navier-Stokes equations, LBM solves for the discrete one-particle
velocity distribution function whose evolution is described by the lattice Boltzmann equation (He & Luo,
1997). It is useful to think of LBM as a solution to the lattice Boltzmann equation, which is obtained
by Hermite-Gauss quadrature of the modeled Boltzmann equation. LBM fundamentally differs from
continuum solutions to Navier-Stokes equations like IBM because it directly solves for a discrete form
of the velocity distribution function at the molecular level. From the LBM solution the hydrodynamic
mean fields such as fluid velocity and pressure can be calculated. Since LBM operations are local in
physical space, it avoids solving the elliptic pressure Poisson equation that is needed in incompressible
continuum flow solvers. This paves way for efficient parallelization of LBM, which has opened door to
solving realistic flow problems (Chen & Doolen, 1998).
However, there are some issues worth considering when using LBM for gas-solids suspension. The
restriction of molecular velocities to discrete values on a lattice is now known to be unnecessary, and
even undesirable for many flow problems, especially in multiphase flow (Fox, 2008). Another feature of
LBM is that it always results in a compressible flow solution, and as a result the solution of incompress-
ible flow at high Reynolds numbers is challenging. In order to reduce the errors due to compressibility
effects at higher Reynolds numbers, the viscosity of the fluid has to be reduced (Ladd, 1994a, 1994b).
When we consider suspension flows, some very important differences arise between IBM and
LBM. In LBM a spherical particle is represented by a stair-step lattice approximation, i.e., the surface
is represented by a set of lattice sites closest to the input diameter D0. Due to this stair-step representa-
tion of the particle surface, the exact value of the particle diameter that appears in the LBM drag law
is not specified a priori. Furthermore, the bounce-back scheme used to implement the no slip bound-
ary condition at the particle-fluid interface does not result in a zero velocity contour coincident with a
stationary sphere boundary. Therefore, in LBM the drag computed directly from the fluid stress at the
particle surface does not correspond to the drag on a sphere of diameter D0. The drag values in LBM
simulations are assumed to correspond to an effective hydrodynamic diameter Dh that is unknown a
priori. The hydrodynamic diameter Dh is obtained a posteriori by calibrating the LBM simulations
against the analytical solution of Hasimoto (Hasimoto, 1959) for Stokes flow in a dilute SC arrange-
ment of spheres. This hydrodynamic diameter depends on the fluid viscosity as well as the particle size.
Therefore, a calibration curve is needed in LBM for every choice of kinematic viscosity and particle
diameter D0. This is a serious limitation for a DNS calculation. In contrast, in IBM the particle surface
is represented as a sphere. The fluid stress on the particle surface is calculated directly from the flow
fields with no intermediate calibration procedure. Therefore, IBM is a first-principles, physics-based
true DNS approach for simulation of gas-solids flows.
270
It is also interesting to note that the drag on the particle reported using D0 gives first order convergence
whereas drag reported using Dh results in approximately quadratic convergence (Ladd & Verberg, 2001).
However, this convergence rate is not independent of the kinematic viscosity of the fluid. Even though
the calibration of hydrodynamic diameter is only done for a single sphere, the same calibration is used
for simulating dense ordered suspensions in the Stokes regime (Ladd, 1994a, 1994b) as well as random
arrays at higher Reynolds number (Hill, et al., 2001a; van der Hoef, et al., 2005).
LBM is a highly efficient and robust solution methodology for gas-solids flow. Overall, it appears
that LBM results for mean drag are relatively insensitive to grid resolution when compared with IBM.
However, this insensitivity of the LBM solution to grid resolution should be carefully interpreted because
LBM yields stable solutions even when the flow is highly under-resolved. For instance, (Beetstra, et al.,
2007) show that the drag force for a dense random packing of 0.5 at Reynolds number equal to 1049
is insensitive to the grid resolution in the range 10 to 50 lattice units per particle diameter. However,
for these grid resolutions it is clear that the boundary layers around the particles cannot be resolved at
such a high Reynolds number. Some studies also report greater sensitivity of LBM to grid resolution.
For example, in the monodisperse simulations of (van der Hoef, et al., 2005) at a volume fraction of 0.5
in the Stokes flow regime, a strong dependence of the drag force on the grid resolution and kinematic
viscosity is observed.
The sensitivity of IBM results to grid resolution has already been discussed, and we find that the IBM
results for the surface viscous stress show the correct increasing trend with increasing grid resolution
as the velocity gradients are better resolved. If IBM is used to simulate high Reynolds number flows
with insufficient resolution (e.g., Re>500 with grid resolution in Table 1), the solution becomes unstable
because of the non-dissipative second-order upwind schemes that have been incorporated for high ac-
curacy. This informs the IBM user that higher grid resolution should be employed to obtain stable and
accurate solutions.
From the preceding discussion we can see that IBM has some unique advantages in solving gas-solids
flow problems that derive from its solution approach to the continuum Navier-Stokes equations. By virtue
of its implementation into structured Cartesian grid solvers, it incurs minimal increase in computational
cost with increasing number of particles. To give a rough idea of the order of magnitude of the increase
in computational cost, the increase is only about 25% going from 2 particles to 97 particles, but the exact
value depends on the Reynolds number and volume fraction. The results presented in this chapter show
that IBM yields numerically convergent solutions to important hydrodynamic problems in gas-solids
flow, which compare well with many established results in the literature. We also find that this powerful
tool is capable of giving additional insight into the important limiting case of steady flow past dilute
random arrays. Also a more thorough exploration of the volume fraction-Reynolds number parameter
space suggests significant changes to existing drag correlations may be required. In the next section
we outline future directions for IBM as a computational method for solving gas-solids flow problems.
future directiOns
Future directions for the use of IBM as a DNS approach for gas-solids flow can be classified into two
broad categories: (i) applications of the IBM approach described in this work to other gas-solids flow
problems, and (ii) development of the IBM formulation for other gas-solids flow problems.
271
Other applications of the ibM approach
The IBM approach described in this work can be used to quantify the terms that appear in the second mo-
ment equations that derive from the transport equation of the one-particle distribution function (Equation
(1.1)). Preliminary work in this direction is already in progress, and our initial studies indicate that such
calculations will require the simulation of freely moving suspensions. In these calculations the particles
need to evolve by the hydrodynamic force that the fluid exerts through pressure and surface viscous
stresses. Particle collisions also need to be accounted for, and for dense suspensions it is appropriate to
use the discrete element method based on the spring-dashpot soft-sphere model (Cundall & Strack, 1979).
In gas-solids suspension flows, the problem of segregation of unlike particles (differing in size or
density) is of great practical interest. The IBM approach described here can be easily extended to simu-
late polydisperse assemblies of particles, and work is ongoing to quantify drag laws for polydisperse
gas-solids flow using IBM.
It is well established in the literature that gas-solids riser flows exhibit particle clustering effects (Col-
lins & Keswani, 2004; Heynderickx, Das, De Wilde, & Marin, 2004; Knowlton, Karri, & Issangya, 2005;
Krol, Pekediz, & de Lasa, 2000; Wylie & Koch, 2000). Also experiments (Moran & Glicksman, 2003)
have shown that gas-phase turbulence interacts differently with particle clusters, than with individual
particles. The IBM method described here has been modified to simulate an inflow-outflow boundary
condition to study the influence of different particle arrangements (uniform and clustered) on upstream
gas-phase turbulence (Xu, 2008). It should be mentioned that the current computational power only
allows for DNS of simple flows, such as flow past homogeneous assemblies considered in this study.
Application of DNS to device scale problems is not envisaged because of the high computational cost
arising from the large range of scales.
extensions to the current ibM formulation
The principal extensions to the IBM formulation lie in the areas of (a) generalizing boundary conditions,
(b) extending the equation set to include other physical effects, such as transport of chemical species
and heat transfer, and (c) improving the solution procedure. The IBM approach described here can be
generalized to incorporate arbitrary boundary conditions and to include complex geometries.
The transport of scalars, such as chemical species or temperature, in gas-solids flow is also a prob-
lem of practical importance. The IBM approach can be applied to the problem of transport of a passive
scalar by a relatively easy augmentation of the existing equation set. This extension is also in progress
and preliminary results show that IBM is a versatile tool that can be used to study heat and mass transfer
in gas-solids flow as well.
The solution procedure employed in this work can be improved by incorporating recent advances
in differencing the pressure equation that have been derived in the context of immersed interface meth-
ods. This differencing formula accounts for a discontinuous jump in the pressure across the solid-fluid
interface (Xu & Wang, 2006, 2008; Lee & Leveque, 2003).
Parallelization strategy
A major advantage of the immersed boundary method is its capability to solve for flow over complex
geometries on a uniform Cartesian grid. Efficient domain decomposition algorithms for Cartesian
272
topologies are provided in almost all the standard MPI implementations – an advantage of Cartesian
topologies compared to body-fitted meshes where one has to rely on third-party libraries like METIS
for domain decomposition. The dominant communication costs in any parallel implementation depend
upon the order of accuracy of the discretization schemes. For example, a second-order central differenc-
ing scheme requires communication of the solution on the first layer of grid nodes that lie adjacent to
the partition boundaries. In the parallel implementation of IBM, no special distribution of the particles
among the processors is required as the particles are assigned to processors based on their spatial loca-
tions. A particle that lies on a partition boundary is partitioned appropriately and the velocity and pres-
sure fields that are exchanged after every time step are sufficient to calculate the immersed boundary
force acting on the portion of the particle that lies within a processor. The immersed boundary field is
communicated at each time step similar to communications for pressure and velocity fields. Therefore,
as discussed before, the parallel implementation preserves the scale up of computational cost with the
number of particles as the communication overhead incurred for forcing field is minimal.
cOnclusiOn
IBM is a powerful and efficient computational method for direct numerical simulation of gas-solids flow.
This contribution connects the quantities computed from DNS using the IBM approach with the inter-
phase momentum transfer term arising in theoretical approaches to gas-solids flow. This correspondence
is described at different levels, starting from the one-particle distribution function and leading naturally
to the averaged equation in that approach. An important connection of IBM quantities with two-fluid
theory is also established. The numerical convergence of IBM is established and its performance in
various validation tests is described. It is shown that IBM simulations reproduce known results for the
average drag in Stokes flow past ordered arrays. For random arrays, the IBM results reveal interest-
ing insights in the dilute limit, and suggest changes to existing drag laws may be required following
comprehensive exploration of the Reynolds number-solid volume fraction parameter space. The IBM
approach is versatile, and can be extended to include effects of gas-phase turbulence, polydispersity
in the size distribution of solid particles, and transport of chemical species and heat due to fluid flow.
acKnOwledgMent
This work is supported in part by the National Energy Technology Laboratory through Ames Laboratory
grant DE-AC02-07CH11358 (RG and SS) and DOE-AR grant DE-FC26-07NT43098 (RG, ST, and SS).
The authors would like to thank Dr. M. G. Pai for insightful discussions during the course of this study,
and useful comments on a draft of the manuscript.
references
Beetstra, R., van der Hoef, M. A., & Kuipers, J. A. M. (2007). Drag force of intermediate Reynolds
number flow past mono- and bidisperse arrays of spheres. AIChE Journal. American Institute of Chemi-
cal Engineers, 53(2), 489–501. doi:10.1002/aic.11065
273
Chapman, S., & Cowling, T. G. (1952). The mathematical theory of non-uniform gases; an account of
the kinetic theory of viscosity, thermal conduction, and diffusion in gases (2nd ed.). Cambridge, UK:
University Press.
Chen, S., & Doolen, G. D. (1998). Lattice Boltzmann method for fluid flows. Annual Review of Fluid
Collins, L. R., & Keswani, A. (2004). Reynolds number scaling of particle clustering in turbulent aero-
sols. New Journal of Physics, 6.
Cundall, P. A., & Strack, O. D. L. (1979). Discrete Numerical-Model for Granular Assemblies. Geo-
technique, 29(1), 47–65. doi:10.1680/geot.1979.29.1.47
Drew, D. A., & Passman, S. L. (1999). Theory of multicomponent fluids. New York: Springer.
Garzo, V., Hrenya, C. M., & Dufty, J. W. (2007). Enskog theory for polydisperse granular mixtures. II.
76(3). doi:10.1103/PhysRevE.76.031304
Goldstein, D., Handler, R., & Sirovich, L. (1993). Modeling a no-slip flow boundary with an external
force field. Journal of Computational Physics, 105(2), 354–366. doi:10.1006/jcph.1993.1081
Hasimoto, H. (1959). On the periodic fundamental solutions of the Stokes equations and their application
to viscous flow past a cubic array of spheres. Journal of Fluid Mechanics, 5(2), 317–328. doi:10.1017/
S0022112059000222
He, X. Y., & Luo, L. S. (1997). Theory of the lattice Boltzmann method: From the Boltzmann equation
to the lattice Boltzmann equation. Physical Review E: Statistical Physics, Plasmas, Fluids, and Related
Interdisciplinary Topics, 56(6), 6811–6817. doi:10.1103/PhysRevE.56.6811
Heynderickx, G. J., Das, A. K., De Wilde, J., & Marin, G. B. (2004). Effect of clustering on gas-solid
drag in dilute two-phase flow. Industrial & Engineering Chemistry Research, 43(16), 4635–4646.
doi:10.1021/ie034122m
Hill, R. J., Koch, D. L., & Ladd, A. J. C. (2001a). The first effects of fluid inertia on flows in ordered and
Hill, R. J., Koch, D. L., & Ladd, A. J. C. (2001b). Moderate-Reynolds-number flows in ordered and
Hill, R. J., Koch, D. L., & Ladd, A. J. C. (2001c). Moderate Reynolds number flows in ordered and
Ikeno, T., & Kajishima, T. (2007). Finite-difference immersed boundary method consistent with wall
conditions for incompressible turbulent flow simulations. Journal of Computational Physics, 226(2),
1485–1508. doi:10.1016/j.jcp.2007.05.028
274
Jenkins, J. T. (1998). Kinetic theory for nearly elastic spheres. Physics of Dry Granular Media, 350,
353–370.
Knowlton, T. M., Karri, S. B. R., & Issangya, A. (2005). Scale-up of fluidized-bed hydrodynamics.
Koch, D. L. (1990). Kinetic-Theory for a Monodisperse Gas-Solid Suspension. Physics of Fluids. A,
Fluid Dynamics, 2(10), 1711–1723. doi:10.1063/1.857698
Koch, D. L., & Hill, R. J. (2001). Inertial effects in suspension and porous-media flows. Annual Review
of Fluid Mechanics, 33, 619–647. doi:10.1146/annurev.fluid.33.1.619
Krol, S., Pekediz, A., & de Lasa, H. (2000). Particle clustering in down flow reactors. Powder Technol-
ogy, 108(1), 6–20. doi:10.1016/S0032-5910(99)00196-5
Ladd, A. J. C. (1994a). Numerical Simulations of Particulate Suspensions Via a Discretized Boltzmann-
Equation. 1. Theoretical Foundation. Journal of Fluid Mechanics, 271, 285–309. doi:10.1017/
S0022112094001771
Ladd, A. J. C. (1994b). Numerical Simulations of Particulate Suspensions Via a Discretized
Boltzmann-Equation. 2. Numerical Results. Journal of Fluid Mechanics, 271, 311–339. doi:10.1017/
S0022112094001783
Ladd, A. J. C., & Verberg, R. (2001). Lattice-Boltzmann simulations of particle-fluid suspensions. Journal
of Statistical Physics, 104(5-6), 1191–1251. doi:10.1023/A:1010414013942
Lee, L., & Leveque, R. J. (2003). An immersed interface method for incompressible Navier-Stokes
equations. SIAM Journal on Scientific Computing, 25(3), 832–856. doi:10.1137/S1064827502414060
Liboff, R. L. (1990). Kinetic theory: classical, quantum, and relativistic descriptions. Englewood Cliffs,
NJ: Prentice Hall.
Mohd-Yusof, J. (1996). Interaction of massive particles with turbulence. Cornell University.
Moran, J. C., & Glicksman, L. R. (2003). Experimental and numerical studies on the gas flow surrounding
a single cluster applied to a circulating fluidized bed. Chemical Engineering Science, 58(9), 1879–1886.
doi:10.1016/S0009-2509(02)00684-X
Pai, M. G., & Subramaniam, S. (2009). A comprehensive probability density function formalism for
multiphase flows. Journal of Fluid Mechanics.
Peskin, C. S. (1982). The Fluid-Dynamics of Heart-Valves - Experimental, Theoretical, and Computational
Methods. Annual Review of Fluid Mechanics, 14, 235–259. doi:10.1146/annurev.fl.14.010182.001315
Pope, S. B. (2000). Turbulent Flows. Cambridge, UK: Cambridge University Press.
Rai, M. M., Gatski, T. B., & Erlebacher, G. (1995). Direct simulation of spatially evolving compressible
turbulent boundary layers. Paper presented at the AIAA 950583, 33rd Aerospace Sciences Meeting and
Exhibit, Reno, NV.
Schiller, L., & Naumann, A. Z. (1933). A Drag Coefficient Correlation. Z. Ver. Deutsch Ing., 77, 318–320.
275
Snyder, L. J., & Stewart, W. E. (1966). Velocity and pressure profiles for newtonian creeping flow in
regular packed beds of spheres. A.I. Ch.E.J., 12(1), 167.
Sorensen, J. P., & Stewart, W. E. (1974). Computation of forced-convection in slow flow through ducts
and packed-beds. 2. velocity profile in a simple cubic array of spheres. Chemical Engineering Science,
29(3), 819–825. doi:10.1016/0009-2509(74)80200-9
Subramaniam, S. (2001). Statistical modeling of sprays using the droplet distribution function. Physics
of Fluids, 13(3), 624–642. doi:10.1063/1.1344893
Syamlal, M., Rogers, W., & O’Brien, T. J. (1993). MFIX Documentation: Theory Guide. Morgantown:
National Energy Technology Laboratory.
van der Hoef, M. A., Beetstra, R., & Kuipers, J. A. M. (2005). Lattice-Boltzmann simulations of low-
drag force. Journal of Fluid Mechanics, 528, 233–254. doi:10.1017/S0022112004003295
Verzicco, R., Mohd-Yusof, J., Orlandi, P., & Haworth, D. (2000). Large eddy simulation in complex
geometric configurations using boundary body forces. AIAA Journal, 38(3), 427–433. doi:10.2514/2.1001
Wylie, J. J., & Koch, D. L. (2000). Particle clustering due to hydrodynamic interactions. Physics of
Fluids, 12(5), 964–970. doi:10.1063/1.870351
Xu, Y. (2008). Modeling and direct numerical simulation of particle laden turbulent flows. Iowa State
University, Ames, IO.
Yusof, J. M. (1996). Interaction of massive particles with turbulence. Ithaca, NY: Cornell University.
Zick, A. A., & Homsy, G. M. (1982). Stokes flow through periodic arrays of spheres. Journal of Fluid
Mechanics, 115, 13–26. doi:10.1017/S0022112082000627
276
277
Chapter 9
The Multiphase Particle-
in-Cell (MP-PIC) Method
for Dense Particle Flow
Dale M. Snider
CPFD, USA
Peter J. O’Rourke
CPFD, USA
abstract
is a derivative of the Particle-in-Cell (PIC) method for multiphase flows and the method employs a fixed
Eulerian grid, and Lagrangian parcels are used to transport mass, momentum, and energy through this
grid in a way that preserves the identities of the different materials associated with the particles. The
main distinction with traditional Eulerian-Lagrangian methods is that the interactions between the par-
ticles are calculated on the Eulerian grid. One of the main advantages of PIC methods is the accuracy
of the convection algorithms (Lagrangian advection is non-diffusive) and this can be important aspect
of gas-solids flows where the overall dynamics is dictated by the instabilities in the system due to sharp
interfaces between the phases. Additional details about this method along with examples are provided
in this chapter. In the following chapter, the application of this method to Circulating Fluidized Beds
(CFBs) is described.
intrOductiOn
For the most part, mathematical models of separated particulate multiphase flow have used either a
continuum approach for both phases (Gidaspow, 1986, Batchelor, 1988, Jackson, 2000) or a continuum
approach for the fluid and a Lagrangian computational-particle model for particle phase (Cundall &
DOI: 10.4018/978-1-61520-651-3.ch009
The Multiphase Particle-in-Cell (MP-PIC) Method for Dense Particle Flow
Strack, 1979, Amsden, et al,, 1989, Snider 2001, Godlieb, et al, 2007). The two-fluid continuum approach
averages both fluids and solids by a statistical procedure (typically ensemble-averaging) and treats the
solids phase as a pseudo continuum. The averaging procedure leads to many unclosed terms which must
be modeled (Jackson, 2000), and much progress has been made in the modeling of such terms as the
particle-particle-stresses and particle viscous-stresses in dense particle flows using spatial gradients of
average particle properties (Batchelor, 1988, Gidaspow, 1994).
The two-fluid continuum approach has trouble modeling flows with a distribution of particle types
and sizes because separate continuity and momentum equations must be solved for each size and type
(Gidaspow, 1994, Risk, 1993). While a continuous-fluid description of the solids phase has application
in some solid-fluid flow regimes it is inaccurate in others. For dilute solid flows, closure models based
on the assumption of high collision frequencies are questionable. In addition, the non-linear behavior
of some solid flows is difficult to model with a Navier-Stokes type momentum equation. The different
characters of fluid and solid flows are seen in an hour-glass and in a U-tube. The hour glass works as
a useful measure of time because the particle flow rate through the orifice is constant and independent
of sand depth. If instead, a fluid were used in the hour glass, the flow rate is initially high because of a
hydrostatic head, and the flow tapers off as the hour glass empties of fluid. In the simple U-tube, a fluid
will fill both arms to the same height (manometer). However, solids poured into one arm of the U-tube
will only fill the one arm.
Modeling the particle phase as discrete particles or elements has a number of variations. The basic
direct numerical solution of resolving the fluid flow around each particle with no-slip condition, and
coupling the motion of particles, can only be done for a small number of particles. Similarly, Lattice
Boltzmann calculations are limited to a small number of particles. Both methods are useful in studying
fundamental solid-fluid flow behavior and can be used in developing closure models for fluid-particle
interactions needed in continuum models. By averaging the fluid over a spatial region containing a num-
ber of particles and using a closure model for the fluid-particle momentum transfer, large systems can
be modeled. The discrete particle or discrete element method (DPM or DEM) (Cundal & Strack, 1979)
is based on a finite number of discrete semi-rigid spherically, or regularly, shaped particles interacting
through contact forces and transferring momentum to and from the fluid by a drag closure model. The
multitude of particle-to-particle contacts and the model for the contact force are the most significant
parts of the direct element model. To efficiently calculate particle interactions, relatively simple models,
often based on a simple coefficient of restitution, are employed. DPM allows solution for flows with a
wide range of particle types, sizes, shapes and velocities. Because, of the high collision frequency for
volume fractions above 5% (O’Rourke, 1981) and the computational complexity of calculating dense
particle-particle interactions, DPM calculations have been limited to on the order of 2x105 particles and
are often restricted to two-dimensional solutions without a fluid phase (Godlieb, et al., 2007).
In this paper, we give a review of the multiphase particle-in-cell (MP-PIC) numerical method for
predicting dense particulate-fluid flow. The MP-PIC method is a discrete particle method (DPM)
where the fluid phase is treated as a continuum (Eulerian) and the solids are modeled by Lagrangian
computational particles. The MP-PIC method grew out of the earlier Particle-in-Cell (PIC) method for
single-phase flows. The PIC method was originally proposed and developed by Harlow (1957) at the
Los Alamos National Laboratory. The method employs a fixed Eulerian grid, and Lagrangian particles
are used to transport mass, momentum, and energy through this grid in a way that preserves the identities
of the different materials associated with the particles. Interactions between the particles are calculated
on the Eulerian grid. Particle properties are interpolated to the grid, and the flow field is updated on
278
the grid. Properties are then interpolated back to particles. A primary motivation for PIC methods was
probably their accuracy in tracking interfaces; however, additional benefits are that the calculation of
convection is very accurate because of the Lagrangian nature of the PIC methods, and PIC methods
offer the promise of subgrid resolution of those properties which are not updated on the grid. Other
particle-in-cell methods developed at Los Alamos were the Marker-and-Cell (MAC) (Harlow & Welch,
1965) method for incompressible flow and the FLIP (Brackbill & Ruppel, 1986) particle-in-cell scheme,
which made significant improvements in the accuracy of the original PIC method. Using a stochastic
computational particle method, the Eulerian-Lagrangian KIVA program models discrete spray droplets
including collision, break-up, vaporization, chemistry coupled with a turbulent gas phase (Amsden, et
al, 1989). O’Rourke, et al (1993) presented particle-in-cell models applied to chemically reacting flow.
Andrews and O’Rourke first proposed the multiphase particle-in-cell (MP-PIC) method and demon-
strated the method with one-dimensional simulations compared to analytical solutions and experimental
data (Andrews and O’Rourke, 1996). The MP-PIC method is an extension of the stochastic particle
method of the KIVA code (Amsden, et al, 1989) in which an isotropic particle stress gradient is added
to the equation of motion of the particles. This addition enabled the method to calculate flows with any
particle volume fraction from the dilute to the close pack limits. The particle stress gradient, which is
difficult to calculate for each particle in dense flow, is calculated as a gradient on the grid, fully coupled
to the other particle and gas acceleration terms, and is then interpolated to the discrete particles. Snider,
et al (1998) extended the method to two-dimensions, with a smooth grid-to-particle interpolation method
and demonstrated the accuracy in calculating polydisperse solid settling. Snider (2001) further extended
MP-PIC to three dimensions and implemented the MP-PIC numerical method into the framework of
the CPFD numerical methodology. The MP-PIC method, as implemented in the CPFD methodology, is
described in the following sections with examples illustrating its accuracy for predicting fluid-particle
flows. MP-PIC deals primarily with modeling the physics of the solid phase; however, the solids and
fluid phases are coupled through mass, momentum, and energy transfer between the phases. The total
solution of solid-fluid physics is under the umbrella of the CPFD methodology.
equatiOns Of MOtiOn
solid Phase equations
The state of the particle field is described in the MP-PIC method by the particle distribution function f.
In this section we define f and give its transport equation. The distribution function f is the local number
density of particles times the local probability distribution function of particle properties that are im-
portant in our particular application. In this section, we take the mass of the particle ms, and the particle
velocity us as the particle properties Thus,
f ( x,ms , u s ,t ) dms du s (1)
is the average number density of particles at spatial location x and at time t, with solid mass in the interval
(ms, ms+dms), and the Cartesian velocity in the interval (us,us+dus). A bold alphabetic letter represents a
vector. Thus, dus represents du1 du2 du3 where the subscripts represent the three orthogonal directions. At
279
times it is clearer to denote a vector with a subscript such that duj represents du1 du2 du3 and repetitions
of the same index follows the common tensor rule of implied summation.
The time rate of change of the particle distribution function is given by a conservation equation in
multi-dimensional space. This transport equation for the particle distribution function is
∂f ∂
+ ∇ x ⋅ ( fu s ) + ∇u ⋅ ( fu s ) + ( fm s ) = 0 (2)
∂t ∂ ms
dms  ∂ ∂ ∂ 
where m s = and ∇u =  , ,  . Equation (2) is a statement of conservation of particle
dt  ∂u1 ∂u2 ∂u3 
numbers assuming no collisions and no breakups (zero on the right-hand side of the equation), and is
derived in an analogous fashion to the derivation of the Boltzmann equation of gas dynamics (Hirshfelder,
et al., 1954, Vincenti & Kruger, 1975). We point out that in a recent extension of the MP-PIC method
collision terms have been added to the right-hand side of Equation (2), and a numerical solution method
for these collision terms has been incorporated in the CPFD methodology (O’Rourke, et al., 2009). These
collision terms increase the accuracy of CPFD in predictions of particle segregation in polydisperse
granular flows.
The evolution in time of the particle distribution function depends on average values of the accel-
eration u s and rate-of-mass-change m s of a single particle. In MP-PIC, the average acceleration on a
particle is postulated to be
dus 1 1
u s = = Ds (u f − us ) − ∇p + g − ∇τ s (3)
dt ρs θ sρs
where us is the particle velocity, uf is the fluid velocity, ρs is the particle material density, g is the ac-
celeration due to gravity, p is the fluid phase pressure, Ds is the so-called drag function (Andrews and
O’Rourke, 1996), and τs is the particle normal stress due to collisions. The particle volume fraction, θs,
and the particle normal stress τs are defined later. The terms in Equation (3) represent acceleration due
to aerodynamic drag, pressure gradient (buoyant forces), gravity and interparticle contact stresses. In
this review, we ignore particle mass change and take m s = 0 .
The Liouville Equation (2) is the mathematical expression of conservation of particle numbers in vol-
umes moving along dynamic trajectories in particle phase space. In the MP-PIC method, we approximate
the Lagrangian equations for particle trajectories in lieu of direct solution of the Liouville Equation (2).
fluid Phase equations
The equations of fluid dynamics can be derived from kinetic theory (dilute gases) or continuum points
of view (Harlow & Amsden, 1971, Vincenti & Krugger, 1975), but the fluid-phase mass and momentum
equations of dense particle flow must be further derived by a statistical averaging procedure similar to
that used in deriving turbulence transport equations from the Navier-Stokes equations. The general forms
of the fluid mass and momentum equations for dense particle flow (Batchelor 1988, Gidaspow 1994) are
280
∂θ f ρ f
+ ∇ ⋅ (θ f ρ f u f ) = 0 (4)
∂t
and
∂ (θ f ρ f u f )
+ ∇ ⋅ (θ f ρ f u f u f ) = −∇p − F + θ f ρ f g + ∇ ⋅ (θ f τ f ) (5)
∂t
where uf is the fluid velocity and θf is the fluid volume fraction, ρf is fluid density, p is fluid pressure,
τf is the fluid stress tensor, and g is the gravitational acceleration. F is the rate of momentum exchange
per unit volume between the fluid and particles phases and will be defined shortly.
The constitutive equation for the non-hydrostatic part of the stress, τf, depends only on the rate of
deformation Sij where Sij is given by
1  ∂ui ∂u j 
Sij =  +  (6)
2  ∂x j ∂xi 
For a Newtonian fluid, the stress is
2 ∂u
τij = 2µSij − µδ ij k
3 ∂uk
where μ is the coefficient of viscosity which depends only on the thermodynamic state of the fluid.
For compressible fluids, the pressure and density are related through an equation-of-state. In the
CPFD scheme, an ideal gas equation-of-state is used where the partial pressure of gas species k is
ρ k RT
pk = (7)
Mwk
and
p= åp k
k
(8)
where R is the universal gas constant, T is mixture temperature, ρk is the fluid species density and Mwk
is the molecular weight of gas species k.
The fluid-phase volume fraction is related to the particle distribution function by
281
ms
θ f = 1 − ∫∫ f dms dus (9)
ρs
The interphase momentum transfer function is
 1 
F= ∫∫ fms  Ds (u f − u s ) − ∇p  dms dus . (10)
 ρs 
When the second term in Equation (10) is combined with the pressure gradient term in the momentum
equation, and use is made of Equation (9), the more familiar form of pressure gradient multiplied by the
fluid volume fraction is obtained.
turbulence equations
Turbulence can produce seemingly chaotic motion with a wide range of fluctuation scales. The computa-
tional cost of resolving these fluctuations in CFD calculations is prohibitive, and fortunately we are not
concerned with predicting detailed fluctuating motions. We are concerned with average flow behavior,
and it is important to include the effect of the unresolved fluctuations on the average flow behavior.
For single phase flow there are two approaches to calculating averaged turbulent flow fields. Reynolds
averaging uses an ensemble averaging or a filter size that is large compared to the scales of fluctua-
tions. Subgrid-scale (SGS) models use filters with as small a size as possible, typically comparable to
the computational grid, and model only the effects of subgrid-scale fluctuations.
Upon averaging, both the Reynolds averaged and the subgrid-scale averaged equations contain cor-
relations of products of fluctuating terms which must be modeled in terms of mean, or averaged, flow
quantities. Empirical closure models, which are based on physical reasoning, dimensional reasoning,
and examination of transport equations for the correlations themselves, are used to close the equations.
For example, the Reynolds stress is an important second-order correlation from the incompressible
Reynolds averaged Navier Stokes equations, and is usually modeled by relating the stress to the mean
rate of deformation as in a Newtonian fluid:
2
−ρui' u'j = 2µ T S ij − ρδ k (11)
3 ij
where k is the turbulent kinetic energy
1 '
()
2
k= u (12)
2 i
and m T is a so-called eddy viscosity. As a result, the averaged Navier-Stokes equations with closure
models usually closely resemble the Navier-Stokes equations for laminar, or non-turbulent, flow.
For fluid mixtures with light solid loading, it is reasonable to use traditional turbulence models such
as the k-ε model. However as the solid loading exceeds volume fractions greater than a few percent, the
282
solids participate in generating and dissipating turbulence. Motion of solids can produce large eddies,
and flow over solids can generate small structures and dissipation. Turbulent fluid motions give rise to
particle dispersion, and work done in dispersing the solids further reduces the fluid turbulence kinetic
energy. These effects cast into doubt the utility of traditional turbulence transport models for modeling
dense particle flows.
In the CPFD scheme a large eddy simulation (LES) model is used. We do this, even though there
are some conceptual difficulties with using LES for particle-fluid flow (Edwards, 1999), because at
least some of the physics of turbulence transport can be included with LES modeling and because the
amount of suspect modeling involved is much less than is required for Reynolds-averaged models.
Large eddies are directly calculated and only the effects of the subgrid-scale turbulence are modeled.
The incompressible averaged LES momentum equation is very similar to the Reynolds averaged Navier
Stokes equation. The subgrid-scale model proposed by Smagorinsky (1963) is used. This model is an
eddy viscosity model which is based on the notion that the effects of the subgrid-scale Reynolds stress
increases transport and dissipation of mean-flow kinetic energy. The subgrid-scale eddy viscosity is
( )
1/ 2
µ T = Cs2∆2 S ij S ij (13)
where Cs is a parameter which may depend on flow conditions and has a typical value of 0.2. The filter
length scale, Δ, is the local computational grid size.
The CPFD model for wall shear stress is somewhat novel. There are two difficulties in applying the
Smagorinsky model to cells next to walls. First, the Smagorinsky model assumes that Δ is an interme-
diate scale in the inertial range of a turbulence cascade. This cannot be true near the wall because the
largest turbulent length scale is limited by the distance from the wall, which is smaller than Δ, i.e. there
are no turbulent length scales as large as Δ. Second, it is difficult to numerically compute the velocity
gradient in cells near the wall.
To overcome these difficulties, a turbulent viscosity consistent with the k-ε model is used by CPFD.
This is consistent with the fact that in k-ε, the transport of all turbulence length scales are modeled, and
as observed above, all scales are less than Δ. In the k-ε model, the turbulent eddy viscosity is (Amsden,
et al 1989)
k2 (14)
µT = Cµ ρ
ε
and in the law-of-the-wall region (Launder and Spalding, 1974),
2
( )
k = Cµ−1/ 2 u* (15)
ε=
1 u ( )
*
(16)
κ y
283
where u* is the shear speed calculated from a wall function, y is the distance from the wall, κ is the Kar-
mann constant (κ=0.43), and Cμ=0.09. Substituting the expression for k and ε into the equation for μT gives
µ T = κρu* y (17)
This is the formula used in conjunction with the subgrid-scale model in cells next to walls. Strictly
speaking, this formula only applies in the law-of-the-wall region. There is turbulence in the laminar
sublayer, but in the laminar sublayer, it is reasonable to use μT=0.
Following Amsden, et al (1989), the normalized shear speed is approximated by 1/7 power-law which
provides an explicit expression for the shear speed
−1
  7 /8  
 1   yρ f u   
u = u  ln  B1 
*  
  + B2  (18)
 κ   µ   
   
where B1=0.15 and B2=5.5.

In calculating the near-wall eddy viscosity, the y length scale is the cell half-width in the direction
normal to the solid wall and the effective tangent velocity u is calculated by averaging fluid velocities
at the cell faces. The wall shear stress is
2
( )
τ wall = u* ρ . (19)
In the Smagorinsky model, the large eddies are directly calculated, and the sub-grid scale turbulence
effects and wall turbulence are modeled. It needs to be kept in mind that the Smagorinsky turbulence model
was developed for single-phase flow, and the application of it to fluid-solid flow should be viewed with
healthy skepticism. In view of the uncertainties in multiphase LES modeling, we feel it is unwarranted
to consider more advanced models than the Smagorinsky model, such as dynamic sub-grid scale models
(Haworth and Jansen, 1996; Pomraning and Rutland, 2001), for calculating dense particulate flows.
Moments of the Particle Phase equations
To motivate definitions of particle properties mapped to and from the Eulerian grid and to provide a
comparison between the MP-PIC equations and the usual continuum-particle equations, the particle
mass and momentum equations are derived from the particle phase equation. The averaged conserva-
tion of mass and momentum equations can be derived by taking the mass and momentum moments of
Equation (2). By multiplying Equation (2) by ms and msus and integrating over particle mass and veloc-
ity coordinates (Travis et al., 1976), the continuum-particle conservation equations are obtained. These
field equations contain terms which have been modeled in traditional particle-continuum equations. It
is important to emphasize that the particle-continuum equations of this section are not explicitly solved
by MP-PIC, but are a consequence of solving Equation (2).
284
Multiplying Equation (2) by ms and integrating over mass and velocity coordinates, results in the
following solids mass conservation equation:
∂
∫∫ fms du s dms + ∇ x ⋅∫∫ fms u s du s dms −∫∫ fm s du s dms = 0 (20)
∂t
The first term is the rate of change of solid mass per volume. Define the solid density by
ρs ≡ ∫∫ fms du s dms (21)
and define the solid velocity by
us ≡
∫∫ fms u s du s dms
(22)
ρs
By using these definitions, the solid conservation-of-mass equation becomes
∂ρ s
+ ∇x ⋅ ρs us = ∫∫ fm s du s dms . (23)
∂t
The solids mass per volume can be replaced by the product of the material density and the solids
volume fraction, where the solids volume fraction is
ms
θs = ∫∫ f
ρs
dms dus (24)
Particles displace the fluid, and the sum of volume fractions of fluid and particle phases must equal
unity, θf+θs=1. Using Equation (24), the solid conservation-of-mass equation is
∂ (θ s ρ s )
∂t
(
+ ∇ x ⋅ θ sρs us = ) ∫∫ fm s du s dms (25)
The term on the right side could result from chemical consumption of solids or other mechanisms
for consumption/generation of solids. Without the rate of mass change on the right side, the conserva-
tion of solid mass is
∂ (θ s ρ s )
∂t
(
+ ∇ x ⋅ θ sρs us = 0) (26)
which is the fluid conservation-of-mass equation of continuum-particle methods.
285
By multiplying Equation (2) by msus and integrating over mass and velocity coordinates, the follow-
ing solids conservation-of-momentum equation can be derived:
( )+∇⋅ ρ u u +∇⋅
∂ ρs us
∂t
( ) ∫∫ s s s f ms u's u's dms du s = ∫ f ms u s dms du s + ∫∫ f ms u s m s dms du s (27)
where the solids velocity is represented by the average velocity u s and a fluctuation component u's is
defined by,
u s = u s + u's (28)
After substituting the acceleration terms from Equation (3) into Equation(27), the solids conservation
of momentum equation becomes
( )+∇⋅ ρ u u +∇⋅
∂ ρs us
∂t
( )  ∫∫
s s s f ms u's u's dms du s  =

∫∫ f ms Ds (u f − u s ) dms du s − θ s∇p + ρ s g − ∇τ s + òò f ms u s m s dms du s (29)
After expressing the average solid density in terms of the product of material density and the solids
volume fraction, and with no solid particle mass change, the above solids conservation-of-momentum
equation becomes
(
∂ θ sρs us )+∇⋅ θ ρ u u +∇⋅
∂t
( )  ∫∫
s s s s f ms u's u's dms du s  =

∫∫ f ms Ds (u f − u s ) dms du s − θ s∇p + θ s ρ s g − ∇τ s (30)
This equation for the conservation of particle momentum resembles the fluid Euler momentum equa-
tion, but has some additional terms. The equation contains a solid collision term (last term on the right-
hand side) and the divergence of the particle kinematic stress (Harris & Crighton, 1994) (last term on the
left-hand side), which arises due to particle fluctuations about the local mean particle velocity. The trace
of this particle kinematic stress is proportional to the so-called granular temperature. In addition, there
is an integral giving the total momentum transfer per unit volume to the particles, from the fluid, due to
aerodynamic drag (first term on the right-hand side). The particle momentum equations for the particle
continuum model must introduce closure-models for the integrals in Equation (30) that are approximate.
For example, the momentum transfer from the fluid to solids depends on the discrete particle velocity,
size and shape, and modeling the integral of fluid-solid drag requires a closure model using average
properties, and disregarding fluctuations in particle properties. Because the MP-PIC method solves for
the full distribution function, the accuracy of the approximations to the integrals are limited only by the
number of computational particles used in approximating the particle distribution function itself.
286
nuMerical sOlutiOn Of tHe MP-Pic equatiOns
The solution method for the equation of motion of solids is given in this section. Physical particles are
grouped into computational particles p (clouds) each containing np particles with identical mass ms,p,
velocity, us,p, and position, xp. The Liouville Equation (2) is the mathematical expression of conservation
of particle numbers in volumes moving along dynamic trajectories in particle phase space. Thus the num-
ber of particles np associated with a computational particle is constant in time. Because we are ignoring
particle mass change in this review, a particle’s mass, ms,p, is also constant. In the MP-PIC method, the
solid interactions (principally collisions) are not directly calculated. The direct coupling of the solids
collisions requires an implicit solution of particle momentum equation which is too computationally
expensive when applied to industrial scale problems. The time-rate-of-change of momentum of solids
is described by the ordinary differential equation Equation(3). While the momentum transfer between
solids is not directly coupled, the acceleration on a particle includes the average effects of neighbor
particles in the particle normal stress. The average effect on a particle from surrounding particles is an
approximation of collective effects. The solids velocity is updated using the following finite difference
approximation of Equation (3):
u ns,+p1 − u ns, p 1 1
= Dsn, p (u
 f , p − u ns,+p1 ) − ∇p f , p + g −  s, p
∇τ (31)
∆t ρs θ s, p ρ s
where u f, p is the interpolated fluid velocity at the particle location, Ñp f, p is the interpolated fluid pressure
gradient at the particle location, Ñτ is the interpolated solid normal stress gradient at the particle loca-
tion, Ds,n p is the drag function for particle p, and tildes over a variable denote an intermediate value of
that variable that closely approximates the advanced-time value of that variable. The numerical particle
positions are updated by
x n+
p
1
= x np + Dtu n+
s, p
1
(32)
The MP-PIC method is a particle-in-cell (PIC) method because it calculates particle properties (e.g.
θs) on the Eulerian grid and then uses the grid properties in calculating the discrete particle motion. The
manner in which particle properties are calculated on the grid is now described.
Particle/grid interpolation
Solids properties are mapped to and from the grid in the MP-PIC method. For a particle located at xs,
where xs=(xs,ys,zs), the x-directional component of the interpolation operator to grid cell ξ, is an even
function, independent of the y and z coordinates, and has the properties
0 xi −1 ≥ xs , xs ≥ xi +1
Sξx ( xs ) = (33)
1 xs = xi
287
where i is the x-index for cell ξ and i+1 is the x-index for neighbor to cell ξ, and
N
nodes
∑ Sξx ( xs ) = 1 (34)
ξ
for any value of xs. The y and z interpolation operators have a similar form. The three-dimensional in-
terpolation operator is formed from the product of the directional operators in the x, y and z directions:
Sξ = Sξx ( xs ) Sξy ( y s ) Sξz ( zs ) . (35)
N
nodes
It can be shown that ∑ Sξ (x s ) = 1 where the node ξ ranges over all Eulerian cells. From experience,
ξ
linear interpolation operators are quick and robust, however other interpolation operators could be used.
For the linear interpolation operator
xi+1 − xs
Sξx = , xi < xs < xi+1
xi+1 − xi
where i is a cell center for cell ξ. In the CPFD numerical scheme, a staggered grid is used where vector
quantities are calculated at cell faces and scalar quantities are calculated at cell centers. This requires
four sets of interpolation operators. A collocated grid, where vector and scalar quantities are at a cell
center, could also require four interpolation operators when using a first order approximation for scalar
gradients. In three-dimensions, properties from eight cells are interpolated to a particle location, and
particles in 27 cells contribute to a particle property mapped to an Eulerian grid cell. As an example,
the fluid velocity interpolated to the particle location xs, which is used in the particle velocity update
Equation (31), is
8
u fp = ∑ u f ξ Sξ (x s ) (36)
ξ=1
Figure 1 shows the interpolation from a two dimensional grid to a particle location, and the fluid
x-velocity at solid location xs is
4
u fp = ∑u fξ
Sξ (x s ) = u f1 S1 + u f2 S2 + u f3 S3 + u f4 S4 (37)
ξ=1
solids volume fraction
To interpolate from the particles to the grid requires that we average particle properties over grid volumes.
Thus, for example, the particle volume fraction at grid point ζ is defined by
288
Figure 1. Two-dimensional interpolation of grid property to a particle
θ s,ζ = ∫ θ S ( x ) dx / V
s ζ ζ
(38)
where θs is defined in Equation (24), Sζ is defined in Equation (35), andVζ = ∫ S (x) dx . The discrete
ζ
probability function is
f d ( x, u ,m,t ) = ∑ n δ (x − x ) δ (u
p
p p s
− u s, p ) δ (ms − ms, p ) (39)
where p ranges over all solid computational particles. Substituting Equation (39) into Equation (24), using
the result to replace θs in Equation (38), and noting that the integration and summation in the resulting
equation can be switched, gives for the solids volume fraction associated with grid point ζ
1 ms
θ s,ζ =
Vζ
∑ n ∫∫∫ δ (x − x ) δ (u
p
p p s
− u s, p ) δ (ms − ms, p ) S (x ) dms du s dx
ρs ζ
(40)
After evaluating the integrals in Equation (40), the computed solids volume fraction is seen to be
N
1 p ms, p
θ s,ζ = ∑ n p Sζ ( x p ) (41)
Vζ p=1 ρs
where the summation is over all numerical particles. In the staggered Eulerian grid used by CPFD
method, the solids volume is mapped to both the cell and face control volumes to get cell and face solids
volume fractions.
It is important to point out that the volume fraction is not a primitive quantity. It depends on the spa-
tial region chosen. For a fixed group of solids, the volume fraction can be larger or smaller depending
on the location and size of the chosen volume. In the MP-PIC method it is assumed that the grid size
is large enough to contain “sufficient” number of solids. This restriction is to some degree relaxed us-
289
ing the smoothed interpolation operator. A small cell may have no particles and still be near close pack
because the volume of particles in neighbor cells are interpolated to the small cell. This is reasonable
considering that particles and numerical-particles have a finite size, and while the center of mass of a
particle or numerical-particle may not lie in the small cell, its size range can extend into the small cell.
It is a good practice to have the Eulerian grid cells large enough to contain sufficient particles. A “suf-
ficient” number of particles is best determined by a particle-number resolution study similar to a grid
resolution study. For engineering analysis, the ideal solution is the calculation with the least number of
numerical-particles and least number of spatial grid-cells which gives an accurate-enough solution of
the problem at hand.
aerodynamic drag force
This section discusses the aerodynamic drag on solids, and the following section discusses the aerody-
namic drag on the fluid phase. Conservation of momentum requires the aerodynamic drag on solids to
be equal to the negative of the drag on the fluid phase. Stokes provided an analytic solution of a sphere
immersed in a fluid stream. The solution gave the drag on the sphere which consists of two-thirds viscous
and one-third pressure force. Stokes drag is valid for an isolated solid at low Reynolds number (ReD < 1).
For higher Reynolds number and a concentration of solids it is common to multiply Stokes drag by an
empirical coefficient from measured data, or empirically fit the measured drag data with the solid-fluid
flow parameters. If the particle size distribution in the experiments used to generate the empirical drag
model has a wide range, it has limited usefulness in a Lagrangian calculation of particle dynamics. Each
particle has a unique size, and thus a unique drag, and a correlation based on one effective size from a
wide particle size distribution is inaccurate. Another consideration in using a drag model is the range of
applicability. Because drag correlations are curve-fits to data, the correlation may not be accurate when
extrapolated to conditions outside the range of measured data from which it was derived.
The drag coefficient in Equation (31) depends on the particle’s properties and local fluid properties.
The drag function Dp of in the MP-PIC Equation (3) is
3 rfp (u fp − u p ) , (42)
D p = Cd
8 rs rs
where Cd is the dimensionless drag coefficient. Because the MP-PIC method predicts the motion of
virtually unlimited distributions in the sizes and types of solid material, the drag coefficient is unique to
each solid, and the MP-PIC naturally predicts the dynamics of polydisperse solids. Many drag correla-
tions use an empirical hindrance function multiplied by the Stokes drag equation, and one of these type
correlations used in CPFD is the the Wen-Yu (1966) drag correlation:
Re < 1000 24
Cd =
Re
( )
1 + 0.15Re0.687 θ−f 2.65
(43)
Re ≥ 1000 Cd = 0.44θ−f 2.65
The Reynolds number is
290
2ρf (u fp − u s ) rs
Re =
µf , (44)
where μf is the gas viscosity at the particle location and the effective particle radius of a solid is
1/ 3
 3V 
rs =  s  (45)
 4π 
ms
whereVs = is the solid volume. In an example we present later of batch settling, calculated results
rs
from three different drag correlations are shown, and all three correlations gave a good prediction of the
solids settling.
fluid Pressure gradient
The fluid pressure gradient is calculated on the grid and interpolated to a particle’s location using a
formula analogous to Equation (36). If the fluid velocity is zero and the solids volume fraction is low,
the pressure gradient gives the traditional hydrostatic buoyancy force.
Particle normal stress
The particle normal stress is an approximation of collective effects of neighbor particles on a particle.
The MP-PIC method makes use of gradients which are readily calculated on the Eulerian grid and ap-
plies the gradients to discrete particles. Gradients in this particle pressure result in particle accelerations
that prevent particle volume fractions from exceeding their close-pack limit. The particle pressure is a
monotonic increasing function of solids volume fraction and vanishes as the solids volume fraction goes
to zero. The solids normal stress taken from Harris and Crighton (1994) is
Ps θ βs
τs = (46)
θ CP − θ s
where Ps is a constant with units of pressure, and θCP is the solids close pack volume fraction. The close-
pack limit is somewhat arbitrary and depends on the size, shape and ordering of the particles. A vessel
filled with solids can be tapped to get a tighter packing of solids. Therefore allowing the particle volume
fraction to slightly exceed an average close-pack value is physically possible considering that shifting
or rearranging of granular materials may occur. The form of the particle normal stress used in CPFD is
Ps θ βs
τs = (47)
max (θ CP − θ s , 0) + ε (1 − θ s )
291
where the denominator is modified to remove the singularity when the solids volume fraction is equal
to θCP. The ε is a small number on the order of 10-6. A typical value for the Ps coefficient used in a CPFD
analysis is Ps =10 Pa, and the power-coefficient ranges from β=1 to 3 (Harris and Crighton, 1994).
The particle velocity given by Equation (31) can be solved directly at each time step using fluid
properties updated from the current time step (new-time fluid velocities and pressure fields) and old time
properties for the normal stress tensor. However, because the particle normal stress is highly nonlinear,
simply using old time values for the particle normal stress and applying the gradient to particles in a cell
does not work when the solid loading is near close pack. A possible solution is to reduce the calculation
time step or use sub-time intervals for the particle momentum equation solution and volume fraction.
Using the small time step is difficult to control and too computationally expensive. To address the stiff
ordinary differential equations for solids acceleration, Andrews and O’Rourke (1996) presented an
Eulerian description of the solids volume from mapping the discrete particles to the grid. The implicit
solution of volume fraction scheme was employed by Snider et. al. (1998) to accurately calculate settling
of polydisperse solids from dilute to close pack.
The MP-PIC scheme for solving an intermediate solids volume fraction is given. The approach is to
develop a description of the solids volume fraction which includes the particle normal stress. This set
of equations is implicitly solved for an intermediate-step solid volume fraction. The Lagrangian particle
momentum equations (solids) are then solved using the Eulerian particle normal stress gradient based
on the intermediate solids volume fraction. The Lagrangian particles are then mapped to the Eulerian
grid to calculate the new-time solids volume fraction.
From Equation (41) the change in solids volume fraction for grid-cell ξ is
N
1 p ms, p 
θ n+1
sξ
−θ =
Vξ
n
sξ ∑n
p=1
p
ρ s  ξ p
( ) ( )
S x n+1 − Sξ x np  .

(48)
We approximate the new-time interpolation operator as
∂S ξ
(
S x n+
p
1
)
= S (x
p)+
∂x
(x n+1
p
−x
p ) (49)
where
x p = x np + u npDt . (50)
Substituting Equation (32), (49) and (50) into Equation (48) gives
θ n+ 1
− θ nsξ 1
N
p ms, p  Sξ (x p ) − Sξ x p
n
( )
∂S 

sξ
∆t
=
Vξ
∑
p=1
np
Ás 

∆t ∂x p
(
+ ξ u np+1 − u np )

(51)
 
292
To complete the Eulerian volume fraction equation, we solve for the new time velocity from Equa-
tion (31), using the old time fluid velocity and pressure gradient interpolated to the particle location,
and substitute the resulting expression into Equation (51), giving
θ n+
sξ
1
− θ nsξ
=
∆t
1
N
p
( )
ms, p  Sξ (x p ) − Sξ x p
n
∆t ∂Sξ  1 1 
n+1 
Vξ
∑ n p
ρ

 ∆ t
+
1 + ∆ tD ∂ x
 D (
u n
 p fp − u n
p
− )
ρ
∇ p n
fp
+ g −
θ ρ
∇ τ sp 

 (52)
p=1 s  p p  s s, p s 

The particle normal stress as a function of volume fraction is
 
∂τ  βθ βs −1 θ βs  ∂θ s
= Ps  +  (53)
∂xi  θ CP − θ s 2  ∂xi
 ( θ CP
− θ s ) 
Substituting Equation (53) into Equation (52), approximating the term in brackets in Equation (53)
∂θ s
using old-time θs values, and numerically approximating using new-time θs values, results in a
∂xi
linear set of equations for the new-time estimated solid volume fractions which involves cells and
neighbor cells. Because of interpolation of the particle normal stress from eight cells to the particle loca-
tion, the numeric stencil may involve 27 cells. From the calculated estimated solids volume fraction, a
particle normal stress is calculated which is applied to each solid. The new-time particle velocity and
particle locations are calculated, and the new-time solids volume fraction, calculated from mapping
solids volume fraction to the grid, replaces the estimated solids volume fraction.
The MP-PIC implicit solid volume fraction scheme works well for a wide range of flow conditions
in batch settling (Snider et. al. 1998). Batch settling begins with an initially homogeneous mixture of
solids. When allowed to settle, solids of different size and density settle out of suspension at different
rates, and if the solids are grouped within a narrow particle-size distributions, distinct regions form
where each region above the previous region contains one less solid species and the top layer is clarified
liquid. Figure 2 shows a CPFD calculation of bimodal settling compared to the bimodal suspension of
a glass beads experiment given by Davis, et al. (1982). The vessel is vertical and the fluid density and
viscosity are 992 kg/m3 and 0.0667 kg/(m-s), respectively. The small glass bead density is 2440 kg/
m3, and beads range in diameter from 125 to 150 μm. The larger bead density is 2990 kg/m3 and beads
range in diameter from 177 to 219 μm. The MP-PIC calculation uses a uniform random distribution
within the reported experimental size ranges. Initial volume fractions for the small and large beads are
3% and 1%, respectively. The implicit calculation of an intermediate solids volume fraction provided
a smooth solution from a dilute solids mixture to settling to close pack. A characteristic of the MP-PIC
method is that a range of solids sizes and densities can be used as easily as one size or one density. In
Figure 2, the settling of two-groups of solids of different sizes and densities is accurately calculated.
While the particle size variation within the two size-ranges is tight, the interface is not perfectly crisp
293
Figure 2. CPFD calculated settling compared to experiment by Davis et al. (1982). Drag hindrance
function times Stokes drag: Gidaspow & Ettehadieh (1983), Cd=θf-2.65Richardson & Zaki (1954), Cd=θf-4,
5θ 
and Barnea & Mizrahi (1973), Cd = θ−f 2 1 + θ1p/ 3 exp  s
( ) 
 3θ f 
because there is a small range in particle size within each group. Two size groups (with a small Gaussian
distribution in size) were used in this study, but a virtually unlimited range of sizes and densities could
as easily have been used.
There are draw-backs to the implicit calculation of the particle volume fraction. First, the implicit
estimate of the new-time volume fraction depends on the gradient of the interpolation operator, and the
extrapolation of the linear interpolation operator on the grid can predict a non-physical, negative volume
fraction. In the implicit calculation of volume fraction, the total volume fraction is conserved even if
the local volume fraction is negative. Therefore a false decrease in volume fraction at one node has an
associated false increase in volume fraction at its neighbor node. Limiting the minimum and maximum
value of the Eulerian calculated volume fraction does not pose a problem in the solution scheme. At the
end of the time step, the estimated volume fraction is discarded and the final volume fraction is calcu-
lated from mapping particles to the grid. A second problem with the implicit scheme is that the Eulerian
equation set is highly non linear and difficult to solve.
The particle normal stress gradients result in particle accelerations that prevent particle volume frac-
tions from exceeding their close-pack limit, and therefore the particle normal stress is important near
close pack and has little effect in dilute flow. As an example of the magnitude of the particle normal
stress, consider dense particle flow of sand at 92% close pack. The sand density (ρp) is 2500 kg/m3; the
close pack particle volume fraction (θCP) is 0.6; the maximum gradient of the particle volume fraction
is 0.55 between cell centers; Ps = 10 Pa; and β = 3 in Equation (46). For this dense concentration of
solids at 92% compaction, the particle acceleration from the normal stress is 0.048 m/s2 which using
gravity as a reference, is small. As the particle volume fraction approaches close pack, the particle stress
becomes extremely large.
294
Figure 3. U-tube experiment
If the solid volume fraction is less than 95% close pack, solid collisions, modeled by Equation (46),
are relatively small, and collision forces are directly applied to solids. Near close pack, the solid normal
stress from of Equation (46) implicitly contains the physics of a high collision frequency associated
with high solids concentration. However, the particle normal stress is independent of solids velocities.
Snider (2001) applied the solid normal stress model with a dependency on each solid’s velocity vector.
The idea is that the collision frequency depends on the solids concentration and on the variation in a
solids velocity relative to the particle-mixture mean velocity. Further the collision frequency near close
pack will be much faster than the hydrodynamic calculation time interval and multiple collisions occur
over the calculation time interval. A small movement in solids location from collision near close pack
(small adjustment in solids volume fraction) gives orders of magnitude change in the particle collision
force on a time scale of a few collisions. The high collision force at close pack does not apply over a
hydrodynamic calculation time step. The model applies the particle normal stress to a solid up to the
point where the solid reaches the particle-mean velocity. The model further assumes that near close pack,
the multiple collisions occurring over the hydrodynamic calculation interval, will move solids, but the
multiple collisions limit motion to near the particle-mixture mean velocity. Therefore particle collisions
near close pack will not impart a solids velocity greater than the mean particle-mixture velocity.
The U-tube example, we present next, shows a number of features of the MP-PIC method including
application of the particle normal stress. The experiment, shown in Figure 3, has sand sitting on a slider
295
Table 1. CPFD calculation parameters for U-tube experiment
Air
Isothermal (300 K)
Compressible flow
Solids
Badger 5574 sand
Density 2760 kg/m3
Solids close pack 0.60
Pressure outlet at top of U-tube legs 101 kPa
119168 numerical particles
28,014 cells
Figure 4. Badger sand 5574 size distribution used in the U-tube expert
plate at the top of one arm of the U-tube. The slider plate is quickly pulled allowing solids to fall driven
by gravity. The experiment provides a validation of the MP-PIC method in calculating solids moving to
close pack and the normal stress model. The experiment also shows a common characteristics of granu-
lar flow where the solids settle to the bottom of only one leg in the U-tube container. The calculation
parameters for the calculation which match the experiment are given in Table 1, and the particle size
distribution is shown in Figure 4.
Figure 5 shows the CPFD calculated sand flow at the bottom of the U-tube compared to experiment
data (images). The experiment shows a common characteristics of granular flow where the solids settle
to the bottom of one leg in the U-tube which is unlike the flow of a fluid where both arms of the U-tube
would fill to a common hydrostatic height. Below the experiment images is the CPFD calculated sand
flow where the solids are colored by solids volume fraction mapped to particle locations. Two sets of
calculated solids are shown. One in gray-scale which matches the gray scale in the experiment, and one
in color where blue is a dilute concentration of solids and red is close pack. Like the experiment, the
MP-PIC numerical method predicts the fundamental characteristics of granular flow where only one leg
fills and the bottom of the U-tube partially fills. The MP-PIC method calculates well the deceleration
of solids to the static close pack column of solids at the bottom of the U-tube using the particle normal
stress gradient given by Equation (46). At rest, the gas pressure must be the atmospheric pressure with
296
Figure 5. Measured sand flow at bottom of U-tube compared to CPFD calculated sand flow
Figure 6. CPFD calculated pressure at the bottom of the U-tube
a small hydrostatic pressure difference of air from top to bottom of the U-tube. Figure 6 shows the CPFD
calculated transient gas pressure at the bottom of the U-tube. Solids moving down the column pressur-
izes the gas and then, as solids go to rest, the gas pressure returns to atmospheric pressure.
Note that one leg of the U-tube is lower than the other and the pressure at the lower leg is higher
from the hydrostatic head or air. This pressure difference is small and is ignored in application of the
boundary conditions.
nuMerical sOlutiOn Of tHe fluid PHase
The CPFD method solves the compressible and incompressible Navier-Stokes fluid equations. While
the CPFD method includes solution of the energy equation, it is not discussed here. There are numer-
ous articles and books on solving the Navier Stokes momentum equation (see Ferziger & Peric, 1996)
and only an overview of the solution of the fluid equations is given here. The focus is on the MP-PIC
coupling of the solids phase in the fluid equations. Solids strongly influence the fluid phase and vice
versa and a tight coupling is required to get a smooth robust solution.
297
Figure 7. Original stl CAD file on left. CPFD gridded geometry in center. control-volume on right
finite volume grid
In the discretization of the momentum equation, space is represented by a finite number of control
volumes. The finite volume approximation uses a staggered mesh where the scalar quantities are cell
centered and vector quantities are face centered. The grid is formed from cut bricks with flow-surfaces in
the orthogonal directions. A brick at a solid boundary is cut to form wall patches which follow the three
dimensional geometry. A cut-brick can have up to 24 wall patches. Figure 7 shows a grid for a cyclone.
One cell from the cyclone is shown, where the control-volume has four flow surfaces and two wall
patches. The choice of a structured orthogonal cut brick grid was driven by having a fast interpolation
of solids to and from the Eulerian grid. The orthogonal grid reduces numerical distortion associated with
skewed grids but suffers from small cells near boundaries giving large aspect ratios between neighbor
cells and a local grid refinement will extend the length of the computation domain.
fluid Momentum equation
The solution of the fluid momentum equation is under the umbrella of the CPFD solution method, and
the MP-PIC description of solids is in the interphase momentum transfer. Equation (5) gives the differ-
ential form of the momentum equation. Integrating Equation (5) over a control-volume and applying the
Reynolds transport theorem gives the control-volume form of the equation (Thompson, 1972, Ferziger
298
Figure 8. Nomenclature for cell and face indexes. A cell index is ζ and a face index is ξ. When an i. j or
k directional index is used it requires the qualifier of cell or face value
and Peric, 1996). A finite volume form of equations are approximated numerically. The momentum
equation for fluid velocity is
∂
∂t ∫ρ f
θ f u f dV + ∫ ρ f θ f u f u f ⋅ ndA = ∫θ f
τ ⋅ ndA − ∫ pndA + ∫ ρ f θ f gdV − ∫ FdV (54)
V A A A V V
and in the i-direction (ii unit normal direction) is
∂
∂t ∫ρ f
θ f ui dV + ∫ ρ f θ f ui u ⋅ ndA = i i ⋅ ∫ θ f τ ⋅ ndA − ∫ pi i dA + ∫ ρ f θ f gi dV − ∫ Fi dV
V A A A V V
where the A is the control-volume surface area and V is the control-volume volume. In the following
discussion, the “f ” subscript, denoting fluid phase, is not included with the velocity term except when
necessary for clarity.
Figure 8 shows a staggered grid with scalars at control-volume centers and vectors at control-volume
surfaces. A cell index is denoted by ζ and a face index is denoted by ξ. If a directional index i, j or k is
used, it requires the qualifier of cell or face value. With a cut-brick grid, a control volume center is at
the control volume mass center. Eulerian control-volume properties that are interpolated to a face are
linearly weighted by the distance from cell centers to the face center. A face volume is the sum of half
the volume of the neighbor control-volumes.
The time derivative for the i-direction is approximated by
n+1 n
d (ρ f
θ f ui ) − (ρ f θ f ui )
∫ρ (55)
ξ ξ
f
θ f ui dV ≈ Vξ
dt V
∆t
299
Figure 9. One dimensional example illustrating geometry for advection operator
where ξ is the face index and Vξ is its volume. The flux through a control-volume is the sum of the flow
through the faces
N
surfaces
∫ gdA = ∑ ∫ g dA
k
k k
(56)
A A
k
where k is a face index, and there can be up to 6 numerical flow boundary surfaces (Nsurfaces). The g is
the convective (ρφu·n) or diffusive vector ( Γ∇φ ⋅ n ) of φ in the direction normal to the face, where Γ
is the diffusivity coefficient for φ. The convective transport of φ is approximated by
∫ρ f
θ f φu ⋅ ndA ≈ ρ f θ f Auφ (57)
A
where the velocity is in the direction of the face normal.

The partial donor cell numerical approximation (Amsden, et al., 1989) is used for the advection.
This convection scheme is a weighted average of central difference and upwind convection. A limiter
is applied to automatically weight the central difference and upwind quantities. Figure 9 illustrates a
one-dimensional grid with the face control-volume surfaces given by e and w and the cell center is at ζ.
The convective transport of φ is approximated by
Gw = uw θ w Awρ w
(58)
Ge = ue θ e Ae ρ e
The donor cell quantity is
uw + ue
fw if >0
fd = 2 (59)
uw + ue
fe if <0
2
The acceptor cell quantity is
300
uw + ue
fe if >0
fa = 2 (60)
uw + ue
fw if <0
2
The partial donor cell convected quantity is
1 1
φζ = φd (1 + α + βC ) + φa (1 − α − βC ) (61)
2 2
where
∆t (uw Aw θ w + ue Ae θ e )
C= (62)
θ wVw + θ eVe
If α=0 and β=0, then the convection is center differencing which is unconditionally unstable without
sufficient amount of physical diffusion. (The amount of physical diffusion required is given in a stability
condition on the grid-based Peclet number (Ferziger & Peric, 1996).) If α=1 and β=0, then the convection
is upwind which tends to be too diffusive. The base values we use for the constants are α=0.1 and β=1.
This scheme is less diffusive than pure upwind differencing but is not monotone. For the momentum
equation Equation (57), φ=ui. Using new-time values for velocities, the advection terms give a nine-point
stencil for the momentum equation velocities.
The i-direction pressure force, is transformed to a volume integral through the divergence theorem
and is approximated as
i i ⋅ ∫ pndA = i i ⋅ ∫ ∇ pdV ≈ VF,i

(p i+1
− pi )
(63)
A V
∆xi
where the pressures are at cell centers i and i+1, VF,i is the face volume, and Δxi is the distance between
control-volume centers. The body force in the i-direction is approximated as
∫ρ f
θ f gi dV ≈ gi (V θ f ρ f ) (64)
ξ
V
The density is interpolated from control volume centers to the face.

The viscous term in the i-direction is
   
i i ⋅∫ θ f τ ⋅ n dA = i i ⋅ ∫  θ µ  ∂ui + ∂u j  − 2 θ µδ ∂ui  ⋅ n dA (65)
 f   
  ∂x j ∂xi  3
f ij
∂u j  j
A A 
301
where the viscosity is the sum of the molecular viscosity and the turbulent eddy viscosity.
The spatial derivatives of velocity are approximated as
∂u1 u − u1,i
∫θ f
µ dA = (θ f µA) 1,i+1 (66)
A
∂x1 Ci δxi
where the fluid viscosity and volume fraction are at the cell center, the area is interpolated from face
nodes i and i+1, and δxi is the distance between faces. The velocities are new-time values which adds
to the nine coefficients per face from advection.
Solids momentum in the MP-PIC method is implicitly coupled into the fluid phase through the drag.
The interphase momentum transfer from solids to fluid in the i-direction is
 1 
∫ Fi dV = ∑  D (u p
n+1
fp,i
− u n+
p,i
1
−) ρs
∇p n+
fp
1
⋅ i i ( p) p p
S x n m (67)
V  
The ufp is the fluid velocity at a particle location and the Ñp fp is the pressure gradient interpolated
to a solids location. Substituting the solids velocity from Equation (31) into Equation (67) gives
 
−g + 1 ∇τ ⋅ i  − 1 ∇p ⋅ i
∑ s fs,i s,i
D u n+1
− u n
+ Ds (
∆ t  i ρ θ

s i)
 ρ i
∫ Fi dV = s s s
S ( x s ) n s ms (68)
V
1 + Ds∆t
Equations (55), (57), (63), (64), (65), (66) and (68) describe the numerical approximations to the
fluid momentum equations. The momentum equations are a set of implicit linear algebraic equations
of coefficients multiplied times fluid velocity. The coefficients on velocity are from the time change of
momentum, convection, viscous diffusion, and interphase momentum transfer. The pseudo discretized
momentum equation for in the i-direction at face index ξ is
 n+1 
 (ρ f V θ f ) Ds Sξ (x s ) ns ms  Nghb Nghb
 (u )n+1 + (
n+1 n+1
i )nghb (


ξ
+ ∑ + C + Dξ i ξ ∑ Cu + ∑ Dui )
 ∆t 1 + Ds∆t ξ
 nghb nghb
nghb
 
n
Vξ (ρ f θ f ui ) Ds usn,i Sξ (x s ) ns ms
 Ds∆t 
= ξ
+∑ + gi ρ f ξ θ f ξVξ + ∑ Sξ (x s ) ns ms 
∆t 1 + Ds∆t  1 + Ds∆t 
 1 1  Ds∆t ∇τ s ⋅ i i
- ∇pξ ⋅ i i Vξ − ∑ Sξ (x s ) ns ms  - ∑ S ξ ( x s ) n s ms (69)
 1 + Ds∆t Ás  1 + Ds∆t ρ s θ s
where Cξ and Dξ are the convective and diffusion coefficients, respectively, for the uξ velocity. The
Nghb Nghb
n+1 n+1
∑ (Cu )
k
i nghb
and ∑ ( Du )
k
i nghb
are the convective and viscous terms, respectively, which are neighbor
302
velocities to face index ξ on a nine-index stencil. The algebraic discretized momentum equations are
solved using a sparse matrix solver.
conservation of Mass and the Pressure correction
The conservation of mass and pressure correction is within the CPFD calculation scheme. The solids
phase is coupled into the pressure equation to give a semi-implicit coupling of solids and a more robust
solution.
The control volume mass continuity equation is
∂
∂t ∫θ f
ρ f dV + ∫ θ f ρ f u f ⋅ ndA = 0
V A
The velocity from solution of the momentum equation does not necessarily satisfy mass continuity.
The approach to satisfying mass conservation is through a pressure-density-velocity correction method
which is applicable to an arbitrary Mach number (Ferziger and Peric 1996). The approach is to solve
a pressure equation derived from the continuity equation, momentum equation and equation-of-state.
The density and velocity which conserve mass are composed of an estimated and a correction value
ρ = ρ* + ρ ' (70)
u = u* + u' (71)
where * is the estimated value and ‘ are correction value. The estimated velocity and correction veloc-
ity is from the momentum equation where an estimated pressure from the previous time or iteration
requires a correction
p = p* + p' (72)
The estimated density is the old-time or last outer iteration density which is pressure dependent. For
a solution satisfying continuity, the correction terms are zero. The product of density and velocity is
ru = r* u* + r'u* + r* u' + r'u'
where the ρ’u’ goes to zero faster than other error terms and is neglected. The mass error from estimated
values is
∂
err =
∂t ∫θ f
ρ*f dV + ∫ θ f ρ*f u*f ⋅ ndA (73)
V A
and the correction conservation of mass equation is
303
Figure 10. Grid notation for east and west cells
∂
∂t ∫θ f
 ( ) (
ρ' f dV + ∫  θ f ρ' f u*f + θ f ρ*f u f ' ) ⋅ ndA + err = 0 (74)
V A
The density is related to the pressure through the equation-of-state, and the density correction from
a pressure correction is
 ∂ρ 
ρ ' ≈   p' = Kp' (75)
 ∂p T
For an ideal gas, p=ρRT, and K is
1
K= . (76)
RT
For discussion, the finite control volume has the cell center index ζ and is connected to an east neighbor
(E) through an east face (e) and similarly to a west cell as shown in Figure10. For consistency with the
previous section, the east face or subscript e corresponds to face index ξ. The numerical approximation
for the correction time term in the continuity Equation (74) for cell ζ is
∂ (
θ f ζ ρ' f ζ − 0 )V
∫θ
'
f
ρ f dV ≈ (77)
∂t V
∆t ζ
where the old-time density correction is zero. Using the equation of-state, the discretized Equation (77),
in terms of pressure, is
∂ θ f ζ Kζ
∂t ∫ θ f ρ'f dV ≈
∆t
Vζ pζ ' (78)
V
The corrected velocity is from the momentum equation. The correction momentum equation from
Equation (69) for the east face is
304
 ( f f )e
 ρ Vθ 
Ds Se ns ms  Nghb Nghb

 ∆t
+∑
1 + Ds∆t
+ Ce + De  ue' + ∑ Cu '
 nghb
( ) nghb
+ ∑ Du '
nghb
( )
nghb
 
 N 
s
1 1
= - ∇p'e ⋅ i e Ve − ∑ Se ns ms  (79)
 s 1 + Ds ∆t ρ s 
In solving the momentum equation, coefficients are combine to form a single coefficient times a
velocity for use in the algebraic equation solver. In doing this, the fluid density in Equation (79), which
has a correction value, is buried in the coefficient and will not be corrected. The velocity correction from
neighbor cells is neglected and the velocity correction in terms of pressure correction is
beVe∇p'e ⋅ i e b
u'e =
ae
(
or u'e = e p'E − pζ'
ae
) (80)
where
 ( f f )e
 ρ Vθ 
DSnm 
ae =  + ∑ s e s s + Ce + De  (81)
 ∆t 1 + Ds∆t 
 
 N 
1 s
1 1
be = −1 + ∑ Se ns ms  (82)
 V e s
1 + D s
∆ t ρ s 
The finite volume approximation for mass advection for the east face is
 b 
∫ (
 θ ρ* + ρ '
 f f f )(u
*
f ) ae
(
+ u' f  ⋅ ndA ≈ Ae θ nfe ρ'fe u nfe + ρ nfe e pζ' − pE' ) (83)
 
and the density correction is
1
ρ'fe = (
K p ' + K E pE'
2 ζ ζ
) (84)
The time and flux correction terms from Equations (78) and (83), respectively, are inserted into
the continuity equation Eq (74) to give the set of pressure equations which are solved using a sparse
matrix solver. The pressure is corrected from Equation (72), the velocity is corrected from the pressure
correction by Equation (70) and Equation (80) and the density is corrected from the pressure and the
equation-of-state.
305
cPfd exaMPles using tHe MP-Pic scHeMe
Two examples using CPFD method are presented which illustrate gas-solids flow using the MP-PIC
scheme. The examples are different applications of granular flow, and both examples have measured
data to validate the MP-PIC method. It must be noted that nothing was tuned in the calculations such as
particle sizes, drag correlations, turbulence model constants, etc. and both calculations compared well
with the experiments even though the problems are vastly different. The first example is the filling of a
sand-core used in metal casting in the automotive industry. Sand-core patterns with complex shapes are
notoriously difficult to fill with sand. The second example is a calculation of 3-ft diameter deep fluid-
ized bed experiment. The bed dynamics are vastly different based on the mix of the particles sizes. A
bed with 3% fines has persistent gas-streamers while a bed with 12% fines has a nice bubbling behavior.
This example shows the importance of solids size on the gas-solids flow in a fluidized bed and shows
the ability of the MP-PIC method to correctly predict the granular-flow physics.
sand casting
In the casting industry, the majority of casting begin with sand cores. The sand cores form a negative of
the pattern to be cast, and in an automotive block there can be 14 to 24 sand cores assembled to form
the casting mold package. Liquid metal poured into the core package fills the open regions. Once cooled
the sand is “shaken-out” leaving the final metal casting. For complex shapes, such as an automotive
cylinder head, the sand core is made by blowing sand into an enclosed metal tooling. Sand is blown
from a magazine through blow-tubes into to the core tool, and the blowing air escapes trough vents on
the tooling surface. Because of the nature of gas solids flow, it is difficult to fill geometrically complex
tools with a uniform bulk density of sand. If the sand core is not formed properly, the result is scrap,
Figure 11. An automotive cooler. (a) Side view showing sand magazine connected through blow-tubes
to the sand core. (b) Top view of sand core. (c) Bottom view of sand core
306
increased labor costs to meet demand, and possibly more equipment to offset inefficiencies. The casting
industry is replacing trial-and-error methods with science-based predictive methods to build core tool-
ing. The following CPFD analysis illustrates the complexity of blowing a sand core and the success of
MP-PIC method in accurately predicting the gas solid flow.
Figure 12 shows a sand core for a automotive-cooler. This is one core of a number of cores used to
make the cooler core-package. The left image shows the side view of the sand magazine connected to the
top of the core, and the center and right images are to top and bottom views of the core. The geometry
is complex and sand must flow down the blow-tubes, into the tooling and spread across the relatively
narrow tool-pattern. Air exits from the tooling through 86 vents. The CPFD calculation parameters are
given in Table 2. The solids size distribution is shown in Figure 11.
Figure 12. Sand size distribution for the automotive cooler
Table 2. CPFD calculation parameters for cooler sand core
Gas
Air
Isothermal
Incompressible
Air density 1.16 kg/m3
Solids
H33 quartz-sand
Density 2650 kg/m3
Bulk density 1400 kg/ m3
Mean sand radius of 130 μm
Magazine boundary (air and solids inlet flow)
Pressure of 463 kPa (gauge)
Air sand mixture enter through pressure boundary
Vent boundaries (air loss only)
86 vents on the model surface
Vent open area fractions ranged from 0.14 to 0.16
Vent sizes from 6 mm and 16 mm in diameter
Vents exhaust gas to atmosphere at 0 kPa (gauge)
1,450,000 numerical particles
Approximately 200,000 cells cut from 575,000 cell
307
Figure 13 shows the CPFD predicted filling of the cooler. The tooling vents, where air exits to atmo-
sphere, are shown at the right in Figure 13. In the case of the cooler for this example, the vents and
blow-tubes were initially located on the tooling based on the tool-designer’s past experience. The result
was a core which did not completely fill.
The accuracy of the CPFD calculation using the MP-PIC method is shown in Figure 14, where the
blown core is compared with the calculation. The calculation shows the solids volume fraction which
results from mapping solids volume to the grid. As seen from the figure, the comparison is excellent.
Images from the CPFD calculation shows a fill pattern which does not have an intuitive relation with
vents. The thin arm on the right fills quicker than the arm on the left although both arms have nearly the
same number of vents. The measured incomplete fill region (circled area in Figure 13) has no vents.
One might think the solution to get a well filled core is to add vents to the problem area to get a better
sand flow to that area. This traditional approach of adding vents at a problem area may fix the immedi-
Figure 13. Time sequence of sand filling of the cooler mold. Solids are colored by mapping particle
volume fraction to particle locations. Red is close-pack and blue is no solids. Figure at right shows
location of vents on the top side of the core where air exits to the atmosphere. Circled region shows
measured region of incomplete sand filling (Blaser and Yeomans 2006)
308
ate problem but often leads to a new fill-problem in another location. From studying the CPFD pre-
dicted gas-solids flow patterns in the cooler, the solution was to add vents so gas-solids flow sweeps
across the problem region which eliminated a flow-stagnation and the incomplete fill. The tooling was
modified based on the CPFD calculations and the incomplete filling was eliminated.
gas bypassing in deep fluidized beds
Gas bypassing can be an issue in deep fluidized beds. If gas streams through a bed, the bed is not well
mixed and the the gas that channels through the bed has reduced contact with the bed solids. Experiments
with deep beds containing Geldart Group a solids have shown that gas bypassing can occur when in the
bubbling bed mode (Karri, et al. 2004). Karri et al (2004) found that increasing the fines content of the
particles reduced the gas bypassing.
Two calculations were made of experiments with beds having FCC solids with 3% fines and the other
having 12% fines. Fines are solids with diameter less than 44 μm. The experiment with low fines content
exhibited streaming. These calculations illustrate the importance of particle size on a beds behavior and
the necessity of including the particle size distribution in predicting fluidized beds. The calculation pa-
rameters are given in Table 3, and the particle size distributions for the two cases are shown in Figure 16.
Figure 14. The blown core (left) and the CPFD predicted core (right). The incomplete fill region is circled
309
Table 3. CPFD calculation parameters for 0.91 m (3 ft) diameter bed with and without fines
Gas
Air
Isothermal (300 K)
Compressible flow
Solids
Heavy FCC catalyst
Density 1488 kg/m3
Mean sand diameter of 79~90 μm (see PSD)
Fines diameter < 44μm
Two calculations at 12% and 3% fines
Geometry
Column height 7.6 m (25 ft)
Vessel diameter 0.9 m (3 ft)
Static bed height 2.44 m (8 ft)
Air sparger
Sparger area 10% open to represent nozzles
Pressure 463 kPa (gauge)
Bed superficial velocity 0.61 m/s (2 ft/s)
Outlet to exterior cyclone
Solids and air exit through boundary
Pressure at outlet
Cyclone dipleg
FCC catalyst fed at bed PSD
Dipleg feed rate to maintain constant total bed mass
Approximately 1.5x106 numerical particles
Approximately 200,000 cells cut from 575,000 cell
Figure 15. Measured minimum bed height without gas bypassing versus the percent content of fines
(Karri, et al. 2004)
From measured data by Karri et al. (2004), shown in Figure 15, the 2.44 m (8 ft) static bed height
used in the CPFD calculations will have gas bypassing with a fines content less than 8.5%. Figure 17
shows the CPFD instantaneous calculated solids distribution for the calculations with 3% and 12% fines.
310
Figure 16. FCC catalyst particle size distribution
Figure 17. CPFD calculated solids distribution in the 0.9 m (3 ft) diameter bed. Solids are colored by
mapping the volume fraction to the solid locations.
The solids are colored by mapping the volume fraction to each solid. Dark red is solids at or near close
pack and blue is a dilute solids region. The bed on the right with 3% fines shows large deep red regions
where solids are near close pack and a stream of relatively dilute gas-solids moving completely up
through the center of the bed. This gas-solids streaming is typical of measured gas bypassing. The CPFD
calculated flow behavior agrees with measured data given in Figure 15 which shows that the 2.44 m (8
311
ft) bed with 12% fines will not have gas bypassing. The MP-PIC method indeed predicts no gas bypass-
ing for the 12% fine case as seen in Figure 15. The high-fines content bed has disbursed bubbles moving
up through the bed and does not severely pack (orange color) compared to the low-fines bed (red color)
which streams.
The solids size is important in fluidized beds and needs to be properly modeled to accurately calcu-
late the gas-solids flow. It is important to note that there was no tuning or adjustment of parameters such
as particle size, turbulence coefficients, or gas properties in running the calculations. The direct applica-
tion of CPFD method using the MP-PIC numerical model of solids dynamics accurately captured the
effects of particle sizes on the bed behavior.
The curves for onset of streaming given by Figure 15 were generated from small, isothermal fluidized
beds with few or no internals. The CPFD calculation, given here, was for one of the experiments used
to generate the curve and thus, the calculation has a sound comparison to the measured data. The curves
shown in Figure 15 are plots between two measured data, and the lab-scale data are not expected to ac-
curately extrapolate to a wide range of large, complex, hot beds. Scale-up from lab scale experiments to
commercial units is known not to work. On the other hand, the MP-PIC scheme in the CPFD numerical
method is based on physics which does scale, and good predictions to a variety of experiments, such
as shown here, provides confidence that the MP-PIC method can correctly and accurately predict large
scale complex granular flow systems.
references
Amsden, A. A., O’Rourke, P. J., & Butler, T. D. (1989). KIVA-II: A computer program for chemically
reactive flows with sprays. LA-11560-MS. Los Alamos, NM: Los Alamos National Lab.
Andrews, M. J., & O’Rourke, P. J. (1996). The multiphase particle-in-cell (MP-PIC) method for dense
particle flow. International Journal of Multiphase Flow, 22, 379–402. doi:10.1016/0301-9322(95)00072-0
Barnea, E., & Mizrahi, J. (1973). A generalized approach to the fluid dynamics of particle systems: Part
1. General correlation for fluidization and sedimentation in solid multiparticle systems. Chem. Engng.
J., 5, 171–189. doi:10.1016/0300-9467(73)80008-5
Batchelor, G. K. (1988). A new theory of the instability of a uniform fluidized bed. Journal of Fluid
Mechanics, 193, 75–110. doi:10.1017/S002211208800206X
Blaser, P., & Yeomans, N. (2006, February). Sand Core Engineering & Process Modeling. Japan Foundry
Society, 2(2), 420–427.
Brackbill, J. U., & Ruppel, H. M. (1986). FLIP: A method for adaptively zoned, particle-in-cell calcula-
tions of fluid flow in two dimensions. Journal of Computational Physics, 65, 314. doi:10.1016/0021-
9991(86)90211-1
Cundall, P. A., & Strack, O. D. L. (1979). A discrete numerical model for granular assemblies. Geotech-
nique, 29, 47–65. doi:10.1680/geot.1979.29.1.47
312
Davis, E., Herbolzheimer, R. H., & Acrivos, A. (1982). The sedimentation of polydisperse suspensions in
vessels having inclined walls. International Journal of Multiphase Flow, 8(6), 571–585. doi:10.1016/0301-
9322(82)90064-7
Edwards, C. F. (1999). Formulating Dense, Large-Eddy Simulations of Multiphase Flows. In ILASS
Americas 12th Annual Conference on Liquid Atomization and Spray Systems, Indianapolis, IN.
Ferziger, J. H., & Peric, M. (1996). Computational methods for fluid dynamics. Berlin: Springer-Verlag.
Gidaspow, D.& Ettehadieh, B. (1983). Fluidization in two-dimensional beds with a jet, Part II: hydro-
dynamic modeling. I&EC Fundamentals, 22, 193-201.
Gidaspow, D. (1986). Hydrodynamics of fluidization and heat transfer supercomputer modeling. Applied
Mechanics Reviews, 39, 1–22. doi:10.1115/1.3143702
Gidaspow, D. (1994). Multiphase flow and fluidization continuum and kinetic theory description. Boston,
MA: Academic Press.
Godlieb, W., Deen, N. G., & Kuipers, J. A. M. (2007). A discrete particle simulation study of solids
mixing in a pressurized fluidized bed. In 2007 ECI Conference on the 12th International Conference on
Fluidization -New Horizons in Fluidization Engineering, Vancouver, Canada.
Harlow, F. H. (1957). Hydrodynamic Problems Involving Large Fluid Distortion. J. Assoc. Comp. Mach.,
4, 137.
Harlow, F. H., & Amsden, A. A. (1971) Fluid Dynamics. Los Alamos National Laboratory Report
LAMS-4700.
Harlow, F. H., & Welch, E. (1965). Numerical Calculation of Time-Dependent Viscous Incompressible
Flow of Fluids with Free Surface. Physics of Fluids, 8, 2182. doi:10.1063/1.1761178
Harris, S. E., & Crighton, D. G. (1994). Solitons, solitary waves and voidage disturbances in gas-fluidized
beds. Journal of Fluid Mechanics, 266, 243–276. doi:10.1017/S0022112094000996
Haworth, D. C., & Jansen, K. (1996). Large-Eddy Simulation on Unstructured Deforming Meshes: To-
wards Reciprocating IC Engines. Center for Turbulence Research, Proceedings of the Summer Program.
Hirshfelder, J. O., Curtiss, C. F., & Bird, R. B. (1954). Molecular Theory of Gases and Liquids. New
York, NY: John Wiley & Sons.
Karri, S.B., Issangya, A. & Knowlton (2004). Gas Bypassing in deep fluidized beds. In U. Arena, R. Chi-
rone, M. Miccio & P. Salatino (eds.), Fluidization XI, (pp. 515-521). New York: Engineering Foundation.
Launder, B. E., & Spalding, D. B. (1974). The Numerical Computation of Turbulent Flows. Computer
Methods in Applied Mechanics and Engineering, 3, 269–289. doi:10.1016/0045-7825(74)90029-2
O’Rourke, P. J. (1981). Collective drop effects on vaporizing liquid sprays. Ph.D. Thesis, Princeton
University, Princeton, NJ.
313
O’Rourke, P. J., Brackbill, J. U., & Larrouturou, B. (1993). On particle grid interpolation and calculat-
ing chemistry in particle-in-cell methods. Journal of Computational Physics, 109, 37–52. doi:10.1006/
jcph.1993.1197
O’Rourke, P. J., Zhao, P., & Snider, D. (2009). A Model for Collisional Exchange in Gas/Liquid/Solid
Fluidized Beds. Chemical Engineering Science, 64, 1784–1797. doi:10.1016/j.ces.2008.12.014
Pomraning, E., & Rutland, C. J. (2001). A Dynamic One-Equation Non-Viscosity LES Model. AIAA
Journal, 25502.
Richardson, J. F., & Zaki, W. N. (1954). Sedimentation an fluidization: part I. Transactions of the Insti-
tution of Chemical Engineers, 32, 35–53.
Risk, M. A. (1993). Mathematical modeling of densely loaded, particle laden turbulent flows. Atomiza-
tion and Sprays, 3, 1–27.
Smagorinsky, J. (1963). General circulation experiments with the primitive equations, part I: the basic
experiment. Monthly Weather Review, 91, 99–164. doi:10.1175/1520-0493(1963)091<0099:GCEWT
P>2.3.CO;2
Snider, D. M. (2001). An Incompressible three dimensional multiphase particle-in-cell model for dense
particle flows. Journal of Computational Physics, 170, 523–549. doi:10.1006/jcph.2001.6747
using multiphase particle-in-cell model for dense particle flow. International Journal of Multiphase
Flow, 24, 1359–1282. doi:10.1016/S0301-9322(98)00030-5
Thompson, P. A. (1972). Compressible-fluid dynamics. New York: McGraw Hill.
Travis, J. R., Harlow, F. H., & Amsden, A. A. (1976). Numerical calculation of two-phase flows. Nuclear
Science and Engineering, 61, 1–10.
Vincenti, W. G., & Kruger, C. H. Jr. (1975). Introduction to physical gas dynamics. Huntington, NY:
Krieger Publishing Company.
Wen, C. Y., & Yu, Y. H. (1966). Mechanics of fluidization. Chem. Eng. Progr. Symp., 62, 100-110.
314
Section 3
Practice
316
Chapter 10
Ray Cocco
PSRI, USA
S.B. Reddy Karri

PSRI, USA
Ted Knowlton
PSRI, USA
abstract
In the last 20 years, significant improvements in the computational fluid dynamics (CFD) modeling
have been made that allow the simulation of large-scale, commercial CFBs. Today, commercial codes
are available that can model some of this behavior in large-scale, commercial units in a reasonable
amount of time. However, the hydrodynamics in a riser or fluidized bed are complex with both micro and
macroscale features. From particle clustering to large streamers to the core-annulus profile, the particle
behavior in these unit operations rarely behaves as a “continuous fluid.” Even the role of particle size
distribution is often neglected and models that do consider particle size distribution don’t always consider
the role of particle size on granular temperature. Many models use insufficient boundary conditions by
assuming uniform or symmetric profiles, which is rarely the case. Furthermore, grid sizing is usually
based on computer limitations instead of model limitations, and many models of commercial systems
extend beyond the capability of the constitutive equations being used. Successful application of today’s
CFD models requires a good understanding of the equations behind the code, the assumptions used for
those equations and the capability or limitations of the code. CFD is nothing more than a guess without
an understanding of the fundamentals, underlying assumptions and code limitations that are part of
every model.
intrOductiOn
The circulating fluidized bed (CFB) has been in large-scale operation since the 1970’s with the advent
of the fluidized catalytic cracking (FCC) riser reactor in the petroleum industry. One of the unique char-
DOI: 10.4018/978-1-61520-651-3.ch010
Figure 1. Simple (left) and more complex (right) schematics of circulating fluidized beds
acteristics of these units is the ability to move catalyst from a reactive environment to a regenerative
environment. As a result, catalyst with high activity but short lifetimes could be regenerated, sometimes
close to a catalyst’s initial activity.
In today’s FCC units, highly active zeolite catalysts are commonly used. However, these catalysts
have a useable lifetimes of only a few seconds due to a buildup of a carbonaceous material (coking) on
the catalyst as a side reaction of the oil cracking process. By circulating the catalyst from the reactor
section to an air regeneration section, coke is continuously removed from the catalyst, activity is resorted,
and the added heat value from regeneration is delivered to the reaction section by way of the returning
catalyst. A typical FCC unit requires about five pounds of catalyst per pound of feedstock, which can
amount to catalyst circulation rates of ten million pounds per hour or more.
Circulating fluidized beds consist of several configurations. The simplest version is a fluidized bed
operated in the fast fluidization or even transport regime where the solids are captured with a primary
cyclone (and perhaps further captured with secondary cyclones). From the cyclone, the solids are returned
to the fluidized bed through a dipleg.
The most common configuration of a circulating fluidized bed is one where solids move from a
riser to one or more fluidized beds. Figure 1 shows schematics of circulating fluidized beds with and
without additional units such as a regenerator. FCC units are based on this configuration where solids
flow from the riser to a stripper to a regenerator and then back to the riser. Solids conveyed in the riser
are separated from the product gas at the top of the riser using cyclones or a riser terminator. The riser
terminator can be a simple vessel expansion or a more sophisticated device such as UOP’s VSS and
RSS (Lomas, 1996), ExxonMobil’s swirl (Tammers, et al., 1993) or Stone and Webster’s Ramshorn
(Yawn, 2006) design. Cyclones are used downstream of these devices for additional separation of the
solids from the product gases.
Interstitial gas is removed from gas/solid suspension using a stripper. Here, an inert gas or steam is
used to further remove the product gases before they are sent to the regenerator. In many cases, strippers
317
contain internals such as a disk and donut design or Koch-Glitsch’s structured packing elements (Rall,
2001) to improve the efficiency of removing the interstitial gases. In fact, there is a plethora of stripper
designs which highlights the importance of this part of the operation.
From the stripper, solids or the spent catalyst in the FCC are fed into a fluidized bed through a
standpipe. These fluidized beds are typically used as catalyst regenerators operating in the bubbling or
turbulent flow regimes. Sparger or grid distributors are used to inject the regenerating gas, usually air,
into the bed. Cyclones are again used to separate the entrained solids in the freeboard region (the region
above the bed) from the flue gas. Additional gas cleaning may be needed further downstream.
From the regenerator, solids are delivered back to the riser. Riser configurations vary according to
the application. For oil cracking, solids are lifted to liquid feed atomizers in the riser reactor. UOP has
proposed that a second solids stream from the stripper be feed directly to the riser to improve the control
of reaction rates (Wegerer & Lomas, 1995) by returning cooler, more deactivated catalyst to the riser reac-
tor. Shell proposed adding corbels (riser inserts) to increase the gas-solid contacting in the riser reactors.
The truth of the matter is that although risers have been used commercially for over 40 years, the gas
and solid hydrodynamics in these units are still poorly understood. Today, with applications of CFBs
increasing, understanding riser hydrodynamics is becoming increasingly important. Whether it is to
improve petroleum yields to apply CFBs to new areas such as in coal gasification or chemical looping,
designing these units is now limited by the understanding of how gas and solids interact on both macro
and micro scales. This chapter will discuss experiments and simulations that explore this gas and solid
relationship as applied to the rise in CFB applications.
aPPlicatiOn Of cfbs
There are only a few commercial applications of CFBs, unlike with their dense-phase fluidized bed coun-
terparts. However, CFBs are one of the more important unit operations in terms of the global economy.
Today, the CFB, as applied to FCC units, account for 30% of the nearly 50 million barrels per day of
crude oil refining (DOE, 2006). FCC units are operating in nearly 400 petroleum refineries worldwide
and are considered state-of-the-art technology for producing high-octane gasoline as well as a wide
range of petroleum derivatives including ethylene, propylene, jet fuel, benzene, toluene, and kerosene.
Petroleum refineries come with large price tags with capital cost ranging from $2 to 5 billion dollars
(USD). In the US, no new oil refineries have been built in over 30 years. However, significant improve-
ments have been to existing units, resulting in significant increases in oil production capacity. Some
retrofits have involved the complete redesign of FCC units.
FCC units come in two different styles; the side-by-side configuration and stacked configurations,
as shown in Figure 2. The side-by-side configuration is more common due to its slightly lower capital
cost, but the stack configuration offers better heat management. Today, the major designers for FCC
units include ABB Lummus, ExxonMobil Research and Engineering, Shell Global Solutions, Stone
and Webster, Total, IFP, KBR and UOP with the side by side configuration and KBR with the stacked
configuration.
Coal combustion is another application of circulating fluidized beds. Unlike other conventional coal
combustors, fluidized bed and circulating fluidized bed coal combustion offer better temperature control,
higher efficiency, and easier in-situ capture of sulfur with sorbents such as limestone. As shown in Fig-
ure 3, coal, sand, and the sorbents are air lifted in the combustion chamber or boiler which is a substan-
318
Figure 2. Schematic of the side-by-side (left) and stacked (right) configuration for the more common
FCC units
tially tall unit lined with boiler tubes. Sand provides a more uniform fluidization and reduces the hydro-
dynamic sensitivity of the incoming feedstock’s particle density and size. Combustion takes place at
800 to 900°C. The spent particles are smaller and/or lighter than the feed particles. With operating ve-
locities higher than the terminal velocities, the spent particles are entrained from the reactor. The active
coal remains near the bottom of the combustor. Any entrained unburned fines and inerts are captured
with one or more cyclones and returned to the bottom of the combustor. The solids flow rate back to the
combustor is usually controlled with an L-valve, J-valve, or the more common loop seal. With this de-
sign, the coal particles are recirculated until they are completely combusted or they reach a particle size
that can not be efficiently captured by the cyclone or cyclones.
Advanced second generation units are similar to the original CFB design except the gas and particle
velocities are higher in the upper combustor with the bottom bed being more in the fast fluidization
regime. This provides an even more turbulent or chaotic flow which increases heat and mass transfer.
Many of these second generation units have the boiling tubes located downstream of the combustor.
This reduces slag build up on the coil surface which can be significant in the bottom region of combus-
tors. Yet, the second generation units provide a higher concentration of entrained particles to the coil so
heat transfer is similar or better than that found in the first generation combustors. Today, second-gen-
eration boilers have been built in the 500 MWe capacity range with larger sizes being proposed.
Coal and biomass gasification can also be carried out in a circulating fluidized bed reactor, although
dense-phase fluidized bed gasifiers are more popular. This is mostly due to the simpler design of a fluidized
bed gasifier. Circulating fluidized bed gasifiers tend to be smaller and offer better carbon conversions.
Currently, these units are seeing more application in the biomass gasification area. Fossil Energy Reserch
Corporation (FERCo) licensed a 40 MWth Battelle Indirect Gasifier that was operation in Burlington,
Vermont (Paisley, 1997). The design consists of two tandem fluidized beds; one for gasification and the
319
Figure 3. Schematic of a CFB coal combustor typically used in the power industry
other for char combustion. KBR has a pressurized CFB gasifier, which consists of a dense bed mixing
zone followed by riser, disengager, cyclone, standpipe, and J-valve.
A new application of circulating fluidized beds for coal combustion or gasification is chemical
looping. Particles used as oxygen carriers are used instead of air or oxygen gas. These oxygen carriers
(iron, nickel or copper oxides) need to circulate from the reducing environment of coal combustion
or gasification to the oxidizing environment of a regenerator. Much like the FCC units, heat from the
oxidizing regenerator is transferred to the combustion reactor. The advantage is that the carbon dioxide
stream is not diluted with nitrogen gas as with air combustion, or an air separation plant is not needed to
make pure oxygen (i.e., oxyblown gasifier). This is a significant benefit if carbon dioxide sequestration
is being considered. It also has the potential benefit of higher hydrogen yields from the coal feedstock
(Gupta, 2005).
In 1996, DuPont de Nemours and Company installed a novel CFB process in Asturias, Spain for
making 180 million pounds per year of maleic anhydride from n-butane (Contractor, 1999). Maleic
anhydride was an intermediate for making tetrahydrofuran and polytetramethylene ether glycol, which
are used to make Spandex™ fibers. This was the first commercial application of CFB reactors for spe-
cialty chemicals and perhaps chemical looping.
The DuPont process used vanadium phosphate as the oxygen carrier for the oxydehydrogenation
of butane to maleic anhydride. Using an oxygen carrier provided lower carbon monoxide, and carbon
dioxide yields (i.e., less deep oxidization) at higher conversions than traditional fixed-bed reactors.
Figure 5 illustrates the advantage of this CFB process compared to the traditional fluidized bed reactor
320
Figure 4. Chemical looping concept (based on Gupta, 2005)
configuration. However, since oxygen was being consumed from the catalyst itself, the catalyst activity
was short lived and the catalyst required frequent regeneration.
The CFB reactor provided DuPont with the ability to continuously regenerate the vanadium phosphate
catalyst and provide the necessary heat transfer for temperature control. Figure 6 shows the process
concept for the DuPont process. This process is remarkably similar to the chemical looping concept,
although it is not typically referred to as a chemical looping process.
As shown in Figure 6, freshly-oxidized catalyst provides the oxygen needed for the oxydehydrogena-
tion of n-butane to maleic anhydride in the riser. After the riser, the reduced catalyst is stripped of inter-
stitial gases and sent to the regenerator. In the regenerator, the catalyst is re-oxidized and any coking
resulting from reactions in the riser is burned off. The oxidized catalyst is sent then back to the riser.
Unlike the FCC process where catalyst costs are on the order of a few dollars per pound, the DuPont
catalyst was significantly more expensive. In addition, the vanadium phosphate catalyst was physically
weak. Thus, attrition in the CFB unit could have been significant, and if so, would have resulted in a
Figure 5. Butane oxidation in a CFB and fluidized bed, (based on Contractor, 1999)
321
Figure 6. Butane oxidation in a CFB and fluidized bed, (from Contractor, 1991)
prohibited operating cost. However, DuPont ingeniously developed a porous silica shell coating to im-
prove the attrition resistance of the catalyst (Contractor, 1987).
Other processes have been proposed similar to the maleic anhydride process. DuPont subsequently
proposed CFB reactor concepts for the oxidization of propylene to acrolein, an intermediate for the
production of acrylic acid (Contractor, 1998). Dow Chemical made a similar proposal for the partial
oxidization of butane to butadiene in a CFB reactor (Khazai, 1989). Phthalic anhydride from o-xylene
oxidization (Gelbein, 1980), xylene isomerization, and partial oxidization of methane (Pugsley, 1997)
are just a few of the other proposed applications.
cfb HydrOdynaMics
The hydrodynamics of circulating fluidized beds are extremely complex. A riser can exhibit a wide range
of axial and radial solids profiles that are highly dependent on the operating conditions, physical properties,
and riser configurations. The riser itself can be separated into three components: the entrance region, the
fully-developed flow region and the exit region. In addition, particle behavior needs to be considered not
only on the macro-scale (where pressure, solids flux, etc. are important) but on the micro-scale (where
particle interactions, clusters, etc. are important) as well. Thus, understanding the axial pressure, veloc-
ity, and flux profiles in a riser may not be enough to ensure successful scale-up. Radial profiles change
322
Figure 7. Transition from fast fluidized bed to dilute-phase conveying based on riser ΔP/L (top) and
solid flux (bottom)
significantly with axial position and asymmetric profiles are more often the norm. Solids backmixing,
clustering and streaming also complicate the understanding of CFB hydrodynamics.
cfb axial solids Profiles
In order to understand the axial solids profile behavior in a riser, we need to understand the transitions
from the fluidized bed to the riser flow regimes. It is generally agreed that the velocity at which the
transition from a bubbling or turbulent fluidized bed to the fast fluidized bed occurs is characterized by
the choking velocity. However, defining the transition from the fast fluidized bed to what is referred to
as dilute phase conveying is less straight forward. Li and Kwauk (1980) noted that there was no exact
boundary for this transition, only a gradual transition that is dependent not only on operating conditions
but the properties of the gases and solids as well.
There are varying criteria to determine this transition with methods ranging from measuring entrain-
ment rates (Reh, 1971; Rhodes, 1989), slip velocities (Yerushalmi, 1978), voidage profiles (Li & Kwauk,
1980) to a minimization in the pressure drop (Zenz, 1949; Karri, 1991]. The last method is the most
popular. It suggests that the transition occurs at the minimum in the riser pressure drop per unit length
(ΔP/L) versus the gas velocity, as shown in Figure 7. Plotting this minimum with respect to the gas ve-
locity and solids flux provides a locus of velocities at the point of minimum pressure drop which can be
used as the transition from fast fluidization to dilute-phase conveying. Often this transitions corresponds
323
to the choking velocity. The benefit of this method is that only simple experiments are needed to deter-
mine the transition. Stromberg (1983) and Takeuchi (1986) provided a similar experimental approach.
Monazam and Shadle (2004) took a different approach to determine the regime transition. They were
able to determine the fluidization regimes of cork particles in a 12-inch (0.3-meter) diameter riser by
stopping the solids flow rate to the riser. As the solids were carried out of the riser, the pressure drop
across the riser decreased with time. An analysis of the transient pressure drop data across the riser dur-
ing a solids flow cut-off against its time derivative indicated three distinct operating regimes and they
were able to discern fast fluidization from dense transport regimes over a wide range of gas velocities.
Fundamentally, the minimum in the pressure drop and its relationship with the fast fluidized bed
to conveying transition makes sense. In a fast fluidized bed, pressure drop is due to gravity, particle
collisions and particle drag with the first two components dominating. As the superficial gas velocity
increases, particle concentration or loading decreases and pressure drop is dominated by the drag force
(Geldart, 1986). Thus, for riser modeling, the choice of the drag model is critical.
By plotting the pressure drop per unit length at various riser heights, the axial profile of the riser
suspension density can be determined, as shown in Figures 8 and 9. These data are from PSRI’s 8-inch
(0.2-meter) ID by 72-foot (21.0-meters) tall CFB riser (fitted with an elbow termination) using FCC
catalyst powder (dp50 ~ 76 μm and particle density of 1500 kg/m3). This axial density profile is highly
sensitive to the solids mass flux, as shown in Figure 8, and the superficial gas velocity, as shown in
Figure 9. The axial density profiles shown here are representative of what is observed in a fast fluidized
bed. It is only at very low solid fluxes or, to a lesser extent, at high superficial gas velocities where the
more uniform density profiles emerge. Others (Reh, 1971; Li & Kwauk, 1980; Weinstein, 1983; Hartge,
1988; Rhodes, 1989; Monazam and Shadle, 2004) have reported a similar “S-shape” behavior in risers.
Horio (1997) suggested this S-shape may be due to large L/Ds and that large, commercial riser may not
exhibit this behavior. Yet, PSRI has operated risers up to 36 inches (0.9 meters) in diameter and has still
observed the “S-shape” density profile.
A close observation of Figures 8 and 9 shows that the profiles are skewed both at the riser bottom
and top. These pressure profiles suggest that fully-developed flow may not be achieved until a riser
height of 2 to 6 meters for the low flux cases. For high flux cases, fully-developed flow was not achieved
in PSRI’s 72-foot (21.0-meter) tall riser height restriction. These profiles are a combination of particle
drag and particle acceleration where article acceleration is significant in the lower part of the column.
The extent of particle acceleration is sensitive to these factors and the results of which is often called
entrance effects.
Similarly, there are distortions in the pressure profile in the top of the riser suggesting that exit
configurations can affect the riser hydrodynamics well below the exit itself. These “exit effects” have
been seen on a wide range riser of test units (Rhodes, 1998; Pugsley, 1997; Brereton & Grace, 1994;
Lackermeier & Werther, 2002) as well as in some commercial FCC units (Martin, 1992; Harris, 2003a].
In addition, the magnitude of this distortion is dependent on the exit configuration. As shown in Figures
10 and 11, blind tees and inverted wyes lead to higher concentration of solids (due to solids refluxing)
in the exit region of the riser. An elbow configuration resulted in a minimal build up of solids material.
Harris et al. (2003b) used phosphorescent tracer particles to study the effects of the riser termination
on the solids residence time distribution. As expected, the increase in solids reflux due to the exit con-
figuration resulted in a longer particle mean residence time, variance and breakthrough time in the riser.
Thus, if backmixing is an issue, the choice of riser termination may need to be addressed.
324
Figure 8. Riser axial density profile for FCC powder in 0.2-meter (8-inch) ID x 21.9-meter (72-feet) tall
riser at varying solid fluxes with a superficial gas velocity of 6.1 m/sec (20 ft/sec)
Figure 9. Riser axial density profile for FCC powder in 0.2-meter (8-inch) ID x 21.9-meter (72-feet) tall
riser at varying superficial gas velocities with solids fluxes at 20 and 560 kg/m2-sec (4.1 and 114 lb/ft2-sec)
cfb radial Profiles
The radial voidage profile in CFBs is even more surprising. At first, it was assumed that the profiles
were uniform with the particle slip velocity corresponding to the single particle terminal velocity. Early
work by Lewis et al. (1949) cast some doubts on this simplification. Later, Geldart (1986) provided
additional data of slip velocity in a riser and further highlighted the disparity between calculated and
325
Figure 10. Exit effects in PSRI’s 0.2-meter (8-inch) ID x 21.9-meter (72-feet) tall riser with a gas velocity
of 9.1 m/sec (30 ft/sec) and solids flux of 196 and 391 kg/m2-sec (40 and 80 lb/ft2-sec)
Figure 11. Exit effects in PSRI s 0.2-meter (8-inch) ID x 21.9-meter (72-feet) tall riser with a gas velocity
of 12.2 m/sec (40 ft/sec) and solids flux of 196 and 391 kg/m2-sec (40 and 80 lb/ft2-sec)
measured slip velocities. Geldart’s experimental measurements were significantly less than the terminal
velocity. This larger-than-expected slip ratio (i.e., solids velocity divided by gas velocity) was explained
by particle clustering in the riser. These cluster behaved more like larger particles with a corresponding
higher terminal velocity.
326
Figure 12. Radial voidage distributions of bubbling, turbulent and circulating fluidized beds (based on
Abed, 1984 and Bader, 1988)
Even more surprising is that the radial solids concentration in risers have a core-annulus profile
with the high concentrations of solids at the wall (van Zoonen, 1962; van Breugel, 1969; Bader, 1988,
Karri & Knowlton, 1996). This behavior not only persists in the pneumatic transport regime but can be
measured in turbulent and bubbling fluidized beds as we;;. Figure 12 shows the radial volume fraction
profile of data collected in a bubbling bed, turbulent fluidized bed, and riser for FCC catalyst. PSRI has
found this behavior to exist in risers with diameters ranging from 0.2-meter (8-inch) ID [Bader et al.]
up to 0.9-meter (36-inch) ID. In fact, studies in commercial FCC units have reported a core-annulus
voidage profile in the risers.
Similarly, this core annulus behavior is retained with increases in the solids flux (even though the
solids flux profiles vary with solids rate). Although solids flux data are a convolution of solids loading
and solids velocity, the significant change in the radial solid flux is assumed to be due to solids loading
since the magnitude of the solids velocity at the wall is expected to be low for all cases. As shown in
Figure 13, increasing the solids flux from 50 to 360 kg/m2-sec (10 to 74 lb/ft2-sec) resulted in a more
dramatic range of solids flux profiles in the riser. Similar results were observed for FCC catalyst in a
0.08-meter (3-inch) ID riser (Issangya, 1998).
Thus, riser hydrodynamics consist of a dilute core region with particles moving upward while a more
dense region resides at the walls. As particles collide with the walls, momentum is lost and particle
concentration increases. With this increase, the slip velocity increases and particles flow more slowly
at the wall. Figure 14 shows experimental data from PSRI of particle downflow behavior at the wall in
terms of radial solids flux measurements for FCC powder with air in an 0.3-meter (12-inch) ID riser. As
the gas velocity increases, the drag force on the particles and particle clusters increases causing them to
flow upward near the wall. It was presumed that this upward particle flow near the wall might destroy
the core-annulus profile; however, PSRI found that even with the solids flowing upward near the wall,
the core-annulus profile was still preserved, as shown in Figure 14. Although these upward flowing
particles near wall were at significantly slower velocities than the particles in the core region. PSRI
found similar behavior for sand in the same system (Karri, 1998). Grassler and Wirth (1999) used x-ray
327
Figure 13. Radial profile of solids flux via solids extraction probe from PSRI’s 0.2-meter (8-inch) ID x
21.9-meter (72-feet) tall riser with FCC powder (dp50 = 72 μm) and air at a superficial gas velocity of
12 m/sec (39 ft/sec) (based on Karri, 2002)
Figure 14. PSRI data for FCC powder (dp50 = 72 μm) with air in a 0.3-meter (12-inch) ID riser at
varying superficial gas velocities and solids flux of 586 kg/m2-sec (120 lb/ft2-sec). Dash lines indicated
extrapolated points where the extraction probe could not reach
tomography to study radial solids profiles in the acceleration, fully developed and exit region of the
riser. They observed a parabolic core-annulus shape in solids concentration in all regions of their
0.19-meter ID x 15-meter tall (7.5-inch ID x 50-foot tall) riser with the core-annulus profile being more
distinct at higher mass fluxes.
328
Figure 15. High speed imaging at 3000 fps with 70 μsec exposure of FCC powder in PSRI’s 0.2-meter
ID x 22-meter tall (8-inch ID x 72-foot) riser at superficial gas velocities and solids fluxes of (a) 9.1 m/
sec (30 ft/sec) and 50 kg/m2-sec (10 lb/ft2-sec), (b) 18.3 m/sec (60 ft/sec) and 50 kg/m2-sec (10 lb/ft2-sec),
(c) 9.1 m/sec (30 ft/sec) and 391 kg/m2-sec (80 lb/ft2-sec), and (d) 18.2 m/sec (60 ft/sec) and 391 kg/
m2-sec (80 lb/ft2-sec), This was collaborative effort between PSRI and NETL
clusters and streams
Visual observation of a riser shows that there is more to the particle hydrodynamics than S-curves and
core-annulus profiles. Early imaging studies by Reh (1971) showed the hydrodynamics was much
more complex with “solids moving mostly in strands.” Today, these strands are commonly referred to
as clusters and streamers.
Recently, PSRI and the National Energy Technology Laboratory (NETL) have conducted high-speed
imaging of particle hydrodynamics in risers (Shaffer, 2009). Figure 15 shows selected images from this
study for FCC powder in an 0.2-meter (8-inch) ID riser at superficial gas velocities of 9.1 and 18.3 m/
sec (30 and 60 ft/sec) and solids fluxes of 49 and 392 kg/m2-sec (10 and 80 lb/ft2-sec). At low veloci-
ties, down-flowing streamers are clearly observed at the wall. The number and size of these clusters
and streamers increase with increasing solids flux. At 18.3 m/sec (60 ft/sec), up-flowing clusters and
streamers were observed at the wall at both low and high solids fluxes. The higher gas velocity reduced
the size of these streamers. In addition, the high-speed videos revealed that the core region in the riser
was dynamic with pulses of gas that would penetrate the annular region.
329
Figure 16. High-speed video of polyethylene clusters and streamers in NETL’s 0.3-meter (12-inch) ID
riser (Shaffer, 2009)
Shaffer of NETL also used high speed imaging to study the nature of these cluster and streamers in
NETL’s 0.3-meter (12-inch) ID riser with polyethylene powder (Shaffer, 2009). Shaffer observed that
the streamers appeared to be the product of shear planes in the riser, as shown in Figure 16. This was in
agreement with the work of Senior and Grace [1998] where boundary layer theory demonstrated that
clusters and streamers are the product of shear-induced lift. At the interface of the downward and upward
moving solids, densification occurred resulting in a significant increase in particle concentration at this
interface. It is possible that these protruding, pulsing gas jets cause, in part, these shear planes and sub-
sequent densification of the particles into clusters and streams.
Shaffer also found that these streamers were more dense at the leading edge. This compacted layer
was flexible, but did not appear to significantly exchange particles in front of the leading edge. Instead,
on-coming particles merely deflected off this compacted boundary. If this explanation is correct, it sug-
gests that clusters and streamers could be the controlling mechanism for the high level of backmixing
often observed in low velocity risers.
Several studies have focused on determining the characteristic size of these clusters. Rhodes et al.
(1992) proposed that the average cluster size in his lab-scaled CFB unit was approximately one cm.
Noyman and Glicksman (1998) confirmed this dimension using infrared thermagraphy measurements.
Guenther and Breault (2007) used wavelet analysis from a backscattered optical probe to discern cluster
characteristics for 812 μm cork particles in a 0.3-meter (12-inch) ID riser. Cluster sizes increased from
one particle diameter to more than ten particle diameters with increasing solids circulation rates. Cluster
sizes were also found to be insensitive to increase in the gas velocity.
330
Figure 17. High-speed video of a polyethylene streamers in NETL’s 0.3-meter (12-inch) ID riser (Shaf-
fer, 2009)
Breault, Ludlow and Yue (2005) where able to use laser Doppler velocimetry of cork particles in a
0.3-meter (12-inch) riser to discern that cluster at the wall region had granular temperatures or velocity
fluctuations an order of magnitude lower than the particles in the dilute regions. In addition, they found
that the granular temperature of the up-flowing clusters was four-times higher than clusters in down-
flow. A later study by Breault, Guenther and Shadle (2008) used time scale criteria to characterize the
variance these particle fluctuations as granular temperature or granular turbulent kinetic energy.
Lint and Glickman (1993) used heat transfer measurements to determine that clusters traveled very
close the wall, often within 100 μm. They later showed that this was a function of the boundary layer
with the wall which is on the order of 1.9 cm. Rashidi, Hetsroni and Banerjee (1990) also showed that
significant particle concentration existed within 2-cm of the wall which could be an indication of this
hydrodynamic boundary layer. Noyman and Glicksman suggested that since a cluster is about 1-cm
thick and travels within 100 μm of the wall and that the boundary layer is approximately 2-cm thick,
the clusters must generally travel within the wall boundary layer in the annulus region. Yet, Guenther
and Breault’s analysis suggests that clusters are present in the core region as well.
The work of Rashidi also showed that the gas velocity gradient around these clusters is small. Noy-
man and Glicksman (1999,2000) noted that this suggests the descending velocity of a cluster should be
unaffected by the average gas velocity. This postulate compliments Guenther and Breault’s findings that
there is little dependency of cluster hydrodynamics on the superficial gas velocity. Other studies suggest
the same findings (Lim, 1996; Noymer, 1999).
Based on the observations of Shaffer and analysis of Noyman and Glicksman, it appears these cluster
and streamers are fairly impervious to both solids and gas penetration. This would explain the transient
331
Figure 18. Particle size distribution in the annular section of a 0.20 meter ID x 21.9-meter tall (8-inch
ID x 72’-foot tall) riser with down-flow at the walls. Superficial gas velocity was 9.1 m/sec (30 ft/sec)
and solids flux was 92 kg/m2-sec (19 lb/ft2-sec)
shape these features have as well as how gas and solids backmixing could be so significant for some
riser operations. Nevertheless, any modeling of a riser needs to consider the behavior of these features
to accurately predict reactor productivity.
Particle segregation effects
Particle segregation is expected in low velocity fluidized beds where larger or heavier particles migrate
towards the bottom. However, higher velocities usually mitigate that effect. So it was somewhat unex-
pected that particle segregation is present in a riser. As shown in Figures 18 and 19, Karri and Knowlton
(1998) found that larger particles migrated towards the annulus region when downward flow at the
wall occurred. There was also an axial dependency with more segregation occurring the wall occurring
near the bottom of the riser. There was no detectable axial segregation in the core region. Also, particle
segregation was not observed at the extreme range of solid flux conditions where upward flow at the
wall was observed. Particle segregation was only observed in the annular region when the flow was
downward at the wall.
Similar findings were found by Das et al. (2008) for axial segregation of binary materials of sand-
FCC catalyst and coal-iron mixtures. Larger particles tend to migrate or remain in the bottom of the
riser. However, Das noted that radial segregation was negligible compared to axial segregation. Perhaps
this is due to the larger disparity between particle size and density for Das’ materials.
Jones, Yurteri and Sinclair (2000) were also able to quantify particle segregation using LDV measure-
ments in a solid-laden gas jet. For glass beads with a mean particle size of 70 μm, they observed that
the larger particles migrated towards the outside of the jet. This resulted in a 10% shift in the particle
size distribution from the outside to the center of the jet. This segregation would thus concentrate fines
in the core region of riser flow.
332
Figure 19. Particle size distribution in the annular section of a 0.20 meter ID x 21.9-meter tall (8-inch
ID x 72’-foot tall) riser with up-flow at the walls. Superficial gas velocity was 9.1 m/sec (30 ft/sec) and
solids flux was 49 kg/m2-sec (10 lb/ft2-sec)
The addition of fines to a bed of Geldart Group A powder can significantly reduce the bed or emul-
sion density in a fluidized bed; but little is known of the role of fines with CFB hydrodynamics. Yet,
it has been reported that fines levels affect attrition in conveying lines (Knowlton, 1994), conversion
in fluidized beds (Pell & Jordan, 1988) and heat transfer in CFB combustors (Lee, 1997). Thus, riser
hydrodynamics is complex with both axial and radial profiles that need to be considered. In addition,
smaller-scaled events such as clustering and streaming as well as particle segregation all play a signifi-
cant role in secondary effects such as backmixing.
Riser design can significantly affect particle hydrodynamics. Some riser termination design can
promote backmixing well below the termination point. Entrance effect can have similar affects. Breault
and Mathur (1989), Bia et al. (1997), Lei and Horio (1998), Kaiser et al. (2000), and Kehlenbeck et al.
(2001) all showed that the fluid or gas injection significantly affected the hydrodynamics. However,
more research is needed in this area.
Hence, when modeling risers all aspects of the riser hydrodynamics and factors affecting the hydro-
dynamics need to be considered. This is especially true with fast reactions where mixing and micro-
mixing are the rate determining step. In addition, a better fundamental understanding of gas-particle and
particle-particle interactions is needed. Empirical relationships are all too often limited to the systems
where they were developed under. Extrapolation beyond that system could result in significant errors.
cfd MOdeling Of cfb systeMs
The development of computational fluid dynamic (CFD) models for fluidized beds and circulating flu-
idized beds has been pursued since the late 1970’s (Arastoopour, 1979; Arastoopour, 1982). However,
such models have been much slower in development than other single and multiphase unit operations.
333
Despite the talent and time that has been given this area of research, particle hydrodynamics in CFBs
remain elusive. This is mostly due to to the fact that fluid-particle interactions are scale dependent, and
fluid-particle experiments can be difficult to control and measure.
Although available data for particle-fluid model development continue to be sparse, recent significant
strides have recently been made in this area. Better constitutive equations exist, and codes are handling
larger and larger domains. Today, large, commercial fluidized beds and CFBs are being modeled (Guen-
ther, 2003; Cocco, 2004; Williams, 2007; Chen, 2008) and CFD models are becoming one of the many
important tools needed in the design and scale up of CFBs.
CFD modeling of CFBs can be divided into three distinct areas: Eulerian-Eulerian, Lagrangian-
Eulerian, and Hybrid methods. The Eulerian-Eulerian method, sometimes called the two-fluid model,
assumes the particle phase can be described by the same Navier-Stokes’ equations as commonly used
to describe gas and liquid phases. The advantage to this type of model is that the number of equations
are small, and time steps for numerical integration can be relatively large. However, particles do not
behave exactly as fluids and shortcomings with fluid-particle drag and particle-particle interactions need
to be addressed. The Lagrangian-Eulerian models, sometimes called discrete element models (DEM) or
discrete particle modeling (DPM), rigorously track each and every particle’s path (and sometime colli-
sions) using Newton’s equation of motion. The fluid is still modeled using the Navier-Stokes’ equation.
Particle motion is described using Newton’s Law of Motion equation. Thus, particle behavior is more
realistic, but even modestly sized circulating fluidized beds contain trillions of particles. With each
particle requiring its own equation of motion, the computational requirements easily exceed today’s
capabilities. Also, modeling the collisions can require extremely small time steps for numerical integra-
tion. Hybrid models are a combination of the Eulerian-Eulerian and Lagrangian-Eulerian models. With
these models, particles are still tracked but collisions are modeled using similar constitutive equations
as that used with the Eulerian-Eulerian models. Details of each of these methods as applied to modeling
CFBs are presented below.
eulerian - eulerian Models
The Eulerian-Eulerian method was the earliest implementation of CFD toward CFBs. Eulerian-Eulerian
models, or two-fluid models, treat particles as one or more continuous phases. It assumes that the ensemble
averaging of all particles in a predefined grid results in a one continuous fluid that can be described by
the same Navier-Stokes’ equations as that used for describing gases and liquids. Equations 1 and 2 show
the momentum expressions for a gas and particle system.
Fluid Phase:

∂v g   β   
ρg + ρ g v g ∇v g = ∇τ g − ∇P − (v g − vs ) + ρ g g [1]
∂t εg
and
334
Solid Phase:

∂v s   β   
ε sρ p + ε s ρ p vs∇vs = ∇τ s − ∇Ps − (vs − v g ) + ε s (ρ p − ρ g ) g [2]
∂t εg
Detailed equations can be found Chapter 7 of this book.

Closure of the momentum equations is needed for the drag and the stresses. The drag force serves as
the momentum exchange between the two phases and is typically closed using one of many empirical
equations (Gidaspow, 1992; Symalal, 1987; Arastoopour, 1990) or a more fundamental model recently
developed by van der Hoef, Beetstra and Kuipers (2005).
The gas-phase stresses are closed using a Newton’s Law of Viscosity type of equation or some other
similar expression (i.e., shear thinning, power law, etc.). Closure for the solid-phase stresses is much less
straight forward. Both the solids normal (i.e., pressure) and shear stress (i.e., viscosity) need closure;
but, what is this solid-phase pressure and viscosity? How can it be measured? How can it be modeled?
Some of these questions were addressed in Ogawa et al. (1980) and Lun et al. (1984) where collisional
stresses were based on assuming particles can behave similar to molecules and be described by the cor-
responding Boltzmann Equation. This was the birth of what is called today as the “Kinetic Theory of
Solids” or “Kinetic Theory of Granular Flow” (KTGF).
Anderson and Jackson (1967) applied the KTGF with a gas phase to produce the first two-phase
computational model for granular-fluid systems. CFD modeling was then able to focus on measurable
quantities such as the captured bubble hydrodynamics in fluidized beds (Garg, 1975; Pritchett, 1978;
Gidaspow, 1983). In all cases, the bubble rise, break up and coalescence in a fluidized bed was predicted,
but bubbles tended to be sharp and pointed at the top. Much later, Guenther and Syamlal (2001) showed
that second order and higher discretization schemes (i.e., SMART, Superbee) were needed to provide
more realistic bubble shapes. In the following decade, many authors (Lun, 1984; Ding & Gidaspow,
1990; Louge, 1991; Ding & Lyczkowski, 1992) improved on the work of Anderson and Jackson.
Closure of the solids pressure and viscosity was now possible, but a new dependent variable was
brought into the picture, granular temperature which was first proposed by Ogawa et al. (1980). Thus,
this granular temperature expression
3  ∂    1
 (ε s ρ pΘ) + ∇(ε s ρ pΘvs ) = (∇Ps I + τ s )∇vs − ∇(k s∇Θ) − γ s − J s Θ= < v '2s > [3]
2  ∂t  3
needed to be integrated with Equations 1 and 2. It also added new relationships such as solids conduc-
tivity, and dissipation resulting from fluctuations from gas phase and particle inelastic collisions that
needed further closure. Lun et al (1984) addressed this with the development of a granular temperature
equation with additional closures for the solids conductivity and dissipation.
However, such efforts fell short when modeling the riser and CFBs where the core-annulus profile
was difficult to simulate. Tsuo and Gidaspow (1990) used solids viscosity measurements to help close
Equation 2, and were able to match the radial profile measurements of Bader (1988). Sinclair and
Jackson (1989) related the granular temperature to the granular pressure and bulk density. Sinclair and
Jackson also used Johnson and Jackson’s (1987) boundary condition where dissipation of particle axial
momentum is related to the tangential stress of the particle’s approach as
335
ρ s πu z φ T
σ r, z = [4]
ε s, packed ε 2s,/packed
3
2 3( − )
εs ε 2s / 3
Granular temperature could now be used as an Equation of State much like temperature with fluid
systems ([Gidaspow & Huilin, 1998). This granular temperature provided a random vibrational component
to the particle hydrodynamics that is independent of the gas velocity and fluctuation. Solids pressure and
viscosity could now be predicted based on the granular temperature. Since the wall boundary conditions
would reduce the granular temperature, solids concentrations would increase and provide the desired
core-annulus behavior in a riser. Yasuna et al. (1995) found good agreement of the Sinclair and Jackson
model with experimental CFB data.
Yet, the derivations were not perfect. Pita and Sundaresan (1991, 1993) noted that small changes in the
coefficient of restitution had a significant effect on the core-annulus profile. The coefficient of restitution
describes the elasticity of a particle to particle collision. A coefficient of 1 would be a perfectly elastic
collision. However, a very small change in the coefficient of restitution from 1.0 to 0.99 reversed the
core-annulus profile, as shown in Figure 20. This was especially concerning, since a typical coefficient
of restitution for particles used in CFB applications was approximately 0.8 to 0.95.
Hrenya and Sinclair (1997) were able to solve this sensitivity to the coefficient of restitution by add-
ing another analog of fluid hydrodynamics to the particle momentum equation, solid phase turbulence.
This was an expansion of a proposal by Dasgupta et al (1994). Hrenya time-averaged the Navier-Stoke’s
momentum equations, which generated a Reynolds stress term. Hrenya used a κ-ε equation to close this
stress term, which also removed the model’s sensitivity to the coefficient of restitution. It was presumed
that solid-phase turbulence could be visualized with the cluster and streamers commonly observed in
riser studies. This is not unfounded as several experimental studies have focused on these fluctuations
in the suspension density (Yerushalmi, 1978; Savage & Sayed, 1984; Plumpe, 1993). It was the cluster
and streamers that provided the dissipation of granular temperature in the annular region and propa-
gated the expected core-annular profile.
Others proposed that this sensitivity was resulting from a deficiency of the drag models and not due to
a solid phase turbulence (O’Brien & Syamlal, 1993; Boemer, 1994). Indeed, several papers [Sun, 1996;
Benyahia, 1998; Samuelsberg, 1996) have shown that the core-annulus profile in a riser can be modeled
with realistic coefficient of restitution without a solid phase turbulence addition. This discrepancy in
modeling methods still exists today.
The role of gas-phase turbulence is another contentious issue with modeling granular fluid systems.
To date, only a few have incorporated gas-phase turbulence in their riser models. Louge et al. (1991) was
the first to incorporate gas-phase turbulence with a KTGF model. Bolio and Sinclair (1995) expanded
on Louge’s efforts with the addition of a low Reynolds number, κ-ε gas phase turbulence model. Their
results suggest that the wake behind particles is important to gas turbulence modulation and can even
enhance turbulence for the larger particles. Smaller particles may have an opposite effect by damping
eddies. Gore and Crowe (1989) also suggested that gas-phase turbulence increases or decreases depend-
ing on the ratio of particle diameter to the characteristic length of the more energetic eddies. Hetsroni
(1989) noted that this was dependent on the particle Reynolds number. All of this suggested that smaller
particles (i.e, < 100 microns) decrease the turbulence intensity by damping eddies whereas larger par-
ticles (i.e., > 100 microns) propagate and maybe enhance turbulence, giving rise to to vortex shedding
336
Figure 20. Radial solids volume fraction with respect to the coefficient of restitution in the Eulerian-
Eulerian model of Sinclair and Jackson (based on Hrenya, 1996)
and wakes from the particles. Vreman et al. (2009) used a Eulerian-Lagrangian technique with Large
Eddy Simulation (LES) to show that turbulence in the normal and spanwise directions was reduced
with particle concentration but increased in the streamwise direction for coarse particles. Kashiwa and
Hull (2004) proposed that gas and solid phase turbulence terms were coupled and added an interphase
transfer term in both the gas and solid phase turbulence expressions. This momentum transfer term was
in addition to the more traditional drag force transfer term. Ahmadi et al. (1990) and Simonin (1996)
have been developing new turbulence models more specific towards particle-laden flows that do not
use the traditional single-phase gas turbulence models of the past. Similar to Kashiwa and Hull, they
propose that the gas-phase turbulence affects particle hydrodynamics directly. Recently, Ahmadi and
Simonin’s work has been directed towards gas phase turbulence in riser-type flows (de Wilde, 2005;
Xu & Subramaniam, 2006).
Still, most CFD models for CFBs consistently neglect the role of gas-phase turbulence, even for
systems having large particles. This mostly results from the limited data available to validate turbulence
and particle interaction. However, the Moran and Glicksman [2003] experimental work on a ¼-scale
2 MWth combustor and hot wire anemometry suggests that the no turbulence assumption may not be
valid, even for dense systems. Their work suggested that large-scale, gas velocity fluctuations were
caused by particle clustering.
Another issue with the Eulerian-Eulerian model is that it tends to under-predict the particle concentra-
tion in a riser. McKeen and Pugsley (2003) attributed this to a deficiency in the drag relationship, and
could correct for it by artificially increasing the particle size. Realistic solid volume fractions could be
obtained with a particle size multiplier of approximately 5. Similar arguments were also proposed by
Yang et al. (2004). Both studies highlight that perhaps empirically-based, single-particle drag correla-
tions fall short with the complex cluster and streamer hydrodynamics observed in risers.
Sundaresan and coworkers (Agrawal, 2001; Andrews, 2005) have been focusing on this deficiency.
They proposed that the current drag laws are only applicable on the micro-scale and can not be realisti-
337
cally used to describe macro-scale events. In other words, a drag law may be applicable for one or a few
particles; but, it may not be applicable when applied to dense systems where grid resolutions are on the
order of inches or even feet. Indeed, Wang, van der Hoef, and Kuipers (2009) showed that reducing the
grid size from 0.5 mm to 0.2 mm resulted in matching bed expansions measurements for a Geldart Group
A powder. Typically, larger grid sizes are used, sometimes on the order of feet. Yet, these models have
almost always over predicted bed expansion in a fluidized bed for Geldart Group A powders. Since this
was not observed with Geldart Group B or D powders, it was assumed not to be attributed to the grid
size but to Group A powders having more cohesive forces (McKeen & Pugsley, 2003; Zimmermann &
Taghipour, 2005). However, Wang’s research showed that once the lack of scale has been addressed,
there are no fundamental differences between Geldart Group A, B or D particles. This, however, presents
a problem. How can large, commercial CFBs with small particle sizes using gird sizes on the scale of
0.2 mm be accurately modeled? The current computational requirements would be prohibitive.
To get around this, Andrews et al. (2005) proposed a systematic approach to develop a coarse-grain
relationships in drag and solid stresses. They proposed that highly-resolved simulations on small rep-
resentative systems be conducted first. The results are filtered at various length scales to provide the
appropriate closure relationships. Chen and Jones (2008) used this approach for large-scale modeling
of commercial cokers. Details of Andrews’ filtering scheme can be found in Chapter 1.
Kuipers and coworkers (van der Hoef, 2005) took a different approach to address the deficiencies in
the drag laws. They derived a drag relationships from the numerical data of a flow field around station-
ary particles using a discretized version of the Boltzmann equation or Lattice Boltzmann method. The
grid resolution was 10 to 30 times smaller than the particle diameters. Similar work was conducted at
Sandia National Lab with a RANS-DEM approach (Lechman, 2006). Benyahia, Syamlal and O’Brien
(2006) later adapted the Kuiper and van der Hoef’s drag law to be applicable over the full range of void
fractions encountered in risers and fluidized beds. Igci et al. (2008) has recently adapted van der Hoef’s
model as a subgrid or filtered model (see Chapter 2).
Despite advances in modeling particle drag and collisional stresses, one underlying factor, particle
size distribution, has been neglected until recently. Particle size has a significant effect on gas and par-
ticle hydrodynamics. Addition of fines (i.e., particles smaller than 44 μm or 325 mesh) has been found
to reduce particle attrition in pneumatic conveying lines (Knowlton, 1994). A small addition of fines can
significantly lower the bed density in a fluidized bed (Pell, 1988; Grace & Sun, 1991) as well as improve
heat transfer (Lee, 997). Similarly, particle segregation has been observed in risers (Karri & Knowlton,
1998; Mitali, 2008). Yet, most Eulerian-Eulerian codes consider only one, maybe two, representative
particle sizes for their simulation models.
The reason for this omission is due to the constitutive equations used in the Eulerian-Eulerian framework.
The KTGF for modeling collisional stresses and to a lesser extent the drag law assume a monodispersed
particle size. Specifically, the KTGF assumes a Maxwellian velocity distribution and the equipartition
of energy with the granular temperature. This underlying assumption limits smaller particles from hav-
ing a sufficiently different granular temperature than the larger particles (Galvin, 2005). In other words,
limitations in the KTGF would prevent the model from accurately representing particle segregation in
a riser or the possibility of fines influencing riser hydrodynamics. Dahl and Hrenya (2005) showed this
with experimental data on radial particle segregation in a fluidized bed.
Fan, Marchisio and Fox (2004) addressed particle size distribution for Eulerian-Eulerian models by
coupling the Navier-Stokes equations to population balances using a direct quadrature method of mo-
ments (DQMOM). This was a significant first step; but, the underlying assumptions in the KTGF were
338
still an issue. Iddir, Arastoopour and Hrenya [2005] rederived the KTGF for binary and ternary mixtures
of particles without the underlying velocity and equalpartition assumptions for a simple shear flow. Their
model showed that fluctuations were strongly tied to fines level. Increases in fines concentrations resulted
in a damping of fluctuations. Their results were in good agreement with the experimental work of Sav-
age and Sayed (1984). Recently, Fox (2009) has developed a quadrature-based third-order momentum
expression with a velocity distribution function. This removes the Maxwellian limits with particle size
distributions. Yin and Sundaresan (2009) expanded on Kuiper and van der Hoef’s drag model for poly-
dispersed systems by including the cross products of each particle size (i.e., drag from one particle to
another and vice versa). Details of this method as applied to polydispersity can be found in Chapter 2.
Still, simulating fluidized beds and risers with the full particle size distribution remains elusive within
the Eulerian-Eulerian framework. Efforts are currently underway to address polydispersed particle
sizes; but until then, doubt will always exists that Eulerian-Eulerian models can capture granular-fluid
hydrodynamics.
lagrangian - eulerian Models
Lagrangian - Eulerian models do not suffer from the particle size limitations of Eulerian-Eulerian mod-
els. Since the trajectory of each and every particle is tracked, a unique particle size and density can be
given to each particle. This benefit, however, is also a “curse” because each particle comes with its own
equation of motion. Typical commercial systems have trillions of particles, and a corresponding num-
ber of equations would be needed just for the particle phase. This is too large of a burden with today’s
computer resources.
Yet, the Lagrangian - Eulerian method has several distinct advantages over Eulerian-Eulerian model-
ing. Particles are modeled individually, and tracking is not tied to any one coordinate system. Trajectories
are not averaged out as in the ensemble integration that is inherent to Eulerian models. Also, Lagrangian
equations are unambiguous and consistent. Diverging solutions are more of an issue within the Eulerian
framework.
Lagrangian - Eulerian models come in many different flavors. The Lagrangian framework refers to the
particles motion using classical mechanics or Newton’s law of motion. In simple terms, each particle’s
trajectory is influenced by the sum of forces on it as

d vs

Pp = β(v g − vs ) − ∇P + ρ p g [5]
dt
Initially, Lagrangian-Eulerian modeling referred to particles under the influence of the fluid only.
In other words, the particle did not impose any momentum on the fluid. This is a valid assumption for
small particles at low concentrations of particles. This method is often called one-way coupling. In CFB
applications, such a model would only be applicable in the second and third stage cyclones. It also has
been applied to dilute-phase pneumatic conveying lines (Tsuji, 1985).
Later, many commercial codes added two-way coupling (i.e., Equation 3 is coupled with Equation 1).
The particle is influenced and influences the fluid phase and vice versa. Meier and Mori (1998) used a
two-way coupling Eulerian-Lagrangian model for particle hydrodynamics in primary cyclones. Tanaka
and Tsuji (1991) also applied a similar model for dilute-phase risers operating with large aluminum
339
spheres (dp,50 ~ 4 mm). They later applied this model to fluidized beds (Tsuji, Kawaguchi, & Tanaka,
1993). Interestingly enough, most Lagrangian- Eulerian models treat the drag as a body force and use
the same empirically-based drag expression used in Eulerian-Eulerian modeling. Hence, the limitation
in the drag force equations also apply to Lagrangian-Eulerian modeling. In addition, drag forces are
treated as body forces. The fluid flow around the particles is not resolved.
Early Lagrangian modeling often neglected particle-particle collisions, which is a valid assumption
for very low particle concentrations. However, many newer models are incorporating particle-particle
and particle-wall collisions. Sometime this is referred to as discrete element modeling (DEM) or direct
numerical simulations (DNS) as well as discrete particle modeling (DPM). It should be cautioned that
granular models with no fluid phase are also called DEM and DNS as well.
Collisions in the Lagrangian - Eulerian framework can be modeled as hard sphere or soft sphere col-
lisions. The rigorous modeling of particle collisions or hard sphere modeling is commonly not applied
to granular fluid problems. This is primarily because discrete time integration of a particle collision is
computationally intensive. How small of a time step is needed to capture the collision of two particles?
What if the time steps are too big and particle trajectories are wrong or missed?
To get around the hard sphere problem, many Lagrangian - Eulerian models use a soft sphere model
where a predetermined degree of particle-to-particle overlap is permitted. This reduces time step require-
ments and allows faster solution times. However, the level of permissible overlap is an unknown and
needs to be resolved in a manner similar to that of resolving a grid with an Eulerian-Eulerian model.
Particle rotation is another aspect that may need to be considered for models of CFBs. Goldschmidt,
et al. (2004)showed that without particle rotation in their DPM simulation, they were unable to capture
the correct bubble coalescence behavior. This led to small zones of particle circulation instead of the
expected large circulation of particles in a fluidized bed. Only by incorporating particle rotation was
good agreement with the data achieved. They also noted that the KTGF used for collisional stresses in
Eulerian-Eulerian models did not have a particle rotation correction factor in the coefficient of restitution.
Lagrangian-Eulerian models offer the advantage of rigorously solving for the particle collisions.
Thus, such a model would not be reliant on the particle stress constitutive equations used in the Eulerian-
Eulerian framework. However, Lagrangian-Eulerian models are still limited by the drag force expres-
sions and computational requirements for such models are still prohibitive for commercial-scale units.
Hybrid Models
The third type of model used for simulating CFBs is commonly called a hybrid model. The fluid phase
is still treated under the Eulerian framework just as in the other two types of modes. The solid phase
is described as a Lagrangian phase, just as the previously discussed model. However, unlike Eulerian-
Lagrangian models where particle collisions are either rigorously captured or ignored. With the hybrid
models, collisions are captured with similar constitutive equations used in Eulerian-Eulerian models.
Hence, each particle can be tracked, but the collisions are approximated with a suitable collisional stress
model such as the KTGF or that of Harris and Crighton (1994).
Much of this work stemmed from stochastic particle methods to capture droplets as a dispersed phase.
Based on Harlow’s (1984) single phase particle-in-cell (PIC) method, O’Rourke et al. (1993) reformulated
Harlow’s method to include a fluid or continuum phase called multiphase-particle-in-cell or MP-PIC.
Snider, O’Rourke and Andrews (1998) later successfully applied their MP-PIC method for sedimentary
flow on inclines. This method was later further developed and commercialized by Snider and Williams
340
Figure 21. Simulated solids volume fraction in 0.9-meter (3-foot) diameter fluidized bed with FCC
catalyst containing 3 or 9% files
under the name Arena-Flow™ and applied towards the automotive industry to predict the integrity of
sand forms used in foundries. The Arena-Flow™ was later reformulated for the chemical and petroleum
industry under the name Barracuda™ by CPFD-Software. A similar formulation was done with a code
called Glacier™ from Reaction Engineering International and Fluent™.
MP-PIC models like Barracuda™ and Glacier™ use the same Navier-Stokes equation as that used
in the previous models (i.e., Equation 1). Momentum interchange between phases is still done through
the particle drag term. The equation of motion for the particles is similar to Equation 5 except for the
addition of the collisional stress term. The time evolution of a particle’s momentum can be obtained
from the Liouville equation for particle distribution functions (Williams, 1985) where

∇P ∇τ
∂φ
+ ∇ ⋅ (φυ s ) + ∇υ (φA) = 0 and A = β(υ g − υ s ) − +g− [6]
∂t ρp ε pρ p
The probability distribution function, ϕ, describes the particle with respect to position, velocity,
particle density and particle volume (Andrews and O’Rourke, 1996). Momentum transfer to and from
each particle in the fluid is done with the integration transfer function of
∇P
F= ∫∫∫ φε ρ [ D (u − υ) −
p p p
ρp
]d ε p d ρ p d υ [7]
341
Figure 22. Simulated dynamic pressure in 0.9-meter (3-foot) diameter fluidized bed with FCC catalyst
containing 3 or 9% files
However, the MP-PIC method still has the same underlying problem with computer requirements
as the the Eulerian-Lagrangian model. Each particle needs to be tracked and commercial units typically
have trillions of particles.
Both the Arena-Flow™ and Barracuda™ codes solve this problem by reducing the number of equa-
tions using computational parcels or clouds. Particles can be grouped into these parcels or clouds where
each contain np particles with identical density, volume, velocity, and position. Drag forces and colli-
sional stresses are simply adjusted with the np multiplier. The Liouville equation (Equation 6) conserves
the particle numbers in volumes moving along dynamic trajectories in particle phase space. Thus, the
number of particles associated with a parcel is constant in time (Snider, 2001).
The real strength of Barracuda™ is it is the first commercial code that can model the entire particle
size distribution for large-scale systems. Thus, gas and particle hydrodynamic behavior affected by fines
levels can be modeled. An example of this behavior can be observed with the gas bypassing data of
Knowlton et al. (2004) and Issangya et al. (2007). Geldart Group A powders with low fines (dp50<44
μm) concentration may be prone to gas bypassing if the bed is deep enough. Figure 21 and 22 show how
Barracuda captured whether gas bypassing can occur with FCC catalyst having 3 or 9% fines (Cocco et
al., 2008). Simulation results suggest that a lower fines level in the bed results in a decrease in the bed
permeability. At some point, pressure exceeds a threshold (or point of instability) and gas bypassing
occurs, as shown in Figure 21. The simulations highlighted that grid pressure drop was not the culprit.
342
Figure 23. Simulation comparison to axial bed density data of Bader [1994] in a 0.2-meter (8-inch) ID
x 21.9-meter (72-foot) tall riser
The higher fines level resulted in a more uniform distribution of gas and pressure in the bed. Results
were in agreement with the published experimental work of Issangya (2007).
This MP-PIC method was also validated with the two riser data sets of Bader et al. (1994) and
Karri et al. (2001). As shown in Figures 23 through 26, simulations were in good agreement in predict-
ing the axial pressure drop and radial solids volume fraction (Cocco & Williams, 2005). Similar valida-
tion efforts have been made with cyclone data (Cocco, 2006a) and jet penetration data in a fluidized bed
including high pressure fluidized beds (Cocco, 2006b). Snider (2008) added kinetics to the Barracuda™
code and compared simulated results with ozone decomposition data in a fluidized bed (Fryer & Potter,
1976). Computational results were also in good agreement with published data as shown in Figures 27
and 28.
With particle technology, particle size distribution is often the major influencing parameter in unit
operation performance. The MP-PIC method is unique in that it models the entire particle size distribu-
tion. Large problems can be simplified by grouping particles in parcels or clouds. Currently Barracuda™
stands alone with that capability as a commercial code, but as noted above, work is continuing in giving
Eulerian-Eulerian models similar capability (Fox, 2008).
suMMary
The hydrodynamics in a riser or fluidized bed are complex with both micro and macro-scale features.
From particle clustering to large streamers to the core-annulus profile, the particle behavior in these
unit operations rarely behaves as a continuous fluid. However, a collection of drag and particle stress
relationships have been developed (and continue to be developed) over the last 20 years that bridge this
343
Figure 24. Simulation comparison to radial solids volume fraction data of Bader [1994] in a 0.2-meter
(8-inch) ID x 21.9-meter (72-foot) tall riser
Figure 25. Simulation comparison to axial bed density data of Karri [2001] in a 0.2-meter (8-inch) ID
x 21.9-meter (72-foot) tall riser
gap between fluid and particle hydrodynamics. Today, commercial codes are available that can model
some of this behavior in large-scale, commercial units in a reasonable amount of time (i.e., weeks).
However all codes are not alike. It is critical for the modeler to understand the basic assumptions
used in model development, including the constitutive equations, and recognize what impact each equa-
344
Figure 26. Simulation comparison to radial solids volume fraction data of Karri [2001] in a 0.2-meter
(8-inch) ID x 21.9-meter (72-foot) tall riser
Figure 27. Calculated O3 mass fraction and catalyst-solids colored by volume fraction in a 0.23 meter
diameter fluidized bed with superficial gas velocity of 0.11 m/sec and bed height of 0.11 meters (based
on Banerjee et al., 2007)
tion would have on the results. If particle size distribution or segregation is a factor, then a model that
uses a one (or two or three) particle size representation may not be appropriate. The domain is another
aspect that needs careful consideration. All too often just the riser is modeled with the assumption that
the inlet boundary condition is known. However, that may not be the case. The simulations in Figures
345
Figure 28. Validation of O3 concentrations with the data of Fryer and Potter (1976) in a 0.23 meter
diameter fluidized bed with a superficial gas velocity of 0.11 m/sec and bed height of 0.11 meters (based
on Snider, 2008)
23 through 26 were only in good agreement once the riser termination and the incoming standpipe were
modeled, with all the aeration as well.
Grid resolution is also an issue that needs careful consideration. Modeling lab-scale or pilot-scale
units may result in grid sizing on the order of a few inches or less. Some studies have suggested the
grid should only be 10 to 20 particle diameters in size. Modeling commercial scale units may push grid
sizing to the order of feet. If a hybrid model is being used, the number of particles represented by a
parcel or cloud needs to be resolved as well. Many of the constitutive equations used in these models
were developed using data from small scale units and validated in small scale units. Thus, more subgrid
models or filters are needed and they need to be validated.
Also, there is more to CFBs than just gas and particle hydrodynamics. Both FCC units and gasifiers
may need to consider radiative heat transfer. Little work has been done in this area for dense particle
flows. Heat transfer in general could use more validation. For FCC units, liquid injection (i.e., oil in-
jection) is another area that would be important to a model. Some commercial codes already have this
capability, but again are the underlying assumption in model development appropriate to the problem?
Simulations with reactions also present additional complications. Slow kinetic reactions can be mod-
eled today. However, fast reactions make micro-mixing the rate determining factor. Are enough of the
micro-hydrodynamics being captured to drive these reactions? The user must always know which time
scale is important.
In the last 20 years, significant improvements have been made that will allow the simulation of
large-scale, commercial CFBs. A talented set of researchers and code developers continues to address
this problem. The next 20 years should be as remarkable as the last.
346
references
Abed, R. (May 29 to June 2, 1983). The Characterization of Turbulent Fluid-Bed Hydrodynamics. In D.

Kunii & R. Toei (Eds.), Internl. Conf. on Fluidization IV, Kashikijima, Japan, (p. 137).
Ahmadi, G., & Ma, D. (1990). A thermodynamical formulation for dispersed multiphase turbulent flows.
I. International Journal of Multiphase Flow, 16(22), 323–340. doi:10.1016/0301-9322(90)90062-N
Anderson, T. B., & Jackson, R. (1968). Fluid Mechanical Description of Fluidized Beds. Stability of Stte
of Uniform Fluidization. Industrial & Engineering Chemistry Fundamentals, 7(1), 12–21. doi:10.1021/
i160025a003
Andrews, A. T., Loezos, P. N., & Sundaresan, S. (2005). Coarse-Grid Simulation of Gas-Particle Flows in
Vertical Risers. Industrial & Engineering Chemistry Research, 44(16), 6022–6037. doi:10.1021/ie0492193
Andrews, M. J., & O’Rourke, P. (1996). The multiphase particle-in-cell (MP-PIC) method for dense
particulate flows. International Journal of Multiphase Flow, 22(2), 379–402. doi:10.1016/0301-
9322(95)00072-0
Arastoopour, H., & Gidaspow, D. (1979). Vertical pneumatic conveying using four hydrodynamic models.
Powder Technology, 18(2), 123–130.
Arastoopour, H., Lin, S., & Weil, S. (1982). Analysis of vertical pneumatic conveying of solids using
multiphase flow models. AIChE Journal. American Institute of Chemical Engineers, 28(3), 467–473.
doi:10.1002/aic.690280315
Arastoopour, H., Pakdel, P., & Adewumi, M. (1990). Hydrodynamics analysis of dilute gas-solids flow
in a vertical pipe. Powder Technology, 62(2), 163–170. doi:10.1016/0032-5910(90)80080-I
Bader, R., Findlay, J., & Knowlton, T. M. (1988). Gas-solid flow pattern in a 30.5 cm diameter circulating
fluidized bed. In Basu, P., & Large, J. F. (Eds.), Circulating Fluidized Bed Technology II (pp. 123–127).
New York: Pergamon Press.
Banerjee, S., Patel, R., Chen, A., Snider, D., Williams, K., O’Hern, T., & Tortora, P. (2007, Nov. 30).
Enhanced Productivity of Chemical Processes Using Dense Fluidized Beds. DOE Award DE-FC36-
04GO14153.
Banyahia, S., Arastoopour, H., & Knowlton, T. M. (1998). Prediction of Solid and Gas Flow Behavior
in a Riser Using a Computational Multiphase Flow Approach. In Fan, L.-S., & Knowlton, T. M. (Eds.),
Fluidization IX (pp. 493–498). New York: Engineering Foundation.
Benyahia, S., Syamlal, M., & O’Brien, T. J. (2006). Extension of Hill–Koch–Ladd drag correlation
over all ranges of Reynolds number and solids volume fraction. Powder Technology, 162(2), 166–174.
doi:10.1016/j.powtec.2005.12.014
Boemer, A., Qi, H., Hannes, J., & Renz, U. (1994). Modeling of solids circulation in a fluidised bed with
Eulerian approach. Paper presented in 29th IEA-FBC Meeting, Paris, France.
347
Breault, R. W., Guenther, C. P., & Shadle, L. T. (2008). Velocity fluctuation interpretation in the near
wall region of a dense riser. Powder Technology, 182(2), 137–145. doi:10.1016/j.powtec.2007.08.018
Breault, R. W., Ludlow, C. J., & Yue, P. C. (2005). Cluster particle number and granular temperature
for cork particles at the wall in the riser of a CFB. Powder Technology, 149(2-3), 68–77. doi:10.1016/j.
powtec.2004.11.003
Brereton, C. M., & Grace, J. R. (1994). End Effects in Circulating Fluidized Bed Hydrodynamics. In
Avidan, A. A. (Ed.), Circulating Fluidized Bed Technology IV (pp. 137–146). Hidden Valley, PA.
Chen, A., & Jones, N. (2008). Application of Coarse Grained Drag Law In Computational Fluid Dynamics
Simulations of Fluidized Beds. Paper presented in AIChE Annual Meeting, Philadelphia, Pennsylvania.
Cocco, R. (2006). Understanding the Gas Residence Time in a Cyclone. Paper presented World Congress
in Particle Technology, Orlando, FL, 2006.
Cocco, R. (2006). Jet Hydrodynamics Around Fluidized Bed Spargers. Orlando, FL: World Congress
in Particle Technology.
Cocco, R., & Williams, K. (2004). Optimization of Particle Residence Time INside Commercial Dryers
with Arena-Flow. Paper presented in AIChE Annual Meeting, Austin, Texas.
Cocco, R., & Williams, K. (2005). Riser Simulations with Arena-flow: Application to the PSRI 1994 and
2001 Challenge Problems. Paper presented in AIChE Annual Meeting, Cincinnati, Ohio.
Contractor, R. (1999). Dupont’s CFB technology for maleic anhydride. Chemical Engineering Science,
54(22), 5627–5632. doi:10.1016/S0009-2509(99)00295-X
Contractor, R., Anderson, M., Campos, D., Hecquet, G., Pham, C., Simon, M., et al. (Oct 30, 2001). Vapor
phase oxidation of acrolein to acrylic acid. DuPont de Nemours and Company, US Patent 6,310,240.
Contractor, R., Bergna, H. E., Horowitz, H. S., Blackstone, C. M., Chowdhry, U., & Sleight, A. W. (1988).
Butane oxidation to maleic anhydride in a recirculating solids reactor. In Ward, J. W. (Ed.), Catalysis
1987 (pp. 645–654). New York: Elsevier Science Publishers.
Contractor, R. M. (1991, June 4).Vapor Phase Catalytic Oxidation of Butane to Maleic Anhydride. U.S.
Patent 5,021,588.
Dahl, S. R., & Hrenya, C. M. (2005). Size segregation in gas–solid fluidized beds with continuous size
distributions. Chemical Engineering Science, 60(23), 6658–6673. doi:10.1016/j.ces.2005.05.057
Das, M., Meilkap, B. S., & Saha, R. K. (2008). Characteristics of axial and radial segregation of single
and mixed particle system based on terminal settling velocity in the riser of a circulating fluidized bed.
Chemical Engineering Journal, 145(1), 32–43. doi:10.1016/j.cej.2008.03.010
Dasgupta, S., Jackson, R., & Sundaresan, S. (1994). Turbulent gas-particle flow in vertical risers. AIChE
de Wilde, J., Van Engelandt, G., Heyndenickx, G. J., & Marin, G. B. (2005). Gas–solids mixing in the
inlet zone of a dilute circulating fluidized bed. Powder Technology, 151(1-3), 96–116. doi:10.1016/j.
powtec.2004.11.037
348
Ding, J., & Gidaspow, D. (1990). A bubbling fluidization model using kinetic theory of granular flow.
Ding, J., & Lyczkowski, R. (1992). Three-dimensional kinetic theory modeling of hydrodynamics and
erosion in fluidized beds. Powder Technology, 73(2), 127–138. doi:10.1016/0032-5910(92)80073-6
Fryer, C., & Potter, O. E. (1976). Experimental investigation of models for fluidized bed catalytic reactors.
Galvin, J. E., Dahl, S. R., & Hrenya, C. M. (2005). On the role of non-equipartition in the dynam-
ics of rapidly flowing granular mixtures. Journal of Fluid Mechanics, 528, 207–232. doi:10.1017/
S002211200400326X
Garg, S., & Pritchett, J. (1975). Dynamics of gas-fluidized beds. Applied Physics (Berlin), 46(10),
4493–4498.
Garzo, V., & Duffy, J. W. (2002). Hydrodynamics for a granular binary mixture at low density. Physics
of Fluids, 14(4), 1476–1490. doi:10.1063/1.1458007
Gelbein, A. (April 14, 1981). Phthalic anhydride reaction system. US Patent 4,261,899.
Geldart, D. (1986). Gas Fluidization Technology. Chichester, UK: Wiley & Sons.
Gidaspow, D., Bezburuah, R., & Ding, J. (1992). Hydrodynamics of Circulating Fluidized Beds, Kinetic
Theory Approach. In Potter, O. E., & Nicklin, D. J. (Eds.), Fluidization VII (pp. 75–82). New York:
Gidaspow, D., & Huilin, L. (1998). quation of state and radial distribution functions of FCC particles
in a CFB. AIChE Journal. American Institute of Chemical Engineers, 44(2), 279–293. doi:10.1002/
aic.690440207
Gidaspow, D., Seo, Y., & Ettehadieh, B. (1983). Hydrodynamics of fluidization: Experimental and
theoretical bubble sizes in a two-dimensional bed with a jet. Chemical Engineering Communications,
22(5), 253–272. doi:10.1080/00986448308940060
Goldschmidt, M. J. V., Beetstra, R., & Kuipers, J. A. M. (2004). Hydrodynamic modelling of dense
gas-fluidised beds: comparison and validation of 3D discrete particle and continuum models. Powder
Technology, 142(1), 23–47. doi:10.1016/j.powtec.2004.02.020
Gore, R. A., & Crowe, C. T. (▪▪▪). Effect of particle size on modulating turbulent intensity. International
Journal of Multiphase Flow, 15(22), 279–285.
Grace, J., & Sun, G. (1991). Influence of particle size distribution on the performance of fluidized bed
reactors. Canadian Journal of Chemical Engineering, 69(5), 1126–1134. doi:10.1002/cjce.5450690512
Grassler, T., & Wirth, K. E. (1999). Radial and Axial Profile of Solids Concentration in a High Loaded
Riser Reactor. In Werther, J. (Ed.), Circulating Fluidized Bed Technology VI (pp. 65–70). Frankfort:
DECHEMA.
349
Guenther, C., & Breault, R. (2007). Wavelet analysis to characterize cluster dynamics in a circulating
fluidized bed. Powder Technology, 173(3), 163–173. doi:10.1016/j.powtec.2006.12.016
Guenther, C., & Syamlal, M. (2001). The effect of numerical diffusion on simulation of isolated bubbles in
a gas–solid fluidized bed. Powder Technology, 116(2-3), 142–154. doi:10.1016/S0032-5910(00)00386-7
Guenther, C., Syamlal, M., Longanback, J., & Smith, P. (2003, Nov.). Two-Fluid Model of an Industrial
Scale Transport Gasifier. Presented at AIChE Annual Meeting, Conference Proceedings, San Francisco,
California.
Gupta, P., Velazquez-Vargas, L. G., Li, F., & Fan, L. S. (2005, Oct.). Chemical looping combustion of
coal. Presented at AIChE Annual Meeting, Conference Proceedings, Cincinnati, Ohio.
Harlow, F. H. (1964). The particle-in-cell computing method of fluid dynamics. In Alder, B., Fembach,
S., & Rotenberg, M. (Eds.), Fundamental Methods in Hydrodynamics. New York: Academic Press.
Harris, A. T., Davidson, J. F., & Thorpe, R. B. (2003a). Influence of exit geometry in circulating fluid-
ized-bed risers. AIChE Journal. American Institute of Chemical Engineers, 49(1), 52–64. doi:10.1002/
aic.690490107
Harris, A. T., Davidson, J. F., & Thorpe, R. B. (2003b). The influence of the riser exit on the particle
residence time distribution in a circulating fluidised bed riser. Chemical Engineering Science, 58(16),
3669–3680. doi:10.1016/S0009-2509(03)00215-X
Harris, S. E., & Crighton, D. G. (1994). Solitons, solitary waves, and voidage disturbances in gas-fluidized
Hartge, E. U., Rensner, D., & Werther, J. (1988). Solid Concentration and Velocity Patterns in Circu-
lating Fluidized Beds. In Basu, P., & Large, J. F. (Eds.), Circulation Fluidized Bed Technology II (pp.
165–179). Oxford, UK: Pergamon Press.
Hetsroni, G. (1989). Particles-turbulence interaction. International Journal of Multiphase Flow, 15(5),
735–746. doi:10.1016/0301-9322(89)90037-2
Horio, M. (1997). Hydrodynamics. In Grace, J. R., Avidan, A. A., & Knowlton, T. M. (Eds.), Circulating
Fluidized Beds (pp. 21–78). London: Blackie Academic & Prof.
Hrenya, C. (1996). Predicting Dense, Turbulent Gas-Solid Flows in Vertical Risers. Doctoral Disserta-
tion, Carnegie Mellon University, Pittsburgh, PA.
Hrenya, C., & Sinclair, J. (1997). Effects of particle-phase turbulence in gas-solid flows. AIChE Journal.
American Institute of Chemical Engineers, 43(4), 853–869. doi:10.1002/aic.690430402
1431–1448. doi:10.1002/aic.11481
Issangya, A. S., Bai, D., Grace, J. R., & Zhu, J. (1998). Solids Flux Profiles in a High Density Circulat-
ing Fluidized Bed Riser. In Fan, L.-S., & Knowlton, T. M. (Eds.), Fluidization IX (pp. 197–204). New
York: Engineering Foundation.
350
Issangya, A. S., Knowlton, T. M., & Karri, S. B. R. (2007). Detection of Gas Bypassing due to Jet Stream-
ing in Deep Fluidized Beds of Group A Particles. In Bi, X., Berruti, F., & Pugsley, T. (Eds.), Fluidization
XII (pp. 775–782). New York: Engineering Conferences International.
Johnson, P., & Jackson, R. (1987). Frictional–collisional constitutive relations for granular materials, with
application to plane shearing. Journal of Fluid Mechanics, 176, 67–93. doi:10.1017/S0022112087000570
Karri, R., & Knowlton, T. M. (2002) Wall Solids Upflow and Downflow Regimes in Risers for Group
A Solids. In J.R. Grace, J.X. Zhu, H. de Lasa (Eds.), Circulating Fluidized Bed Technology VII, (pp.
310-316). Ottawa: CSChE.
Karri, S. B. R., & Knowlton, T. M. (1991). A Practical Definition of the Fast Fluidization Regime. In
Basu, P., Horio, M., & Hasatani, M. (Eds.), Circulating Fluidized Bed III (pp. 67–72). Oxford, UK:
Pergamon Press.
Karri, S. B. R., & Knowlton, T. M. (1998). Flow Direction and Size Segregation of Annulus Solids in a
Riser. In Fan, L.-S., & Knowlton, T. M. (Eds.), Fluidization IX (pp. 189–195). New York: Engineering
Foundation.
Karri, S. B. R., & Knowlton, T. M. (August 29-31, 2001). The PSRI Challenge Problem. Paper presented
at MFDRC Meeting, Salt Lake, Utah.
Kashiwa, B. A., & Hull, L. M. (Aug 9-12, 2004). Multifield Closure Modeling for Metal-Loaded High
Explosives. Paper presented in 7th Joint Classified Bobs/Warheads & Ballistics Symposium, Monterey,
CA, Los Alamos National Lab Report LA-UR-045081.
Khazai, B., Vrieland, E., Murchison, C., Dixit, R., & Weihl, E. (Sept 8, 1992). Process of oxidizing
aliphatic hydrocarbons employing a molybdate catalyst composition. US Patent 5,146,031.
Knowlton, T. M., Carson, J. W., Klinzing, G. E., & Yang, W.-C. (1994). The Importance of Storage,
Transfer and Collection. Chemical Engineering Progress, 90, 44–54.
Knowlton, T. M., Karri, S. B. R., & Issangya, A. (2005). Scale-up of fluidized-bed hydrodynamics.
Lackermeier, U., & Werther, J. (2002). Flow phenomena in the exit zone of a circulating fluidized bed.
Chemical Engineering Progress, 41(9), 771–783. doi:10.1016/S0255-2701(02)00008-9
Lechman, J. (2006). Computational and Numerical Approaches to Particle Flow. Paper presented at the
AIChE Annual Meeting, San Francisco, California.
Lee, Y. Y. (1997). Design considerations for CFB boilers. In Grace, J. R., Avidan, A. A., & Knowlton,
T. M. (Eds.), Circulating Fluidized Beds (pp. 417–422). New York: Blackie Academic & Prof.
Lewis, W. K., Gilliland, E. R., & Bauer, W. C. (1949). Characteristics of Fluidized Particles. Industrial
& Engineering Chemistry, 41(6), 1104–1117. doi:10.1021/ie50474a004
Li, Y., & Kwauk, M. (1980). The Dynamics of Fast Fluidization. In Grace, J. R., & Matsen, J. M. (Eds.),
Fluidization (pp. 537–542). New York: Plenum.
351
Lints, M. C., & Glicksman, L. R. (1993). Parameters governing particle-to-wall heat transfer in a circu-
lating fluidized bed. In A.A. Avidan (Ed.), Proceeding of the International Conference on Circulating
Fluidized Beds (pp. 297-203), Somerset, PA.
Lomas, D. (Dec. 17, 1996). FCC Separation Method and Apparatus with Improved Stripping. US Pat-
ent 5,584,985.
Louge, M. Y., Mastorakos, E., & Jenkins, J. T. (1991). The role of particle collisions in pneumatic trans-
port. Journal of Fluid Mechanics, 231, 345–359. doi:10.1017/S0022112091003427
Lun, C., & Savage, S. (1986). The Effects of an Impact Velocity Dependent Coefficient of Restitution on
Stresses Developed by Sheared Granular Materials. Acta Mechanica, 63, 15–44. doi:10.1007/BF01182538
Lun, C., Savage, S., Jeffrey, D., & Chepurniy, N. (1984). Kinetic theories for granular flow: inelastic
particles in Couette flow and slightly inelastic particles in a general flowfield. Journal of Fluid Mechan-
ics, 140, 223–256. doi:10.1017/S0022112084000586
Martin, M. P., Turlier, P., Wild, G., & Bernard, J. R. (1992). Gas and solid behavior in cracking circulat-
ing fluidized beds. Powder Technology, 70(3), 249–258. doi:10.1016/0032-5910(92)80060-A
Mckeen, T., & Pugsley, T. (2003). Simulation and experimental validation of a freely bubbling bed of
Meier, H. F., & Mori, M. (1998). Gas-solid flow in cyclones: The Eulerian-Eulerian approach. Comput-
ers & Chemical Engineering, 22(1), S641–S644. doi:10.1016/S0098-1354(98)00114-8
Monazam, E. R., & Shadle, L. J. (2004). A transient method for characterizing flow regimes in a circu-
lating fluid bed. Powder Technology, 139(1), 89–97. doi:10.1016/j.powtec.2003.10.007
Noymer, P. D., & Glicksman, L. R. (1998). Cluster motion and particle-convective heat transfer at the
wall of a circulating fluidized bed. International Journal of Heat and Mass Transfer, 41(1), 147–158.
doi:10.1016/S0017-9310(97)00065-3
Noymer, P. D., & Glicksman, L. R. (1999). Near-wall hydrodynamics in a scale-model circulating flu-
idized bed. International Journal of Heat and Mass Transfer, 42(8), 1389–1403. doi:10.1016/S0017-
9310(98)00238-5
Noymer, P. D., & Glicksman, L. R. (2000). Descent velocities of particle clusters at the wall of a circulating
fluidized bed. Chemical Engineering Science, 55(22), 5283–5289. doi:10.1016/S0009-2509(00)00171-8
O’Brien, T. J., & Syamlal, M. (1993). Particle Cluster Effects in the Numerical Simulation of a Circulat-
ing Fluidized Bed. In Avidan, A. A. (Ed.), Circulating Fluidized Beds IV (pp. 345–350). Somerset, PA.
O’Rourke, P., Brackbill, J., & Larrouturou, B. (1993). On Particle-Grid Interpolation and Calculating
Chemistry in Particle-in-Cell Methods. Journal of Computational Physics, 109(1), 37–52. doi:10.1006/
jcph.1993.1197
Ogawa, S., Umemura, A., & Oshima, N., J. (1980). On the equations of fully fluidized granular materi-
als. [ZAMP]. Applied Math. and Physics, 31(4), 483–493. doi:10.1007/BF01590859
352
Paisley, M. A., Farris, G., Slack, W., & Irving, J. (1997). Commercial Development of Battelle/FERCO
Biomass Gasification Process: Initial Operation of the McNeil Gasifier. In R.P. Overend, E. Chornet
(Eds.), Making a Business from Biomass in Energy, Environment, Chemicals, Fibers and Materials: Pro-
ceedings of the Third Biomass Conference in Americas (pp. 579-592) Oxford: Elsevier Science, Oxford.
Pell, M., & Jordan, S. P. (1988). Effects of fines and velocity on fluid bed reactor performance. Fluidiza-
tion Engineering: Fundamentals and Application: AIChE Symp. Ser., 84 (262) 68-71.
Pita, J., & Sundaresan, S. (1991). Gas-solid flow in vertical tubes. AIChE Journal. American Institute
Pita, J., & Sundaresan, S. (1993). Developing flow of a gas-particle mixture in a vertical riser. AIChE
Plumpe, J. G., Zhu, C., & Soo, S. L. (1993). Measurement of fluctuations in motion of particles in a dense
gas—solid suspension in vertical pipe flow. Powder Technology, 77(2), 209–214. doi:10.1016/0032-
5910(93)80057-H
Pritchett, J., Blake, T., & Garg, S. (1978). A Numberical Model of Gas Fluidized Beds. Fluidization:
Application to Coal Conversion Processes: AIChE Symp. Ser. 74 (176) 134-148.
Pugsley, T. S., Lapointe, D., Hirschberg, B., & Werther, J. (1997). Exit effects in circulating fluidized
bed risers. Canadian Journal of Chemical Engineering, 75(6), 1001–1010. doi:10.1002/cjce.5450750602
Pugsley, T. S., & Malcus, S. (1997). Partial Oxidation of Methane in a Circulating Fluidized-Bed Cata-
lytic Reactor. Industrial & Engineering Chemistry Research, 36(11), 4567–4571. doi:10.1021/ie970088y
Rall, R. (May 1, 2001). Apparatus for Contacting of Gases and Solids in Fluidized Beds. US Patent
6,224,833.
Rashidi, M., Hetsroni, G., & Banerjee, S. (1999). Particle-turbulence interaction in a boundary layer.
International Journal of Heat and Mass Transfer, 42, 935–949.
Reh, L. (1971). Fluid bed combustion in Processing, Environmental Protection and Energy Supply.
Chemical Engineering Progress, 67(2), 63–67.
Rhodes, M. J. (1989). The upward flow of gas/solids suspensions: Part 2: A practical quantitative flow
regime diagram for the upward flow of gas/solid suspensions. Chemical Engineering Research & De-
sign, 67, 30–37.
Rhodes, M. J., Mineo, H., & Hirama, T. (1992). Particle motion at the wall of a circulating fluidized
Rhodes, M. J., Sollaart, M., & Wang, X. S. (1998). Structure of the Dense-Dilute Interface in Fast Fluidi-
zation. In Fan, L.-S., & Knowlton, T. M. (Eds.), Fluidization IX (pp. 141–148). New York: Engineering
Foundation.
Samuelsberg, A., & Hjertager, B. J. H. (1996). Computational modeling of gas/particle flow in a riser.
353
Savage, S. B., & Sayed, M. (1984). Stresses developed by dry cohesionless granular materials sheared
in an annular shear cell. Journal of Fluid Mechanics, 142, 391–430. doi:10.1017/S0022112084001166
Senior, R. C., & Grace, J. R. (1998). Formation of particle streamers in the wall region of circulating
fluidized bed risers. In Fan, L.-S., & Knowlton, T. M. (Eds.), Fluidization IX (pp. 165–172). New York:
Shaffer, F. (2008). Personal communication. National Energy Technology Laboratory, Pittsburgh, PA.
Sinclair, J., & Jackson, R. (1989). Gas-particle flow in a vertical pipe with particle-particle interactions.
Snider, D. M. (2001). An incompressible three-dimensional multiphase particle-in-cell model for dense
particle flows. Journal of Computational Physics, 170(2), 523–549. doi:10.1006/jcph.2001.6747
Snider, D. M. (2008). Ozone Decomposition. Paper presented at AIChE Annual Meeting, Philadelphia,
Pennsylvania.
using a multiphase particle-in-cell model for dense particle flows. International Journal of Multiphase
Flow, 24(8), 1359–1382. doi:10.1016/S0301-9322(98)00030-5
Stromberg, L. (1983). Operational modes for fluidized beds. Studsvik Report E4-79/83.
Sun, B. (1996). Simulations of Gas-Liquid and Gas-Solid Two Phase Flows. Doctoral Dissertation, Il-
linois Institute of Technology, Chicago.
Syamlal, M., & O’Brien, T. (April 1987) Derivation of a drag coefficient from velocity-voidage correc-
tion. US Dept. of Energy, Office of Fossil Energy, National Energy Technology Laboratory, Morgantown,
West Virginia,
Takeuchi, H., Hirama, T., Chiba, T., Biswas, J., & Leung, L. S. (1986). A quantitative definition and
flow regime diagram for fast fluidization. Powder Technology, 47(2), 195–199. doi:10.1016/0032-
5910(86)80115-2
Tammers, R. F., Shaw, D. F., Reinman, K. J., & Melfi, G. (March 2, 1993). Tangential Solids Separation
Transfer Tunnel. US Patent 5,190,650.
Tanaka, T., Yonemura, S., & Tsuji, Y. (1995). Effects of particle properties on the structure of clusters.
ASMD Publications FED, 228, 297–302.
Tsuji, T., Kawaguchi, T., & Tanaka, T. (1993). Discrete particle simulation of two-dimensional fluidized
Tsuji, Y., Oshima, T., & Morikawa, Y. (1985). Numerical simulations of pneumatic conveying in a
horizontal pipe. Kona, 3, 38–44.
Tsuo, Y. P., & Gidaspow, D. (1990). Computation of flow patterns in circulating fluidized beds. AIChE
354
US DOE, Energy Information Administration. (2006). Retrieved from http://tonto.eia.doe.gov/dnav/

pet/hist/mcrccus2A.htm
van Breugel, J., Stein, J. J. M., & de Vries, R. (1969). Isokinetic sampling in a dense gas-solids stream.
Proceedings - Institution of Mechanical Engineers, 184, 18–23.
van Zoonen, D. (1962). Measurement of diffusional phenomena and velocity profiles in a vertical riser.
In Proceeding of the Symposium on the Interaction between Fluids and Particles, London, (pp. 64-71).
Vreman, B., Geurts, B. J., Deen, N. G., Kuipers, J. A. M., & Kuerten, J. G. M. (2009). Two- and Four-
Way Coupled Euler–Lagrangian Large-Eddy Simulation of Turbulent Particle-Laden Channel Flow.
Flow, Turbulence and Combustion, 82(1), 47–71. doi:10.1007/s10494-008-9173-z
Wang, J., van der Hoef, M. A., & Kuipers, J. A. M. (2009). Why the two-fluid model fails to predict the
bed expansion characteristics of Geldart A particles in gas-fluidized beds: A tentative answer. Chemical
Wegerer, D., & Lomas, D. (Sept. 19, 1995). FCC Feed Contacting with Catalyst Recycle Reactor. US
Patent 5,451,313.
Weinstein, H., Graff, R. F., Meller, M., & Shao, M. J. (1983). The Influence of the Imposed Pressure
Drop Across a Fast Fluidized Bed. In Kunii, K., & Toei, R. (Eds.), Fluidization IV (pp. 299–306). New
Williams, F. A. (1985). Combustion Theory. 2nd. Reading, MA: Addison-Wesley.
Williams, K., & Cocco, R. (2007). Validation of Barracuda CPFD Dense-Phase Particle Physics Models
Using Unique Data From a Large Deep-Bed Experiment. Paper presented at the AIChE Annual Meet-
ing, Salt Lake City, Utah.
Xu, Y., & Subramanian, S. (2006). A multiscale model for dilute turbulent gas–particle flows based on
the equilibration of energy concept. Physics of Fluids, 18, 21–38. doi:10.1063/1.2180289
Yang, N., Wang, W., Ge, W., Wang, L., & Li, J. (2004). Simulation of heterogeneous structure in a cir-
culating fluidized-bed riser by combining the two-fluid model with the EMMS approach. Industrial &
Engineering Chemistry Research, 43(18), 5548–5561. doi:10.1021/ie049773c
Yasuna, J., Moyer, H., Elliot, S., & Sinclair, J. (1995). Quantitative predictions of gas-particle flow in
a vertical pipe with particle-particle interactions. Powder Technology, 84(1), 23–34. doi:10.1016/0032-
5910(94)02971-P
Yerushalmi, J., Cankurt, N., Geldart, D., & Liss, B. (1978). Flow regimes in vertical gas-solid contact
systems. Fluidization: Application to Coal Conversion Processes: AIChE Symp. Ser. 74 (176) 1-13.
Yerushalmi, J., & Cankurt, N. T. (1979). Further studies of the regimes of fluidization. Powder Technol-
ogy, 24(2), 187–205. doi:10.1016/0032-5910(79)87036-9
Yin, X., & Sundaresan, S. (2009). Drag law for bidisperse gas-solid suspensions containing equally
sized spheres. Industrial & Engineering Chemistry Research, 48(1), 227–241. doi:10.1021/ie800171p
355
Yuan, E., Letzsch, W., Jackson, G., Evans, J., & Hood, J. (2006, Aug. 10). Riser Termination Device.
US Patent Application 2006/0177357.
Zenz, F. A. (1949). Two-phase fluid-solid flow. Industrial & Engineering Chemistry, 41(12), 2801–2806.
doi:10.1021/ie50480a032
Zimmermann, S., & Taghipour, F. (2005). CFD modeling of the hydrodynamics and reaction kinetics of
FCC fluidized-bed reactors. Industrial & Engineering Chemistry Research, 44, 9818–9827. doi:10.1021/
ie050490+
356
357
Chapter 11
CFD Modeling of
Bubbling Fluidized Beds
of Geldart A Powders
T. Pugsley
The University of Saskatchewan, Canada
S. Karimipour
Z. Wang
abstract
Fluidized beds have applications in a range of industrial sectors from oil refining and coal combustion
to pharmaceutical manufacture and ore roasting. In spite of more than 80 years of industrial experi-
ence and a tremendous amount of academic attention, the fundamental understanding of fluidized bed
hydrodynamics is still far from complete. Advanced modeling using computational fluid dynamics is one
tool for improving this understanding. In the current chapter the focus is on the application of CFD to a
particularly challenging yet industrially relevant area of fluidization: fluidized beds containing Geldart A
powders where interparticle forces influence the bed behavior. A critical step in modeling these systems is
proper representation of the interfacial drag closure relations. This is because the interparticle cohesive
forces lead to the formation of clusters that reduce the drag below that for non-cohesive particles. The
influence on the predictions of the macroscale fluidized behavior due to mesoscale phenomena such as
clustering is discussed. At present in the literature, approaches representing the reduced drag arising
from mesoscale phenomena are ad hoc. There is a pressing need for an improved understanding of
cluster formation and for robust models describing it that can be incorporated into coarse grid models
of industrial scale fluidized beds.
DOI: 10.4018/978-1-61520-651-3.ch011
1. intrOductiOn
1.1 industrial applications of fluidized bed technology
Fluidized beds are widely used throughout various industrial sectors. The first application of fluidized
bed technology is generally considered to be the Winkler gasifier for coal gasification in the late 1920s
and new fluidized bed processes have been developed in nearly every decade since then. With the ongo-
ing interest in fluidized bed technology, it seems likely that new applications will continue to appear in
this century. For a description of the range of fluidized bed processes the reader is referred to the classic
textbook of Kunii and Levenspiel (1991).
The industrial success of fluidized bed technology is undeniable, however in spite of 8 decades of
operational experience and research that has spanned the range from fundamental to highly applied, the
development of a new fluidized bed process or the troubleshooting of existing processes continues to be
an exercise that can be costly and fraught with challenges. Hence there is a continued need for research
that will improve the fundamental understanding of the hydrodynamics of fluidized beds. One means
of improving this understanding is through advanced computational fluid dynamics (CFD) modeling,
which is the topic of this chapter.
1.2 fluidization defined and the concept of Minimum fluidization
When a bed of solid particles contained in a vessel is contacted with upward-flowing air, water, or
other fluid, the fluid imparts a drag force on the particles in the bed. As the volumetric flow of fluid is
increased, the drag force increases until a point is reached where the upward force of the fluid on the
particle bed equals the downward force of gravity (i.e. the bed weight). At that point, the bed “unlocks”
and may expand and is said to be incipiently fluidized. This process is known as fluidization or a fluid-
ized bed. In what follows, the focus will be on gas-solids fluidized beds. It is important to note that the
behavior of liquid-solid fluidized beds can be very different from that of gas-solids systems and hence
the information presented in the current chapter is only applicable to the latter.
The main operating variable in gas-solids fluidized systems is the superficial gas velocity, defined as
the volumetric flow of gas divided by the cross-sectional area of the vessel. This ignores the presence
of the bed solids, hence the terminology “superficial”.
Figure 1 illustrates the trend in the bed pressure drop (ΔP) as a function of superficial gas velocity,
Uo. As gas velocity increases, the bed remains in a packed state and the pressure drop associated with
forcing the gas through the bed increases. Eventually the point of incipient or minimum fluidization is
reached beyond which the pressure drop becomes constant with further increases in gas velocity. The
corresponding velocity is called the minimum fluidization velocity, Umf and the bed voidage at this point
is called the minimum fluidization voidage (εmf).
It should be pointed out that certain very fine particles or very coarse particles may not fluidize at
all. Furthermore, for those particles that can be fluidized, what happens at gas velocities beyond mini-
mum fluidization depends on the size and density of the particles. The fluidization behavior of particles
of differing sizes and densities is known as the Geldart classification (Geldart, 1973).
358
Figure 1. Pressure drop vs. superficial gas velocity denoting the onset of fluidization at Umf
1.3 geldart classification of Powders
Figure 2 presents the Geldart classification of powders, which divides powders into four groups based
on their mean size and particle density. Powders in each of these four groups behave differently when
fluidized. For instance, Geldart A powders are aeratable (hence the ‘type A’ designation) and fluidize
smoothly. Since these powders are aeratable, a bed of Geldart A powder expands by 20 to 40% of its
initial static bed height once minimum fluidization is reached. Once this expansion is complete, additional
gas passes through the bed in the form of bubbles. A classic Geldart A powder is fluid catalytic cracking
(FCC) catalyst. Geldart B powders are coarser, but still fluidize well and unlike Geldart A powders, a
bed of Geldart B powder does not expand when fluidized. Instead, bubbles start to form immediately
once incipient fluidization is reached (hence the ‘type B’ designation, indicating bubbling tendencies).
Figure 2. Geldart classification of powders (Geldart, 1973)
359
Sand is a widely used Geldart B powder. Geldart C powders, of which flour is a good example, are very
fine and highly cohesive powders that do not fluidize well unless some sort of fluidization aid is used.
Geldart D powders are extremely coarse; gas-solids contacting of these coarse powders is typically car-
ried out in a spouted bed configuration or a very shallow conical fluidized bed.
The difference in the fluidized behavior of these 4 classes of Geldart powders has been explained by
Molerus (1982) and Seville (1997) as due to the relative importance of interparticle forces in comparison
to the force of gravity on the particles in the bed. A recent paper from our group has summarized the
role of these forces in tabular form (see Table 1). The italicized section in the explanation of the transi-
tion from A to B behavior has been added here for emphasis. As will be shown later in this chapter,
proper representation of the drag force has been critical to the success of our work as well as that of
other research groups, when using CFD to model beds of Geldart A powders.
1.4 summary
This brief introduction has established the industrial relevance and associated challenges of fluidized
bed technology. Because of these challenges, there is a need for continued research aimed at improving
the fundamental understanding of gas-solids fluidized beds. CFD modeling is one means of improving
this understanding. The present chapter focuses on the CFD modeling of bubbling fluidized beds of
Geldart A powders. CFD modeling of this case is made more complex than for Geldart B powders due
to the role of interparticle forces when Geldart A powders are fluidized. The chapter therefore focuses
on the modeling approaches used by our group and other groups to deal with this added complexity.
2. gOverning equatiOns and clOsure relatiOnsHiPs
Numerical models of gas-solids two phase flow may be classified as either Lagrangian or Eulerian.
In the former, the motion of individual particles in the flow field is determined. This requires separate
equations for each particle; thus in large fluidized systems containing billions of particles, the number
of equations and the corresponding computational requirement become so large as to be impractical. In
the Eulerian approach, the two phases (gas and solids) are treated as two separate, but interpenetrating
continua. Thus only one equation of motion is required for each phase and computational requirements
are substantially reduced. This is not to say that computing requirements are insignificant for the Eu-
lerian approach, but with parallel processors, industrial scale units can be modeled using coarse grids.
The focus of this chapter is the Eulerian approach.
Table 1. Molerus’ interpretation of the Geldart classification (after Wormsbecker et al., 2008)
Transition Between Explanation

C→A Free particle motion is suppressed by the dominance of cohesive forces with Geldart C powders as compared to A.
A→B Geldart A powders have significant adhesive forces compared to the fluid drag force exerted on the particle.
Geldart B powders are negligibly influenced by adhesive forces as compared the drag force.
B→D Geldart D behaviour exists when the dynamic pressure of the fluidization gas exceeds a distinct value based on
the size and density of the powder at minimum fluidization.
360
The governing equations describing continuity of mass and momentum for two-phase gas-solids flow
are presented in Chapter 6. Closure relations for the gas-phase stress t g , granular stress S s , and gas-
solids drag Fgs are needed in order to solve the governing equations. Unlike the governing equations,
which exhibit little variability between modeling studies, the closure relations used in the open literature
exhibit more variability. A complete review of these relations is beyond the scope of the present chapter;
however the reader is referred to the paper of van Wachem et al. (2001) for a description of the closure
relations and a consideration of their influence on CFD predictions of dense (i.e. bubbling) beds. Van
Wachem et al. (2001) conclude that gravity and drag are the dominant terms in the solid-phase momen-
tum balance. Indeed, a sensitivity analysis that we have performed on our CFD model for a dense fluid-
ized bed of Geldart A powder has led us to the same conclusion. Therefore, in the discussion that follows
in this chapter, we will focus exclusively on the role of the drag force relationship on CFD model pre-
dictions of bubbling beds of Geldart A powders.
3. literature review: PreviOus cfd MOdeling

effOrts Of dense beds Of geldart a POwder
Prior to the work of our group, which will be discussed in more detail in the next section, no successful
CFD modeling efforts of bubbling beds containing Geldart A powders could be identified in the open
literature. Given the importance of the fluidized bed contacting of such powders in catalytic reactor ap-
plications, this represented a substantial gap in the fluidization knowledge.
Ferschneider and Mege (1996) attempted a simulation of a freely bubbling bed of FCC particles at
a superficial gas velocity of 0.2 m/s using an Eulerian (two-fluid) CFD model. They observed severe
overestimation of bed expansion, with an average emulsion phase voidage of 0.8. This is much differ-
ent than that typically observed experimentally, where the emulsion phase voidage is usually near the
minimum fluidization value (approximately 0.45 to 0.5). Recognizing that the existing two-fluid CFD
models did not adequately represent the particle interactions in Geldart A fluidized beds, Krishna and van
Baten (2001) proposed the use of “pseudo-fluids” in their model. They used a two-fluid CFD model that
set the bubble phase and emulsion phase as the two fluids (instead of the usual gas and solids phases).
Constant properties were assigned to the emulsion phase, and empirical correlations for mean bubble
diameter and rise velocity were used. Limited validation was performed by comparing simulated gas
holdup with data from the same experiments used to derive the empirical correlations for bubble proper-
ties. The major shortcoming of this model is in the use of mean bubble and emulsion properties, which
is not a true physical representation of the fluidized bed. In reality, bubble sizes and velocities exhibit a
distribution about the mean value, which will significantly affect the predicted hydrodynamics. Bubble
splitting and coalescence is an important phenomenon associated with Geldart A bubbling beds that
cannot be predicted with this model.
Preliminary simulations of a turbulent fluidized bed of FCC particles were reported by Bayle et al.
(2001) to give 20-30% over prediction of bed expansion. Although this is not a bubbling bed simulation,
the high bed expansion is consistent with that observed by Ferschneider and Mege (1996). Bayle et al.
(2001) speculated that the drag laws and solids viscosity relations in their model needed improvement.
The need to modify the drag relation was consistent with the paper of van Wachem et al. (2001).
361
More recently, Gelderbloom et al. (2003) undertook a simulation of the collapse test of a fluidized
bed of Geldart A powder. No special modifications to the drag relationships were made. Instead they
focused on a term called the solid-phase modulus. Although not discussed by the authors in their paper,
their simulated results show that the initial bed height (i.e. the starting point of the collapse test, cor-
responding to fully fluidized conditions) is 75 to 80% greater than the final consolidated bed height at
the end of the simulated collapse test. So although the predicted collapse time compared well with ex-
periment, the simulation of a fluidized bed of Geldart A powder again suffered from an over-prediction
of bed expansion.
The group led by Professor Kuipers at the University of Twente has undertaken a substantial mod-
eling effort known as multi-level modeling. In this approach, a fluidized bed of Geldart A powder is
simulated using Discrete Particle Modeling (DPM). As the name implies, interactions at the particle-
particle level are modeled in detail, with the notion being that these results are then fed into an Eulerian
model that is more practical for larger scale simulations (the meso and macro-levels). The efforts of this
group are documented in a series of papers (Ye et al., 2008; 2005; 2003) and in the complete thesis of
Ye (2005). The most recent publication (Ye et al., 2008) takes a similar approach to that of our group
whereby the drag force models available in the open literature are reduced through the use of a simple
empirical coefficient in the range of 0.2 to 0.3. As Ye et al. (2008) note, this is a rather ad hoc solution
to the problem, but as will be discussed in the next section, there is a theoretical basis for this approach
based on a consideration of the cohesive forces between particles. Furthermore, our group has collected
experimental evidence of the reduced drag force in a fluidized bed of Geldart A powder.
4. drag fOrce MOdificatiOn in tHe eulerian MOdel

Of a fluidized bed Of geldart a POwder
In the fluidization literature, the most commonly used drag force models in CFD studies are those of
Wen and Yu (1966), Gidaspow (1994), Syamlal and O’Brien (1989), and Gibilaro et al. (1985). While
these models vary in their functional form and lead to somewhat different predictions of the drag force
as a function of bed voidage, little influence on the predictions of bed properties such as expansion and
emulsion voidage are observed when the different drag models are used interchangeably in a CFD code.
In the case of Geldart A powders, the same unrealistic bed expansion and emulsion voidage is predicted
regardless of the drag model. This is likely due to the fact that they were all developed for Geldart B
powders and hence the influence of interparticle forces on the effective gas-solids drag for Geldart A
powders is not captured.
Realizing that all drag models lead to similar CFD predictions, our group has chosen the Gibilaro
et al. (1985) model since it is continuous over the entire range of voidages found in a fluidized bed and
because it is a single compact equation. The Gibilaro et al. (1985) drag force model is given below:
 
17.3  ρ g vs − v g

Fgs =C  + 0.336 (1 − ε g )ε−g 1.8
 Re p  dp
Where the Reynolds number is based on mean particle diameter and the gas-solids slip velocity,
362
 
ρ g d p vs − v g ε g
Re p =
µg
The coefficient, C, at the front of the Gibilaro et al. (1985) equation is the fractional constant that is
needed to reduce the drag coefficient, thus accounting for interparticle forces in beds containing Geldart
A powders.
The system of governing equations and closure relations has been solved numerically for a 14-cm
ID, fluidized bed in two-dimensional Cartesian coordinates using the Multiphase Flow with Interphase
eXchanges (MFIX) available from the United States Department of Energy (www.mfix.org). Boundary
and initial conditions are summarized in Figure 3.
As shown in Figure 4, manipulation of the drag force equation has a strong influence on the pre-
dicted bed expansion. When the drag force model is left unchanged (C = 1.0) the bed height almost
doubles from its initial condition, but when the drag force is reduced (C=0.25), the bed exhibits about
a 25% expansion from its initial condition, which is more realistic.
Figure 5 presents instantaneous voidage profiles corresponding to the same values of the drag force
correction factor, C. The figure illustrates quite clearly how the drag force modification leads to predic-
tions of discrete bubbles that would be predominant at this superficial gas velocity (approximately 10
times umf). Without this modification, bed expansion is high and the emulsion phase is too dilute.
In addition to this qualitative analysis of the influence of reducing the drag force, McKeen and Pug-
sley (2003a) also provided a quantitative analysis by comparing model predictions with published data
Figure 3. Boundary and initial conditions for 2-D model of a 14-cm ID fluidized bed (numerical results
shown in Figures 4 and 5)
363
Figure 4. Transient profiles of bed expansion at a superficial gas velocity of 0.1 m/s illustrating the effect
of modifying the drag force equations on the predictions of the CFD model
Figure 5. Instantaneous CFD predictions of voidage in the bed at a superficial gas velocity of 0.1 m/s:
C = 0.25; (b) C = 1.0.]
364
for bed expansion as well as with published correlations for bubble diameter and bubble rise velocity.
The predictions of the CFD model with reduced drag coefficient compared very well with this data and
the correlations.
Hence our approach to reducing the drag coefficient in a CFD model of a bubbling bed of Geldart
A powder has proven able to overcome the issues of bed expansion and high emulsion phase voidage
reported in section 3 of this chapter. While this approach has been referred by other authors as ad hoc,
there is a physical basis for the magnitude of the correction factor in the range of 0.2 to 0.3. This has
been discussed in the paper of das Sharma et al. (2006) and the reader is therefore referred to that work
for complete details. Briefly, the force balance on a particle in a fluidized bed states that the drag force
is equal to the sum of the buoyant weight plus the cohesive force. Following the statement of Molerus
(1982) that the cohesive force on a particle is proportional to its buoyant weight, an expression for the
drag coefficient of a cohesive particle (CDC) can be obtained:
CDC = (1 + K )CD
Where CD is the drag coefficient for non-cohesive particles and K is a constant of proportionality (K =
0 for non-cohesive particles and K >1 for cohesive particles).
Consistent with the approach of Gibilaro et al. (1985), the drag coefficient in a fluidized bed of
cohesive particles may be related to the drag on a single particle under terminal conditions as follows:
CDC = CDt ε1−2 n
Equating these two expressions for CDC leads to the following result:
1 C ε1−2 n 
CD =
(1 + K )  Dt 
Note that with K = 0, this expression reverts to the drag coefficient in a fluidized bed of non-cohesive
particles. For Geldart A powders, the value of K is approximately 3 (das Sharma et al., 2006), giving a
value of 1/(1+K) of 0.25. This analysis provides a theoretical basis for the coefficient C that we have
implemented in the expression of Gibilaro et al. (1985).
5. cfd MOdeling Of tHe striPPer sectiOn Of a

fluid catalytic cracKing unit (fccu)
The fluid catalytic cracking unit (FCCU) in a crude oil refinery is a major unit operation for the produc-
tion of gasoline. A schematic of the FCCU process is given in Figure 6.
The FCCU is a fluidized bed process in which hot catalyst contacts fresh liquid feed at the base of
the riser. The liquid is cracked to lighter hydrocarbons in the gasoline range as well as light gases. Car-
bon is also rejected in the form of coke which deposits on the catalyst. The coked catalyst is therefore
circulated into the regenerator where it is burned off and the hot catalyst is re-injected into the riser.
365
Figure 6. Schematic of the FCCU highlighting the stripper section
Between the riser and the regenerator is the stripper section where entrained hydrocarbons are removed
by countercurrently contacting the circulating catalyst with steam in a dense fluidized bed containing
inserts or baffles. The baffles promote gas-solids contacting through enhanced mixing and the break-up
of large bubbles.
In recent years, as existing FCCUs have been pushed beyond their design limits in an effort to maxi-
mize gasoline production, the stripper has become a bottleneck. Operational difficulties associated with
this have been documented by Senior et al. (1998). CFD modeling is a valuable tool for troubleshooting
the stripper section and for evaluating new designs to alleviate operational issues. CFD modeling of a
stripper also highlights a key advantage of commercial CFD codes, namely the ability to model complex
geometries and to screen different hardware configurations without the need to resort to experimentation.
However, to be confident in the ability of CFD to perform these tasks, detailed model development and
validation is critical.
McKeen and Pugsley (2003b) present a 2-D model of an FCCU stripper, developed using the MFIX
software. The base-case stripper baffle geometry was the well-known staged-internal “disc-and-donut”
design. In addition, the baffles were modified from the base case to simulate the stripper performance
with Flux TubesTM installed. The Flux TubesTM were reported by Senior et al. (1998) to mitigate the
problem of flooding in FCCU strippers experiencing high solids mass fluxes. Flooding refers to the
inability of the stripping steam to flow upward against this high downward flux. As a result, steam ac-
cumulates at the base of the stripper. This is associated with poor stripping efficiencies and a loss of
valuable hydrocarbon product to the regenerator where it is simply combusted.
366
Figure 7. Simulated FCCU geometry (all dimensions in cm)
Table 2. Summary of boundary conditions for stripper model
BC Location Type Variables Specified

Dipleg Mass Inflow εg = 0.45, Xg(2) = 1.0, Vg = Vs = depends on Gs, Ug = Us = 0
Standpipe Mass Outflow εg = 0.45, Vg = Vs = depends on Gs, Ug = Us = 0
Distributor Mass Inflow εg = 1.0, Xg(1) = 1.0, Vg = depends on U0, Ug = 0
Freeboard Outlet Pressure Outlet Pg = 101.3 kPa
Figure 7 illustrates the computational domain for the 2-D stripper model of McKeen and Pugsley
(2003b) as well as details on the standard and modified (Flux TubeTM) baffle dimensions. This geometry
is similar to that in the work of Rivault et al. (1995), which represents one of the few published works
on experimental measurement in a cold-flow FCCU stripper. Table 2 presents the boundary conditions
for the stripper simulation.
A quantitative comparison between model predictions and the data of Rivault et al. (1995) is pre-
sented in Figure 8. The positions 2 and 4 refer to measurement locations between the second and third
baffles from the bottom and above the uppermost baffle, respectively as shown in Figure 6. The agree-
ment between the CFD model predictions and the experimental data is very good.
Other comparisons between the CFD stripper model and the data of Rivault et al. (1995) are pre-
sented in the work of McKeen and Pugsley (2003b). In general experimental trends are well predicted,
providing a level of confidence in the ability of the model. One of the most successful aspects of the
367
Figure 8. Comparison between model predictions and experimental data for bubble frequency at a strip-
per superficial gas velocity of 0.18 m/s and solids mass flux of 28 kg/(m2·s)
model is the ability to predict stripper flooding and the mitigation of flooding with the modified baffles.
This is illustrated in Figure 9. In Figure 9a, the accumulation of gas can be seen at the stripper base over
the 20 seconds of real time captured by the simulation. Figure 9b clearly shows that this accumulation
of gas no longer occurs, thus elimination of stripper flooding is predicted with the modified baffle ge-
ometry.
These results on the simulation of modified baffle geometry point out what is the greatest potential
of CFD: the capacity to perform computational experiments related to hardware modifications that would
otherwise require costly and time-consuming cold-flow experimentation. Clearly we have not reached
that stage yet in the development of CFD modeling; no fluidized bed process would ever be designed
at full scale without cold flow experiments at multiple scales. Also, troubleshooting of existing indus-
trial scale fluidized beds often relies on cold flow experimentation as well, with some additional inter-
pretation through the use of CFD. In order to further advance this potential of CFD, there needs to be
more model validation work such as that presented in Figure 8. Also, the issue of long simulation times
for large-scale commercial vessels and the requirement for a more versatile means of handling the reduc-
tion in the drag force needs to be addressed. This is discussed in the next section.
6. sub-grid MOdels
Work in this area is being led by the group of Professor Sundaresan at Princeton University. In a series
of papers (Igci et al., 2008, Andrews et al., 2005, and Agrawal et al., 2001), they effectively illustrate
that CFD model predictions of macroscale behavior are influenced by mesoscale structures. In particu-
lar, they use the formation of clusters of 10 or more individual particles as an example. The presence
of the clusters at the mesoscale reduces the gas-solids interfacial drag and hence influences macroscale
predictions of key parameters of interest such as bed pressure drop or bed expansion. Their analysis
is completely consistent with our need to reduce the drag coefficient to obtain realistic predictions of
368
Figure 9. CFD predictions of flooding (a) and mitigation of flooding with Flux TubesTM (b) at a super-
ficial gas velocity of 0.18 m/s and a mass flux of 90 kg/(m2·s). Frames correspond to 5 s intervals, from
t = 0 at the left to t = 20 s on the right.]
369
Geldart A bubbling bed behavior, although it should be pointed out that the Princeton group has focused
almost exclusively on riser flow rather than dense phase fluidization.
The main problem introduced by events at the mesoscale has been quite clearly laid out by Sun-
daresan’s group: CFD modeling of industrial scale reactors relies on coarse grids to simulate the large
overall computational domain. However, if one is to resolve the clusters that are known to influence
the macroscale predictions, a very fine grid is required and the associated computation time becomes
prohibitively large for any realistic size of fluidized bed. Given that the long term goal is to be able to
model industrial scale fluidized bed reactors, this problem needs to be overcome. In this regard, the
work of Agrawal et al. (2001) is highly recommended as it provides a thorough review of the issue and
also presents a preliminary approach for sub-grid modeling. In this work, Agrawal et al. (2001) define a
computational domain whose size is comparable to the size of a cell in a typical coarse-grid simulation
of an industrial reactor. They then perform a highly resolved simulation of the computational domain,
essentially focusing in on the mesoscale behavior within that coarse-grid cell. This allows for the de-
velopment of computer-generated “ad hoc” (Andrews et al., 2005) closure relations at the mesoscale
that could then be implemented into the coarse grid simulation. As noted by Andrews et al. (2005), this
approach has its limitations in that such sub-grid closure relations would be dependent on the type of
particle used and the regime of fluidization. Also, the influence of boundaries such as solid walls on the
gas-solids flow is not incorporated. However, this work appears to be continuing and important advances
will no doubt be made.
7. cOnclusiOn
The emphasis in this chapter has been on the application of CFD to an important industrial problem,
namely the fluidization of Geldart A powders where interparticle forces are important. The results on
the modeling of a lab-scale bubbling bed and a cold-flow stripper illustrate the potential of CFD model-
ing to handle these problems. However, the work of our group as well as the work of other researchers
highlights some existing shortcomings. First, existing drag models for fluidized beds are not applicable
for beds containing Geldart A powders due to the formation of clusters with such powders and the cor-
responding reduction in the drag force. At present in the open literature, ad hoc solutions have been
applied to reduce the drag for modeling purposes, although the solutions do have a physical basis.
Clearly more robust models that accurately reflect the physics behind cluster formation are needed.
The second shortcoming is related to the first: the need for a highly resolved numerical grid to capture
the mesoscale behavior of cluster formation makes computational time for CFD models of full-scale
industrial reactors prohibitively long. Coarse grid simulations of industrial reactors that do not include
this mesoscale behavior and its influence on the macroscale behavior will be called into question. Thus
there is a pressing need for robust sub-grid models that can be implemented into coarse grid models for
the design and troubleshooting of large scale fluidized beds.
references
Agrawal, K., Loezos, P. N., Syamlal, M., & Sundaresan, S. (2001). The role of meso-scale structures in
rapid gas-solids flows. Journal of Fluid Mechanics, 445, 151–185. doi:10.1017/S0022112001005663
370
Andrews, P., Loezos, N., & Sundaresan, S. (2005). Coarse-grid simulation of gas-particle flows in verti-
cal risers. Industrial & Engineering Chemistry Research, 44(16), 6022–6037. doi:10.1021/ie0492193
Bayle, J., Mege, P., & Gauthier, T. (2001). Dispersion of bubble flow properties in a turbulent FCC
fluidized bed. In Kwauk, M., Li, J., & Yang, W. C. (Eds.), Fluidization X (pp. 125–132). New York:
Das Sharma, S., Pugsley, T., & Delatour, R. (2006). Three-dimensional CFD model of the deaeration
rate of FCC particles. AIChE Journal. American Institute of Chemical Engineers, 52(7), 2391–2400.
doi:10.1002/aic.10858
Ferschneider, G., & Mege, P. (1996). Eulerian simulation of dense phase fluidized beds. Revue de l’Institut
Français du Pétrole, 51(2), 301–307.
Geldart, D. (1973). Types of gas fluidization. Powder Technology, 7, 285–292. doi:10.1016/0032-
5910(73)80037-3
Gelderbloom, S. J., Gidaspow, D., & Lyczkowski, R. W. (2003). CFD Simulations of bubbling/collaps-
ing fluidized beds for three Geldart group. AIChE Journal. American Institute of Chemical Engineers,
49(4), 844–858. doi:10.1002/aic.690490405
Gibilaro, L. G., Di Felice, R., & Waldram, S. P. (1985). Generalized friction factor and drag coeffi-
cient correlations for fluid-particle interactions. Chemical Engineering Science, 40(10), 1817–1823.
doi:10.1016/0009-2509(85)80116-0
Gidaspow, D. (1994). Multiphase flow and fluidization: continuum and kinetic theory descriptions. New
York: Academic Press.
Igci, Y., Andrews, A. T. IV, Sundaresan, S., Pannala, S., & O’Brien, T. (2008). Filtered two-fluid models
1431–1448. doi:10.1002/aic.11481
Krishna, R., & van Baten, J. M. (2001). Using CFD for scaling up gas-solids bubbling fluidized bed
reactors with Geldart A powders. Chemical Engineering Journal, 82, 247–257. doi:10.1016/S1385-
8947(00)00369-7
Kunii, D., & Levenspiel, O. (1991). Fluidization Engineering. Boston: Butterworth-Heinemann.
McKeen, T., & Pugsley, T. (2003a). Simulation and experimental validation of a freely bubbling bed of
McKeen, T. R., & Pugsley, T. S. (2003b). Simulation of a cold flow FCC stripper at small scale using
computational fluid dynamics. Int. J. Chem. Reactor Eng. 1, paper A18.
Molerus, O. (1982). Interpretation of Geldart’s type A, B, C and D powders by taking into account in-
terparticle cohesion forces. Powder Technology, 33, 81–87. doi:10.1016/0032-5910(82)85041-9
Rivault, P., Nguyen, C., Laguerie, C., Bernard, J. R., & Aquitaine, E. (1995). Countercurrent stripping
dense circulating beds effect of the baffles. In Large, J. F., & Laguerie, C. (Eds.), Fluidization VIII (pp.
491–499). New York: Engineering Foundation.
371
Senior, R. C., Smalley, C. G., & Gbordzoe, E. (1998). Hardware modifications to overcome common
operating problems in FCC catalyst strippers. In Fan, L. S., & Knowlton, T. M. (Eds.), Fluidization IX
(pp. 725–732). New York: Engineering Foundation.
Seville, J. P. K., Tuzun, U., & Clift, R. (1997). Processing of Particulate Solids. New York: Blackie
Academic.
Syamlal, M., & O’Brien, T. (1989). Computer simulation of bubbles in a fluidized bed. AIChE Symp.
Ser. 85(27) 22-31.
Van Wachem, B. G. M., Schouten, J. C., Krishna, R., & van den Bleek, C. M. (1991). Validation of the
Eulerian simulated dynamic behaviour of gas-solids fluidized beds. Chemical Engineering Science, 54,
2141–2149. doi:10.1016/S0009-2509(98)00303-0
Wen, C. Y., & Yu, Y. H. (1966). Mechanics of fluidization. AIChE Symp. Ser. 62, 100-111.
Wormsbecker, M., & Pugsley, T. (2008). The influence of moisture on the fluidization behaviour of porous
pharmaceutical granule. Chemical Engineering Science, 63, 4063–4069. doi:10.1016/j.ces.2008.05.023
Ye, M. (2005). Multi-level modeling of dense gas-solids two-phase flows. PhD Dissertation, University
of Twente, Twente, The Netherlands.
Ye, M., van der Hoef, M. A., & Kuipers, J. A. M. (2004). A numerical study of fluidization behavior of
Geldart A particles using a discrete particle model. Powder Technology, 139(2), 129–139. doi:10.1016/j.
powtec.2003.10.012
Ye, M., van der Hoef, M. A., & Kuipers, J. A. M. (2005). From discrete particle model to a continuous
model of Geldart A particles. Chemical Engineering Research & Design, 83(7), 833–843. doi:10.1205/
cherd.04341
Ye, M., Wang, J., van der Hoef, M. A., & Kuipers, J. A. M. (2008). Two-fluid modeling of Geldart A
particles in gas-fluidized beds. Particuology, 6(6), 540–548. doi:10.1016/j.partic.2008.07.005
372
373
Chapter 12
Computational Modeling of
Gas-Solids Fluidized-Bed
Polymerization Reactors
Ram G. Rokkam
Rodney O. Fox
Michael E. Muhle
Univation Technologies, USA
abstract
Gas-solids flows have numerous industrial applications and are also found in natural processes. They
are involved in industries like petrochemical, polymer, pharmaceutical, food and coal. Fluidization is
a commonly used gas-solids operation and is widely used in production of polyethylene. Polyethylene
is one of the most widely used thermoplastics. Over 60 million tons are produced worldwide every year
by both gas-phase and liquid-phase processes. Gas-phase processes are more advantageous and use
fluidized-bed reactors (e.g., UNIPOLTM PE PROCESS and Innovene process) for the polymerization
reactions. In this work a chemical-reaction-engineering model incorporating a given catalyst size dis-
tribution and polymerization kinetics along with the quadrature method of moments is used to predict
the final polymer size distribution and temperature. An Eulerian-Eulerian multi-fluid model based on
the kinetic theory of granular flow is used to solve the fluidized-bed dynamics and predict behavior such
as particle segregation, slug flow and other non-ideal phenomena.
intrOductiOn
Polymers have various applications in automotive industry, packaging, chemical, construction, electrical,
packaging, and agriculture. Widely used polymers fall in the category of thermoplastics and thermosets.
The major thermoplastics include low density polyethylene (LDPE), polypropylene (PP), poly vinyl
DOI: 10.4018/978-1-61520-651-3.ch012
Computational Modeling of Gas-Solids Fluidized-Bed Polymerization Reactors
chloride (PVC), high density polyethylene (HDPE), polyethylene terephthalate (PET), polystyrene (PS)
and polyamide (PA). Important thermoplastics include polyurethanes (PU), phenolics and epoxy resins.
Various commercial reactors are used for the production of thermosets and thermoplastics. These include
stirred tank reactors, fluidized-beds, plug flow reactors, loop reactors, bubble columns and multizone
circulating reactors. These fall in the category of gas-solids systems (fluidized-beds, stirred tanks), gas-
liquids (bubble columns), and gas-solid-liquids (slurry reactors).
When one surveys the total polymerization marketplace today, thermoplastics represent the major
portion of the total production volume. Polyolefins, in turn, represent roughly 60 percent of all the ther-
moplastics produced and sold in the world today. The major types of polyolefins include polypropylene,
high-density polyethylene, linear low-density polyethylene, low-density polyethylene, metallocene
polyethylene and polypropylene, and various co-polymers and elastomers. The polyolefin family of
products serves a wide variety of end-use markets in the major sectors of packaging, automotive, con-
struction, medical, wire and cable and others. Unlike the incremental technological developments more
common in other polymers, polyolefin technology developments are significant and routinely leapfrog
the existing ones. These tremendous developments in technology impact the whole industry as a unit
as well as the high profit sectors. Multiphase reactors and the associated catalyst technology play a key
role in polyolefin businesses.
Polyethylene is the most widely used polyolefin in a wide range of applications. For example, it is
used to manufacture plastic bags, electrical insulation, plastic tubing, bottles and packaging materials,
and has the advantages of low price, flexibility of molding, and ease of disposal and recycling (Kaneko
et al., 1999). Over 60 million tons of polyethylene is produced annually worldwide. It can be produced
using gas-phase and liquid-phase processes. In general the gas-phase processes are more advantageous
than liquid-phase processes. The activity of most catalysts is approximately 1000 Kg polymer per gram
of catalyst and the final polymer need not be separated from the catalyst. In liquid-phase processes the
olefin needs to be dried and separated from the solvent (Xie et al., 1994). Earlier technologies produced
polyethylene at high pressures (~ 3000 atm). With the advent of catalysts like Ziegler-Natta and metal-
locene, polyethylene can be produced at low pressures (~ 20 atm). The available commercial gas-phase
reactors are fluidized beds (UNIPOL and Innovene process), vertical stirred-beds (NOVOLEN process),
horizontal stirred-beds (AMOCO process) and multizone circulating reactors (Basell process). Of all reac-
tor types, fluidized-beds have become popular due to its excellent mass and heat transfer characteristics.
Although there are several advantages with gas-solids fluidized-bed polymerization reactors, there are
also some disadvantages, which researchers are trying to understand and solve. In fluidization we gener-
ally encounter particles of different sizes and/or densities. When gas flows through a bed of such particles
different phenomena are observed. Consider a gas-solids system where the density of the particles is the
same (as is the case for gas-phase polyethylene polymerization), but they have two different sizes. The
dynamic response of one particle size is different from the other primarily due to the size dependence of
the drag force exerted by the gas phase. As a general rule, the drag force for a given gas velocity on small
catalyst particles will be much larger than on large polymer particles. When the gas flow rate is operated
at velocities between the minimum fluidization velocity of the smaller and larger particles a phenomenon
known as segregation (Bokkers et al, 2004; Fan, 2006; Fan & Fox, 2008; Kim & Choi, 2001; Mahdi
et al., 2002) is observed. Fluidized beds operated at velocities higher than the minimum fluidization of
all particles results in a well-mixed system. However, at high gas velocities particles can escape out of
the top of the reactor. This phenomenon is called entrainment or elutriation (Baerns, 1966; Baron et
al., 1987; Briens et al., 1992). The other important phenomena occurring in fluidized-bed polymeriza-
374
tion is electrostatics (Bafrnec & Bena, 1972; Boland & Geldart, 1972; Ciborowski & Wlodarski, 1962;
Guardiola et al., 1996; Mehrani et al., 2005, 2007). Charge generation in a fluidized bed takes place due
to particle-particle and particle-wall collisions. This phenomenon is called tribo-electrification (Elsdon
& Mitchell, 1976; Gajewski, 1985; Mountain et al., 2001; Revel et al., 2003; Wolny & Ka Zmierczak,
1989; Zhao et al., 2000). Particle agglomerates and sheets forming on the walls of the reactor are specific
problems associated with electrostatics in fluidized-bed reactors (Hendrickson, 2006). Polymerization
reactions are highly exothermic reactions. If the polymer temperature rises above the melting point, the
polymer can melt and be a source of thermal agglomeration (Behjat et al., 2008; Mckenna & Soares,
2001) and hot spots (McKenna et al., 1999; McKenna & Soares, 2001; Song, 2004). On the other hand,
if the temperature is too low then particles can become brittle producing unwanted fines that must be
separated from the outlet gas.
In this review only CFD models for olefin polymerization reactors will be discussed. Heat transfer
plays an important role in polymer particle overheating. Behjat et al. (2008), Dehnavi et al. (2008) and
McKenna et al. (1999) used CFD simulations to compute convective heat-transfer coefficients between
polymer particles and the gas phase. They compared the single particle heat-transfer coefficient results
with the Ranz-Marshall correlation and obtained good comparisons. They also analyzed the effects of
gas velocity, particle size and shape, and different configurations on the heat-transfer coefficient. Eriks-
son & McKenna, (2004) employed CFD to study the influence of initial catalyst particle size on particle
growth and the effects of other parameters such as particle interactions on heat transfer. Kaneko et al.
(1999) used a discrete element model to simulate a small gas-phase olefin polymerization reactor. They
did two-dimensional, constant particle size simulations and studied the effects of gas distributor design
and particle behavior on hot-spot formation. The works on CFD simulation of pilot-plant and industrial-
scale fluidized-bed polymerization reactors are very few. Gobin et al. (2003) used a two-fluid model-
ing approach to simulate a pilot plant and industrial reactors. They utilized two and three-dimensional
time-dependent simulations and compared the simulation results of bed height, pressure drop and mean
flow properties with experimental results. The work described above assumes a unimodal, single particle
size for the polymer. In reality, fluidized-bed polyolefin processes have a wide range of particle sizes.
In order to describe phenomena such as segregation and entrainment, a particle size distribution needs
to be considered. To describe this case, the multifluid model and chemical reaction engineering model
are combined to simulate a pilot-plant-scale fluidized-bed polymerization reactor.
At present, the art of scale-up and design of industrial fluidized-bed reactors is largely empirical
(Glicksman, 1988). The correlations derived from laboratory-scale reactors are used to build the pilot-
plant scale reactors. If these correlations work well at the pilot-scale plant, they are extended to the
commercial scale. This process is trial and error since the hydrodynamics in larger units can be vastly
different from that in smaller units. For example bubble sizes and wall effects will be different in larger
units compared to smaller units, and conversion of the reactants will thus be different. This empirical
approach has been adopted because of the difficulty in understanding the gas-solids and solid-solid in-
teractions. In recent years, computational fluid dynamics has emerged as a useful tool for understanding
such behavior. With the advent of high-performance computers and improved computational resources,
the complex behavior of the underlying gas-solids flows can be understood and used to help in the design
process of fluidized-bed polymerization reactors.
The chapter is organized intwhere -ΔHRx is the heat of reaction, Mf the monomer concentration, ρ the
fluid density and Cp the fluid heat capacity. Data are shown in Table 1 comparing the adiabatic tempera-
ture rise (Ray & Villa, 2000) for different monomers.
375
Table 1. Adiabatic temperature rise for typical polymerization reactions
Monomer Heat Poly (Kcal/mol) ΔTad (°C) Heat Duty (Kcal/Kg)

Ethylene 25.9 1609 922
Propylene 20.1 850 476
1,3-Butadiene 17.4 589 322
Vinyl Chloride 17.2 803 275
Tetrafluoroethylene 38.9 1447 390
Acrylonitrile 18.3 774 344
Vinyl Acetate 21.0 519 244
It can be seen that the temperatures that can be reached are quite high, and the associated heat loads
very large requiring novel and complex heat removal systems. A better appreciation of the scope of the
heat removal problem can perhaps be had by comparing these data with those of the adiabatic flame
temperatures for many common combustion reactions. Typical values are in the range of 1900-2100°
for hydrocarbon mixtures in air (Griffiths et al., 1995) demonstrating the highly exothermic nature of
polymerization reactions and operating challenges one faces in these processes. Coupled with this is the
fact that the solid polymer particles in multiphase processes are operated close to the melting point of
the polymers.
The development of engineered catalysts designed for efficient heat and mass transfer at the meso-
scale provided the enabling technology for the utilization of multiphase reactors. Once adopted they
rapidly expanded. For example, today there are over 100 reactor lines devoted to UNIPOL polyethylene
gas phase process producing in excess of 18 million tons per annum representing 25% of world wide
polyethylene capacity (Borruso, 2008). Re-circulating fluid-bed reactors are particularly advantageous
due to their uniform composition and temperature, low investment and operating cost, excellent mixing
and product uniformity. Monomer serves as the reactant, fluidizing gas and a heat-transfer medium to
remove the heat of polymerization in this highly exothermic reaction. The polymer particles grow on the
very active single particle catalyst so that there is no need to remove catalyst residues from the granular
product. The specific nature of the transition metal catalyst and the ligand environment for the more
recent metallocene catalyst developments allows molecular features to be tailored to meet a targeted
resin structure including molecular weight and chemical composition distributions.
benefits of cfd
Computational fluid dynamics (CFD) has only recently been explored in multiphase polymerization
reactors. This is due, in large part, to the complexity of modeling these systems. There are several dif-
ferent length and time scales. Large-scale convection and diffusion occur on a macroscale, mid-scale
mixing and particle interactions take place at the mesoscale, while complex polymerization chemistries
and molecular transport are evident at the microscale. The individual catalyst particles are modeled as
batch microreactors (Hutchinson et al., 1992; Song & Luss, 2004) and incorporate interphase heat and
mass transport as well as intraparticle diffusion. The dynamics of these different scales, the gas and
solid phase interactions, heterogeneity of the different sizes of both the polymer and catalyst, and the
coupling of the chemical kinetics must be considered in the CFD model. Interparticle forces include the
376
Figure 1. CFD implementation strategy for rapid scale up and commercialization
familiar van der Waals, drag and gravity forces, but there is a growing realization of the importance of
electrostatic and capillary factors as well.
There have been several new catalyst discovery innovations the form of high throughput experimentation
using small-scale laboratory systems. Highly automated reactors with volumes of 10 cubic centimeters or
less are coupled with rapid analysis, spectroscopy, molecular modeling and large data bases to develop
novel new catalyst systems (Tuchbreiter et al., 2003). Hundreds of experiments can be ran in the course
of a week and require rapid screening for potential commercial application. CFD holds great promise
for commercialization of these catalysts by avoiding the need for costly and time-consuming pilot-plant
evaluations. Coupling of CFD with a chemical reaction engineering model has the potential to reduce
the need for expensive and time-consuming larger scale tests. This is shown figuratively in Figure 1.
CFD when coupled with today’s high-performance computers, newly developed numerical and vi-
sualization algorithms and improved user interfaces has the potential to truly transform the current
commercial paradigm. In today’s ever more competitive environment, it important that process engineers
have access to technology that allows them to make well-informed design, scale-up, and operational
choices in a timely and cost-efficient manner.
uniPOl Pe PrOcess
A sketch of a typical pilot-plant fluidized-bed reactor is shown in Figure 2. The reactor consists of a
fluidized-bed zone, a disengagement zone and a dome section. The top of the tapered disengagement
zone has a larger diameter (~2-4D) than the fluidized zone (D) so as to reduce the gas velocity to prevent
entrainment and elutriation of polymer particles. A gas mixture of monomer (ethylene), comonomer,
hydrogen and an inert gas (typically nitrogen) is used as a fluidizing medium. The height of the fluidized-
bed zone is ~4-8 D and the disengagement zone height is ~4-6 D. The fluidized bed always contains
polymer particles so as to distribute and to prevent elutriation of the injected catalyst particles (10-150
microns). At start up, the reactor is filled with polymer particles from a previous run. The gas is passed
through the distributor plate at a velocity where the drag force on each particle is significantly greater
377
Figure 2. Sketch of a unipol pilot scale gas-phase polymerization fluidized-bed reactor
than its weight, resulting in a highly fluidized bed. At the level of individual particles, the monomer dif-
fuses into the porous polymer structure until it reaches the active sites of the catalyst where the reaction
takes place. Due to rapid initial growth of the polymer particles, the original catalyst breaks into small
pieces, and the active sites are distributed throughout the polymer particle. The polymerization reaction
is highly exothermic and the excess heat is removed by conduction to the fluidization gas at the particle
surface. The polymer particles grow in size (500-3000 microns) until the catalyst is deactivated. The
larger particles move towards the bottom of the reactor near the distributor plate and are removed periodi-
cally through the product outlet. The catalyst particles are injected at a rate equal to their consumption
(i.e., each catalyst particle produces one polymer particle). The heated, monomer-rich gas (due to low
conversion per pass) exiting the reactor is compressed, cooled and recycled.
Multi-scale behavior of fluidized-bed Polymerization reactors
Figure 3 shows the different phenomena and length scales associated with olefin polymerization in
fluidized-bed reactors. The length scales are divided into three scales of interest: microscale (1-100 A),
mesoscale (10-3000 microns) and macroscale (1-10 m). On the microscale, the active sites are where the
reaction takes place. The important steps at this scale are: initiation of active sites, chain propagation,
chain transfer and deactivation of active sites. The polymerization mechanism has to be modeled cor-
rectly as it affects the mesoscale processes. The molecular weight distribution and particle composition
depends on the polymerization chemistry. During the early stage of polymerization, particle fragmenta-
tion takes place due to hydraulic forces. The mesoscale accounts for features at the particle length scale.
378
Figure 3. Multi-scale phenomena associated with olefin polymerization fluidized-bed reactor
At this scale the heat and mass transfer to and from the particle are important. There are several models
that deal with processes at the single particle level. Also, as described earlier, there are effects of thermal
and electrostatic phenomena at the mesoscale. At the macroscale, there are additional features like seg-
regation, mixing of polymer particles due to bubbles and elutriation/entrainment that must be modeled.
Phenomena over a wide range of length scale are involved in fluidized-bed polymerization reactors.
The reactor diameter is on the order of meters, the particle diameter varies from 10-3000 microns, sub-
fragments are on the order of 1-10 nanometers and catalyst active sites are of the order of angstroms. In
order for reaction to occur, the monomer must diffuse across the boundary layer surrounding the par-
ticle, through the macropores in the particle, and to an active site of the catalyst where the polymeriza-
tion reaction takes place. Because the catalyst is injected continuously, the residence time of each par-
ticle is different. Hence, the particles will have different sizes resulting in a particle size distribution
(PSD). Under unfavorable operating conditions, particles can agglomerate or break into fragments de-
pending on the local temperature and produce a wide distribution of particle sizes. A wide distribution
of particle sizes can cause phenomena such as segregation and elutriation. Large particles produced by
polymerization and agglomeration move towards the bottom of the reactor where they are removed.
This phenomenon can result in particle segregation. At high gas velocities smaller particles are elutri-
ated through the top of the reactor. Heat and mass transfer to and from the particle controls the local
particle temperature and phenomena like agglomeration and breakage. The reaction kinetics depends
upon the heat and mass transfer to the active sites. All of the above stated phenomena are highly coupled
and have a strong influence on the hydrodynamics of the fluidized bed. At present, it is difficult to have
a single model that can explain all of the multiscale phenomena of fluidized-bed polymerization reactors.
Because the time scale of the polymerization reaction is of the order of hours and the fluid dynamic time
scales are on the order of seconds, it is too expensive to run a three-dimensional CFD simulation that
captures the entire range of time scales using the current computational resources and codes. Thus, in
379
the present work, the reaction processes occurring on long time scales are decoupled from the fluid
mechanics as described below.
Microscale Model
At the microscale, the important steps are the polymerization kinetics and monomer diffusion through
the particle to active sites. The polymerization kinetics account for initiation of active sites, propagation
by insertion of monomer, chain transfer and deactivation of active sites. These steps need to be modeled
correctly as they determine the molecular weight distribution, polymer composition, properties of the
final polymer and the heat-release rate (Vladimir et al., 2003).
Mesoscale Model
The mesoscale model serves as an interface between the macroscale and microscale models. At these
length scales, the heat and mass transfer to/from the particle are important (Debling & Ray, 1995; Fer-
nandes & Lona, 2002; Floyd et al., 1987; Hoel et al., 1994). Thermal and/or electrostatic agglomeration
are also important phenomena that occur at these scales and the models need to capture such phenomena
when present. The mesoscale model needs information from the microscale model to describe particle
growth and heat generation, and from the macroscale model to fix the gas-phase boundary conditions
at the particle surface. There are several mesoscale models available in literature. The most widely
used models are the uniformly distributed model (McKenna & Soares, 200l), the polymeric flow model
(Schmeal & Street, 1971), the multilayer model (Veera et al., 2002) and the multigrain model (Floyd et
al., 1987). These models differ in how the active sites are distributed inside the growing particles. The
uniformly distributed model used in this work is relatively simple, while the other models give a more
detailed intraparticle description of the phenomena occurring during polymerization.
Macroscale Model
The macroscale model accounts for phenomena that include hydrodynamics, macromixing, particle size
distributions, segregation and elutriation/entrainment. The models that describe the macroscale behavior
of polymerization fluidized beds can be classified into two categories: kinetic models and CFD models.
Kinetic models (Ahmmed et al., 2009; Kiashemshaki et al., 2007) can be classified into three categories:
well-mixed models (single phase), constant bubble size models (emulsion-bubble phase) and bubble
growth models (emulsion-bubble). These models assume a constant bubble size or empirical relations
for bubble velocities and bubble diameter. In general, bubble velocities and diameters change spatially
and temporally in a fluidized bed and, hence, kinetic models have many drawbacks for explaining the
dynamics of the fluidized bed.
There are two main approaches for modeling gas-solids flows: Eulerian and Lagrangian models.
Under these approaches there are three main categories depending on how the Eulerian and Lagrangian
methods (Van der Hoef et al., 2008) are used to treat the gas and solid phases. Direct numerical simula-
tion (Mittal & Iaccarino, 2005) solves the microscopic transport equations for the gas and solid phases,
and the gas-solids coupling is through the boundary conditions at the surface of the solid phase. These
models are highly accurate but at present are restricted to laboratory scales where the length scales are of
380
the order of 0.01 m. The next level of models is the discrete element model (Hoomans et al., 1996; Tsuji
et al., 1993) that solves the Newtonian equations for individual point particles and the Navier-Stokes
equations for the gas phase. The gas-solids interactions are modeled by engineering correlations for the
drag force and heat/mass transfer. Due to the large number of particles in a fluidized bed, these models
are restricted to length scales of the order of 0.1 m. The highest level of models is the Eulerian multi-
fluid model (Fan & Fox, 2008; Gidaspow, 1994; Van der Hoef et al., 2008) that describes the solid and
gas phases as pseudo fluids, and solves mass and momentum equations for each phase. Because they
can be eventually scaled up to model pilot-plant and commercial-scale reactors, Eulerian multi-fluid
models are used in the present work.
cHeMical reactiOn engineering MOdel fOr POlyMer Particles
If the catalyst size distribution and the polymerization kinetics are known a mesoscale chemical reac-
tion engineering (CRE) model can be used to predict the final polymer size distribution and particle
temperature. The results from the CRE model can then be used in the macroscale multi-fluid model to
predict segregation, elutriation and hot spots in the reactor. This macroscale information is useful in
design and scale-up of fluidized-bed polymerization reactors.
Polymerization Kinetics for Metallocene catalyst
A simple multi-step mechanistic scheme is used to describe the microscale polymerization chemistry.
The fundamental reactions considered are the initiation, propagation and termination of the active sites.
Chain transfer is ignored in this work since it plays a minor role in the heat generation of a polymeriza-
tion. The resulting kinetic scheme is
k
Initiation : c  
i
→ c* (2)
k
Propagation : Pn*(c* ) + M  
p
→ Pn*+1 (3)
k
Decay : Pn*(c*)  
d
→ Pn + c 0 (4)
where ki, kp, kd are reaction-rate constants for the initiation, propagation and termination reactions, re-
spectively. c denotes a potential catalyst active site, c* an active catalyst site, Pn* a living polymer of
length n, M monomer, Pn dead polymer and c0 a dead site. An Arrhenius rate equation is used for each
rate constant:
 −∆E 
ki = ki 0 exp  d 
 (5)
 RTsα 
381
Figure 4. Mass and heat transfer model for single growing particle
 −∆E 
kd = kd 0 exp  d 
 (6)
 RTsα 
 −∆E 
k p = k p 0 exp  d 
 (7)
 RTsα 
where ki0, kp0, kd0 and Ei, Ep, Ed are the pre-exponential factors and activation energies of initiation, propa-
gation and termination, respectively. The pre-exponential factors are seldom measured directly and are
found by fitting the experiment rate profiles to rate profiles obtained from CRE model.
cre Model
A uniformly distributed lumped thermal model is used to describe the mesoscale phenomena. The model
assumes that the particle temperature is uniform but different from the gas phase, and that the active
sites are uniformly distributed inside the polymer particle. The mass and heat transfer are assumed to
occur as schematically represented in Figure 4.
The heat removal is controlled by the external boundary layer and described by a convective heat
transfer coefficient. The intraparticle heat transfer is assumed to be negligible (Song, 2004). The control-
ling factor in this case is the Nusselt number, the ratio of the convective to conductive heat transfer
coefficient. The heat generated by the polymerization reaction is given by the product of the reaction
rate and heat of polymerization. The heat balance equation combines the heat removal and heat genera-
tion into Equation (14) shown below. The mass transfer is controlled by the convective mass transfer in
the boundary layer and the internal particle diffusive mass transfer. Similar to the convective heat trans-
fer case, the convective mass transfer is described by a convective mass transfer coefficient. The control-
ling factor is the Sherwood number, the ratio of the convective to diffusive mass transfer in the external
boundary layer. Diffusion within the particle is an important consideration. Polymers have very low
diffusion coefficients and can limit the rate of reaction. This factor is accounted for using the isothermal
382
effectiveness factor, η, and is defined in terms of the Thiele modulus, ψ. The resulting equations are
given by
3(ψcothψ − 1)
η= (8)
ψ2
k 'pC *
ψ=R (9)
De
where R is the particle radius, k 'p is the propagation rate constant corrected for the monomer partial
pressure and De the effective monomer diffusivity. The mass balance equation combines the external
mass transfer and rate of reaction into Equation (15) also shown below.
The CRE model (Fan et al., 2007) combines the mass, species and energy balances for an individual
growing particle into Equations (10)-(16). The gas phase far from the particle is considered to have a
fixed temperature and composition, and the mass/energy-transfer coefficients are used to describe the
fluxes through the boundary layer to the particle surface as described above. The resulting balance equa-
tions solved in the CRE model for the solid phase are given by
d εs
ρs = ε s kc av (C g − Cm )M w (10)
dτ
d
(ε [c ]) = −ε s ki [c ] (11)
dτ s
d
(ε [c* ]) = −ε s ki [c ] − ε s ki [c* ] (12)
dτ s
d
(ε s [c 0 ]) = −ε s ki [c* ] (13)
dτ
dTs
ε s ρ sC ps = ε s h f av (Tg − Ts ) − ε s ηk 'p [c* ]CM ∆H r (14)
dτ
d
ρs (ε X ) = ε s kc av (C g − CM )M w − ηε s ρ s k p' X sM [c* ] (15)
d τ s sM
ε s = aL3s (16)
where ρs is the polymer density, εs the volume fraction of the solid phase, τs the residence time of the
particles, kc the mass-transfer coefficient, av the ratio of particle external surface area to volume, Cg
383
and CM the monomer concentration in the gas and solid phase, respectively, Mw the molecular weight
of the monomer, Cps the specific heat capacity of the solid, Ts and Tg the temperatures of the solid and
gas phase, respectively, hf the heat-transfer coefficient, ΔHr the heat released due to polymerization, XsM
the monomer mass fraction in the solid phase, Ls the particle diameter and a a constant relating particle
diameter to the solid volume fraction (i.e., proportional to particle number density in the fluidized bed).
Note that the heat/mass-transfer coefficients depend on the particle diameter through the Reynolds,
Nusselt and Sherwood numbers.
Equation (10) is the mass balance for the solid phase, and the right-hand-side (RHS) term describes
the mass transfer of monomer from the gas phase to the solid phase. (Note that the solid phase includes
the polymer and the monomer inside the external surface of the particle.) The growth of the solid particle
is given by the increase of volume fraction of the particle. The volume fraction of the particle is related
to its size by a simple relation as given in Equation (16), which is valid when there is no aggregation
and breakage. Equations (11), (12) and (13) describe the species balances for the potentially available
active sites, active sites and dead sites, respectively. Equation (14) is the particle thermal energy balance.
The terms on the RHS describe the heat transfer from the solid to the gas and the heat produced by the
exothermic polymerization reaction. Equation (15) is the monomer mass balance in the solid phase. The
terms on the RHS describe the mass transfer from the gas phase and the consumption of the monomer
due to the propagation reaction. The heat and mass-transfer coefficients are found using simple empirical
relations for the Nusselt and Sherwood numbers.
In this work the set of ordinary differential equations (ODE’s) given by equations (10)-(16) are
solved using The Mathworks Matlab® software with initial conditions given by the catalyst diameter,
the gas-phase temperature, and assuming that all catalyst sites are type c. The CRE model is used to fit
the experimental data to obtain the kinetic parameters. Figure 5 shows the reaction rates from the CRE
model and associated experimental data. A two-site kinetic model was utilized to fit the experimental data.
The rate profiles have a peak at the start of the polymerization, and then the reaction rate diminishes due
to mass-transfer limitations and site deactivation. In the two-site model, site 1 is the major contributor
to the temperature rise. The CRE model provides the particle diameter and temperature as a function of
the particle residence time for a given initial catalyst particle diameter.
Figure 5. Example reaction rate profiles from an experiment and from the two-site CRE model. The
reaction rates and time are each scaled by a constant
384
Model for Particle size distribution
After the kinetic parameters of a metallocene olefin polymerization reaction have been determined as
described above, the CRE model can be combined with quadrature method of moments (QMOM) and
the particle residence time distribution (RTD) to find the polymer size distribution in the fluidized-bed
reactor. The Discrete method (also known as classes method (Ramkrishna, 2000) computes the PSD
directly but it is computationally expensive and for this reason QMOM is used. The flow chart for this
process is shown in the Figure 6. Given the catalyst particle size distribution (PSD) function f(L) found
from experiments, the integer-order catalyst-size moments are calculated. In this work, three quadrature
nodes, computed from the moments using QMOM (Fan et al., 2004; Marchisio et al., 2003a, 2003b)
are used to represent the catalyst PSD. The CRE model is applied to each node by assuming that the
initial size Ls0 has the diameter of the quadrature node. From the CRE model the evolution of particle
diameter (Figure 8) and particle temperature (Figure 9) with age for each node is obtained. According
to the definition of moments, the moments of final product for each node can be obtained by integrating
each curve with the particle RTD function. The moments predicted from the model compared well with
experimental results and the probability plots are shown in Figure 7. The evolution of particle tempera-
ture change with size is plotted in Figure 10.
Figure 6. Flow chart for process used to determine polymer PSD. The final polymer PSD is obtained
by combination of CRE model and QMOM
385
Figure 7. The PSD of catalyst and polymer from experiment and CRE model. Particle size log scaled
relative to a constant
Figure 8. Evolution of particle diameter with age. The particle diameter is scaled by the final average
particle size and age by the average residence time
Multi-fluid eulerian MOdel
In the multi-fluid model the gas and solid phases are treated as interpenetrating continua. The model
treats the solid phase as a pseudo fluid, and solves the mass and momentum conservation equations for
each phase. The drag force plays a dominant role in gas-solids fluidized beds, and the Gidaspow drag
law is used in this work. Solid-phase stresses and pressure are calculated from kinetic theory of granular
flow (KTGF) and frictional theory. A granular temperature equation is also solved for the solid phase.
The general equations for the multi-fluid model will not be discussed here and can be found elsewhere
(Fan & Fox, 2008; Gidaspow, 1994; Goldschmidt et al., 2001; Van der Hoef et al., 2008).
grid generation
Grid generation is done using the ANSYS software Gambit 2.3. A Cartesian coordinate system is used
for grid generation, hexahedron elements are formed for the three-dimensional geometry and quadrilat-
386
Figure 9. Evolution of particle temperature with age. The particle temperature rise is scaled by a refer-
ence temperature and age by the average residence time
Figure 10. Evolution of particle temperature with particle diameter. The particle diameter is scaled by
the final average particle size and temperature rise by a reference temperature
eral elements are formed for the two-dimensional geometry. Some specifics about the grid generation
are shown in Table 2. A sample grid is shown in Figure 11. The X and Z directions are along the cross
section of the reactor, and the Y direction is along the height of the reactor.
The center of the base of the pilot-plant reactor is considered the origin. The X, Z axis are used to
represent the cross section of the reactor and their values range from -D/2 to D/2. The Y axis is used to
represent the height of the reactor. Z= 0 represents a plane cut from the three-dimensional geometry.
This information is useful in understanding the results.
simulations
Simulations (two-dimensional and three-dimensional) were done for the pilot-plant reactor using the
ANSYS FLUENT 6.3 software platform on the high performance computing (HPC) machine at Iowa
State University. One gas phase and three solid phases are considered in this work. The three solid phases
with different particle diameters are used to represent the PSD, and the diameters are found using QMOM
387
Figure 11. Front view of the grid used for pilot plant reactor simulations. Two and three-dimensional
grids are used in the simulations
Table 3. Parameters used in the pilot-plant simulations
Solid1 Solid2 Solid3 Gas

Diameter(mm) 0.523 1.176 1.751 -
Density(kg/m )
3
843.0 843.0 843.0 22.1
Volume fraction 0.00453 0.194 0.310 0.491
from the CRE model. The particle properties are shown in Table 3. The diameter and volume fractions
are obtained from the CRE model and QMOM. Lift forces and virtual mass effects are neglected in the
simulations as they are small compared to the drag force. One hundred seconds of flow time is simu-
lated and the last ninety seconds of flow time is time averaged to find the pressure drop and holdup for
comparison with experiments.
results and discussion
Three-dimensional CFD (pilot-plant reactor) simulations are computationally expensive and have to be
run on multiprocessors to reduce the computation time. Second-order discretization schemes are more
accurate than first-order methods. First-order methods are numerically diffusive and so the structures
are not seen clearly. Hence, many details are not resolved. Bubbles simulated by second-order schemes
388
Figure 12. Variation of mean static pressure versus the height of the reactor. Three dimensional simula-
tions predicted the bed height and pressure drop close to the experimental values. The static pressure
and reactor height are scaled relative to the maximum
Table 4. Value of bed height and pressure drop from simulations and experiment
Relative Bed Height Pressure Drop % Increase in Bed Height

[Pascal/m] (relative to experiments)
Experiment (for reference) 1.00 3045 -
2D simulation 1.26 2349 26.0
3D – Coarse Grid 1.31 2101 31.5
3D – Refined Grid 1.05 2900 5.20
look more realistic than simulated by first-order schemes. For these reasons, second-order schemes are
used in this work.
Bed Height and Pressure Drop
CFD models have to be verified and validated with experimental data. Experimental data are obtained
from a pilot-plant reactor at Univation Technologies, Baytown, Texas. The scaled bed height measured
from experiments is 0.55. Figure 12 shows the variation of mean static pressure versus the height of the
reactor for three different cases: a three-dimensional coarse-grid simulation, a refined-grid simulation
and a two-dimensional simulation. Bed height and pressure drop per unit length are tabulated in Table 4.
Pressure decreases linearly in the two- and three-dimensional simulations and the slope (pressure
drop/unit length) is highest for three-dimensional–refined case. The relation between pressure drop and
the buoyant weight of solid particles is
∆p = −(ρ p − ρ g )(1 − ε s )gH (17)
389
Figure 13. Contour plots of gas-phase volume fraction for (a) coarse grid and (b) refined grid. Fine
structures are resolved by the refined grid
where ρp, ρg are the density of solid (polymer) and gas-phase, 〈ε〉 the mean volume fraction of gas-phase,
g acceleration due to gravity and H the bed height.
The pressure drop calculated from Equation (17) is 3151 Pascal/m. The three-dimensional refined
grid simulations predicted a bed height reasonably close to the experimental result and also matched
well with simulations done in MFIX (Fan, 2006). The pilot-plant geometry and flow in the reactor is
three dimensional. Strictly speaking two-dimensional simulations for the pilot plant are not applicable
but can be useful for sensitivity analysis. The bubble coalescence and break up make the flow three
dimensional. The two-dimensional simulations predicted a 26% higher bed height than the experimental
value similar to those of Enwald et al. (1996).
Bed height predicted with the coarse mesh is 31% higher than the experimental value. Errors in the
simulations increase with increased grid size. Coarse grids do not resolve the spatial structures at small
scales. To resolve small-scale structures the grid should be refined. Since the small-scale structures affect
the overall flow, the results obtained using a coarse mesh case may not be accurate.
Coarse vs. Refined Grid
The instantaneous contour plots of gas-phase volume fraction for different grids are shown in Figure 13.
Three dimensional simulations are done using the coarse and refined grids as shown in Table 2.From
the contour plots of the coarse-grid simulation, we can see that the gas volume fraction is uniform and,
in reality, this is not the case. Refined grids are able to capture the small-scale behavior and simula-
tions are close to the complex flow behavior in fluidized beds (Andrews et al., 2005). In the coarse-grid
simulation the bed height is high due to higher effective drag, and the solids volume fraction is nearly
uniform in the bed. In contrast, the solids near the wall are better resolved with the refined grid, and
390
Table 2. Grid sizes
Dimensions X [cm] Y [cm] Z [cm]

2D 0.95 2.45 -
3D Coarse 8.00 8.00 8.00
3D Refined 2.50 2.50 2.50
the bed height is closer to the experimental value. Fine-grid simulations also predict behavior such as
downward gas and solid flow near the walls. The grid is resolved till the macroscopic properties like
pressure drop, bed height, etc., matched well with experimental results and further refinement did not
change the simulation results to any appreciable extent.
Segregation
Study of segregation is necessary for the design and scale-up of fluidized beds (Fan, 2006). In the case
of fluidized-bed polymerization reactors, the larger particles move to the bottom of the reactor and fines,
and catalyst can be elutriated with the gas. Three dimensional simulations are done using refined grid
as shown in Table 2.The average particle size normalized by the average particle size in the bed as a
function of the bed height is shown in Figure 14. The average particle size at the bottom of the reactor
is 1.1 times the average particle size for the whole bed. The bottom of the reactor is well mixed, and
little segregation is observed. In the expansion and the dome area the average particle size at the top is
only 50-60% of the average particle size in the bed. These results show segregation in the expansion
zone and dome section.
Slug Flow
Slug flow (Lee et al., 2002) is a multiphase flow regime in which large bubbles are formed and the bubble
size approaches the diameter of the reactor. Slug flow is generally observed in fluidized-bed reactors
Figure 14. Average particle size along the bed height. The bed height is scaled by the height of the
fluidized-bed. The particle size is scaled by the average particle size
391
that have a small diameter to height ratio. Three dimensional simulations are done using refined grid as
shown in Table2.Figure 15 shows the instantaneous contour plots of gas volume fractions of the three-
dimensional (Z=0) simulation results. Areas marked with grey are where the gas volume fraction is close
to unity and the black regions show high solid concentrations. Small bubbles are formed at the bottom
of the reactor, and the bubble size increases as they rise. Large bubbles are formed from coalescence of
small bubbles. The simulation result is shown at four different times. At t =0 gas is introduced into the
reactor and at t=4.75 seconds of flow time a large bubble close to the diameter of the reactor is seen in
the simulation. Experiments show that slugs are formed in the pilot-plant reactor, which are predicted
well by the three-dimensional simulations.
cOnclusiOn and future directiOns
In this work, results for two- and three-dimensional CFD simulations of a pilot-plant fluidized-bed reactor
are reported. The simulation results for pressure drop and bed height are compared with experimental
results. Significant differences are found between two- and three-dimensional, refined and coarse grids.
The multi-fluid model is able to predict the slug-flow behavior observed in experiments. Polymer size
distribution obtained from CRE model and QMOM matched well with experiments. Overall, the model
results are very promising.
Electrostatic charges play an important role in the operation of fluidized-bed polymerization reactors.
The presence of charge influences the bed dynamics and creates problems in reactor operation. When the
Figure 15. Contour plots of volume fraction of gas (a) t = 0.75 sec (b) t = 2.75 sec (c) t =4.75 sec (d)
t=6.5sec.Slugs are formed at t = 4.75 sec
392
electrostatic charges exceed a critical value, the electrostatic force created causes the particles to stick to
the wall. This is known as wall sheeting. A significant amount of wall sheeting in the reactor can cause
defluidization and plant shut down. An excellent review on electrostatics and wall sheeting in fluidized-
bed polymerization reactors can be found in Hendrickson (2006). At present there are no computational
models that can describe the electrostatic behavior in fluidized-bed polymerization reactors (Mahdi,
2002). Future work should be directed at this problem, and our current efforts are directed at incorpora-
tion of an electrostatic force component into the CFD framework. The other problem involved is the
difficulty in obtaining experimental results for electrostatic model validation. The Bi group (Mehrani
et al., 2005, 2007) has performed many electrostatic experiments and is studying the mechanism for
charge generation and charge dissipation. They are also exploring ways to reduce charges effects in the
bed. More detailed experimental electrostatic measurements will be required for model validation and
refinement. The other issue is the scale-up from pilot-plant to commercial reactors. Coarse-grid simula-
tions (Andrews et al., 2005; Igci et al., 2008) that include the effect of unresolved flow structures on the
bed dynamics have the potential to reduce computation time. Future work in this field should include
development of a multi-fluid model to describe the electrostatic behavior in fluidized-bed polymerization
reactors and scale-up to commercial reactors in a more reasonable time frame than is currently required.
references
Ahmmed, S. I., Mohamed, A. H., & Nayef, M. G. (2009). Modified mathematical model for gas phase
olefin polymerization in fluidized-bed catalytic reactor. Chemical Engineering Journal.
Andrews, A. T. IV, Loezos, P. N., & Sundaresan, S. (2005). Coarse-grid simulation of gas-particle flows in
vertical risers. Industrial & Engineering Chemistry Research, 44(16), 6022–6037. doi:10.1021/ie0492193
Baerns, M. (1966). Effect of interparticle adhesive forces on fluidization of fine particles. Industrial &
Engineering Chemistry Fundamentals, 5(4), 508–516. doi:10.1021/i160020a013
Bafrnec, M., & Bena, J. (1972). Quantitative data on the lowering of electrostatic charge in a fluidized
bed. Chemical Engineering Science, 27(5), 1177–1181. doi:10.1016/0009-2509(72)80030-7
Baron, T., Briens, C. L., Bergougnou, M. A., & Hazlett, J. D. (1987). Electrostatic effects on entrainment
from a fluidized bed. Powder Technology, 53(1), 55–67. doi:10.1016/0032-5910(87)80125-0
Behjat, Y., Dehnavi, M. A., Shahhosseini, S., & Hashemabadi, S. H. (2008). Heat transfer of polymer
particles in gas phase polymerization reactors. International Journal of Chemical Reactor Engineering,
6(A81). doi:10.2202/1542-6580.1649
Bokkers, G. A., van Sint Annaland, M., & Kuipers, J. A. M. (2004). Mixing and segregation in a bidis-
perse gas-solids fluidised bed: A numerical and experimental study. Powder Technology, 140, 176–186.
doi:10.1016/j.powtec.2004.01.018
Boland, D., & Geldart, D. (1972). Electrostatic charging in gas fluidized beds. Powder Technology, 5(5),
289–297. doi:10.1016/0032-5910(72)80033-0
Borruso, A. V. (2008). CEH Marketing Research Report Linear Low-Density Polyethylene (LLDPE)
Resins. Chemical Economics Handbook. Menlo Park, CA: SRI Consulting.
393
Briens, C. L., Bergougnou, M. A., Inculet, I. I., Baron, T., & Hazlett, J. D. (1992). Size distribution
of particles entrained from fluidized beds: Electrostatic effects. Powder Technology, 70(1), 57–62.
doi:10.1016/0032-5910(92)85054-Y
Ciborowski, J., & Wlodarski, A. (1962). On electrostatic effects in fluidized beds. Chemical Engineering
Science, 17(1), 23–32. doi:10.1016/0009-2509(62)80003-7
Debling, J. A., & Ray, W. H. (1995). Heat and mass transfer effects in multistage polymerization processes:
Impact polypropylene. Industrial & Engineering Chemistry Research, 34, 3466–3480. doi:10.1021/
ie00037a035
Dehnavi, A. M., Shahhosseini, S., Hashemabadi, S. H., & Ghafelebashi, S. M. (2008). CFD based evalu-
ation of polymer particles heat transfer coefficient in gas phase polymerization reactors. International
Communications in Heat and Mass Transfer, 35, 1375–1379. doi:10.1016/j.icheatmasstransfer.2008.07.017
Elsdon, R., & Mitchell, F. R. G. (1976). Contact electrification of polymers. Journal of Physics. D, Ap-
plied Physics, 9, 1445–1460. doi:10.1088/0022-3727/9/10/010
Enwald, H., Peirano, E., & Almstedt, A. E. (1996). Eulerian two phase flow theory applied to fluidiza-
tion. International Journal of Multiphase Flow, 22, 21–66. doi:10.1016/S0301-9322(96)90004-X
Eriksson, E. J. G., & Mckenna, T. F. (2004). Heat transfer phenomena in gas olefin polymerization us-
ing computational fluid dynamics. Industrial & Engineering Chemistry Research, 43(23), 7251–7260.
doi:10.1021/ie034328n
Fan, R. (2006). Computational fluid dynamics simulation of fluidized bed polymerization reactors.
Unpublished doctoral dissertation, Iowa State University, IA.
Fan, R., & Fox, R. O. (2008). Segregation in polydispersed fluidized beds: Validation of a multi-fluid
model. Chemical Engineering Science, 63(1), 272–285. doi:10.1016/j.ces.2007.09.038
Fan, R., Fox, R. O., & Muhle, M. E. (2007). Role of intrinsic kinetics and catalyst particle size distribu-
tion in CFD simulations of polymerization reactors. Fluidization 12 conference, (Article 122), 994-1000.
Fan, R., Marchisio, D. L., & Fox, R. O. (2004). Application of the direct quadrature method of moments to
polydisperse gas-solids fluidized beds. Powder Technology, 139, 7–20. doi:10.1016/j.powtec.2003.10.005
Fernandes, F. A. N., & Lona, L. M. F. (2002). Heterogeneous modeling of fluidized bed polymerization
reactors. Influence of mass diffusion into the polymer particle. Computers & Chemical Engineering,
26, 841–848. doi:10.1016/S0098-1354(02)00010-8
Floyd, S., Heiskanen, T., Taylor, W. T., Mann, G. E., & Ray, W. H. (1987). Polymerization of olefins through
heterogenous catalysis. VI. Effect of particle and mass transfer on polymerization behavior and polymer
properties. Journal of Applied Polymer Science, 33(4), 1021–1065. doi:10.1002/app.1987.070330402
Gajewski, A. (1985). Investigation of the electrification of polypropylene particles during the fluidization
process. Journal of Electrostatics, 17(3), 289–298. doi:10.1016/0304-3886(85)90029-4
Gidaspow, D. (1994). Multiphase Flow and Fluidization: Continuum and Kinetic Theory Descriptions.
Boston: Academic Press.
394
Glicksman, L. (1988). Scaling relationships for fluidized beds. Chemical Engineering Science, 43(6),
1419–1421. doi:10.1016/0009-2509(88)85118-2
Gobin, A., Neau, H., Simonin, O., Llinas, J. R., Reiling, V., & Seilo, J. L. (2003). Fluid dynamic numeri-
cal simulation of a gas phase polymerization reactor. International Journal for Numerical Methods in
Fluid Dynamics, 43, 1199–1220. doi:10.1002/fld.542
Goldschmidt, M. J. V., Kuipers, J. A. M., & van Swaaij, W. P. M. (2001). Hydrodynamic modeling of
dense gas-fluidised beds using the kinetic theory of granular flow: effect of coefficient of restitution
on bed dynamics. Chemical Engineering Science, 56, 571–578. doi:10.1016/S0009-2509(00)00262-1
Griffiths, J. F., Barnard, J. A., & Bradley, J. N. (1995). Flame and Combustion, (3rd). New York: Chap-
man & Hall.
Guardiola, J., Rojo, V., & Ramos, G. (1996). Influence of particle size, fluidization velocity and relative
humidity on fluidized bed electrostatics. Journal of Electrostatics, 37(1-2), 1–20. doi:10.1016/0304-
3886(96)00002-2
Hendrickson, G. (2006). Electrostatics and gas phase fluidized bed polymerization reactor wall sheeting.
Chemical Engineering Science, 61(4), 1041–1064. doi:10.1016/j.ces.2005.07.029
Hoel, E. L., Cozewith, C., & Byrne, G. D. (1994). Effect of diffusion on heterogeneous ethylene pro-
pylene copolymerization. AIChE Journal. American Institute of Chemical Engineers, 40, 1669–1684.
doi:10.1002/aic.690401009
Hoomans, B. P. B., Kuipers, J. A. M., Briels, W. J., & van Swaaij, W. P. M. (1996). Discrete particle
simulation of bubble and slug formation in a two-dimensional gas-fluidized bed: A hard sphere approach.
Chemical Engineering Science, 51(1), 99–118. doi:10.1016/0009-2509(95)00271-5
Hutchinson, R. A., Chen, C. M., & Ray, W. H. (1992). Polymerization of olefins through heterogeneous
catalysis X: Modeling of particle growth and morphology. Journal of Applied Polymer Science, 44,
1398–1414. doi:10.1002/app.1992.070440811
1431–1448. doi:10.1002/aic.11481
Kaneko, Y., Shiojima, Y. T., & Horio, M. (1999). DEM simulation of fluidized beds for gas-phase olefin
polymerization. Chemical Engineering Science, 54, 5809–5821. doi:10.1016/S0009-2509(99)00153-0
Kiashemshaki, A., Rahmat, S., & Mostoufi, N. (2007). Reactor modeling of fluidized bed ethylene
polymerization using dynamic two phase behavior. Iranian Journal of Chemistry and Chemical Engi-
neering, 26, 51–60.
Kim, J. Y., & Choi, K. Y. (2001). Modeling of particle segregation phenomena in a gas phase fluidized
bed olefin polymerization reactor. Chemical Engineering Science, 56, 4069–4083. doi:10.1016/S0009-
2509(01)00078-1
Lee, S. H., Lee, D. H., & Kim, S. D. (2002). Slug characteristics of polymer particles in a gas-solids
fluidized bed. Korean Journal of Chemical Engineering, 19(2), 351–355. doi:10.1007/BF02698428
395
Mahdi, F. A., Dudley, A. S., & Sundaresan, S. (2002). The effect of static electrification on gas-solids
flows in vertical risers. Industrial & Engineering Chemistry Research, 41(25), 6224–6234. doi:10.1021/
ie010982w
Marchisio, D. L., Pikturna, J. T., Fox, R. O., Vigil, R. D., & Barresi, A. A. (2003a). Quadrature method of
moments for population-balance equations. AIChE Journal. American Institute of Chemical Engineers,
49, 1266–1276. doi:10.1002/aic.690490517
Marchisio, D. L., Vigil, R. D., & Fox, R. O. (2003b). Quadrature method of moments for aggregation-
breakage processes. Journal of Colloid and Interface Science, 258, 322–334. doi:10.1016/S0021-
9797(02)00054-1
McKenna, T. F., & Soares, J. B. P. (2001). Single particle modeling for olefin polymerization on sup-
ported catalysts: A review and proposals for future developments. Chemical Engineering Science, 56(13),
3931–3941. doi:10.1016/S0009-2509(01)00069-0
McKenna, T. F., Spitz, R., & Cokljat, D. (1999). Heat transfer from catalysts with computational fluid
dynamics. AIChE Journal. American Institute of Chemical Engineers, 45, 2392–2410. doi:10.1002/
aic.690451113
Mehrani, P., Bi, H. T., & Grace, J. R. (2005). Electrostatic charge generation in gas-solids fluidized beds.
Journal of Electrostatics, 63, 165–173. doi:10.1016/j.elstat.2004.10.003
Mehrani, P., Bi, H. T., & Grace, J. R. (2007). Electrical behaviour of different fines added to beds of glass
beads and polyethylene particles. Journal of Electrostatics, 65(1), 1–10. doi:10.1016/j.elstat.2006.05.002
Mittal, R., & Iaccarino, G. (2005). Immersed boundary methods. Annual Review of Fluid Mechanics,
37, 239–261. doi:10.1146/annurev.fluid.37.061903.175743
Mountain, J. R., Mazumder, M. K., Sims, R. A., Wankum, D. L., Chasser, T., & Pettit, P. H. Jr. (2001).
Triboelectric charging of polymer powders in fluidization and transport processes. IEEE Transactions
on Industry Applications, 37(3), 778–784. doi:10.1109/28.924759
Ramkrishna, D. (2000). Population balances: theory and applications to particulate systems in engi-
neering. San Diego, CA: Academic Press.
Ray, W. H., & Villa, C. M. (2000). Nonlinear dynamics found in polymerization processes – a review.
Chemical Engineering Science, 55, 275–290. doi:10.1016/S0009-2509(99)00323-1
Revel, J., Gatumel, C., Dodds, J. A., & Taillet, J. (2003). Generation of static electricity during fluidi-
zation of polyethylene and its elimination by air ionization. Powder Technology, 135–136, 192–200.
doi:10.1016/j.powtec.2003.08.015
Schmeal, W. R., & Street, J. R. (1971). Polymerization in expanding catalyst particles. AIChE Journal.
American Institute of Chemical Engineers, 17(5), 1188–1197. doi:10.1002/aic.690170526
Song, H. (2004). Bounds on operating conditions leading to melting during olefin polymerization. Doc-
toral disseration. Houston, TX: University of Houston.
396
Song, H., & Luss, D. (2004). Bounds on operating conditions leading to melting during olefin polym-
erization. Industrial & Engineering Chemistry Research, 43, 270–282. doi:10.1021/ie020981j
Tsuji, Y., Kawaguchi, T., & Tanaka, T. (1993). Discrete particle simulation of two dimensional fluidized
Tuchbreiter, A., Marquardt, J., Kappler, B., Honerkamp, J., Kristen, M. O., & Muhlhaupt, R. (2003).
High-output polymer screening: exploiting combinatorial chemistry and data mining tools in catalyst and
polymer development. Macromolecular Rapid Communications, 24, 47–62. doi:10.1002/marc.200390014
Van der Hoef, M. A., Van Sint Annaland, M., Deen, N. G., & Kuipers, J. A. M. (2008). Numerical
simulation of dense gas-solids fluidized beds: A multiscale modeling strategy. Annual Review of Fluid
Veera, U. P., Weickert, G., & Agarwal, U. S. (2002). Modeling monomer transport by convection dur-
ing olefin polymerization. AIChE Journal. American Institute of Chemical Engineers, 48, 1062–1070.
doi:10.1002/aic.690480515
Vladimir, B. S., Vladimir, A. Z., & Valerii, A. K. (1996). Investigation of kinetics of ethylene polymer-
ization support titanium-magnesium catalyst of various composition. Macromolecular Chemistry and
Physics, 197(5), 1615–1631. doi:10.1002/macp.1996.021970504
Wolny, A., & Ka Zmierczak, W. (1989). Triboelectrification in fluidized bed of polystyrene. Chemical
Engineering Science, 44(11), 2607–2610. doi:10.1016/0009-2509(89)85204-2
Xie, T., McAuley, K. B., Hsu, J. C. C., & Bacon, J. W. (1994). Gas phase ethylene polymerization:
Production processes, polymer properties, and reactor modeling. Industrial & Engineering Chemistry
Research, 33, 449–479. doi:10.1021/ie00027a001
Zhao, H., Castle, G. S. P., Inculet, I. I., & Bailey, A. G. (2000). Bipolar charging in polydisperse poly-
mer powders in industrial processes. In Proceedings of the IEEE-IAS Annual Meeting, Rome, October
2000, (pp. 835–841).
397
398
Chapter 13
Validation Approaches
to Volcanic Explosive
Phenomenology
Sébastien Dartevelle
Los Alamos National Laboratory, USA
abstract
Large-scale volcanic eruptions are inherently hazardous events, hence cannot be described by detailed
and accurate in situ measurements. As a result, volcanic explosive phenomenology is poorly understood
in terms of its physics and inadequately constrained in terms of initial, boundary, and inflow conditions.
Consequently, little to no real-time data exist to validate computer codes developed to model these
geophysical events as a whole. However, code validation remains a necessary step, particularly when
volcanologists use numerical data for assessment and mitigation of volcanic hazards as more often per-
formed nowadays. We suggest performing the validation task in volcanology in two steps as followed.
First, numerical geo-modelers should perform the validation task against simple and well-constrained
analog (small-scale) experiments targeting the key physics controlling volcanic cloud phenomenology.
This first step would be a validation analysis as classically performed in engineering and in CFD sci-
ences. In this case, geo-modelers emphasize on validating against analog experiments that unambigu-
ously represent the key-driving physics. The second “geo-validation” step is to compare numerical
results against geophysical-geological (large-scale) events which are described ―as thoroughly as pos-
sible― in terms of boundary, initial, or flow conditions. Although this last step can only be a qualitative
comparison against a non-fully closed system event —hence it is not per se a validation analysis—, it
nevertheless attempts to rationally use numerical geo-models for large-scale volcanic phenomenology.
This last step, named “field validation or geo-validation”, is as important in order to convince policy
maker of the adequacy of numerical tools for modeling large-scale explosive volcanism phenomenology.
DOI: 10.4018/978-1-61520-651-3.ch013
1. intrOductiOn
Large-scale explosive volcanic eruption cloud is one of the most enthralling yet hazardous phenomena
one can witness in Nature (see Figures 1 and 2). Such catastrophic events potentially pose a major threat
to human life, livestock, the environment at large, and aircraft safety. They can also potentially disrupt
all social and economical activities for many years after the eruption. Typically, these volcanic clouds
consist of hot magmatic fragments and lithic clasts dispersed in a carrying gas phase. Initially, this hot
multiphase mixture is expelled subvertically from a volcanic vent at speeds up to a few hundred of seconds
and with densities greater than the surrounding atmosphere (negative buoyancy). As this momentum-
driven jet “thrusts” upwards into the atmosphere, it expands, hence dilutes itself and decreases its own
bulk density w.r.t. the ambient atmosphere. Consequently, the jet becomes a buoyancy-driven plume
(Valentine, 1998;Dartevelle et al., 2004;Dartevelle, 2005). The exact fate of this buoyant plume will be
controlled by a balance between three major forces, viz., (1) the buoyancy force, which pulls the cloud
upward to higher altitudes, (2) the gravity force, which exerts a downwards pull, and (3) turbulence,
which has an overall dissipative effects on the clouds and slows it down (this is often characterized as
the “atmospheric drag” effect). In addition to the natural dissipative effects, turbulence may also have
important supplementary non-linear effects upon the rising plume. For instance, turbulence causes im-
portant entrainment of atmospheric “fresh” ambient into the volcanic dusty cloud. As such, turbulence
further dilutes the flow, which potentially increases its buoyancy; yet, at the same time, turbulence
entrains colder air into the cloud, which decreases the buoyancy of the plume w.r.t. atmospheric ambi-
ent (Dartevelle et al., 2004). Hence, either the plume further rises to higher altitudes till it exhausts its
excess of buoyancy and radially spreads like a gigantic mushroom (the cloud is named “plinian”), or
the plume is not buoyant enough and collapses back to the ground forming destructive high-velocity
hot ash-and-gas avalanches propagating around the volcano (these avalanches are named “pyroclastic”
flows and surges) (Valentine and Wohletz, 1989;Druitt, 1998;Dartevelle et al., 2004;Dartevelle, 2005).
The whole phenomenology can last from a few minutes to a few hours and covers spatial scales from a
few kilometers to tens of kilometers.
Since the pioneer works of volcanologists from Los Alamos National Laboratory (Wohletz et al.,
1984;Valentine and Wohletz, 1989;Valentine et al., 1991), multiphase codes have been used more and
more often to capture the whole volcanic phenomenology (e.g., Dobran et al., 1993;Neri et al., 2003;Ober-
huber et al., 1998;Dartevelle et al., 2004;Suzuki et al., 2005), yet with little evidences that the produced
numerical results accurately capture the physics of these eruptions. So far, numerical “validation” in
volcanology tends to be more qualitative rather than to be a true quantitative and rigorous validation
analysis, as one would expect. However, because of the enormous scale of the event and its rather de-
structive and lethal nature, only afar and indirect methods can be used to infer some information about
their dynamic and physical properties (e.g., with satellite remote sensor, photographic methods, acous-
tic pressure sensors, etc.). Consequently, little is known about the exact dynamic of these gigantic vol-
canic clouds and too little data can be usefully used to validate computer codes (Dartevelle et al., 2004).
Yet, more and more often, these codes and numerical results are used for assessing volcanic hazards and
for mitigating the associated volcanic risks (e.g., Todesco et al., 2002; Esposti et al., 2002;Dartevelle
and Valentine, 2005,2008). Without any thorough validation studies, one may question the intrinsic
value of such invalidated numerical studies. Validation studies are needed; not only volcanologists would
gain more confidences in their newly developed numerical tools but would also be empowered to better
convince policy-makers of the usefulness of their approaches to mitigate potential volcanic hazards.
399
Figure 1. Mt. Pinatubo volcanic jet, Philippines, 12 June 1991. Altitude: ~12 km. Notice the well struc-
tured underexpanded jet and the proto-developed turbulent plume above the jet
Figure 2. Ascending eruption cloud from Redoubt Volcano. View is to the west from the Kenai Peninsula.
Notice that the main plume is offset from the vent. Altitude: ~10 km. (Photograph by J. Warren, April
21, 1990, USGS)
400
Validating codes within this specific volcanic contest is possible if one recognizes the key physics that
dominate and control the dynamic of these clouds, viz., (1) expansion of supersonic and underexpanded
jets and (2) development of turbulence within multiphase jets (Valentine, 1998;Dartevelle, 2005). In a
typical plinian cloud, the lower, thrusting, momentum-driven part has all the properties of an underex-
panded jets expanding into the atmosphere (Kieffer and Sturtevant, 1984;Valentine, 1998), while the
upper, turbulent, buoyantly-driven part is controlled by the atmospheric drag and the dissipation induced
by multiphase turbulence (Dartevelle, 2005). As a matter of fact, the physics of underexpanded jets
(e.g., Ladenburg et al., 1949;Kieffer and Sturtevant, 1984) and multiphase turbulent jets (e.g., Hishida
et al., 1987; Violet et al., 1992) are well known and documented with accurate measurements to be used
for any validation purposes. However, validation against small-scale experiments cannot be enough to
“qualitatively” demonstrate the adequacy of numerical models for geo-physical phenomenology; hence,
geo-modelers must also compare their numerical results against such large-scale events. In this latter
case, validation is intrinsically qualitative; yet, this “field validation” exercise is as important to show
the ability to reproduce large-scale phenomenology as a whole.
In the following, we explore validation test cases of an open-source code developed by U.S. National
Labs, GMFIX and MFIX, against analog experiments for underexpanded jets of Ladenburg et al. (1949)
and for turbulent multiphase jets of Hishida et al. (1987). Then, we perform two “field validation exer-
cises” against well-documented large-scale volcanological events.
2. PHysical and nuMerical aPPrOacH
The multifield approach treats each phase (in our case, air for the gas phase and dispersed particles)
as a fluid field; the two fields interpenetrate, occupying the same control volume as volume fractions.
Each instantaneous local point variable (e.g., mass, velocity, temperature, pressure) must be treated
in a manner consistent with the volume fraction of the phase for which the variable is defined, by a
“smoothing” process (Dartevelle, 2005). The approach we use to derive the multifield equations is laid
out in detail in Dartevelle (2005), and is particularly useful because it allows flexibility in whether
turbulence is modeled by an ensemble averaging approach (so-called Reynolds Average Navier Stokes
or RANS framework, which is most appropriate for internal shear flows such as volcanic conduits) or
a large eddy simulation approach (LES, which is most appropriate for unbounded flows with a large
range of turbulence length scales, such as eruption columns and plumes). The core idea of deriving a
set of “universal” multifield Navier-Stokes Partial Differential Equations (PDE) formally compatible
with different approaches to turbulence (RANS vs. LES) is based upon the “function of presence” of a
given phase at any point in space and time (Dartevelle, 2005) (see also Chapter 1.1). The function of
presence acts as a unique mathematical identifier of the presence of any phase in time and space, while
the gradient of the function of presence acts as a unique identifier of the interface between phases, which
allows to define mass and heat fluxes between phases at their respective interfaces. Dartevelle (2005)
demonstrates that if one carefully defines an ensemble-average process (RANS) or a filtering process
(LES) abiding by the mathematical properties of conservation of constant, linearity, and of commutativ-
ity with respect to the derivations, then the resulting multiphase RANS and LES Navier-Stokes PDEs
would be strictly identical for an eventual implementation in any multiphase computer codes. Only the
turbulence closures would differ depending on whether the modeler works within the LES or RANS
401
framework of turbulence. The turbulence closure and the set of Partial Different Equations will not be
detailed herewith but can be found in Dartevelle (2005,2007) and in Dartevelle and Valentine (2007).
GMFIX (Geophysical Multiphase Flow with Interphase eXchanged, version 1.62) is a set of mul-
tiphase computational fluid dynamics codes evolved from the MFIX family of codes (Syamlal et al.,
1993; Syamlal, 1994, 1998; see also www.mfix.org) in order to model geophysical multiphase fluid
dynamic problems, particularly multiphase explosive events at the proposed first U.S. high level nuclear
waste underground repository at Yucca Mountain, Nevada (Dartevelle and Valentine, 2005,2008). Like
the MFIX code, GMFIX relies on the Implicit Multiphase Formalism in which each phases are mod-
eled as continuum. GMFIX solves full Navier-Stokes equations for each phase following Dartevelle
(2005,2006) with appropriate turbulence and interfacial coupling between phases (Dartevelle, 2005,2006).
The new turbulence model is based on the work of Simonin (1996) and Benyahia et al. (2005) which
couples production and dissipation of turbulence between the dispersed and the gas phase in solving
four coupled scalar turbulence quantities. This coupled multiphase turbulence model basically merges
together classical Reynolds-averaged Navier-Stokes approach (e.g.,Simonin, 1996; Benyahia et al.,
2005) of gas turbulence with a kinetic-collisional model for the dispersed phase (Dartevelle, 2004). All
partial differential equations are solved with a Monotonic Upstream-centered Scheme for Conservation
Law (MUSCL, van Leer (1979)) finite-volume scheme using Deferred Correction Methods (DCM, see
Guenther and Syamlal (2001)). DCM allies stability of a first order method with the accurateness of a
second order scheme. MUSCL is particularly recommended for strong shock dominated flows. GMFIX
1.62 has been extensively Verified and Validated to ensure that the results of simulations are accurate
representations of the physical systems modeled. Details of verification and validation tests are provided
in Dartevelle (2007) and in Dartevelle and Valentine (2007).
2.a. Multifield Partial differential equation
Similarly as in Chapter 1.1, the conservation of mass (continuity) used in this chapter reads:
 ∂ˆρ
 g + ∇ ⋅ ˆρ u  =0
 ∂t g g
 (1)
 ∂ˆρ s
 + ∇ ⋅ ˆρ s u
s = 0
 ∂t
where u is the Favre mass-weighted averaged velocity field as defined by Dartevelle (2005); r̂ is the
macroscopic bulk density of a given phase. We assume there is no exchange of mass between these two
phases. The following relations hold in the two phase system:
εs + εg = 1 a.
ˆρ s = ε s ρs b. (2)
ˆρ g = ε g ρg c.
402
where rs and rg are respectively the ensemble-averaged (within the RANS framework) or filtered
(within the LES framework) microscopic density of the particle field and of the carrier phase; and εs and
εg are respectively the volumetric concentration of the tephra material and of the carrier phase.
Momentum conservation reads:
 ∂ˆρ u
 g  g
 ∂t + ∇ ⋅ ˆρ g u gu ( )
 g = −ε g ∇Pg − ∇ ⋅ ε g t g + tur / SG Tg + I drag
g
+ ˆρ g g

 ∂ˆρ s u
s (3)

 ∂t
+ ∇ ⋅ ˆρ s u
 su ( )
 s = −ε s ∇Pg − ∇ ⋅ ε s f Ts + tur / SG Ts + I drag
s
+ ˆρ s g
where Pg is the ensemble-averaged or filtered thermodynamic pressure of the gas phase (assumed to be
ideal gas, (Dartevelle, 2004, 2005)), Idrag is the interfacial momentum transfer rate between phases; g
represents the body force (e.g., gravity); t g is a “molecular” viscous shear stress tensor; f Ts is a fric-
tional stress tensor defined from visco-plastic theory (Dartevelle, 2004,2005;Dartevelle et al., 2004);
T is the turbulence (RANS) or subgrid (LES) stress tensors. At the gas-particle interface:
tur/SG
M drag
s
+ M drag
g
=0 (4)
The key phenomenology to capture is the extra-dissipation either induced by the statistical “turbulent”
variant motions of gas and grains around their respective ensemble-averaged mean value (provided by
tur
T within the RANS framework) or, within the LES framework, induced by the unresolved motions
within the subgrid (provided by SGT). Within a specific turbulence framework (RANS vs. LES), differ-
ent constitutive equations must be specified for the turbulence/subgrid stress tensor of the gas phase
(tur/SGTg), the stress tensor of the solid phase (tur/SGTs), and the drag vector for all the phases (Iidrag). The
closures for these functions are explained in details by Dartevelle (2004,2005,2006,2007), Dartevelle
and Valentine (2007), and Chapter 1.1.
Let T be the Favre mass-weighted averaged (RANS) or filtered (LES) temperature of a given phase,
then conservation of specific internal energy is:
  ∂ˆρ T 
Cv, g  g g + ∇ ⋅ ˆρ g Tg u
 ∂ t
 g  = −ε g wg − ε g Pg ∇ ⋅ u ( )
 g − ∇ ⋅ ε g q g + tur / SG q g + T H g
  (6)

  ∂ˆρ T 
Cv, s 

s s    = −∇ ⋅ ε q +
 ∂t + ∇ ⋅ ˆρ sTs u s (s s
tur /
)
SG T
qs + H s
 
 

where Cv is the specific heat at constant volume; q is the heat flux; wg is a mean or filtered viscous
dissipation of the gas phase; and TH is the mean rate of interfacial heat transfer between phases. Equation
(6) differs with Equations 6.1 and 6.2 of Chapter 1.1 because we include the irreversible work due to
viscous dissipation of the gas phase. The term in Eq 6, ε g Pg ∇ ⋅ u  g , is the reversible work of the gas
phase because of compressibility effects. At the interface between phases:
403
T
Hs + T Hg = 0 (7)
The heat fluxes are particularly important because they define a molecular heat flux ( q ≈ −k ∇T ,
where k is the thermal conduction coefficient) and additional heat fluxes from either turbulence within
RANS (turq) or from the subgrid within the LES framework (SGq). Closures for k, H, and q functions are
given in Dartevelle (2004, 2005, 2006, 2007), in Chapter 1.1, and are not repeated here.
In Eq. 3, the “viscous” contribution to the gas momentum, , is modeled as:
(8)
while, for the solid phase, the “viscous” contribution to the momentum, −∇ ⋅ ε s ( f
Ts + tur / SG
)
Ts , is
modeled as:
  
∇ ⋅ εs ( f
Ts + tur / SG
Ts ) ≈ −∇ ( f
Ps + tur
)  s I
Ps − ∇ ⋅ 2 eff µ s D s − eff µ bs ∇ ⋅ u (9)
 
In these equations, effμ and effμb are respectively an effective shear and bulk viscosity modeled from

1 1
the RANS framework of turbulence; D = − ∇u  + ∇u  T  + ∇ ⋅ u
 I is the deviator of the rate-of-
2   3
strain tensor where the superscript ‘T’ denotes the operation of transpose of matrix and I is the unit
tensor. The reasons of the approximatively equal sign, ≈, in Eq. (8) and (9) can be found in Appendix 5
of Dartevelle (2005).
2.b. equation of state and thermodynamical Properties
In a multiphase system, the averaged multiphase thermodynamic properties must be properly set to ac-
count for the specific contribution of each phase making up the system. Ensemble-averaged or filtered
phase densities must be specified for each phase, knowing that, ρ̂i = εi ρi with rs = ρs, constant (see
Table 4 and 8) and with an Equation of State (EoS) specified for the gas phase which is assumed ideal:
Pg
ρg = (10)
 T
R g
where R is the ratio of the universal gas constant (R = 8314.56 J/kmol K) and the molar mass of a gas
mixture of m species,
m yj
 = R ∑
R (11)
j =1 Mj
404
where Mj is the molar mass of the jth gas species. In Eq.(10) the over-bar indicates a Favre phase opera-
tion which is not distributive in either the LES or RANS context. Hence in the turbulence context, an
exact formulation for the gas EoS can be fairly complicated, we will assume ―as universally assumed―
that the turbulent fluctuations upon pressure and temperature are small and that they only varies
smoothly and progressively (i.e., T » T ), hence
Pg
ρg ≈ (12)
T
R g
(the same simplification would apply to Eq. 11).

The specific heat ratio for the gas phase is:
Cpg
γg = (13)
Cv g
where Cp and Cv are the specific heats at constant pressure and constant volume which are given by
 for ideal gas. For the solid phase, both Cv and Cp are assumed
(Dartevelle, 2004) and Cv g = Cpg − R
constant, 954 J/kg·K and 1300 J/kg·K respectively.
The adiabatic speed of sound in a pure gas phase can be calculated as:
T
ag = γ g R (14)
g
The loading ratio in a two-phase gas-dust system is:
ˆρ s u
s
η= (15)
ˆρ g u
g
One can note that whenever εs → 0, η → 0. This loading ratio is used to calculate the multiphase
specific heat ratio and the speed of sound in this multiphase (dusty) system (Dartevelle, 2007):
 
 1 + η Cps 
 Cpg 
γ mix = γ g  

1 + γ η Cps 
 

g
Cpg 
(16)
γ mix
amix = a g
γ g (1 + η)
405
Table 1. Cylindrical geometrical setup: Analog validation against underexpanded jets
Radial Length X (m), including the casing wall1 0.105

Radial Resolution ΔX (m) 1
0.5x10-3
Number of Grid-Points in the X-Direction 210
Vertical Length Y (m)1 0.17
Vertical Resolution ΔY (m)1 0.5x10-3
Number of Grid-Points in the Y-Direction 340
Inlet radius (m)1,2 3.5x10-3
Top outlet radius (m)1 0.1045
1
: see Figure 3
2: All analog validation test-cases were obtained with this exact configuration described here, except, for illustration purposes, the simu-
lations shown on Figure 3 (right side), Figure 4, and Figure 7; for these simulations, the inlet radius was 5 mm. In Figure 6C, the dot for K =
3 is the one from Figure 4.
And, again, if the dispersed phase vanished, εs → 0, amix → ag, and γmix → γg. The multiphase Mach
number is therefore:
εgu
 g + εsu
s
M mix = (17)
amix
3. analOg validatiOn aPPrOacHes
As explained in Dartevelle (2007), following a thorough Verification analysis, the goal of validation is to
ensure that the conceptual model and its implementation into a numerical code are the right theoretical
and numerical tools for modeling the “real world” (e.g.,Roache, 1998;Dartevelle et al., 2004;Dartevelle
and Valentine, 2005,2008). In the science of explosive volcanology, “real world” physics covers a wide
range of spatial and temporal scales and of flow dynamisms. In addition, given the inherent danger of
such events, the physical phenomenology is poorly constrained by in situ measurements and real-time
field data. Hence, validation exercise against volcanological field data remains a difficult and mostly
descriptive and qualitative in its nature (see next paragraph, §4) (Dartevelle et al., 2004;Dartevelle and
Valentine, 2007). However, quantitative validations against well-constrained experiments —although not
always directly connected to explosive volcanology— are still needed in order to properly “quantify”
the credibility and validity of a conceptual model and its implementation. Rather than validating a model
against the whole volcanic phenomenology, we also suggest —as a first validation step—validating
against small-scale, highly constrained, and well-understood analog experiments covering the same
physics as the ones found in the large-scale natural event: viz., supersonic and turbulent dynamisms.
Table 1, 2, 3, and 4 present the geometrical, boundary, and numerical properties used for all the ana-
log validation simulation test cases (which are also summarized in Figure 3 for the supersonic jet case
and Figure 8 for the turbulent jet case). The illustration on the right in each Figure further describes a
typical case of jets generated during these validation simulations. Table 4 details the common physical
properties used for all simulations.
406
Table 2. Cylindrical geometrical setup: Analog validation against turbulent jets
Radial Length X (m), including the casing wall1 3.2x10-2

Radial Resolution ΔX (m) 1
Variable: from 0.61x10-3 to 2.0x10-3
Vertical Length Y (m)1 0.52
Vertical Resolution ΔY (m)1 Variable: from 2.8x10-3 to 1.3x10-2
Core inlet radius (m)1 1.1x10-2
Top outlet radius (m)1 3.0x10-2
1
: see Figure 8
Table 3. Numerical properties for all analog validation simulations
Underexpanded Jets Turbulent Jets

Geometry
Cylindrical
Spatial Discretization
MUSCL (2nd order accurate)
Time Discretization 1st order accurate
BI-CIGSTAB
Linear-Equation Solver
constant Mass Flux inflow (MI)
Inlet boundary
constant Pressure & Temperature outflow adaptive Pressure1 outflow
Outlet boundary
(PO) (OF)
Wall Free-slip wall (all phases) No-slip wall (all phases)
1
: the turbulent jet is treated as an isothermal flow, hence only the Pressure changes at the outlet.
3.a. against underexpanded Jets
Thorough laboratory studies of momentum-driven single- to multi-phase supersonic and under-expanded

jets (e.g.,Ladenburg et al., 1949;Lewis and Carlson, 1964; Kieffer, 1984) provide quality and well-
constrained analog experimental data to validate numerical codes against such phenomenology. An
under-expanded jet is defined by a pressure at the nozzle, Pn, higher than the pressure in the expansion
room, P∞:
Pn
K= >1 (18)
P∞
Similarly, an over-expanded jet is when K < 1, and a matched jet is when K = 1.

Table 1, 4, and Figure 3 present the geometrical setup, boundary, and initial conditions for these
simulations.
On a qualitative level, Figure 4 shows all the classic properties of under-expanded jets one would
expect from a code like GMFIX. As the gas leaves the overpressurized nozzle, it over-expands to form
the expansion fans. The ambient gas in the chamber then acts as a piston and pushes the jet gas backwards
407
Table 4. Initial and boundary physical properties for all analog validation simulations
Underexpanded Turbulent Turbulent

Simulations & supersonic Jets Jets
Jets (light particles) (heavy particles)
Variable
Pressure (Pa) (depends on the 1.01x105 1.01x105
simulation)1
Temperature (K) 298 298 298
Domain Volumetric Solid Concentration (vol.%) 2
0 1x10 -8
1x10-8
Kappa (gas phase turbulent production) (m2/s2) NA3 0.1 0.1
Epsilon (gas phase turbulent dissipation) (m /s )
2 3
NA 3
1.6 1.6
Theta (granular temperature) (m /s )
2 2
NA 3
0.01 0.01
Temperature (K) 298 298 298
Gas Pressure (Pa) 6.90×10 5
1.01x10 5
1.01x105
Grain Diameter4 (m) 25x10-6 81x10-6 64x10-6
Grain Density4 (kg/m3) 2,500 280 2,590
Variable
Variable
Volumetric Solid Concentration4 (vol.%) (depends on the
(depends on the radial position)5
simulation)1
Variable
Gas Vertical Speed (m/s) 346
Inlet
Variable
Particle Vertical Speed (m/s) 173
Theta, Granular Temperature Variable
NA3
(solid phase turbulent production) (m2/s2) (depends on the radial position)5
Solid Inelastic Collisional Dissipation coefficient NA3 0.8
Variable
Kappa (gas phase turbulent production) (m2/s2) NA3
Variable
Epsilon (gas phase turbulent dissipation) (m2/s3) NA3
Variable
Gas Pressure (Pa) (depends on the NA
Outlet simulation)1
Gas Temperature (K) 298 NA
1
: The underexpanded and supersonic jet simulation have been performed in maintaining the pressure at the nozzle (Pn) the same for all
experiments but in changing the pressure in the domain (P∞) so that K = Pn/ P∞ = 2, 5, 10, 20, and 30
2
: A volumetric concentration of 1.0x10-8 is equivalent to no particle in the domain, yet this low concentration is nevertheless specified to
maintain a smoother and easier convergence of the turbulence PDEs.
3
: NA, Not Applicable, the particle-gas turbulence model is not used for the supersonic underexpanded jets
4
: This is only for the multiphase flow cases, all experiments were also achieved for single phase case (pure gas, no grains)
5
: Instead of one fixed value, these are radial profiles given by Figure 9 and 10.
along characteristic paths towards the centerline of the jet to form a converging conical shock called the
incident shock. If the value of K is close to one, this incident shock reaches the jet axis at some punctual
positions. The incident shock thereafter undergoes a regular reflection to form a diverging outward shock
called the reflected shock. However, if K<<1 or K>>1 as in Figure 4, then, rather than converging to a
unique point on the jet axis, this incident shock reflects itself at the edge of the jet to form the so-called
408
Figure 3. Analog validations against supersonic and under-expanded single- to multi-phase jet analog
experiments. Left: Cylindrical geometry, grid configuration, initial and boundary conditions. All simu-
lations were performed on a uniform grid of 0.5 x 0.5 mm (see Table 1 & 4). Right: A typical jets which
show three Mach disk. Note all the classical structure of these jets: incidents, reflected shocks, slip
lines, and expansion fans (see also Figure 4). Only the exact position of the first Mach disk is relevant
for the validation tests
Mach disk. The Mach disk, which is a shock normal to the flow direction, can only be found in strongly
under-expanded/over-expanded jets. As soon as this reflected shock reaches the jet boundary, it pushes
outward the jet boundaries to create new rarefaction (expanding) fans and to cause the process to begin
again (see Figures 4 & 5).
409
Figure 4. Qualitative properties of an under-expanded jet obtained by GMFIX. Inlet has a radius of 5
mm with a pressure of 5.52·105 Pa (the pressure in the chamber is 3 times lower). Axis are dimensionless
(normalized by the diameter of the inlet)
In Figure 4, a clear and important property of these jets is the emergence of slip discontinuities from
the shock triple point where the Mach, the incident, and reflected shocks all meet. This discontinuity
arises because the thermodynamical pathway through the incident and reflected shocks does not equal
the pathway through the sole Mach disk. Consequently, the flow velocity, density, and temperature are
not equal across the slip discontinuity. From the Bernoulli’s principle, the total specific energy (i.e.,
kinetic and internal) along any streamline must remain equal. In addition, at any downstream location,
two adjacent fluid parcels along a streamline must also have the same pressure. However, a fluid parcel
which crossed the Mach disk is much more shock-heated along a Hugoniot than a fluid parcel crossing
the weaker incident and reflected shocks. Hence the Mach-heated fluid parcel has higher internal spe-
cific energy; in order to preserve the total specific energy (Bernoulli’s principle), it must have lower
kinetic specific energy than the adjacent fluid parcel on the other side of the slip discontinuity. Hence
this discontinuity results from a velocity difference on each side of the slip lines.
410
Figure 5. Different single-phase jets obtained with different K, viz., 2, 5, 10, 20, and 30. Notice how the
first Mach disk moves downstream and becomes wider as K increases (see Table 1 & 4 for the initial
and boundary conditions for these simulations)
On a quantitative validation level, the position of the first Mach disk for under-expanded jets is known
empirically as a function of both the pressure ratio, K, and the particle mass fraction at the nozzle, χ:
ρ ρ 
 s g −ρ 
 ρ g 
χ = 1 −  mix ,
 (19)
 ρs − ρg 
 
 
where the mixture density of a two-phase dust-gas flow must be, ρmix = ˆρ s + ˆρ g . Ladenburg et al. (1949)
and Lewis and Carlson (1964) have experimentally showed that the higher the value of K, the further
downstream (i.e., the further away from the nozzle) the position of the first Mach disk. The higher the
value of χ, the further upstream (i.e., the closer to the nozzle) the position of the first Mach disk. In our
validation experiment we used values for K of 2, 5, 10, 20, and 30, where the inlet pressure (Pn) remains
the same (6.9∙105 Pa) for all experiments; i.e., the pressure in the domain decreases as K increases (see
Table 1 and Figure 5).
411
Figure 5 shows the shape and size of single phase (gas) jets fully developed along with the position
of their first Mach discs. As expected, as K increases, the position of the first Mach disk moves down-
stream and the overall jets becomes, wider and higher (Ladenburg et al., 1949;Lewis and Carlson, 1964).
This is further illustrated in Figure 6C which shows the downstream position of the first Mach disk vs.
K empirically obtained from analog experiments of Lewis and Carlson (1964) (plain blue curve) and
numerically obtained from GMFIX codes (red dot). The agreement is clearly excellent.
Another important feature of under-expanded jets (especially in the volcanological context) is the
exact effects of particles upon the position and shape of the Mach disk within the jet. Figure 6A shows
the identical jet (K = 10) in the single-phase case (left) and in the multiphase case (right with χ=0.7).
GMFIX simulations clearly show that not only the particles pull “down” the first Mach disk further
upstream, but also change its overall shape: the Mach disk becomes wider and rounder.
On a quantitative level, Figure 6B shows the first Mach disk position vs. χ empirically obtained from
the analog experiments of Lewis and Carlson (1964) (plain blue curve) and numerically calculated by
GMFIX codes (red dot). The “error bars” around each red dot indicates the small uncertainty in exactly
measuring the position of the Mach disk because of the tendency of the jet and its Mach disk to wiggle
and fluctuate around an average position (shown by the dot itself). Within this uncertainty range, the
agreement between analog and numerical experiments is again excellent.
Finally, the effects of the grid size chosen by modelers on the final numerical results can be shown
as in Figure 7. Although the overall quality decreases with the grid size, the position of the first mach
disc remains the same with the precision of the grid size.
Figure 6. A: Two jets obtained with a Pressure ration of K = 10. Left-side, for a single phase flow (pure
gas); right-side, for a multiphase flow (gas and particles) made of particles of 2500 kg/m3, 25 μm, and
a particle mass ratio, χ = 0.7. Notice how the first Mach disk, in the multiphase case, moved upstream
and became wider and rounder. Axes are dimensionalized by the diameter length of the inlet. B: Position
(dimensionless) of the first Mach disk vs. the mass fraction of particles in the jet (from 0, pure gas to
0.7). Blue plain curve is the position given by Lewis and Carlson (1964)’s empirical formula, red plain
dot are the position given by GMFIX codes. The errors bars reflect the fact that there is no definitive
stead-state and the Mach disk tends to slightly fluctuates around an averaged position. C: Position (di-
mensionless) of the first Mach disk vs. the pressure ratio at the nozzle, K. Note that the red dot at K = 3
is represented in Figure 4. The Higher K, the further away (downstream) the Mach disk. Same legend
as in B. The agreement between experimental and numerical data is excellent
412
Figure 7. Effects of different grid-resolutions upon a same simulation’s numerical solution. As already
demonstrated in the Verification test cases, the solution (i.e., in this Figure, the position of the Mach
disk) is independent of the grid resolution. However, the quality (rather than the accurateness) of the
numerical solutions overall depends on the grid-size chosen. Axes are dimensionless
413
3.b. against turbulent Multiphase Jets
These validation test cases show the technical and physical adequacy of GMFIX for calculating the
correct velocity and turbulent energy coupling between the gas phase and the dispersed phase within a
matched, single to two-phases, highly turbulent (Re ~ 22,000), cylindrical, confined jet (Hishida et al.,
1987;Viollet et al., 1992). The experimental data were obtained by laser Doppler velocimetry measure-
ments of a particle-laden jet discharged in a clean-gas stagnant surrounding. This Doppler technique is
capable of particle-size discrimination in order to measure two-component velocities of gas and particles,
and their fluctuations (Hishida et al., 1987). The GMFIX turbulence model is based on separate transport
equations for the components of the particulate stress tensor, and takes the inter-particle collisions into
account using granular kinetic theory (as detailed in Dartevelle and Valentine (2007)).
Three sets of experiments were performed, viz., single phase gas, multiphase with low- and high-
density particles but with particles within the same size range (i.e., 80 mm, 280 kg/m3 and 64.4 mm,
2590 kg/m3), with velocity and turbulent energies sampled at various downstream locations (viz., 0 cm,
6.5 cm, 13 cm, and 26 cm). The initial and the boundary conditions are set to exactly match the ones
given by Hishida et al. (1987) (see Table 2, Table 4, Figure 8 to 10). The experimental error measure-
ments for both mean and fluctuating velocities are ~5% at the centerline of the jet and ~10% at the
edges of the jet (Hishida, personal communication, 2006). Therefore, the experimental error upon the
Figure 8. Validations against turbulent single- to multi-phase jet analog experiments. Left: Cylindrical
geometry, grid configuration, initial and boundary conditions. Note that the grid is non-uniform (see
Tables 2 & 4) and the jet is falling down under gravity. Right: Snapshot of the jets in the multiphase case
with low- (left, 280 kg/m3, 80 μm) and high-inertial (right, 2590 kg/m3, 64 μm) particles. The Horizontal
lines at positions 0 cm, 6.5 cm, 13 cm, and 26 cm represent the sampling positions in the analog experi-
mental and numerical experiments (sampling at 6.5 cm was only achieved in the high-inertial particles
case). Note: the redder, the more particle concentrated
414
turbulent energy is anywhere between 25% and 100%. The reader ought to be careful in the reading of
Figure 9 & 10 as the error bars on the analog experimental data can be rather large. The analog experi-
mental turbulent energy production for a given phase was calculated by:
vr′′ 2 + 2 ⋅ v ′′y 2
k= , (20)
2
where vr″ and vy″ are the measured velocity fluctuation of a given phase in the radial and axial direction
respectively (Hishida, personal communication, 2006).
The comparison of mean axial gas and particle speed distribution with the experimental data (Figure
9) show excellent agreement within ~10% at all downstream locations for the single-gas phase and mul-
tiphase, except small discrepancies at the 13 cm downstream location for the heavy particle case. One
can also note that the light particles have a better coupling than the heavy particles, which is also well-
captured by GMFIX. In addition, as experimentally (Hishida et al., 1987) and numerically confirmed
(Figure 9), the addition of particles (particularly high density particles) tends to increase the axial speed
of the gas phase, particularly, at downstream locations.
In Figure 10, we compare the experimental data (plain curve) with GMFIX numerical data (opened
square) of turbulent kinetic energy for both the gas phase (blue) and dispersed phase (red). If one keeps
in mind the rather large error bar upon the kinetic energy measurements (~25% to ~100%), the numeri-
cal data agrees very favorably with the experimental ones. As in the analog experiments (Hishida et al.,
1987), GMFIX also predicts –perhaps, even exacerbates– the reduction of the production rate of the gas
phase turbulent energy when mixed with particles; the solid particles tend to disturb the production rate
of the turbulent energy from the gas mean flow; this is particularly true for the heavy particles (Figure
10). However, because of the large experimental errors, one can also argue that for the multiphase cases,
GMFIX may underestimate the production of kinetic energy, especially for the particulate phase. This
could be mostly due because of an underprediction of the velocity fluctuation in the axial direction
exacerbated by (i) the assumption that the local shaking of particles is “only” due to the gas phase tur-
bulence (Viollet et al., 1992) (see Eq.(A.2) in Dartevelle and Valentine (2007)) and (ii) a simplistic (and
questionable) approximation of the particle-gas covariance model as in the current version of GMFIX
1.62. Indeed, the production of fluctuating velocity cross-correlation turbulent energy, k12=<u’g·u’s>
(see Eq.(A.3) in Dartevelle and Valentine (2007)), is not resolved by a full PDE approach but by a more
simplistic algebraic expression. Obviously, in order to improve the multiphase turbulent prediction, it
will be required to set a full PDE for the gas-particle covariance model.
4. field validatiOn aPPrOacHes
As we mentioned, validating against geophysical/volcanological data is rather challenging since no in situ

and live measurements can be performed. However, it is nevertheless critical to make every effort pos-
sible to qualitatively validate —in a geo-scientific sense—the model against geophysical-volcanological
flows, e.g., against the height of plinian columns (Dartevelle et al., 2004) or against the 1977 explosive
eruption of a basaltic magma through an Icelandic geothermal borehole (Dartevelle and Valentine, 2007).
Such a “field validation” practice is not as self-contained as those seen in the previous paragraph (§3)
415
Figure 9. Profiles of axial speeds along the radial direction within the jet. Left, single phase (gas) flow;
Middle, for multiphase flow with high inertial particles; Right, for multiphase flow with low inertial
particles. Samplings at 0 cm (inlet), 6.5 cm, 13 cm, and 26 cm downstream. Blue is the gas phase, red
is the particle phase, plain curve from the analog experiments, opened square from GMFIX numerical
codes. Note: the striking difference in the coupling between the gas phase with the low-inertia particle
(tightly coupled) or high-inertia particles (loosely coupled). Generally speaking, the agreement between
numerical and experimental data is excellent and within 10%. The only exception is at 13 cm for the
high-inertia particles. The experimental errors on speed is 5% within the jet core and up to 10% towards
its edges {Hishida, personal communication, 2006)
416
Figure 10. Profiles of the turbulent kinetic energy along the radial direction within the jet. Left, for a
single phase (gas) flow; Middle, for multiphase flow with high inertial particles; Right, for multiphase
flow with low inertial particles. Samplings at 0 cm (inlet), 6.5 cm, 13 cm, and 26 cm downstream. Blue
is the gas phase (k1), red is the particle phase (k2), plain curve from the analog experiments, opened
square from GMFIX numerical codes. Generally speaking, the agreement between numerical and experi-
mental data is excellent and within 10% for single phase flow. There is more discrepancy for k2 between
numerical and analog data however. However, one should keep in mind that the experimental errors
in the determination of k1 and k2 is between 25% and 100%. This is because the turbulent energy is not
directly measured but rather indirectly inferred from the fluctuating velocity fields which is measured
with experimental errors between 5 to 10%
417
—given the geophysical unknowns in terms of boundary, initial, and inflow attributes—,yet it remains
an essential practice in Earth Sciences in order to demonstrate the ability of the validated numerical tool
to reproduce the whole geophysical event.
4.a validation against the Height of a Plinian column
Plinian column upper-heights (HT) have been often related to the mass flux of material released at the
vent; this flux represents the amount of energy released and available to the plinian column. Figure 11
represents HT of the plinian column vs. the inferred mass flux at the vent for different historical erup-
tions and three plinian simulations performed with GMFIX where HT is measured at 3600 s (results
from Dartevelle et al., 2004; see Table 5 & 6 for simulation setups). Also shown on Figure 11, the best
fit curve between the past eruptions (Wilson et al., 1978;Settle, 1978;Sparks et al. 1997) and two curves
from Morton et al.’s theory (1956) for two magma temperatures at the vent (from Wilson et al., 1978).
Knowing the uncertainties for historical eruptions to infer the exact HT and, most importantly, the mass
flux at the vent, the top-altitude predicted by our model is in excellent agreement with past eruptions
and quite surprisingly with Morton et al (1956) theory which was initially developed for plume within
the troposphere (Sparks, 1986). Figure 11 allows comparing numerical results with classical plume
theory (e.g., Morton et al., 1956; Wilson et al, 1978;Sparks, 1986) and most importantly real observa-
tions. In addition, such “geo-validation” exercise (as shown in Figure 11) not only offers an easy way
to constrain the whole plinian cloud simulation in terms of maximum height reached by the clouds but
also indicates whether the atmospheric drag and buoyancy as being modeled hold altogether as whole
for such phenomenology.
4.b validation against the geothermal explosive basaltic event at námafjall,

iceland, 1977
During the night of September 8, 1977, a basaltic dike, associated with minor local volcanic episode,
intersected a borehole at a depth between 635-1038 m (most likely ~1000 m) in the Námafjall geother-
mal field, Iceland. According to Larsen et al. (1979), the eruptive event at the wellhead began with an
audible explosion followed by an incandescent column 15-25 m high (note that due to the night time
occurrence, only incandescent flows and objects were observed, and no observations of the lower tem-
perature part of the eruptive column are available). As the column grew in width over a period of about
1 minute, “sparks and cinders” were ejected and a constant roaring sound was heard. This was followed
by a period of 10-20 minutes during which apparently “there was little to no activity at the eruption
site”; toward the end of this second phase, red flashes were observed. The final phase consisted of “a
series of rapid explosions or shots of glowing scoria,” wherein explosions were focused in groups over
a total of about one minute. Flow velocities of 20-30 m/s were estimated by Larsen et al. (1979) based
upon observed ejecta heights. This event provides a unique test case for multiphase volcanic processes,
given that its vertical extent (~1 km) is similar to that of natural volcanic conduits and its geometry is
exactly known. Dartevelle and Valentine (2007) modeled this eruption using the exact same model as
validated in paragraph §3.a &.b. This event, not only serves as a typical geo-scientific “field validation”
test case but, also, provide an interesting and unique case of a natural kilometric-long fluidized system.
418
Table 5. Standard atmospheric properties for plinian simulations1
Pressure at vent level 105 Pa

Temperature at vent level 298 K
Calculated gas density at vent level 1.169 kg/m3
Vapor mixing ratio at vent level 0 (dry atmosphere)
Tropospheric temperature gradient (0 - 11 km) -7 K/km (temperate atmosphere)
Lower stratospheric temperature gradient (19 - 32 km) +1.8 K/km
Upper stratospheric temperature gradient (32 - 47 km) +2.8 K/km
Tropopause 11 - 19 km
1
: A temperate, dry, idle standard atmosphere.
Table 6. Plinian simulation setups1
Plinian Eruption: PL1 PL2 PL3

Geometry 2D Cylindrical
Radial/Horizontal length X (km) 20 40 60
Radial/Horizontal resolution ΔX (m) 30 to 1000 50 to 1000 80 to 1000
Number of grid-points in the X-direction 145 168 150
Vertical length Y (km) 18 25 36
Vertical resolution ΔY (m) 30 50 80
Number of grid-points in the Y-direction 601 501 401
Vent radius r (m) 60 200 400
Mixture vertical speed Vy (m/s) 110 110 160
Volumetric solid concentration εs (vol.%) 0.1 0.1 0.1
Grain diameter d (μm) 50 50 50
Grain microscopic density ρs (kg/m ) 3
1500 1500 1500
Mixture temperature at the vent Tm (K) 900 900 900
Gas pressure at the vent Pg (Pa) 105 105 105
Mass fraction of water vapor at the vent 1.0 1.0 1.0
Calculated mixture density ρm (kg/m ) 3
1.74 1.74 1.74
Calculated mass flux (kg/s) 3.15x106 2.41x107 1.39x108
1
: In Cylindrical geometry, the mass flux at the vent is calculated by π·r2·Vy·ρm, where r is the volcanic vent radius, Vy is the mixture verti-
cal speed and ρm is the mixture density defined as:
Figure 12, Table 7 and 8 describe the initial and boundary conditions used to model the Námafjall
volcanic event: viz., particle size of 18.4 mm, 3 wt.% water mass fraction as an initial condition, an
average constant mass flux of 25 kg/s, and free slip boundaries at the wall.
Dartevelle and Valentine (2007) showed that early on instabilities set in the multiphase mixture at
the bottom of the fluidized bed borehole to eventually grow to form particulate train waves (ash clusters)
followed by voids (bubbles) propagating upward in the borehole towards the wellhead of the borehole.
Each ash cluster arriving at the wellhead triggers a particle ash outburst (individual explosions) consis-
tent with observations of Larsen et al. (1979). Figure 13 shows wave trains which are essentially high
419
Figure 11. Top altitude of the plinian cloud (HT in km) vs. mass flux at the volcanic vent (kg/s). Triangle
are for historical eruptions from which HT and the mass flux has been inferred from field studies and
remote sensing observations (data from Wilson et al., 1978;Settle, 1978;Sparks et al. 1997); dash-curve
is the best regression fit between these historical eruption data; plain curve are from Morton et al. (1956)
theory calculated for two initial magma temperatures at the vent (600 K and 1200 K); and circles are
for GMFIX’s three plinian simulations (from Dartevelle et al., 2004). Knowing all the uncertainties of
historical eruptions for determining the mass flux at the vent and HT, we may conclude that there is an
excellent agreement between GMFIX’s simulations and past historical eruptions.
concentration, narrow slugs of ashes, separated by relatively clean gas. The slugs reach sufficiently high
particle concentration to begin to enter into a plastic-like rheological behavior (Dartevelle et al.,
420
Figure 12. Boundary and initial conditions; and geometrical and spatial discretization setups for the
Borehole volcanic event. Drawing not to scale. As indicated, at the inlet, the multiphase flow has a gas
pressure of 241.8 atm, a temperature of 1428 K, a solid volumetric concentration of 30.8 vol.%, a phasic
weighted vertical speed of 1.41 m/s, and a phasic mass flux of 25 kg/s
Table 7. Borehole explosive simulation: cylindrical geometrical setup
Radial Length X (m), including the casing wall 0.12

Radial Resolution ΔX (m) 0.04
Vertical Length Y (m) 1001
Vertical Resolution ΔY (m) 1.0
Borehole radius (m) 0.08
Inlet radius (m) 0.08
Outlet radius (m) 0.08
2004;Dartevelle 2004,2005), a state in which the resistance to upward movement increases substan-
tially. Phase weighted velocity shows one effect of the segregation of particles into slugs; as the gas
expands and accelerates upward, it encounters slugs of particles that exert large drag forces and rapidly
decelerate the gas, which then re-accelerates once it passes through a slug. The coupling between gas
and particle momentum creates a nonlinear feedback; increasing particle concentration slows the gas,
which reduces drag on the particles and further slows a slug and increases its concentration. Compress-
ible flow effects are also complex (Figure 13). Sound speed in the mixture is very sensitive to particle
concentration, and as a result the particle slugs result in narrow zones where the flow is supersonic.
These zones are shock waves that travel upward with the slugs of high concentration. Comparison of
421
Table 8. Borehole explosive simulation: initial and boundary physical properties
Simulations
Pressure (Pa) 1.01x105
Temperature (K)1 529.15
Calculated Gas Density (kg/m3) 0.4136
Borehole The Mass Fraction of Water in Vapor Phase 1
Volumetric Solid Concentration (vol.%) 0
Kappa (gas phase turbulent production) (m2/s2) 0.01
Epsilon (gas phase turbulent dissipation) (m2/s3) 100.0
Mixture Temperature (K) 1
1428.15
Gas Pressure (Pa)2 245.0×105
Calculated Gas Density (kg/m3) 37.174
The Mass Fraction of Water in Vapor Phase3 1.0
Grain Diameter (m) 1
18.4x10-3
Grain/Magma Density (kg/m3)1 2,700
Magmatic Mass Fraction of Water (wt.%) 3
Inferred Volumetric Solid Concentration (vol.%) 30.804
Mass flux (kg/s) 1
25
Inlet Inferred Vertical Speed (m/s) 1.45014
Calculated Mixture Density (kg/m3) 857.4308
Calculated Mixture Speed of Sounds (m/s) 141.37
Calculated Mixture Static Pressure (Pa) 169.53x105
Calculated Mixture Specific Heat Ratio 1.0108
Theta, Granular Temperature
0.01
(solid phase turbulent production) (m2/s2)
Solid Inelastic Collisional Dissipation coefficient 0.9
Kappa (gas phase turbulent production) (m /s )2 2
0.01
Epsilon (gas phase turbulent dissipation) (m /s ) 2 3
100.0
Gas Pressure (Pa) 1.01x105
Outlet Gas Temperature (K)1 529.15
Calculated Gas Density (kg/m ) 3
0.4136
1
: given by Larsen et al. (1979).
2
: This corresponds to lithostastic plus magmatic water vapor pressure.
3
: This corresponds to magmatic gas made of pure water vapor.
profiles at successive times shows that each wave or slug tends to grow in amplitude as it propagates
upward until it finally exits the top of the borehole. Dartevelle and Valentine (2007) noted the complex
feedback mechanism caused by the large variations of the mixture speed of sound on the overall unsteady
dynamic of this multiphase system and on the particle burst frequency (e.g., wave amplification). Indeed,
the initial variations of the sonic regime in the conduit (subsonic to supersonic and vice-versa) is the
initial result of the inherent decoupling between phases (for instance caused by the larger particle size);
yet, at a later stage, these mixture sound speed variations must be also one of the leading cause of the
422
Figure 13. Time sequence (from 50 s to 180 s) of, A., particle volumetric concentration (dimensionless);
B., phase weighted vertical speed (m/s); and, C., multiphase Mach number (dimensionless) profiles
vs. the borehole height (m). Notice, with time, the vertical propagation and amplification of “particle
waves”. There is a one-on-one relation between higher concentration wave and higher Mach number
(the speed of sound is smaller in more concentrated system)
423
amplifications of the particle burst frequency and, hence, cause of further developments of unsteadiness.
In addition, it be shown that the frequency of particle bursts is strongly dependent on the particle size,
i.e., the larger the particle sizes, the more decoupled the phases, hence the more amplified the unsteadi-
ness and explosive bursts.
At the wellhead (Figure 14A & B), the numerical results qualitatively reproduced phenomena simi-
lar to those documented at the Námafjall event by Larsen et al. (1979), namely: “The first phase (…)
started with an explosion and a thin incandescent column was seen. (…). This phase, accompanied by
a continuous roar, lasted no longer than 1 minute”. From Figure 14B, one can see an initial blast of pure
steam exiting the wellhead at speeds ~220 m/s, which is followed ~25 s later by the ejection of the
tephra (Figure 14A). “The second phase, lasting 10-20 minutes, there was little or no activity. Occa-
sional red flashes may have occurred during the latter part of this phase.” From Figure 14A & B, one
can see that, after the initial blast explosion, between 35 and 70 s, there is a period of relative quietness
with decreasing speed during which the borehole “mimics” a steady state period during which insta-
bilities gradually set in at the bottom of the borehole and finally rise towards the wellhead. The modeled
period of relative steady state quietness is much shorter than the 10 minutes observed by Larsen et al.
(1979). This difference in time might be caused by some supplementary unsteadiness in the borehole
system such as temporary disruption of magma supply from the dyke, or from water entering the bore-
hole and temporarily stopping upward magma flow (i.e., potentially representing a phase of pre-mixing
of melt and water). This quiet period precedes the last stage described as “the final phase consisted of
a series of very rapid explosions or shots of glowing scoria. A few groups of explosions were observed
each consisting of several individual shots.” This final phase of the eruption, which qualitatively com-
pares well with the last stage in Figure 14, corresponds to the arrival of the magmatic slugs followed by
voidage (bubble) formed initially at the bottom of the borehole and amplifying as this succession of
magmatic pulses rise towards the surface.
This simulation in a geothermal borehole not only reproduced well the field observations made by
Icelandic geologists but also is in agreement with the engineer literatures on fluidized bed instabilities.
Vertical gas-particle flows generally involve non-linear, chaotic, unsteady multi-phase dynamics. Fluid-
ized systems manifest fluctuations over a wide range of length and time scales with a tendency for strong
phase segregation, which leads to particle-free regions often referred to as “bubbles” (Homsy, 1998) and
particle-rich regions referred to as “clusters” of solids (Breault et al., 2005). In narrow fluidized beds
and in liquid fluidized systems, these successions of voidage fronts with discontinuous jumps in particle
concentration form traveling disturbance bands (El-Kaissy and Homsy, 1976;Didwania, and Homsy,
1981). Several investigators (Pigford and Baron, 1965;Anderson and R. Jackson, 1968;El-Kaissy and
Homsy, 1976;Didwania, and Homsy, 1981) have shown that a small disturbance imposed on a uni-
formly fluidized bed can grow with time to eventually form particle concentration heterogeneities. Based
on linear instability analysis, some of these investigators (Pigford and Baron, 1965;Anderson and R.
Jackson, 1968) have shown that the formation of voidage heterogeneities arises spontaneously from
instabilities in the initial state of uniform fluidization. Anderson and Jackson (1968) hypothesized that
these instabilities can be related to compression wave propagating upward through the fluidized bed and
amplifying with time, also resulting in particle-free regions. Pigford and Baron (1965) were the first to
suggest that the development of these heterogeneities could be interpreted as being analogous to shock-
wave structures. This shock wave analogy was confirmed by Fenucci et al. (1979) who use the method
of characteristics applied to a full set of nonlinear two-phase flow equations. They show that a small
disturbance changes with time and distance and can, eventually, produce a flow discontinuity similar to
424
Figure 14. A. Time sequence profiles of the particle volume concentration (dimensionless) and, B., of
the phase weighted vertical speed (m/s) sampled at the top outlet of the borehole every second. Note at
4 s, an initial water vapor explosion (no ash) with an ejection vertical speed of 215 m/s. The ash will
be ejected from the borehole at 31 s with a phase weighted speed of 31 m/s. Also, it is worth noting that
a quiet phase up to 75 s, then this eruptive system becomes pulsating with an explosion or ashy-shot
every 6 s or so
425
a shock wave in gases, i.e., the shock front and the bubble front are mathematically analogous. Hence,
there is little difference in the behavior of granular multiphase system between a kilometric long and
laboratory fluidized beds.
6. cOnclusiOn
Natural geo-systems are inherently opened (i.e., unconstrained) in terms of boundary, initial and inflow
conditions; hence validation of numerical geo-models is a challenging task to perform, yet a necessary
one (Roache, 1998). First, we suggest singling-out the key driving physics of a given geo-physical
events in order to achieve validations against well-constrained analog experiments encompassing such
specific-physics (e.g., multiphase turbulence, Mach disk in under-expanded jets, etc.). This first valida-
tion analysis is what would be traditionally performed in CFD and in engineering sciences. Yet geo-
modelers must still demonstrate the adequacy of their “analog-validated” numerical tool for simulating
large-scale geo-physical events. This second “geo-validation” step consists to compare numerical results
against large-scale, natural, and thoroughly (as much as possible) described phenomenology. Although
this “field validation” approach cannot be as closed and strict as an “analog validation” —given the
inherent large-scale unknowns—, it is nevertheless as important to prove the suitableness of numerical
tool for simulating natural systems.
references
Anderson, T.B., & Jackson, R. (1968). Fluid mechanical description of fluidized beds. Stability of the
state of uniform fluidization. I&EC Fundamentals, 7, 12-2.
Benyahia, S., Syamlal, M., & O’Brien, T. J. (2005). Evaluation of boundary conditions used to model dilute
turbulent gas/solid flows in a pipe. Powder Technology, 156, 62–72. doi:10.1016/j.powtec.2005.04.002
Breault, R., Ludlow, C. J., & Yue, P. C. (2005). Cluster particles number and granular temperature for cork
particles at wall in the riser of a CFB. Powder Technology, 149, 68–77. doi:10.1016/j.powtec.2004.11.003
Dartevelle, S. (2004). Numerical modeling of geophysical granular flows: 1. A comprehensive
approach to granular rheologies and geophysical multiphase flows. G-cubed, 5, Q08003. doi:.
doi:10.1029/2003GC000636
Dartevelle, S. (2005). Comprehensive Approaches to Multiphase Flows in Geophysics. Application to
non-isothermal, non-homogenous, unsteady, large-scale, turbulent dusty clouds. I. Basic RANS and LES
Navier-Stokes equations, LA-14228 (p. 51). Los Alamos, New Mexico: Los Alamos National Laboratory.
Dartevelle, S. (2006). Geophysical Multiphase Flow with Interphase eXchanges. Hydrodynamic and
Thermodynamic Models and Numerical Techniques. Version GMFIX-1.62, pp. 62, Design Document
Attachment 1, U.S. Department of Energy, Office of Repository Development, 11192-DD-1.62-00.
Dartevelle, S. (2007). From Model Conception to Verification and Validation. A Global Approach to
Multiphase Navier-Stoke Models with an Emphasis to Volcanic Explosive Phenomenology, LA-14346
(p. 86). Los Alamos, New Mexico: Los Alamos National Laboratory.
426
Dartevelle, S., Rose, W. I., Stix, J., Kelfoun, K., & Vallance, J. W. (2004). Numerical modeling of geo-
physical granular flows: 2. Computer simulations of plinian clouds and pyroclastic flows and surges.
G-cubed, 5, Q08004. doi:.doi:10.1029/2003GC000637
Dartevelle, S., & Valentine, G. A. (2005). Early-time multiphase interactions between basaltic magma
and underground repository openings at the proposed Yucca Mountain radioactive waste repository.
Geophysical Research Letters, 32, L22311..doi:10.1029/2005GL024172
Dartevelle, S., & Valentine, G. A. (2007). Transient multiphase processes during the explosive eruption
of basalt through a geothermal borehole (Námafjall, Iceland, 1977) and implications for natural volcanic
flows. Earth and Planetary Science Letters, 262, 363–384..doi:10.1016/j.epsl.2007.07.053
Dartevelle, S., & Valentine, G. A. (2008). Multiphase magmatic flows at Yucca Mountain, Nevada.
Journal of Geophysical Research, 113, B12209..doi:10.1029/2007JB005367
Didwania, A. K., & Homsy, G. M. (1981). Flow regimes and flow transitions in liquid fluidized beds.
International Journal of Multiphase Flow, 7, 563–58. doi:10.1016/0301-9322(81)90031-8
Dobran, F., Neri, A., & Macedonio, G. (1993). Numerical simulations of collapsing volcanic columns.
Journal of Geophysical Research, 98, 4231–4259. doi:10.1029/92JB02409
Druitt, T. H. (1998). Pyroclastic density currents, in The physics of explosive volcanic eruptions. Geol.
Soc., 145, 145–182.
El-Kaissy, M. M., & Homsy, G. M. (1976). Instability waves and the origin of bubbles in fluidized
beds, Part 1: Experiments. International Journal of Multiphase Flow, 2, 379–395. doi:10.1016/0301-
9322(76)90021-5
Eposti Ongaro, T., Neri, A., Todesco, M., & Macedonio, G. (2002). Pyroclastic flow hazard assessment
at Vesuvius (Italy) by using numerical modeling. I. Analysis of flow variables. Bulletin of Volcanology,
64, 178–191..doi:10.1007/s00445-001-0190-1
Fanucci, J. B., Ness, N., & Yen, R.-H. (1979). On the formation of bubbles in gas-particulate fluidized
Guenther, C., & Syamlal, M. (2001). The effects of numerical diffusion on simulation of isolated bubbles
in a gas-solid fluidized bed. Powder Technology, 116, 142–154. doi:10.1016/S0032-5910(00)00386-7
Harlow, F. H., & Amsden, A. (1975). Numerical calculation of multiphase flow. Journal of Computational
Physics, 17, 19–52. doi:10.1016/0021-9991(75)90061-3
Hishida, K., Takemoto, K., & Maeda, M. (1987). Turbulence characteristics of gas-solids two-phase
confined jet (effect of particle density). Japanese J. Multiphase Flow, 1, 56–68.
Homsy, G. M. (1998). Nonlinear waves and the origin of bubbles in fluidized beds. Applied Scientific
Research, 58, 251–274. doi:10.1023/A:1000787803463
Kieffer, S. W., & Sturtevant, B. (1984). Laboratory studies of volcanic jets. Journal of Geophysical
Research, 89, 8253–8268. doi:10.1029/JB089iB10p08253
427
Ladenburg, R., van Voorhis, C. C., & Winckler, J. (1949). Interferometric studies of faster than sound
phenomena. Part II. Analysis of supersonic air jets. Physical Review, 76, 662–677. doi:10.1103/Phys-
Rev.76.662
Larsen, G., Gronvold, K., & Thorarinson, S. (1979). Volcanic eruption through a geothermal borehole
at Namafjall, Iceland. Nature, 278, 707–710. doi:10.1038/278707a0
Lewis, C. H. Jr, & Carlson, D. J. (1964). Normal shock location in underexpanded gas and gas-particle
jets. AIAA Journal, 2, 776–777. doi:10.2514/3.2409
Morton, B. R., Taylor, G. F. R. S., & Turner, J. S. (1956). Turbulent gravitational convection from
maintained and instantaneous sources. Proceedings of the Royal Society of London. Series A, 234, 1–23.
doi:10.1098/rspa.1956.0011
Neri, A., Ongaro, T. E., Macedonio, G., & Gidaspow, D. (2003). Multiparticle simulation of collapsing
volcanic pyroclastic flow. Journal of Geophysical Research, 108, 2202..doi:10.1029/2001JB000508
Oberhuber, J. M., Herzog, M., Graf, H.-F., & Schwanke, K. (1998). Volcanic plume simulation on large
scales. Journal of Volcanology and Geothermal Research, 87, 29–53. doi:10.1016/S0377-0273(98)00099-7
Pigford, R.L. & Baron, T. (1965). Hydrodynamic stability of a fluidized bed. I&EC Fundamentals, 4,
81-87.
Roache, P. J. (1998). Verification and Validation in Computational Science and Engineering (p. 446).
Albuquerque, NM: Hermosa Publishers.
Settle, M. (1978). Volcanic eruption clouds and the thermal power output of explosive eruptions. Journal
of Volcanology and Geothermal Research, 3, 309–324. doi:10.1016/0377-0273(78)90041-0
Simonin, O. (1996). Continuum modeling of dispersed two-phase flows. Combustion and turbulence
in two-phase flows. Von Karman Institute of Fluid Dynamics, VKI Lecture Series, 1-47, Rhodes-St-
Geneses, Belgium.
Sparks, R. S. J. (1986). The dimension and dynamics of volcanic eruption columns. Bulletin of Volcanol-
ogy, 48, 3–15. doi:10.1007/BF01073509
Sparks, R. S. J., Bursik, M. I., Carey, S. N., Gilbert, J. S., Glaze, L. Z., Sigurdson, H., & Woods, A. W.
(1997). Volcanic Plumes (p. 574). Hoboken, NJ: John Wiley.
Suzuki, Y. J., Koyaguchi, T., Ogawa, M., & Hachisu, I. (2005). A numerical study of turbulent mixing
in eruption clouds using a three-dimensional fluid dynamics model. Journal of Geophysical Research,
110, B08201..doi:10.1029/2004JB003460
Syamlal, M. (1994). MFIX documentation. In User’s manual, (pp. 87). Washington, DC: U.S. Dept. of
Energy, DOE/METC-95/1013, DE95000031.
Syamlal, M. (1998). MFIX documentation. In Numerical Technique, (pp. 80). Washington, DC: U.S.
Dept. of Energy. DOE/MC/31346-5824, DE98002029.
Syamlal, M., Rogers, W., & O’Brien, T. J. (1993). MFIX documentation. In Theory Guide, (pp. 49).
Washington, DC: U.S. Dept. of Energy. DOE/METC-94/1004, DE94000097.
428
Todesco, M., Neri, A., Eposti Ongaro, T., Papale, P., Macedonio, G., Santacroce, R., & Longo, A. (2002).
Pyroclastic flow hazard assessment at Vesuvius (Italy) by using numerical modeling. I. Large-scale
dynamics. Bulletin of Volcanology, 64, 155–177..doi:10.1007/s00445-001-0189-7
Valentine, G. A. (1998). Eruption column physics. In Freundt, A., & Rosi, M. (Eds.), From Magma to
Tephra: Modeling physical processes of explosive volcanic eruptions (pp. 91–138). New York: Elsevier
Science.
Valentine, G. A., & Wohletz, K. H. (1989). Numerical models of Plinian eruption columns and pyroclastic
flows. Journal of Geophysical Research, 94, 1867–1887. doi:10.1029/JB094iB02p01867
Valentine, G. A., Wohletz, K. H., & Kieffer, S. W. (1991). Sources of unsteady column dynamics in
pyroclastic flow eruptions. Journal of Geophysical Research, 93, 21887–21892. doi:10.1029/91JB02151
Viollet, P. L., Simonin, O., Olive, J., & Minier, J. P. (1992). Modeling turbulent two-phase flows in
industrial equipments. In Hirsch, C. (Ed.), Computational Methods in Applied Sciences. New York:
Elsevier science.
Wilson, L., Sparks, R. S. J., Huang, T. C., & Watkins, N. D. (1978). The control of volcanic column heights
by eruption energetic and dynamics. Journal of Geophysical Research, 83, 1829–1836. doi:10.1029/
JB083iB04p01829
Wohletz, K. H., McGetchin, T. R., Standford, M. T. II, & Jones, E. M. (1984). Hydrodynamics aspects
of caldera-forming eruptions: Numerical models. Journal of Geophysical Research, 89, 8269–8286.
doi:10.1029/JB089iB10p08269
429
430
Compilation of References
Agrawal, K., Loezos, P. N., Syamlal, M., & Sundaresan, Almedeij, J. (2008). Drag coefficient of flow around
S. (2001). The role of meso-scale structures in rapid gas- a sphere: matching asymptotically the wide trend.
solid flows. Journal of Fluid Mechanics, 445, 151–185. Powder Technology, 186, 218–223. doi:10.1016/j.pow-
doi:10.1017/S0022112001005663 tec.2007.12.006
Ahmadi, G., & Ma, D. (1990). A thermodynamical Ammara, I., & Masson, C. (2004). Development of a
formulation for dispersed multiphase turbulent flows. fully coupled control-volume finite element method for
I. International Journal of Multiphase Flow, 16(22), the incompressible navier-stokes equations. International
323–340. doi:10.1016/0301-9322(90)90062-N Journal for Numerical Methods in Fluids, 44, 621–644.
doi:10.1002/fld.662
Ahmmed, S. I., Mohamed, A. H., & Nayef, M. G. (2009).
Modified mathematical model for gas phase olefin po- Amsden, A. A., O’Rourke, P. J., & Butler, T. D. (1989).
lymerization in fluidized-bed catalytic reactor. Chemical KIVA-II: A computer program for chemically reactive
Engineering Journal. flows with sprays. LA-11560-MS. Los Alamos, NM: Los
Alamos National Lab.
Alam, M., & Luding, S. (2003). Rheology of bidis-
perse granular mixtures via event-driven simulations. Anderson, T. B., & Jackson, R. (1967). A Fluid-mechanical
Journal of Fluid Mechanics, 476, 69–103. doi:10.1017/ Description of Fluidized Beds. I&EC Fundam., 6, 527.
S002211200200263X doi:10.1021/i160024a007
Alam, M., & Luding, S. (2005). Energy nonequiparti- Anderson, T. B., & Jackson, R. (1968). Fluid Mechanical
tion, rheology, and microstructure in sheared bidisperse Description of Fluidized Beds. Stability of Stte of Uni-
granular mixtures. Physics of Fluids, 17, 063303. form Fluidization. Industrial & Engineering Chemistry
doi:10.1063/1.1938567 Fundamentals, 7(1), 12–21. doi:10.1021/i160025a003
Alam, M., Willits, J. T., Arnarson, B. O., & Luding, S. Andrews, M. J., & O’Rourke, P. J. (1996). The multiphase
(2002). Kinetic theory of a binary mixture of nearly elastic particle-in-cell (MP-PIC) method for dense particle flow.
disks with size and mass disparity. Physics of Fluids, International Journal of Multiphase Flow, 22, 379–402.
14(11), 4085–4087. doi:10.1063/1.1509066 doi:10.1016/0301-9322(95)00072-0
Alexander, F., & Garcia, A. (1997). Computers in Physics, Andrews, A. T., Loezos, P. N., & Sundaresan, S. (2005).
11, 588. doi:10.1063/1.168619 Coarse-Grid Simulation of Gas-Particle Flows in Vertical
Risers. Industrial & Engineering Chemistry Research,
Alley, W., Alder, B., & Yip, S. (1983). The Neutron Scat-
44(16), 6022–6037. doi:10.1021/ie0492193
tering Function for Hard Spheres. Physical Review A., 27,
3174. doi:10.1103/PhysRevA.27.3174
Andries, P., Aoki, K., & Perthame, B. (2002). A consistent Bafrnec, M., & Bena, J. (1972). Quantitative data on
BGK-type model for gas mixtures. Journal of Statistical the lowering of electrostatic charge in a fluidized bed.
Physics, 106, 993–1018. doi:10.1023/A:1014033703134 Chemical Engineering Science, 27(5), 1177–1181.
doi:10.1016/0009-2509(72)80030-7
Arastoopour, H., & Gidaspow, D. (1979). Vertical
pneumatic conveying using four hydrodynamic models. Bagnold, R. A. (1954). Experiments on a gravity-free
Powder Technology, 18(2), 123–130. dispersion of large solid spheres in a Newtonian fluid
under shear. Proc. Roy. Soc., A225, 49.
Arastoopour, H., Lin, S., & Weil, S. (1982). Analysis
of vertical pneumatic conveying of solids using multi- Bai, D., Shibuya, E., & Nakagawa, N. (1997). Fractal
phase flow models. AIChE Journal. American Institute characteristics of gas-solids flow in a circulating fluidized
of Chemical Engineers, 28(3), 467–473. doi:10.1002/ bed. Powder Technology, 90, 205. doi:10.1016/S0032-
aic.690280315 5910(96)03212-3
Arastoopour, H., Pakdel, P., & Adewumi, M. (1990). Bai, D. R., Jin, Y., & Yu, Z. (1991). Momentum exchange
Hydrodynamics analysis of dilute gas-solids flow in between gas and solids in fast fluidized bed. Journal of
a vertical pipe. Powder Technology, 62(2), 163–170. Chemical Industry and Engineering (China), 5, 548–553.
doi:10.1016/0032-5910(90)80080-I
Balzer, G., Boelle, A., & Simonin, O. (1995). Eulerian
Aris, R. (1962). Vectors, Tensors, and the Basic Equations gas-solid flow modelling of dense fluidized beds. Paper
of Fluid Mechanics. London: Dover. presented at the Fluidization VIII.
Arnarson, B. O., & Willits, J. T. (1998). Thermal diffusion Banerjee, S., Scott, D. S., & Rhodes, E. (1968). Mass
in binary mixtures of smooth, nearly elastic spheres with Transfer To Falling Wavy Liquid Films In Turbulent Flow.
and without gravity. Physics of Fluids, 10, 1324–1328. Industrial & Engineering Chemistry Fundamentals, 7(1),
doi:10.1063/1.869658 22–27. doi:10.1021/i160025a004
Arnarson, B. O., & Jenkins, J. T. (2000). Particle seg- Banerjee, S., Patel, R., Chen, A., Snider, D., Williams,
regation in the context of species momentum balances. K., O’Hern, T., & Tortora, P. (2007, Nov. 30). Enhanced
In Helging, D., Herrmann, H. J., Schreckenberg, M., & Productivity of Chemical Processes Using Dense Fluid-
Wolf, D. E. (Eds.), Traffic and Granular Flow ‘99: Social, ized Beds. DOE Award DE-FC36-04GO14153.
Traffic, and Granular Dynamics. Berlin: Springer.
Banyahia, S., Arastoopour, H., & Knowlton, T. M. (1998).
Bader, R., Findlay, J., & Knowlton, T. M. (1988). Gas- Prediction of Solid and Gas Flow Behavior in a Riser
solid flow pattern in a 30.5 cm diameter circulating fluid- Using a Computational Multiphase Flow Approach. In
ized bed. In Basu, P., & Large, J. F. (Eds.), Circulating Fan, L.-S., & Knowlton, T. M. (Eds.), Fluidization IX (pp.
Fluidized Bed Technology II (pp. 123–127). New York: 493–498). New York: Engineering Foundation.
Pergamon Press.
Barnea, E., & Mizrahi, J. (1973). A generalized approach
Baerns, M. (1966). Effect of interparticle adhesive forces to the fluid dynamics of particle systems: Part 1. General
on fluidization of fine particles. Industrial & Engineering correlation for fluidization and sedimentation in solid
Chemistry Fundamentals, 5(4), 508–516. doi:10.1021/ multiparticle systems. Chem. Engng. J., 5, 171–189.
i160020a013 doi:10.1016/0300-9467(73)80008-5
Baeyens, J., & Geldart, D. (1974). An investigation into Baron, T., Briens, C. L., Bergougnou, M. A., & Hazlett,
slugging fluidized beds. Chemical Engineering Science, J. D. (1987). Electrostatic effects on entrainment from
29, 255–259. doi:10.1016/0009-2509(74)85051-7 a fluidized bed. Powder Technology, 53(1), 55–67.
doi:10.1016/0032-5910(87)80125-0
431
Barrat, A., & Trizac, E. (2002). Lack of energy equipar- Benyahia, S., Syamlal, M., & O’Brien, T. (2006). Exten-
tition in homogeneous heated binary granular mixtures. sion of Hill–Koch–Ladd drag correlation over all ranges
Granular Matter, 4(2), 57–63. doi:10.1007/s10035-002- of Reynolds number and solids volume fraction. Powder
0108-4 Technology, 162, 166. doi:10.1016/j.powtec.2005.12.014
Barrett, J. C., & Webb, N. A. (1998). A comparison of Benyahia, S. (2008). Verification and validation study of
some approximate methods for solving the aerosol general some polydisperse kinetic theories. Chemical Engineering
dynamic equation. Journal of Aerosol Science, 29(1-2), Science, 63, 5672–5680. doi:10.1016/j.ces.2008.08.016
31–39. doi:10.1016/S0021-8502(97)00455-2
Benyahia, S., Syamlal, M., & O’Brien, T. J. (2006).
Baskaran, A., Dufty, J., & Brey, J. (2007). Kinetic Theory Extension of Hill–Koch–Ladd drag correlation over all
of Response Functions for the Hard Sphere Granular ranges of Reynolds number and solids volume fraction.
Fluid. Journal of Statistical Mechanics, 12, 12002. Powder Technology, 162(2), 166–174. doi:10.1016/j.
doi:10.1088/1742-5468/2007/12/P12002 powtec.2005.12.014
Baskaran, A., Dufty, J., & Brey, J. (2008). Transport Co- Benyahia, S., Syamlal, M., & O’Brien, T. J. (2005).
efficients for the Hard Sphere Granular Fluid. Physical Evaluation of boundary conditions used to model dilute
Review E: Statistical, Nonlinear, and Soft Matter Physics, turbulent gas/solid flows in a pipe. Powder Technology,
77, 031311. doi:10.1103/PhysRevE.77.031311 156, 62–72. doi:10.1016/j.powtec.2005.04.002
Batchelor, G. K. (1988). A new theory of the instability Bhatnagar, P. L., Gross E. P., & Krook M. (1954). A model
of a uniform fluidized bed. Journal of Fluid Mechanics, for collisional processes in gases. I. Small amplitude
193, 75–110. doi:10.1017/S002211208800206X processes in charged and neutral one-component systems.
Physical Reviews 94, 4(3), 511 – 525.
Bayle, J., Mege, P., & Gauthier, T. (2001). Dispersion of
bubble flow properties in a turbulent FCC fluidized bed. Bi, H. T., & Grace, J. R. (1995). Effect of measurement
In Kwauk, M., Li, J., & Yang, W. C. (Eds.), Fluidization method on the velocities used to demarcate the onset of
X (pp. 125–132). New York: Engineering Foundation. turbulent fluidization. Chemical Engineering Journal,
57, 261.
Beetstra, R., Hoef, M. A. d., & Kuipers, J. A. M. (2007).
Numerical study of segregation using a new drag force Bird, G. (1994). Molecular Gas Dynamics and the Direct
correlation for polydisperse systems derived from lattice- Simulation of Gas Flows. Oxford, UK: Clarendon.
Boltzmann simulations. Chemical Engineering Science,
Bird, R. B., Stewart, W. E., & Lightfoot, E. N. (2002).
62, 246–255. doi:10.1016/j.ces.2006.08.054
Transport Phenomena. New York: John Wiley and Sons.
Beetstra, R., van der Hoef, M. A., & Kuipers, J. A. M.
Blaser, P., & Yeomans, N. (2006, February). Sand Core
(2007). Drag force of intermediate Reynolds number
Engineering & Process Modeling. Japan Foundry Society,
flow past mono- and bidisperse arrays of spheres. AIChE
2(2), 420–427.
Journal. American Institute of Chemical Engineers, 53(2),
489–501. doi:10.1002/aic.11065 Boemer, A., Qi, H., Hannes, J., & Renz, U. (1994). Model-
ing of solids circulation in a fluidised bed with Eulerian
Behjat, Y., Dehnavi, M. A., Shahhosseini, S., &
approach. Paper presented in 29th IEA-FBC Meeting,
Hashemabadi, S. H. (2008). Heat transfer of polymer
Paris, France.
particles in gas phase polymerization reactors. Interna-
tional Journal of Chemical Reactor Engineering, 6(A81). Bogoliubov, N. N. (1962). Problems in a Dynamical
doi:10.2202/1542-6580.1649 Theory in Statistical Physics, (English translation by E.
Gora). In Uhlenbeck, G. E., & de Boer, J. (Eds.), Studies
in Statistical Mechanics I. Amsterdam: North Holland.
432
Bokkers, G. A., van Sint Annaland, M., & Kuipers, Breault, R. W., Ludlow, C. J., & Yue, P. C. (2005). Cluster
J. M. (2004). Mixing and segregation in a bidisperse particle number and granular temperature for cork particles
gas-solid fluidized bed: a numerical and experimental at the wall in the riser of a CFB. Powder Technology,
study. Powder Technology, 140, 176–186. doi:10.1016/j. 149(2-3), 68–77. doi:10.1016/j.powtec.2004.11.003
powtec.2004.01.018
Breault, R. W., Shadle, L. J., & Pandey, P. (2005). Granular
Boland, D., & Geldart, D. (1972). Electrostatic charging Temperature, Turbulent Kinetic Energy and Solids Frac-
in gas fluidized beds. Powder Technology, 5(5), 289–297. tion of Cork Particles at the wall in the Riser of a CFB.
doi:10.1016/0032-5910(72)80033-0 In Cen, K. (Ed.), Circulating Fluidized Bed Technology
VIII (pp. 755–761). Hangzhou, China: Int. Academic
Bolland, O. (1998). Describing Mass transfer in Circulat-
Publishers.
ing Fluidized Beds by Ozone Decomposition. Trondheim,
Norway: Norwegian University of Science and Technol- Breault, R. W., & Guenther, C. (2005). Sensitivity of
ogy, Department of Thermal Energy and Hydropower. Gas-Solids Dispersion and Mass Transfer Coefficient in
an Eulerian-Eulerian CFD Modeling. Paper presented at
Boon, J.-P., & Yip, S. (1991). Molecular Hydrodynamics.
the Spring Meeting of the American Institute of Chemical
New York: Dover.
Engineers, Atlanta, GA.
Borruso, A. V. (2008). CEH Marketing Research Report
Brereton, C. M. H., & Grace, J. R. (1993). Microstruc-
Linear Low-Density Polyethylene (LLDPE) Resins.
tural aspects of the behavior of circulating fluidized
Chemical Economics Handbook. Menlo Park, CA: SRI
beds. Chemical Engineering Science, 48, 2565–2568.
Consulting.
doi:10.1016/0009-2509(93)80267-T
Boublik, T. (1970). Hard-sphere equation of state. The Jour-
Brereton, C. M., & Grace, J. R. (1994). End Effects in
nal of Chemical Physics, 53, 471. doi:10.1063/1.1673824
Circulating Fluidized Bed Hydrodynamics. In Avidan, A.
Boyalakuntla, D. S., & Pannala, S. (2006). Summary A. (Ed.), Circulating Fluidized Bed Technology IV (pp.
of Discrete Element Model (DEM) implementation in 137–146). Hidden Valley, PA.
MFIX. Retrieved from http://www.mfix.org/documents/
Brey, J., Dufty, J., Kim, C.-S., & Santos, A. (1998). Hy-
MFIXDEM2006-4-1.pdf
drodynamics for Granular Flow at Low Density. Physi-
Brackbill, J. U., & Ruppel, H. M. (1986). FLIP: A method cal Review E: Statistical Physics, Plasmas, Fluids, and
for adaptively zoned, particle-in-cell calculations of fluid Related Interdisciplinary Topics, 58, 4638. doi:10.1103/
flow in two dimensions. Journal of Computational Phys- PhysRevE.58.4638
ics, 65, 314. doi:10.1016/0021-9991(86)90211-1
Brey, J. J., Dufty, J. W., & Santos, A. (1997). Dissipative
Breault, R. W. (2006). A review of gas–solid dispersion Dynamics for Hard Spheres. Journal of Statistical Phys-
and mass transfer coefficient correlations in circulating ics, 87, 1051. doi:10.1007/BF02181270
fluidized beds. Powder Technology, 163(1-2), 9–17.
Brey, J. J., Ruiz-Montero, M. J., & Cubero, D. (1996).
doi:10.1016/j.powtec.2006.01.009
Homogeneous Cooling State of a Low Density Granular
Breault, R. W., Guenther, C., & Shadle, L. J. (2008). Flow. Physical Review E: Statistical Physics, Plasmas,
Velocity fluctuation interpretation in the near wall region Fluids, and Related Interdisciplinary Topics, 54, 3664.
of a dense riser. Powder Technology, 182(2), 137–145. doi:10.1103/PhysRevE.54.3664
doi:10.1016/j.powtec.2007.08.018
433
Brey, J. J., & Ruiz-Montero, M. J. (2004). Simulation study Carlos, C. R., & Richardson, J. F. (1968). Solids move-
of the Green-Kubo relations for dilute granular gases. ment in liquid fluidised beds-I Particle velocity distri-
Physical Review E: Statistical, Nonlinear, and Soft Matter bution. Chemical Engineering Science, 23, 813–824.
Physics, 70, 051301. doi:10.1103/PhysRevE.70.051301 doi:10.1016/0009-2509(68)80016-8
Brey, J. J., Ruiz-Montero, M. J., Maynar, P., & Garcia de Cercignani, C. (1998). The Boltzmann equation and its
Soria, M. I. (2005). Hydrodynamic modes, Green-Kubo applications. New York: Springer.
relations, and velocity correlations in dilute granular gases.
Chaouki, J., Godfrey, L., & Larachi, F. (1999). Position
Journal of Physics Condensed Matter, 17, S2489–S2502.
and Velocity of a Large Particle in a Gas/Solid Riser using
doi:10.1088/0953-8984/17/24/008
the Radioactive Particle Tracking Technique. Canadian
Brey, J. J., Ruiz-Montero, M. J., & Moreno, F. (2001). Journal of Chemical Engineering, 77, 253. doi:10.1002/
Hydrodynamics of an open vibrated system. Physical cjce.5450770210
Review E: Statistical, Nonlinear, and Soft Matter Physics,
Chapman, S., & Cowling, T. G. (1970). The Mathematical
63, 061305. doi:10.1103/PhysRevE.63.061305
Theory of Non-Uniform Gases (3rd ed.). Cambridge, UK:
Brey, J. J., Ruiz-Montero, M. J., & Moreno, F. (2005). Cambridge University Press.
Energy partition and segregation for an intruder in a
Chen, N. S. (2009). Comparison of formulas for drag
vibrated granular system under gravity. Physical Review
coefficient and settling velocity of spherical particles.
Powder Technology, 189, 395–398. doi:10.1016/j.pow-
Briens, C. L., Bergougnou, M. A., Inculet, I. I., Baron, tec.2008.07.006
T., & Hazlett, J. D. (1992). Size distribution of particles
Chen, Y. M., Jang, C. S., & Cai, P. (1991). On the forma-
entrained from fluidized beds: Electrostatic effects.
tion and disintegration of particle clusters in a liquid--solid
Powder Technology, 70(1), 57–62. doi:10.1016/0032-
transport bed. Chemical Engineering Science, 46, 2253.
5910(92)85054-Y
doi:10.1016/0009-2509(91)85124-G
Brilliantov, N., & Pöschel, T. (2004). Kinetic Theory of
Chen, S., & Doolen, G. D. (1998). Lattice Boltzmann
Granular Gases. New York: Oxford University Press.
method for fluid flows. Annual Review of Fluid Mechan-
doi:10.1093/acprof:oso/9780198530381.001.0001
ics, 30, 329–364. doi:10.1146/annurev.fluid.30.1.329
Brilliantov, N., & Pöschel, T. (2006). Breakdown of the
Chen, A., & Jones, N. (2008). Application of Coarse
Sonine Expansion for the Velocity Distribution of Granular
Grained Drag Law In Computational Fluid Dynamics
Gases. Europhysics Letters, 74, 424. doi:10.1209/epl/
Simulations of Fluidized Beds. Paper presented in AIChE
i2005-10555-6
Annual Meeting, Philadelphia, Pennsylvania.
Brito, R., Enriquez, H., Godoy, S., & Soto, R. (2008).
Chen, P. (2004). Modeling the fluid dynamics of bubble
Segregation induced by inelasticity in a vibrofluidized
column flows. Doctoral dissertation, Washington Univer-
granular mixture. Physical Review E: Statistical, Nonlin-
sity, Saint Louis, LA.
ear, and Soft Matter Physics, 77, 061301. doi:10.1103/
PhysRevE.77.061301 Choi, S.-K., Kim, S.-O., Lee, C.-H., & Choi, H.-K. (2003).
Use of the momentum interpolation method for flows with
Brodkey, R. S. (1967). The Phenomena of Fluid Motions.
a large body force. Numerical Heat Transfer Part B, 43,
Reading, MA: Addison-Wesley.
267–287. doi:10.1080/713836204
Campbell, C. S. (1990). Rapid Granular Flows. Annual
Review of Fluid Mechanics, 22, 57. doi:10.1146/annurev.
fl.22.010190.000421
434
Chou, C. S., & Richman, M. W. (1998). Constitutive theory Collins, L. R., & Keswani, A. (2004). Reynolds number
for homogeneous granular shear flows of highly inelastic scaling of particle clustering in turbulent aerosols. New
spheres. Physica A. Statistical and Theoretical Physics, Journal of Physics, 6.
259, 430–448. doi:10.1016/S0378-4371(98)00265-9
Contractor, R. (1999). Dupont’s CFB technology for
Chou, C.S. (2001). Collisional source of the second maleic anhydride. Chemical Engineering Science, 54(22),
moment and pressure tensor for inhomogeneous rapid 5627–5632. doi:10.1016/S0009-2509(99)00295-X
granular flows of highly inelastic spheres. Physica A:
Contractor, R., Bergna, H. E., Horowitz, H. S., Black-
Statistical Mechanics and its Applications, 290, 341-359.
stone, C. M., Chowdhry, U., & Sleight, A. W. (1988).
Ciborowski, J., & Wlodarski, A. (1962). On electrostatic Butane oxidation to maleic anhydride in a recirculating
effects in fluidized beds. Chemical Engineering Science, solids reactor. In Ward, J. W. (Ed.), Catalysis 1987 (pp.
17(1), 23–32. doi:10.1016/0009-2509(62)80003-7 645–654). New York: Elsevier Science Publishers.
Clelland, R., & Hrenya, C. M. (2002). Simulations of a Contractor, R. M. (1991, June 4).Vapor Phase Catalytic
binary-sized mixture of inelastic grains in rapid shear flow. Oxidation of Butane to Maleic Anhydride. U.S. Patent
Physical Review E: Statistical, Nonlinear, and Soft Matter 5,021,588.
Physics, 65, 031301. doi:10.1103/PhysRevE.65.031301
Contractor, R., Anderson, M., Campos, D., Hecquet, G.,
Clift, R., Grace, J. R., & Weber, M. E. (1978). Bubbles, Pham, C., Simon, M., et al. (Oct 30, 2001). Vapor phase
drops and particles. New York: Academic Press. oxidation of acrolein to acrylic acid. DuPont de Nemours
and Company, US Patent 6,310,240.
Clift, R., Seville, J. P. K., Moore, S. C., & Chavarie,
C. (1987). Comments on buoyancy in fluidized beds. Cox, J. D., & Clark, N. N. (1991). The effect of particle
Chemical Engineering Science, 42(1), 191–194. drag relationships on prediction of kinematic wave veloc-
doi:10.1016/0009-2509(87)80228-2 ity in fluidized beds. Powder Technology, 66, 177–189.
doi:10.1016/0032-5910(91)80099-5
Cocco, R. (2006). Jet Hydrodynamics Around Fluidized
Bed Spargers. Orlando, FL: World Congress in Particle Cruz, E., Steward, F. R., & Pugsley, T. (2006). New clo-
Technology. sure models for CFD modeling of high-density circulat-
ing fluidized beds. Powder Technology, 169, 115–122.
Cocco, R. (2006). Understanding the Gas Residence Time
doi:10.1016/j.powtec.2006.08.005
in a Cyclone. Paper presented World Congress in Particle
Technology, Orlando, FL, 2006. Cundall, P. A., & Strack, O. D. L. (1979). A discrete
numerical model for granular assemblies. Geotechnique,
Cocco, R., & Williams, K. (2004). Optimization of Par-
29, 47–65. doi:10.1680/geot.1979.29.1.47
ticle Residence Time INside Commercial Dryers with
Arena-Flow. Paper presented in AIChE Annual Meeting, Curtis, J. S., & van Wachem, B. (2004). Modeling particle-
Austin, Texas. laden flows: a research outlook. AIChE Journal. American
Institute of Chemical Engineers, 50(11), 2638–2645.
Cocco, R., & Williams, K. (2005). Riser Simulations
doi:10.1002/aic.10394
with Arena-flow: Application to the PSRI 1994 and 2001
Challenge Problems. Paper presented in AIChE Annual Dahl, S., Hrenya, C., Garzo, V., & Dufty, J. (2002). Kinetic
Meeting, Cincinnati, Ohio. Temperatures for a Granular Mixture. Physical Review
E: Statistical, Nonlinear, and Soft Matter Physics, 66,
Cohen, E.G.D. (1962). On the Generalization of the
041301. doi:10.1103/PhysRevE.66.041301
Boltzmann Equation to General Order in the Density.
Physica 28, 1025. ibid. 28, 1045. ibid 28 1060.
435
Dahl, S. R., Clelland, R., & Hrenya, C. M. (2002). The Dartevelle, S., & Valentine, G. A. (2005). Early-time
Effects of Continuous Size Distributions on the Rapid multiphase interactions between basaltic magma and
Flow of Inelastic Particles. Physics of Fluids, 14(6), underground repository openings at the proposed Yucca
1972–1984. doi:10.1063/1.1476917 Mountain radioactive waste repository. Geophysical Re-
search Letters, 32, L22311..doi:10.1029/2005GL024172
Dahl, S. R., Clelland, R., & Hrenya, C. M. (2003).
Three-dimensional, rapid shear flow of particles with Dartevelle, S., & Valentine, G. A. (2007). Transient mul-
continuous size distributions. Powder Technology, 138, tiphase processes during the explosive eruption of basalt
7–12. doi:10.1016/j.powtec.2003.08.036 through a geothermal borehole (Námafjall, Iceland, 1977)
and implications for natural volcanic flows. Earth and
Dahl, S. R., & Hrenya, C. M. (2004). Size segregation in
Planetary Science Letters, 262, 363–384..doi:10.1016/j.
rapid, granular flows with continuous size distributions.
epsl.2007.07.053
Physics of Fluids, 16(1), 1–13. doi:10.1063/1.1626682
Dartevelle, S., & Valentine, G. A. (2008). Multi-
Dahl, S. R., & Hrenya, C. M. (2005). Size segregation
phase magmatic flows at Yucca Mountain, Nevada.
in gas-solid fluidized beds with continuous particle
Journal of Geophysical Research, 113, B12209..
size distributions. Chemical Engineering Science, 60,
doi:10.1029/2007JB005367
6658–6673. doi:10.1016/j.ces.2005.05.057
Dartevelle, S. (2006). Geophysical Multiphase Flow
Danckwerts, P. V. (1951). Significance of liquid-film
with Interphase eXchanges. Hydrodynamic and Ther-
coefficients in gas absorption. Industrial & Engineering
modynamic Models and Numerical Techniques. Version
Chemistry, 43, 1460–1467. doi:10.1021/ie50498a055
GMFIX-1.62, pp. 62, Design Document Attachment 1,
Dartevelle, S. (2004). Numerical modeling of geophysical U.S. Department of Energy, Office of Repository Devel-
granular flows: 1. A comprehensive approach to granular opment, 11192-DD-1.62-00.
rheologies and geophysical multiphase flows. G-cubed,
Das, M., Meilkap, B. S., & Saha, R. K. (2008). Character-
5, Q08003. doi:.doi:10.1029/2003GC000636
istics of axial and radial segregation of single and mixed
Dartevelle, S. (2005). Comprehensive Approaches to particle system based on terminal settling velocity in the
Multiphase Flows in Geophysics. Application to non- riser of a circulating fluidized bed. Chemical Engineering
isothermal, non-homogenous, unsteady, large-scale, Journal, 145(1), 32–43. doi:10.1016/j.cej.2008.03.010
turbulent dusty clouds. I. Basic RANS and LES Navier-
Das Sharma, S., Pugsley, T., & Delatour, R. (2006). Three-
Stokes equations, LA-14228 (p. 51). Los Alamos, New
dimensional CFD model of the deaeration rate of FCC
Mexico: Los Alamos National Laboratory.
particles. AIChE Journal. American Institute of Chemical
Dartevelle, S. (2007). From Model Conception to Verifi- Engineers, 52(7), 2391–2400. doi:10.1002/aic.10858
cation and Validation. A Global Approach to Multiphase
Dasgupta, S., Jackson, R., & Sundaresan, S. (1994).
Navier-Stoke Models with an Emphasis to Volcanic Ex-
Turbulent gas-particle flow in vertical risers. AIChE
plosive Phenomenology, LA-14346 (p. 86). Los Alamos,
New Mexico: Los Alamos National Laboratory.
215–228. doi:10.1002/aic.690400204
Dartevelle, S., Rose, W. I., Stix, J., Kelfoun, K., & Val-
Davidson, J. F., & Harrison, D. (1971). Fluidization.
lance, J. W. (2004). Numerical modeling of geophysical
London: Academic Press.
granular flows: 2. Computer simulations of plinian clouds
and pyroclastic flows and surges. G-cubed, 5, Q08004.
doi:.doi:10.1029/2003GC000637
436
Davis, E., Herbolzheimer, R. H., & Acrivos, A. (1982). Deng, G., Piquet, J., Vasseur, X., & Visonneau, M. (2001).
The sedimentation of polydisperse suspensions in vessels A new fully coupled method for computing turbulent
having inclined walls. International Journal of Multiphase flows. Computers & Fluids, 30, 445–472. doi:10.1016/
Flow, 8(6), 571–585. doi:10.1016/0301-9322(82)90064-7 S0045-7930(00)00025-6
de Wilde, J., Van Engelandt, G., Heyndenickx, G. J., & Desai, P. & Wen, C. (1978). Computer modeling of the
Marin, G. B. (2005). Gas–solids mixing in the inlet zone MERC fixed bed gasifier, MERC/CR-78/3.
of a dilute circulating fluidized bed. Powder Technology,
Desjardin, O., Fox, R. O., & Villedieu, P. (2008). A quadra-
151(1-3), 96–116. doi:10.1016/j.powtec.2004.11.037
ture-based moment method for dilute fluid-particle flows.
De Wilde, J. (2005). Reformulating and quantifying the Journal of Computational Physics, 227(4), 2514–2539.
generalized added mass in filtered gas-solid flows models. doi:10.1016/j.jcp.2007.10.026
Physics of Fluids, 17, 113304. doi:10.1063/1.2131925
Di Felice, R. (1994). The void function for fluid-particle
De Wilde, J. (2007). Filtered gas-solid momentum interaction systems. International Journal of Multiphase
transfer models and their applications to 3D steady-state Flow, 20, 153–159. doi:10.1016/0301-9322(94)90011-6
riser simulations. Chemical Engineering Science, 62,
Didwania, A. K., & Homsy, G. M. (1981). Flow regimes
5451–5457. doi:10.1016/j.ces.2006.12.028
and flow transitions in liquid fluidized beds. International
Debling, J. A., & Ray, W. H. (1995). Heat and mass Journal of Multiphase Flow, 7, 563–58. doi:10.1016/0301-
transfer effects in multistage polymerization processes: 9322(81)90031-8
Impact polypropylene. Industrial & Engineering Chem-
Ding, J., & Gidaspow, D. (1990). A bubbling fluidiza-
istry Research, 34, 3466–3480. doi:10.1021/ie00037a035
tion model using kinetic theory of granular flow. AIChE
Deen, N. G., Van Sint Annaland, M., Van der Hoef, M. A., Journal. American Institute of Chemical Engineers, 36(4),
& Kuipers, J. A. M. (2007). Review of discrete particle 523–538. doi:10.1002/aic.690360404
modeling of fluidized beds. Chemical Engineering Sci-
Ding, J., & Lyczkowski, R. (1992). Three-dimensional
ence, 62, 28–44. doi:10.1016/j.ces.2006.08.014
kinetic theory modeling of hydrodynamics and erosion
Dehnavi, A. M., Shahhosseini, S., Hashemabadi, S. H., in fluidized beds. Powder Technology, 73(2), 127–138.
& Ghafelebashi, S. M. (2008). CFD based evaluation of doi:10.1016/0032-5910(92)80073-6
polymer particles heat transfer coefficient in gas phase
Dobran, F., Neri, A., & Macedonio, G. (1993). Nu-
polymerization reactors. International Communications
merical simulations of collapsing volcanic columns.
in Heat and Mass Transfer, 35, 1375–1379. doi:10.1016/j.
Journal of Geophysical Research, 98, 4231–4259.
icheatmasstransfer.2008.07.017
doi:10.1029/92JB02409
Deltour, P., & Barrat, J. L. (1997). Quantitative Study of
Dorfman, J., & Kirkpatrick, T. (1980). Kinetic Theory
a Cooling Granular Medium. Journal de Physique. I, 7,
of Dense Gases not in Equilibrium. In Garrido, L. (Ed.),
137. doi:10.1051/jp1:1997130
Systems Far from Equilibrium, (Lecture Notes in Physics
Deng, G., Piquet, J., Queutey, P., & Visonneau, M. 132). Berlin: Springer.
(1994). Incompressible flow calculations with a con-
Drew, D. A. (1983). Mathematical modeling of two-
sistent physical interpolation finite volume approach.
phase flow. Annual Review of Fluid Mechanics, 15, 261.
Computers & Fluids, 23, 1029–1047. doi:10.1016/0045-
doi:10.1146/annurev.fl.15.010183.001401
7930(94)90003-5
Drew, D. A., & Passman, S. L. (1999). Theory of multi-
component fluids. New York: Springer.
437
Druitt, T. H. (1998). Pyroclastic density currents, in The Edwards, C. F. (1999). Formulating Dense, Large-Eddy
physics of explosive volcanic eruptions. Geol. Soc., 145, Simulations of Multiphase Flows. In ILASS Americas
145–182. 12th Annual Conference on Liquid Atomization and Spray
Systems, Indianapolis, IN.
Dufty, J., Baskaran, A., & Brey, J. (2008). Linear Response
and Hydrodynamics for a Granular Fluid. Physical Review El-Kaissy, M. M., & Homsy, G. M. (1976). Instability
E: Statistical, Nonlinear, and Soft Matter Physics, 77, waves and the origin of bubbles in fluidized beds, Part 1:
031310. doi:10.1103/PhysRevE.77.031310 Experiments. International Journal of Multiphase Flow,
2, 379–395. doi:10.1016/0301-9322(76)90021-5
Dufty, J. W. (2001). Kinetic Theory and Hydrodynamics
for a Low Density Granular Gas. Advances in Complex Elsdon, R., & Mitchell, F. R. G. (1976). Contact elec-
Systems, 4, 397. doi:10.1142/S0219525901000395 trification of polymers. Journal of Physics. D, Applied
Physics, 9, 1445–1460. doi:10.1088/0022-3727/9/10/010
Dufty, J. W. (2002). Shear Stress Correlations in Hard
and Soft Sphere Fluids. Molecular Physics, 100, 2331. Enwald, H., Peirano, E., & Almstedt, A. E. (1996).
doi:10.1080/00268970110109934 Eulerian two phase flow theory applied to fluidization.
International Journal of Multiphase Flow, 22, 21–66.
Dufty, J. W., & Ernst, M. H. (2004). Exact Short Time
doi:10.1016/S0301-9322(96)90004-X
Dynamics for Steeply Repulsive Potentials. Molecular
Physics, 102, 2123. doi:10.1080/00268970412331292858 Eposti Ongaro, T., Neri, A., Todesco, M., & Macedonio,
G. (2002). Pyroclastic flow hazard assessment at Vesu-
Dufty, J., & Baskaran, A. (2005). Hard Sphere Dynamics
vius (Italy) by using numerical modeling. I. Analysis of
for Normal and Granular Fluids. In Gottesman, S. (Ed.),
flow variables. Bulletin of Volcanology, 64, 178–191..
Nonlinear Dynamics in Astronomy and Physics. Annals
doi:10.1007/s00445-001-0190-1
of the New York Academy of Sciences 1045, 93.
Epstein, N. (1984). Comments on a unified model for
Dufty, J. (2009). Granular Fluids. In Meyers, R. (Ed.),
particulate expansion of fluidized beds and flow in
Encyclopedia of Complexity and Systems Science. Hei-
fixed porous media. Chemical Engineering Science, 39,
delberg, Germany: Springer.
1533–1534. doi:10.1016/0009-2509(84)80020-2
Dufty, J., & Brey, J. (2005). Origins of Hydrodynamics
Ergun, S. (1952). Fluid Flow through Packed Columns.
for a Granular Gas. In Pareschi, L., Russo, G., & Toscani,
Chemical Engineering Progress, 48, 89–94.
G. (Eds.), Modelling and Numerics of Kinetic Dissipative
Systems. New York: Nova Science. Eriksson, E. J. G., & Mckenna, T. F. (2004). Heat transfer
phenomena in gas olefin polymerization using computa-
Duran, J., Rajchenach, J., & Clement, E. (1993). Arching
tional fluid dynamics. Industrial & Engineering Chemistry
effect model for particle size segregation. Physical Review
Research, 43(23), 7251–7260. doi:10.1021/ie034328n
Ernst, M. H. (2000). Kinetic Theory of Granular Fluids:
Dutta, A., & Basu, P. (2004). An improved cluster-renewal
Hard and Soft Spheres. In Karkheck, J. (Ed.), Dynam-
model for the estimation of heat transfer coefficients on
ics: Models and Kinetic Methods for Non-Equilibrium
the furnace walls of commercial circulating fluidized
Many Body Systems. NATO ASI Series 371. Dordrecht,
bed boilers. Transactions of the ASME, 126, 1040–1043.
Germany: Kluwer.
doi:10.1115/1.1833360
Español, P. (1998). Fluid particle model. Physical Review
Eaton, J. K., & Fessler, J. R. (1994). Preferential con-
E: Statistical Physics, Plasmas, Fluids, and Related
centration of particles by turbulence. International
Interdisciplinary Topics, 57, 2930–2948. doi:10.1103/
Journal of Multiphase Flow, 20, 169. doi:10.1016/0301-
PhysRevE.57.2930
9322(94)90072-8
438
Español, P., & Revenga, M. (2003). Smoothed dissipative Fernandes, F. A. N., & Lona, L. M. F. (2002). Heteroge-
particle dynamics. Physical Review E: Statistical, Non- neous modeling of fluidized bed polymerization reactors.
linear, and Soft Matter Physics, 67, 026705. doi:10.1103/ Influence of mass diffusion into the polymer particle.
PhysRevE.67.026705 Computers & Chemical Engineering, 26, 841–848.
doi:10.1016/S0098-1354(02)00010-8
Fan, L. S., Han, L. S., & Brodkey, R. S. (1987). Comments
on the buoyancy force on a particle in a fluidized suspen- Ferschneider, G., & Mege, P. (1996). Eulerian simulation
sion. Chemical Engineering Science, 42(5), 1269–1271. of dense phase fluidized beds. Revue de l’Institut Français
doi:10.1016/0009-2509(87)80087-8 du Pétrole, 51(2), 301–307.
Fan, R., Marchisio, D. L., & Fox, R. O. (2004). Application Ferziger, J. H., & Kaper, H. G. (1972). Mathematical
of the direct quadrature method of moments to polydisperse theory of transport processes in gases. New York: Elsevier.
gas-solid fluidized beds. Powder Technology, 139(1),
Ferziger, J. H., & Peric, M. (2001). Computational methods
7–20. doi:10.1016/j.powtec.2003.10.005
for fluid dynamics. New York: Springer.
Fan, R., & Fox, R. O. (2008). Segregation in polydispersed
Fessler, J. R. (1994). Preferential concentration of heavy
fluidized beds: Validation of a multi-fluid model. Chemi-
particles in a turbulent channel flow. Physics of Fluids,
cal Engineering Science, 63(1), 272–285. doi:10.1016/j.
6, 3742. doi:10.1063/1.868445
ces.2007.09.038
Flemmer, R. L. C., & Banks, C. L. (1986). On the drag
Fan, R. (2006). Computational fluid dynamics simulation
coefficient of a sphere. Powder Technology, 48, 217–221.
of fluidized bed polymerization reactors. Unpublished
doi:10.1016/0032-5910(86)80044-4
doctoral dissertation, Iowa State University, IA.
Floyd, S., Heiskanen, T., Taylor, W. T., Mann, G. E., &
Fan, R., Fox, R. O., & Muhle, M. E. (2007). Role of
Ray, W. H. (1987). Polymerization of olefins through het-
intrinsic kinetics and catalyst particle size distribution in
erogenous catalysis. VI. Effect of particle and mass trans-
CFD simulations of polymerization reactors. Fluidization
fer on polymerization behavior and polymer properties.
12 conference, (Article 122), 994-1000.
Journal of Applied Polymer Science, 33(4), 1021–1065.
Fanucci, J. B., Ness, N., & Yen, R.-H. (1979). On the doi:10.1002/app.1987.070330402
formation of bubbles in gas-particulate fluidized beds.
Forterre, Y., & Pouliquen, O. (2001). Longitudinal vor-
Journal of Fluid Mechanics, 94, 353–367. doi:10.1017/
tices in granular flows. Physical Review Letters, 86(26),
S0022112079001063
5886–5889. doi:10.1103/PhysRevLett.86.5886
Feitosa, K., & Menon, N. (2002). Breakdown of en-
Fortes, A. F., Joseph, D. D., & Lundgren, T. S. (1987).
ergy equipartition in a 2D binary vibrated granular gas.
Nonlinear mechanics of fluidization of beds of spherical
Physical Review Letters, 88(19), 198301. doi:10.1103/
particles. Journal of Fluid Mechanics, 177, 467–483.
PhysRevLett.88.198301
doi:10.1017/S0022112087001046
Feng, Y., & Yu, A. (2004). Assessment of model formula-
Foscolo, P. U., & Gibilaro, L. G. (1984). A fully predic-
tions in the discrete particle simulation of gas-solid flow.
tive criterion for the transition between particulate and
Industrial & Engineering Chemistry Research, 43, 8378.
aggregate fluidization. Chemical Engineering Science,
doi:10.1021/ie049387v
39(12), 1251–1260. doi:10.1016/0009-2509(84)80100-1
Feng, Y. Q., & Yu, A. (2004). Comments “Discrete particle-
continuum fluid modelling of gas-solid fluidsized beds”
by Kafui et al. Chemical Engineering Science, 59(3), 719.
doi:10.1016/j.ces.2003.11.003
439
Foscolo, P. U., Gibilaro, L. G., & Waldram, S. P. (1983). Garside, J., & Al-Dlbouni, M. R. (1977). Velocity-
A unified model for particulate expansion of fluidized- voidage relationships for fluidization and sedimentation
beds and flow in fixed porous-media. Chemical Engi- in solid-liquid systems. Industrial & Engineering Chem-
neering Science, 38(8), 1251–1260. doi:10.1016/0009- istry Process Design and Development, 16(2), 206–214.
2509(83)80045-1 doi:10.1021/i260062a008
Fox, R. O. (2008). A quadrature-based third-order moment Garzo, V. (2008). Brazil-nut effect versus reverse Brazil-
method for dilute gas-particle flows. Journal of Com- nut effect in a moderately dense granular fluid [R]. Physical
putational Physics, 227(12), 6313–6350. doi:10.1016/j. Review E: Statistical, Nonlinear, and Soft Matter Physics,
jcp.2008.03.014 78, 020301. doi:10.1103/PhysRevE.78.020301
Fox, R. O., & Vedula, P. (2010). Quadrature-based moment Garzo, V., Hrenya, C. M., & Dufty, J. W. (2007). Enskog
model for moderately dense polydisperse gas-particle theory for polydisperse granular mixtures. II. Sonine
flows. Industrial & Engineering Chemistry Research, polynomial approximation. Physical Review E: Statistical,
49(11), 5175–5187. doi:10.1021/ie9013138 Nonlinear, and Soft Matter Physics, 76(3). doi:10.1103/
PhysRevE.76.031304
Fryer, C., & Potter, O. E. (1976). Experimental investiga-
tion of models for fluidized bed catalytic reactors. AIChE Garzo, V., & Duffy, J. W. (2002). Hydrodynamics for a
Journal. American Institute of Chemical Engineers, 22(1), granular binary mixture at low density. Physics of Fluids,
38–47. doi:10.1002/aic.690220104 14(4), 1476–1490. doi:10.1063/1.1458007
Gajewski, A. (1985). Investigation of the electrifica- Garzó, V., & Dufty, J. (1999). Homogeneous Cooling State
tion of polypropylene particles during the fluidization for a Granular Mixture. Physical Review E: Statistical
process. Journal of Electrostatics, 17(3), 289–298. Physics, Plasmas, Fluids, and Related Interdisciplinary
doi:10.1016/0304-3886(85)90029-4 Topics, 60, 5706. doi:10.1103/PhysRevE.60.5706
Galvin, J. E., Hrenya, C. M., & Wildman, R. D. (2007). Garzó, V., Dufty, J., & Hrenya, C. (2007). Enskog Theory
On the role of the Knudsen layer in rapid granular flows. for Polydisperse Granular Fluids. I. Navier Stokes Order
Journal of Fluid Mechanics, 585, 73–92. doi:10.1017/ Transport. Physical Review E: Statistical, Nonlinear,
S0022112007006489 and Soft Matter Physics, 76, 031303. doi:10.1103/Phys-
RevE.76.031303
Galvin, J. E., Dahl, S. R., & Hrenya, C. M. (2005). On the
role of the non-equipartition in the dynamics of rapidly Garzó, V., & Dufty, J. W. (1999). Dense Fluid Transport
flowing granular mixtures. Journal of Fluid Mechanics, for Inelastic Hard Spheres. Physical Review E: Statistical
528, 207–232. doi:10.1017/S002211200400326X Physics, Plasmas, Fluids, and Related Interdisciplinary
Topics, 59, 5895. doi:10.1103/PhysRevE.59.5895
Galvin, J. E. (2007). On the hydrodynamic description of
binary mixtures of rapid granular flows and gas-fluidized Garzó, V., Hrenya, C., & Dufty, J. (2007). Enskog Theory
beds. PhD Thesis, University of Colorado, Boulder, CO. for Polydisperse Granular Fluids. II. Sonine Polynomial
Approximation. Physical Review E: Statistical, Nonlin-
Garg, S., & Pritchett, J. (1975). Dynamics of gas-fluidized
ear, and Soft Matter Physics, 76, 031304. doi:10.1103/
beds. Applied Physics (Berlin), 46(10), 4493–4498.
PhysRevE.76.031304
Garg, R., Passalacqua, A., Subramaniam, S., & Fox, R. O.
Garzó, V., Santos, A., & Montanero, J. (2007). Modified
(2008). Comparison of Euler-Euler and Euler-Lagrange
Sonine Approximation for the Navier-Stokes Transport
simulations of finite-Stokes-numbers gas-particle flows
Coefficients of a Granular Gas. Physica A, 376, 94.
in a lid-driven cavity. AIChE Annual Meeting, November
doi:10.1016/j.physa.2006.10.081
16th – 21st, Philadelphia.
440
Garzó, V., & Vega Reyes, F. (2009). Mass Transport of Garzó, V., Vega Reyes, F., & Montanero, J. (2008).
Impurities in a Moderately Dense Granular Gas. Physical Evaluation of the Navier-Stokes Transport Coefficients
Review E: Statistical, Nonlinear, and Soft Matter Physics, of a Granular Binary Mixture from a Modified Sonine
79, 041303. doi:10.1103/PhysRevE.79.041303 Approximation. arXiv:0806.1858.
Garzó, V. (2006). Segregation in granular binary mixtures: Ge, W. (2005). Particle methods for multiscale simula-
Thermal diffusion. Europhysics Letters, 75, 521–527. tion of complex flows. Chinese Science Bulletin, 50,
doi:10.1209/epl/i2006-10143-4 1057–1069. doi:10.1360/04wb0108
Garzó, V., & Dufty, J. (1999). Homogeneous cooling state Ge, W., & Li, J. (2001). Macro-scale pseudo-particle
for a granular mixture. Physical Review E: Statistical Phys- modeling for particle-fluid systems. Chinese Science
ics, Plasmas, Fluids, and Related Interdisciplinary Topics, Bulletin, 46, 1503–1507. doi:10.1007/BF02900568
60(5), 5706–5713. doi:10.1103/PhysRevE.60.5706
Ge, W., & Li, J. (2002). Physical mapping of fluidization
Garzó, V., & Dufty, J. W. (2002). Hydrodynamics for a regimes-the EMMS approach. Chemical Engineering Sci-
granular binary mixture at low density. Physics of Fluids, ence, 57, 3993–4004. doi:10.1016/S0009-2509(02)00234-
14(4), 1476–1490. doi:10.1063/1.1458007 8
Garzó, V., Dufty, J. W., & Hrenya, C. M. (2007). Enskog Ge, W., & Li, J. (2003). Macro-scale phenomena repro-
theory for polydisperse granular mixtures. I. Navier-Stokes duced in microscopic systems-pseudo-particle model-
order transport. Physical Review E: Statistical, Nonlin- ing of fluidization. Chemical Engineering Science, 58,
ear, and Soft Matter Physics, 76, 031303. doi:10.1103/ 1565–1585. doi:10.1016/S0009-2509(02)00673-5
PhysRevE.76.031303
Ge, W., Wang, W., Dong, W., et al. (2008) Meso-scale
Garzó, V., Hrenya, C. M., & Dufty, J. W. (2007). Enskog structure-a challenge of computational fluid dynamics
theory for polydisperse granular mixtures. II. Sonine for circulating fluidized bed risers. 9th International
polynomial approximation. Physical Review E: Statisti- Conference of Circulating Fluidized Beds. May 13-16,
cal, Nonlinear, and Soft Matter Physics, 76, 031304. Hamburg, Germany.
doi:10.1103/PhysRevE.76.031304
Gelbein, A. (April 14, 1981). Phthalic anhydride reaction
Garzó, V., & Montanero, J. M. (2002). Transport coeffi- system. US Patent 4,261,899.
cients of a heated granular gas. Physica A, 313, 336–356.
Geldart, D. (1986). Gas Fluidization Technology. Chich-
doi:10.1016/S0378-4371(02)00994-9
ester, UK: Wiley & Sons.
Garzó, V., & Montanero, J. M. (2003). Shear viscosity
Geldart, D. (1973). Types of gas fluidization. Powder Tech-
for a moderately dense granular mixture. Physical Review
nology, 7, 285–292. doi:10.1016/0032-5910(73)80037-3
041302. doi:10.1103/PhysRevE.68.041302 Gelderbloom, S. J., Gidaspow, D., & Lyczkowski, R. W.
(2003). CFD Simulations of bubbling/collapsing fluidized
Garzó, V., & Montanero, J. M. (2004). Diffusion of
beds for three Geldart group. AIChE Journal. Ameri-
impurities in a granular gas. Physical Review E: Statis-
can Institute of Chemical Engineers, 49(4), 844–858.
tical, Nonlinear, and Soft Matter Physics, 69, 021301.
doi:10.1002/aic.690490405
doi:10.1103/PhysRevE.69.021301
Ghia, U., Ghia, K., & Shin, C. (1982). High-re solutions
Garzó, V., Reyes, F. V., & Montanero, J. M. (Manuscript
for incompressible flow using the Navier-Stokes equations
submitted for publication). Modified Sonine approxi-
and a multigrid method. Journal of Computational Phys-
mation for granular binary mixture. Journal of Fluid
ics, 48, 387–411. doi:10.1016/0021-9991(82)90058-4
Mechanics.
441
Gibilaro, L. G., Di Felice, R., & Waldram, S. P. (1985). Gen- Gladden, L. F. (2003). Magnetic resonance: Ongoing
eralized friction factor and drag coefficient correlations for and future role in chemical engineering research. AIChE
fluid-particle interactions. Chemical Engineering Science, Journal. American Institute of Chemical Engineers, 49,
40(10), 1817–1823. doi:10.1016/0009-2509(85)80116-0 2–9. doi:10.1002/aic.690490102
Gidaspow, D. (1994). Multiphase Flow and Fluidization: Gladden, L. F., Anadon, L. D., Dunckley, C. P., Mantle,
Continuum and Kinetic Theory Description. San Diego, M. D., & Sederman, A. J. (2007). Insights into gas-
CA: Academic press. liquid-solid reactors obtained by magnetic resonance
imaging. Chemical Engineering Science, 62, 6969–6977.
Gidaspow, D., Jung, J., & Singh, R. (2004). Hydrody-
doi:10.1016/j.ces.2007.08.084
namics of fluidization using kinetic theory: an emerging
paradigm. Powder Technology, 148, 123. doi:10.1016/j. Glicksman, L. (1988). Scaling relationships for fluidized
powtec.2004.09.025 beds. Chemical Engineering Science, 43(6), 1419–1421.
doi:10.1016/0009-2509(88)85118-2
Gidaspow, D., & Mostofi, R. (2003). Maximum Carry-
ing Capacity and Granular Temperature of A, B and C Glicksman, L. R. (1997). Heat transfer in circulating flu-
Particles. AIChE Journal. American Institute of Chemical idized beds. In Grace, J. R., Avidan, A. A., & Knowlton,
Engineers, 49(4), 831–843. doi:10.1002/aic.690490404 T. M. (Eds.), Circulating Fluidized Beds (pp. 261–307).
London: Blackie Academic & Professional.
Gidaspow, D. (1986). Hydrodynamics of fluidization and
heat transfer supercomputer modeling. Applied Mechanics Glowinski, R., Pan, T. W., & Hesla, T. I. (1999). A dis-
Reviews, 39, 1–22. doi:10.1115/1.3143702 tributed Lagrange multiplier/fictitious domain method
for particulate flows. International Journal of Multiphase
Gidaspow, D., & Huilin, L. (1998). quation of state and
Flow, 25, 755–794. doi:10.1016/S0301-9322(98)00048-2
radial distribution functions of FCC particles in a CFB.
AIChE Journal. American Institute of Chemical Engineers, Glowinski, R., Hestla, T., Joseph, D. D., Pan, T. W., &
44(2), 279–293. doi:10.1002/aic.690440207 Periaux, J. (1997). Distributed lagrange multiplier meth-
ods for particulate flows. In Bristeau, M. O., Etgen, G.,
Gidaspow, D., Seo, Y., & Ettehadieh, B. (1983). Hy-
Fitzgibbon, W., Lions, J. L., Periaux, J., & Wheeler, M.
drodynamics of fluidization: Experimental and theo-
F. (Eds.), Computational Science for the 21st Century
retical bubble sizes in a two-dimensional bed with a jet.
(pp. 270–279). New York: Wiley.
Chemical Engineering Communications, 22(5), 253–272.
doi:10.1080/00986448308940060 Gobin, A., Neau, H., Simonin, O., Llinas, J. R., Reiling,
V., & Seilo, J. L. (2003). Fluid dynamic numerical simula-
Gidaspow, D., Bezburuah, R., & Ding, J. (1992). Hydro-
tion of a gas phase polymerization reactor. International
dynamics of Circulating Fluidized Beds, Kinetic Theory
Journal for Numerical Methods in Fluid Dynamics, 43,
Approach. In Potter, O. E., & Nicklin, D. J. (Eds.), Fluidiza-
1199–1220. doi:10.1002/fld.542
tion VII (pp. 75–82). New York: Engineering Foundation.
Godlieb, W., Deen, N. G., & Kuipers, J. A. M. (2007).
Gidaspow, D.& Ettehadieh, B. (1983). Fluidization in
A discrete particle simulation study of solids mixing in
two-dimensional beds with a jet, Part II: hydrodynamic
a pressurized fluidized bed. In 2007 ECI Conference on
modeling. I&EC Fundamentals, 22, 193-201.
the 12th International Conference on Fluidization -New
Gingold, R. A., & Monaghan, J. J. (1977). Smoothed Horizons in Fluidization Engineering, Vancouver, Canada.
particle hydrodynamics: theory and application to
non-spherical stars. Monthly Notification of the Royal
Astronamy Society, 181, 375.
442
Goldhirsch, I., Tan, M. L., & Zanetti, G. (1993). A Molecu- Grad, H. (1960). Theory of Rarified Gases. In Rarified
lar Dynamics Study of Granular Fluids 1: The Unforced Gas Dynamics. Oxford, UK: Pergamon Press.
Granular Gas in 2 Dimensions. Journal of Scientific
Grad, H. (1949). On the kinetic theory of rarified gases.
Computing, 8, 1. doi:10.1007/BF01060830
Communications on Pure and Applied Mathematics, 2,
Goldhirsch, I., Noskowicz, S. H., & Bar-Lev, O. (2004). 329. doi:10.1002/cpa.3160020403
Theory of granular gases: some recent results and some
Grad, H. (1958). Principles of the kinetic theory of gases.
open problems. Journal of Physics Condensed Matter,
In Flügge, S. (Ed.), Handbuch der Physik XII: Thermo-
17, 2591–2608. doi:10.1088/0953-8984/17/24/015
dynamic der Gase. Berlin, Germany: Springer.
Goldhirsch, I. (2003). Rapid granular flows. Annual
Grandell, J. (1976). Doubly stochastic Poisson processes.
Review of Fluid Mechanics, 35, 267–293. doi:10.1146/
Berlin: Springer-Verlag.
Grassler, T., & Wirth, K. E. (1999). Radial and Axial
Goldschmidt, M. J. V. (2002). Hydrodynamic modelling
Profile of Solids Concentration in a High Loaded Riser
of dense gasfluidised beds: comparison of the kinetic
Reactor. In Werther, J. (Ed.), Circulating Fluidized Bed
theory of granular flow with 3D hard-sphere discrete
Technology VI (pp. 65–70). Frankfort: DECHEMA.
particle simulations. Chemical Engineering Science, 57,
2059. doi:10.1016/S0009-2509(02)00082-9 Griffiths, J. F., Barnard, J. A., & Bradley, J. N. (1995).
Flame and Combustion, (3rd). New York: Chapman & Hall.
Goldschmidt, M. J. V., Beetstra, R., & Kuipers, J. A. M.
(2004). Hydrodynamic modelling of dense gas-fluidised Gu, W. K., & Chen, J. C. (1998). A model for solid con-
beds: comparison and validation of 3D discrete particle centration in circulating fluidized beds. In Fan, L. S., &
and continuum models. Powder Technology, 142(1), Knowlton, T. M. (Eds.), Fluidization (p. 501). Durago,
23–47. doi:10.1016/j.powtec.2004.02.020 CO: Engineering Foundation.
Goldschmidt, M. J. V., Kuipers, J. A. M., & van Swaaij, Guardiola, J., Rojo, V., & Ramos, G. (1996). Influence of
W. P. M. (2001). Hydrodynamic modeling of dense gas- particle size, fluidization velocity and relative humidity
fluidised beds using the kinetic theory of granular flow: on fluidized bed electrostatics. Journal of Electrostatics,
effect of coefficient of restitution on bed dynamics. 37(1-2), 1–20. doi:10.1016/0304-3886(96)00002-2
Chemical Engineering Science, 56, 571–578. doi:10.1016/
Guenther, C., & Breault, R. (2007). Wavelet analysis to
S0009-2509(00)00262-1
characterize cluster dynamics in a circulating fluidized
Goldstein, D., Handler, R., & Sirovich, L. (1993). Model- bed. Powder Technology, 173(3), 163–173. doi:10.1016/j.
ing a no-slip flow boundary with an external force field. powtec.2006.12.016
Journal of Computational Physics, 105(2), 354–366.
Guenther, C., & Syamlal, M. (2001). The effect of nu-
doi:10.1006/jcph.1993.1081
merical diffusion on simulation of isolated bubbles in a
Grace, J. R., & Tuot, J. (1979). A theory for cluster for- gas–solid fluidized bed. Powder Technology, 116(2-3),
mation in vertically conveyed suspensions of intermedi- 142–154. doi:10.1016/S0032-5910(00)00386-7
ate density. Transactions of the Institution of Chemical
Guenther, C., & Breault, R. W. (2005). Two and three
Engineers, 57, 49–54.
dimensional simulations investigating dispersion and mass
Grace, J., & Sun, G. (1991). Influence of particle size transfer coefficients in an Eulerian-Eulerian CFD model,
distribution on the performance of fluidized bed reac- In S. T. Johansen & I. Gamst Page, (Ed.), Proceedings of
tors. Canadian Journal of Chemical Engineering, 69(5), the Fourth International Conference on Computational
1126–1134. doi:10.1002/cjce.5450690512 Fluid Dynamics in the Oil and Gas, Metallurgical &
Process Industries, Paper 53, Trondheim, Norway.
443
Guenther, C., Syamlal, M., Longanback, J., & Smith, P. Harris, A. T., Davidson, J. F., & Thorpe, R. B. (2002).
(2003, Nov.). Two-Fluid Model of an Industrial Scale The prediction of particle cluster properties in the near
Transport Gasifier. Presented at AIChE Annual Meet- wall region of a vertical riser. Powder Technology, 127,
ing, Conference Proceedings, San Francisco, California. 128–143. doi:10.1016/S0032-5910(02)00114-6
Gunn, D. J. (1978). Transfer of Heat or Mass to Particles Harris, S. E., & Crighton, D. G. (1994). Solitons, solitary
in Fixed and Fluidized Beds. International Journal of waves and voidage disturbances in gas-fluidized beds.
Heat and Mass Transfer, 21, 467–476. doi:10.1016/0017- Journal of Fluid Mechanics, 266, 243–276. doi:10.1017/
9310(78)90080-7 S0022112094000996
Gunn, D. J., & Malik, A. A. (1967). The structure of fluid- Harris, S. E., & Crighton, D. G. (1994). Solitons, solitary
ized beds in particulate fluidization. In A. A. Dringkenbrug waves, and voidage disturbances in gas-fluidized beds.
(Ed.), Proceedings of the International Symposium on Journal of Fluid Mechanics, 266, 243–276. doi:10.1017/
Fluidization (pp. 52-65). Eindhoven, The Netherlands: S0022112094000996
Netherlands University Press.
Hartge, E. U., Rensner, D., & Werther, J. (1988). Solid
Gupta, P., Velazquez-Vargas, L. G., Li, F., & Fan, L. Concentration and Velocity Patterns in Circulating Fluid-
S. (2005, Oct.). Chemical looping combustion of coal. ized Beds. In Basu, P., & Large, J. F. (Eds.), Circulation
Presented at AIChE Annual Meeting, Conference Pro- Fluidized Bed Technology II (pp. 165–179). Oxford, UK:
ceedings, Cincinnati, Ohio. Pergamon Press.
Haff, P. K. (1983). Grain Flow as a Fluid Mechanical Hasimoto, H. (1959). On the periodic fundamental
Phenomenon. Journal of Fluid Mechanics, 134, 401. solutions of the Stokes equations and their application
doi:10.1017/S0022112083003419 to viscous flow past a cubic array of spheres. Jour-
nal of Fluid Mechanics, 5(2), 317–328. doi:10.1017/
Happel, J. (1958). Viscous flow in multi-particle systems:
S0022112059000222
Slow motion of fluids relative to beds of spherical par-
ticles. AIChE Journal. American Institute of Chemical Haworth, D. C., & Jansen, K. (1996). Large-Eddy Simula-
Engineers, 4, 197. doi:10.1002/aic.690040214 tion on Unstructured Deforming Meshes: Towards Re-
ciprocating IC Engines. Center for Turbulence Research,
Harlow, F. H. (1957). Hydrodynamic Problems Involving
Proceedings of the Summer Program.
Large Fluid Distortion. J. Assoc. Comp. Mach., 4, 137.
Hays, R., & Karri, S. B. Reddy, Cocco, R. & Knowlton,
Harlow, F. H., & Welch, E. (1965). Numerical Calcula-
T.M. (2008) Small Particle Cluster Formation in Fluid-
tion of Time-Dependent Viscous Incompressible Flow
ized Beds and its Effect on Entrainment. In J. Werther, W.
of Fluids with Free Surface. Physics of Fluids, 8, 2182.
Nowak, K-E. Wirth & E-U. Hartge, (Ed.), Proceedings of
doi:10.1063/1.1761178
the 9th International Conference on Circulating Fluidized
Harlow, F. H., & Amsden, A. (1975). Numerical calculation Bed, (pp. 93), Hamburg, Germany, May 13-16, 2008.
of multiphase flow. Journal of Computational Physics,
He, X. Y., & Luo, L. S. (1997). Theory of the lattice
17, 19–52. doi:10.1016/0021-9991(75)90061-3
Boltzmann method: From the Boltzmann equation to
Harlow, F. H. (1964). The particle-in-cell computing the lattice Boltzmann equation. Physical Review E:
method of fluid dynamics. In Alder, B., Fembach, S., & Statistical Physics, Plasmas, Fluids, and Related In-
Rotenberg, M. (Eds.), Fundamental Methods in Hydro- terdisciplinary Topics, 56(6), 6811–6817. doi:10.1103/
dynamics. New York: Academic Press. PhysRevE.56.6811
Harlow, F. H., & Amsden, A. A. (1971) Fluid Dynamics.

Los Alamos National Laboratory Report LAMS-4700.
444
Helland, E., Bournot, H., Occelli, R., & Tadrist, L. (2007). Hishida, K., Takemoto, K., & Maeda, M. (1987). Turbu-
Drag reduction and cluster formation in a circulating flu- lence characteristics of gas-solids two-phase confined
idised bed. Chemical Engineering Science, 62, 148–158. jet (effect of particle density). Japanese J. Multiphase
doi:10.1016/j.ces.2006.08.012 Flow, 1, 56–68.
Helland, E., Occelli, R., & Tadrist, L. (2000). Numerical Hoel, E. L., Cozewith, C., & Byrne, G. D. (1994). Ef-
study of cluster formation in gas-particle flow in a circu- fect of diffusion on heterogeneous ethylene propylene
lating fluidized bed. Powder Technology, 110, 210–221. copolymerization. AIChE Journal. American Institute
doi:10.1016/S0032-5910(99)00260-0 of Chemical Engineers, 40, 1669–1684. doi:10.1002/
aic.690401009
Hendrickson, G. (2006). Electrostatics and gas phase fluid-
ized bed polymerization reactor wall sheeting. Chemical Holway, L. H. (1966). New statistical models for kinetic
Engineering Science, 61(4), 1041–1064. doi:10.1016/j. theory: methods of construction. Physics of Fluids, 9(9),
ces.2005.07.029 1658–1673. doi:10.1063/1.1761920
Hetsroni, G. (1989). Particles-turbulence interaction. In- Holzer, A., & Sommerfeld, M. (2008). New simple cor-
ternational Journal of Multiphase Flow, 15(5), 735–746. relation formula for the drag coefficient of non-spherical
doi:10.1016/0301-9322(89)90037-2 particles. Powder Technology, 184(3), 361. doi:10.1016/j.
powtec.2007.08.021
Heynderickx, G. J., Das, A. K., De Wilde, J., & Marin, G.
B. (2004). Effect of clustering on gas-solid drag in dilute Homsy, G. M. (1998). Nonlinear waves and the origin of
two-phase flow. Industrial & Engineering Chemistry bubbles in fluidized beds. Applied Scientific Research,
Research, 43(16), 4635–4646. doi:10.1021/ie034122m 58, 251–274. doi:10.1023/A:1000787803463
Higbie, R. (1935). The rate of absorption of a pure gas into Hoogerbrugge, P. J., & Koelman, J. M. V. A. (1992).
a still liquid during short periods of exposure. Transac- Simulating microscopic hydrodynamic phenomena with
tions of the A.I. Ch.E., 31, 365–387. dissipative particle dynamics. Europhysics Letters, 19(3),
155–160. doi:10.1209/0295-5075/19/3/001
Hill, R., Koch, D., & Ladd, J. (2001). Moderate-Reynolds-
number flows in ordered and random arrays of spheres. Hoomans, B. P. B., Kuipers, J. A. M., Briels, W. J., &
Journal of Fluid Mechanics, 448, 243. doi:10.1017/ Van Swaaij, W. P. M. (1996). Discrete particle simula-
S0022112001005936 tion of bubble and slug formation in a two-dimensional
gas-fluidised bed: a hard-sphere approach. Chemical
Hill, R. J., Koch, D. L., & Ladd, A. J. C. (2001). Moderate-
Engineering Science, 51(1), 99–118. doi:10.1016/0009-
Reynolds-number flows in ordered and random arrays
2509(95)00271-5
of spheres. Journal of Fluid Mechanics, 448, 243–278.
doi:10.1017/S0022112001005936 Hoomans, B. P. B. (2000). Granular dynamics of gas-solid
two-phase flows. Ph.D. dissertation, Twente University,
Hirshfelder, J. O., Curtiss, C. F., & Bird, R. B. (1954).
The Netherlands.
Molecular Theory of Gases and Liquids. New York, NY:
John Wiley & Sons. Horio, M., & Kuroki, H. (1994). Three-dimensional flow
visualization of dilutely dispersed solids in bubbling and
Hirt, C. W., & Nichols, B. D. (1981). Volume of fluid (VOF)
circulating fluidized beds. Chemical Engineering Science,
method for the dynamics of free boundaries. Journal of
49, 2413. doi:10.1016/0009-2509(94)E0071-W
Computational Physics, 39, 201–226. doi:10.1016/0021-
9991(81)90145-5 Horio, M. (1997). Hydrodynamics. In Grace, J. R., Avidan,
A. A., & Knowlton, T. M. (Eds.), Circulating Fluidized
Beds (pp. 21–78). London: Blackie Academic & Prof.
445
Hrenya, C. M., Galvin, J. E., & Wildman, R. D. (2008). Iddir, H., & Arastoopour, H. (2005). Modeling of Multitype
Evidence of higher-order effects in thermally-driven, Particle Flow Using the Kinetic Theory Approach. AIChE
rapid granular flows. Journal of Fluid Mechanics, 598, Journal. American Institute of Chemical Engineers, 51(6),
429–450. doi:10.1017/S0022112007000079 1620–1632. doi:10.1002/aic.10429
Hrenya, C., & Sinclair, J. (1997). Effects of particle-phase Iddir, H. (2004). Modeling of the multiphase mixture of
turbulence in gas-solid flows. AIChE Journal. Ameri- particles using the kinetic theory approach. Ph.D. Thesis,
can Institute of Chemical Engineers, 43(4), 853–869. Illinois Institute of Technology, Chicago.
doi:10.1002/aic.690430402
Igci, Y., Andrews, A. T., Sundaresan, S., Pannala, S., &
Hrenya, C. (1996). Predicting Dense, Turbulent Gas-Solid O’Brien, T. (2008). Filtered two-fluid models for fluid-
Flows in Vertical Risers. Doctoral Dissertation, Carnegie ized gas-particle suspensions. AIChE Journal. American
Mellon University, Pittsburgh, PA. Institute of Chemical Engineers, 54(6), 1431–1448.
doi:10.1002/aic.11481
Hsiau, S. S., & Hunt, M. L. (1996). Granular thermal dif-
fusion in flows of binary-sized mixtures. Acta Mechanica, Ikeno, T., & Kajishima, T. (2007). Finite-difference
114, 121–137. doi:10.1007/BF01170399 immersed boundary method consistent with wall con-
ditions for incompressible turbulent flow simulations.
Hu, H. H. (1996). Direct simulation of flows of solid-fluid
Journal of Computational Physics, 226(2), 1485–1508.
mixtures. International Journal of Multiphase Flow, 22,
doi:10.1016/j.jcp.2007.05.028
335–352. doi:10.1016/0301-9322(95)00068-2
Issa, R. I. (1986). Solution of the implicitly discretised
Hu, H. H., Patankar, N. A., & Zhu, M. Y. (2001). Direct
fluid flow equations by operator splitting. Journal of
numerical simulations of fluid-solid systems using the
Computational Physics, 61(1), 40–65. doi:10.1016/0021-
arbitrary Lagrangian-Eulerian technique. Journal of
9991(86)90099-9
Computational Physics, 169, 427–462. doi:10.1006/
jcph.2000.6592 Issangya, A. S., Grace, J. R., & Bai, D. R. (2000). Fur-
ther measurements of flow dynamics in a high-density
Huan, C., Yang, X., Candela, D., Mair, R., & Walsworth,
circulating fluidized bed riser. Powder Technology, 111,
R. (2004). NMR Measurements on a Three-Dimensional
104. doi:10.1016/S0032-5910(00)00246-1
Vibrofluidized Granular Medium. Physical Review E:
Statistical, Nonlinear, and Soft Matter Physics, 69, 041302. Issangya, A. S., Knowlton, T. M., & Karri, S. B. R. (2007).
doi:10.1103/PhysRevE.69.041302 Detection of Gas Bypassing due to Jet Streaming in Deep
Fluidized Beds of Group A Particles. In Bi, X., Berruti,
Huilin, L., Gidaspow, D., & Manger, E. (2001). Kinetic
F., & Pugsley, T. (Eds.), Fluidization XII (pp. 775–782).
theory of fluidized binary granular mixtures. Physical
New York: Engineering Conferences International.
64, 061301. doi:10.1103/PhysRevE.64.061301 Issangya, A. S., Bai, D., Grace, J. R., & Zhu, J. (1998).
Solids Flux Profiles in a High Density Circulating Fluid-
Hutchinson, R. A., Chen, C. M., & Ray, W. H. (1992).
ized Bed Riser. In Fan, L.-S., & Knowlton, T. M. (Eds.),
Polymerization of olefins through heterogeneous catalysis
Fluidization IX (pp. 197–204). New York: Engineering
X: Modeling of particle growth and morphology. Journal
Foundation.
of Applied Polymer Science, 44, 1398–1414. doi:10.1002/
app.1992.070440811 Issangya, A. S. (1998). Flow dynamics in high density
circulating fluidized beds. Ph.D thesis. The University
of British Columbia, Canada.
446
Jackson, R. (2000). The Dynamics of Fluidized Particles. Jiradilok, V., Gidaspow, D., & Breault, R. W. (2007).
Cambridge, UK: Cambridge University Press. Computation of gas and solid dispersion coefficients in
turbulent risers and bubbling beds. Chemical Engineering
Jackson, R. (1963). The mechanics of fluidized beds. I:
Science, 62, 3397–3409. doi:10.1016/j.ces.2007.01.084
the stability of the state of uniform fluidization. Transac-
tions of the Institution of Chemical Engineers, 41, 13–21. Johnson, P. C., & Jackson, R. (1987). Frictional-collisional
constitutive relations for granular materials, with applica-
Jackson, R. (1997). Locally averaged equations of mo-
tions to plane shearing. Journal of Fluid Mechanics, 176,
tion for a mixture of identical spherical particles and
67–93. doi:10.1017/S0022112087000570
a newtonian fluid. Chemical Engineering Science, 52,
2457–2469. doi:10.1016/S0009-2509(97)00065-1 Joseph, G. G., Leboreiro, J., Hrenya, C. M., & Stevens, A.
R. (2007). Experimental segregation profiles in bubbling
Jean, R. H., & Fan, L. S. (1992). On the model equations
gas-fluidized beds. AIChE Journal. American Institute
of Gibilaro and Foscolo with corrected buoyancy force.
of Chemical Engineers, 53, 2804–2813. doi:10.1002/
Powder Technology, 72, 201–205. doi:10.1016/0032-
aic.11282
5910(92)80038-X
Kafui, K. D., Thornton, C., & Adams, M. J. (2002). Discrete
Jenkins, J. T., & Mancini, F. (1987). Balance laws and
particle-continuum fluid modelling of gas-solid fluidised
constitutive relations for plane flows of a dense, binary
beds. Chemical Engineering Science, 57, 2395–2410.
mixture of smooth, nearly elastic, circular disks. Journal
doi:10.1016/S0009-2509(02)00140-9
of Applied Mechanics, 54, 27–34. doi:10.1115/1.3172990
Kafui, K. D., Thornton, C., & Adams, M. J. (2004).
Jenkins, J. T., & Mancini, F. (1989). Kinetic theory for
Reply to comments by Feng and Yu on “Discrete particle-
binary mixtures of smooth, nearly elastic spheres. Phys.
continuum fluid modelling of gas-solid fluidsized beds”
Fluids A, 1(12), 2050–2057. doi:10.1063/1.857479
by Kafui et al. Chemical Engineering Science, 59(3), 723.
Jenkins, J. T., & Yoon, D. K. (2002). Segregation in Binary doi:10.1016/j.ces.2003.11.004
Mixtures under Gravity. Physical Review Letters, 88(19),
Kajishima, T. (2004). Influence of particle rotation on the
194301. doi:10.1103/PhysRevLett.88.194301
interaction between particle clusters and particle-induced
Jenkins, J. T. (1998). Kinetic theory for nearly elastic turbulence. International Journal of Heat and Fluid Flow,
spheres. Physics of Dry Granular Media, 350, 353–370. 25, 721–728. doi:10.1016/j.ijheatfluidflow.2004.05.007
Jenkins, J. T. (1998). Particle segregation in collisional Kalil, J., Chu, J. C., & Wetteroth, W. A. (1953). Mass
flows of inelastic spheres. In Hermann, H. J., Hovi, J. P., Transfer in a Fluidized Bed. Chemical Engineering
& Luding, S. (Eds.), Physics of Dry Granular Media (pp. Progress, 49(3), 141–149.
645–658). Amsterdam: Kluwer.
Kandhai, D., Derksen, J. J., & Van den Akker, H. E. A.
Jenkins, J., & Mancini, F. (1989). Kinetic Theory for (2003). Interphase drag coefficients in gas-solid flows.
Binary Mixtures of Smooth, Nearly Elastic Spheres. AIChE Journal. American Institute of Chemical Engineers,
Physics of Fluids, A 1, 2050. 49(4), 1060–1065. doi:10.1002/aic.690490423
Jiradilok, V., Gidaspow, D., Damronglerd, S., Koves, W. Kaneko, Y., Shiojima, Y. T., & Horio, M. (1999). DEM
J., & Mostofi, R. (2006)... Chemical Engineering Science, simulation of fluidized beds for gas-phase olefin polym-
61, 5544. doi:10.1016/j.ces.2006.04.006 erization. Chemical Engineering Science, 54, 5809–5821.
doi:10.1016/S0009-2509(99)00153-0
447
Karema, H., & Lo, S. (1999). Efficiency of interphase Khakhar, D. V., McCarthy, J. J., & Ottino, J. M. (1999).
coupling algorithms in fluidized bed conditions. Com- Mixing and segregation of granular materials in
puters & Fluids, 28, 323–360. doi:10.1016/S0045- chute flows. Chaos (Woodbury, N.Y.), 9(3), 594–610.
7930(98)00028-0 doi:10.1063/1.166433
Karion, A., & Hunt, M. L. (2000). Wall streses in granular Khan, A. R., & Richardson, J. F. (1990). Pressure gradient
couette flows of mono-sized particles and binary mixtures. and friction factor for sedimentation and fluidisation of
Powder Technology, 109, 145–163. doi:10.1016/S0032- uniform spheres in liquids. Chemical Engineering Science,
5910(99)00233-8 45(1), 255–265. doi:10.1016/0009-2509(90)87097-C
Karlis, D., & Xekalaki, E. (2005). Mixed Poisson Khazai, B., Vrieland, E., Murchison, C., Dixit, R., &
distributions. International Statistical Review, 73, 35. Weihl, E. (Sept 8, 1992). Process of oxidizing aliphatic
doi:10.1111/j.1751-5823.2005.tb00250.x hydrocarbons employing a molybdate catalyst composi-
tion. US Patent 5,146,031.
Karri, S. B. R., & Knowlton, T. M. (1991). A Practical
Definition of the Fast Fluidization Regime. In Basu, P., Kiashemshaki, A., Rahmat, S., & Mostoufi, N. (2007).
Horio, M., & Hasatani, M. (Eds.), Circulating Fluidized Reactor modeling of fluidized bed ethylene polymeriza-
Bed III (pp. 67–72). Oxford, UK: Pergamon Press. tion using dynamic two phase behavior. Iranian Journal
of Chemistry and Chemical Engineering, 26, 51–60.
Karri, S. B. R., & Knowlton, T. M. (1998). Flow Direc-
tion and Size Segregation of Annulus Solids in a Riser. Kieffer, S. W., & Sturtevant, B. (1984). Laboratory studies
In Fan, L.-S., & Knowlton, T. M. (Eds.), Fluidization of volcanic jets. Journal of Geophysical Research, 89,
IX (pp. 189–195). New York: Engineering Foundation. 8253–8268. doi:10.1029/JB089iB10p08253
Karri, R., & Knowlton, T. M. (2002) Wall Solids Upflow Kim, J. Y., & Choi, K. Y. (2001). Modeling of particle
and Downflow Regimes in Risers for Group A Solids. In segregation phenomena in a gas phase fluidized bed olefin
J.R. Grace, J.X. Zhu, H. de Lasa (Eds.), Circulating Fluid- polymerization reactor. Chemical Engineering Science,
ized Bed Technology VII, (pp. 310-316). Ottawa: CSChE. 56, 4069–4083. doi:10.1016/S0009-2509(01)00078-1
Karri, S. B. R., & Knowlton, T. M. (August 29-31, 2001). Kirkpatrick, T., Das, S. P., Ernst, M. H., & Piasecki, J.
The PSRI Challenge Problem. Paper presented at MFDRC (1990). Kinetic Theory of Transport in a Hard Sphere
Meeting, Salt Lake, Utah. Crystal. The Journal of Chemical Physics, 92, 3768.
doi:10.1063/1.457835
Karri, S.B., Issangya, A. & Knowlton (2004). Gas Bypass-
ing in deep fluidized beds. In U. Arena, R. Chirone, M. Knowlton, T. M., Carson, J. W., Klinzing, G. E., & Yang,
Miccio & P. Salatino (eds.), Fluidization XI, (pp. 515-521). W.-C. (1994). The importance of storage, transfer, and
New York: Engineering Foundation. collection. Chemical Engineering Progress, 90(4), 44–54.
Kashiwa, B. A., & Hull, L. M. (Aug 9-12, 2004). Multifield Knowlton, T. M., Karri, S. B. R., & Issangya, A. (2005).
Closure Modeling for Metal-Loaded High Explosives. Scale-up of fluidized-bed hydrodynamics. Powder Tech-
Paper presented in 7th Joint Classified Bobs/Warheads nology, 150(2), 72–77. doi:10.1016/j.powtec.2004.11.036
& Ballistics Symposium, Monterey, CA, Los Alamos
Koch, D. L. (1990). Kinetic theory for a monodisperse
National Lab Report LA-UR-045081.
gas-solid suspension. Phys. Fluids A, 2, 1711–1723.
Khain, E., Meerson, B., & Sasorov, P. V. (2008). Knudsen doi:10.1063/1.857698
temperature jump and the Navier-Stokes hydrodynamics
of granular gases driven by thermal walls. Physical Review
041303. doi:10.1103/PhysRevE.78.041303
448
Koch, D. L., & Sangani, A. S. (1999). Particle pressure Kunii, D., & Levenspiel, O. (1991). Fluidization En-
and marginal stability limits for a homogenous mono- gineering (2nd ed.). Boston: Butterworth- Heinemann.
disperse gas-fluidized bed: kinetic theory and numerical
Kunz, R., Cope, W., & Venkateswaran, S. (1999). De-
simulations. Journal of Fluid Mechanics, 400, 229–263.
velopment of an implicit method for multi-fluid flow
doi:10.1017/S0022112099006485
simulations. Journal of Computational Physics, 152,
Koch, D. L. (1990). Kinetic-Theory for a Monodisperse 78–101. doi:10.1006/jcph.1999.6235
Gas-Solid Suspension. Physics of Fluids. A, Fluid Dynam-
Kurose, R., & Komori, S. (1999). Drag and lift forces on a
ics, 2(10), 1711–1723. doi:10.1063/1.857698
rotating sphere in a linear shear flow. Journal of Fluid Me-
Koch, D. L., & Hill, R. J. (2001). Inertial effects in chanics, 384, 183–206. doi:10.1017/S0022112099004164
suspension and porous-media flows. Annual Review of
Lackermeier, U., Rudnick, C., & Werther, J. (2001).
Fluid Mechanics, 33, 619–647. doi:10.1146/annurev.
Visualization of flow structures inside a circulating
fluid.33.1.619
fluidized bed by means of laser sheet an image process-
Koshizuka, S., & Oka, Y. (1996). Moving-particle semi- ing. Powder Technology, 114, 71. doi:10.1016/S0032-
implicit method for fragmentation of incompressible fluid. 5910(00)00265-5
Nuclear Science and Engineering, 123, 421.
Lackermeier, U., & Werther, J. (2002). Flow phenomena
Koshizuka, S., Tamako, Y., & Oka, Y. (1995). A paticle in the exit zone of a circulating fluidized bed. Chemical
method for incompressible viscous flow with fluid frag- Engineering Progress, 41(9), 771–783. doi:10.1016/
mentation. Journal of Computational Fluid Dynamics, S0255-2701(02)00008-9
4, 29.
Ladd, A. J. C., & Verberg, R. (2001). Lattice-
Kostinski, A. B., & Jameson, A. R. (1997). Fluctuation Boltzmann simulations of particle-fluid suspensions.
properties of precipitation. Part 1: On the deviations Journal of Statistical Physics, 104(5-6), 1191–1251.
of single-size drop counts from the Poisson distribu- doi:10.1023/A:1010414013942
tion. Journal of the Atmospheric Sciences, 54, 2174.
Ladenburg, R., van Voorhis, C. C., & Winckler, J. (1949).
doi:10.1175/1520-0469(1997)054<2174:FPOPPI>2.0
Interferometric studies of faster than sound phenomena.
.CO;2
Part II. Analysis of supersonic air jets. Physical Review,
Krishna, R., & van Baten, J. M. (2001). Using CFD for 76, 662–677. doi:10.1103/PhysRev.76.662
scaling up gas-solid bubbling fluidized bed reactors with
Lapple, C. E., & Shepherd, C. B. (1940). Calculation of
Geldart A powders. Chemical Engineering Journal, 82,
particle trajectories. Industrial & Engineering Chemistry,
247–257. doi:10.1016/S1385-8947(00)00369-7
32(5), 605. doi:10.1021/ie50365a007
Krol, S., Pekediz, A., & de Lasa, H. (2000). Particle clus-
Larsen, G., Gronvold, K., & Thorarinson, S. (1979). Volca-
tering in down flow reactors. Powder Technology, 108(1),
nic eruption through a geothermal borehole at Namafjall,
6–20. doi:10.1016/S0032-5910(99)00196-5
Iceland. Nature, 278, 707–710. doi:10.1038/278707a0
Kuipers, J., Hoomans, B., & van Swaaij, W. (1998),
Launder, B. E., & Spalding, D. B. (1974). The Numerical
Hydrodynamic modeling of gas-fluidized beds and their
Computation of Turbulent Flows. Computer Methods
role for design and operation of fluidized bed chemical
in Applied Mechanics and Engineering, 3, 269–289.
reactors. In Proceedings of the Fluidization IX conference,
doi:10.1016/0045-7825(74)90029-2
(p.15), Durango, USA.
Kumaran, V. (2004). Constitutive Relations and Linear

Stability of a Sheared Granular Flow. Journal of Fluid
Mechanics, 506, 1. doi:10.1017/S0022112003007602
449
Leboreiro, J., Joseph, G. G., & Hrenya, C. M. (2008). Re- Li, J. (2000). Compromise and Resolution — exploring the
visiting the standard drag law for bubbling, gas-fluidized multi-scale nature of gas-solid fluidization. Powder Tech-
beds. Powder Technology, 183, 385–400. doi:10.1016/j. nology, 111, 50–59. doi:10.1016/S0032-5910(00)00238-2
powtec.2008.01.008
Li, J., Cheng, C., Zhang, Z., Yuan, J., Nemet, A., &
Leboreiro, J. (2008). Influence of drag laws on segrega- Fett, F. N. (1999). The EMMS model—its application,
tion and bubbling behavior in gas-fluidized beds. Ph.D. development and updated concepts. Chemical Engi-
Thesis, University of Colorado, Boulder. neering Science, 54, 5409–5425. doi:10.1016/S0009-
2509(99)00274-2
Lebowitz, J. L. (1964). Exact solution of generalized
Percus-Yevick equation for a mixture of hard spheres. Li, J., & Kwauk, M. (1994). Particle-fluid two-phase
Physical Review, 133, 895–899. doi:10.1103/Phys- flow: the energy-minization multi-scale method. Beijing:
Rev.133.A895 Metallurgical Industry Press.
Lechman, J. (2006). Computational and Numerical Ap- Li, J., & Kwauk, M. (2001). Multiscale nature of complex
proaches to Particle Flow. Paper presented at the AIChE fluid-particle systems. Industrial & Engineering Chem-
Annual Meeting, San Francisco, California. istry Research, 40, 4227–4237. doi:10.1021/ie0011021
Lee, L., & Leveque, R. J. (2003). An immersed interface Li, J., & Kwauk, M. (2003). Exploring complex systems
method for incompressible Navier-Stokes equations. in chemical engineering—the multi-scale methodology.
SIAM Journal on Scientific Computing, 25(3), 832–856. Chemical Engineering Science, 58, 521–535. doi:10.1016/
doi:10.1137/S1064827502414060 S0009-2509(02)00577-8
Lee, S. H., Lee, D. H., & Kim, S. D. (2002). Slug char- Li, J., Wen, L., Ge, W., Cui, H., & Ren, J. (1998). Dissipa-
acteristics of polymer particles in a gas-solid fluidized tive structure in concurrent-up gas-solid flow. Chemical
bed. Korean Journal of Chemical Engineering, 19(2), Engineering Science, 53(19), 3367–3379. doi:10.1016/
351–355. doi:10.1007/BF02698428 S0009-2509(98)00130-4
Lee, Y. Y. (1997). Design considerations for CFB boilers. Li, Y., & Kwauk, M. (1980). The dynamics of fast fluidiza-
In Grace, J. R., Avidan, A. A., & Knowlton, T. M. (Eds.), tion. In Grace, J. R., & Matsen, J. M. (Eds.), Fludization
Circulating Fluidized Beds (pp. 417–440). New York: (pp. 537–544). New York: Plenum Press.
Blackie Academic & Professional.
Li, Y. (1994). Hydrodynamics. In Kwauk, M. (Ed.),
Lewis, W. K., Gilliland, E. R., & Bauer, W. C. (1949). Advances in Chemical Engineering (pp. 85–196). San
Characteristics of Fluidized Particles. Industrial and Diego: Academic Press.
Engineering Progress, 41(6), 1104–1117. doi:10.1021/
Li, J., & Wang, L. (2002). Concentration Distributions
ie50474a004
During Mass Transfer in Circulating Fluidized Beds. In
Lewis, C. H. Jr, & Carlson, D. J. (1964). Normal shock 7th International Conference on Circulating Fluidized
location in underexpanded gas and gas-particle jets. AIAA Beds, Niagra Falls, Ontario, Canada.
Journal, 2, 776–777. doi:10.2514/3.2409
Li, J., Chen, A., Yan, Z., Xu, G., & Zhang, X. (1993).
Li, H., Xia, Y., Tung, Y., & Kwauk, M. (1991). Micro-visu- Particle-fluid contacting in circulating fluidized beds.
alization of clusters in a fast fluidized bed. Powder Technol- In A. A. Avidan (Ed.), Preprint Volume for Circulating
ogy, 66, 231–235. doi:10.1016/0032-5910(91)80035-H Fluidized Beds IV (pp. 49-54). Somerset: AIChE.
450
Li, J., Tung, Y., & Kwauk, M. (1988a). Axial voidage Lu, B., Wang, W., & Li, J. (2009). Searching for a
profiles of fast fluidized beds in different operating re- mesh-independent sub-grid model for gas-solids riser
gions. In P. Basu, & J.F. Large (Eds.), 2nd international flows. Chemical Engineering Science, 64, 3437–3447.
conference on circulating fluidized beds(pp. 193). Oxford, doi:10.1016/j.ces.2009.04.024
UK: Pergamon Press.
Lu, B., Wang, W., Li, J., Wang, X., Gao, S., & Lu,
Liboff, R. L. (1990). Kinetic theory: classical, quantum, W. (2007). Multi-scale CFD simulation of gas-solid
and relativistic descriptions. Englewood Cliffs, NJ: flow in MIP reactors with a structure-dependent drag
Prentice Hall. model. Chemical Engineering Science, 62, 5487–5494.
doi:10.1016/j.ces.2006.12.071
Lin, Q., Wei, F., & Jin, Y. (2001). Transient density
signal analysis and two-phase micro-structure flow in Lucas, D., Prasser, H. M., & Manera, A. (2005). Influence
gas-solids fluidization. Chemical Engineering Science, of the lift force on the stability of a bubble column. Chemi-
56(6), 2179–2189. doi:10.1016/S0009-2509(00)00499-1 cal Engineering Science, 60, 3609–3619. doi:10.1016/j.
ces.2005.02.032
Lints, M. C., & Glicksman, L. R. (1993). Parameters
governing particle-to-wall heat transfer in a circulating Lucy, L. B. (1977). A numerical approach to the testing
fluidized bed. In A.A. Avidan (Ed.), Proceeding of the of the fission hypothesis. The Astronomical Journal, 83,
International Conference on Circulating Fluidized Beds 1013–1024. doi:10.1086/112164
(pp. 297-203), Somerset, PA.
Luding, S., Strauss, O., & McNamara, S. (2000). Segrega-
Liu, X., Metzger, M., & Glasser, B. J. (2007). Couette tion of polydisperse granular media in the presence of a
flow with a bidisperse particle mixture. Physics of Fluids, temperature gradient. In A. D. Rosato & D. L. Blackmore
19, 073301. doi:10.1063/1.2741245 (Eds.), IUTAM Symposium on Segregation in Granular
Flows. Boston: Kluwer Academic Publishers.
Liu, X. (2005). Aggregation and lateral transfer behav-
iour of particles in gas-solid fluidized beds. Ph.D thesis. Lun, C. K. (1991). Kinetic Theory for Granular Flow
Institute of Process Engineering, Chinese Academy of of Dense, Slightly Inelastic, Slightly Rough Spheres.
Science, Beijing, P. R. China. Journal of Fluid Mechanics, 233, 539. doi:10.1017/
S0022112091000599
Lomas, D. (Dec. 17, 1996). FCC Separation Method and
Apparatus with Improved Stripping. US Patent 5,584,985. Lun, C., & Savage, S. (1986). The Effects of an Impact
Velocity Dependent Coefficient of Restitution on Stresses
López de Haro, M., Cohen, E. G. D., & Kincaid, J. M.
Developed by Sheared Granular Materials. Acta Me-
(1983). The Enskog Theory for Multicomponent Mixtures
chanica, 63, 15–44. doi:10.1007/BF01182538
I. Linear Transport Theory. Journal of Chemical Phys-
ics, 78, 2746. Koopman, B. (1969). Relaxed Motion in Lun, C., Savage, S., Jeffrey, D., & Chepurniy, N. (1984).
Irreversible Molecular Statistics. Stochastic Processes in Kinetic theories for granular flow: inelastic particles in
Chemical Physics, 15, 37. Couette flow and slightly inelastic particles in a general
flowfield. Journal of Fluid Mechanics, 140, 223–256.
Loth, E. (2008). Drag of non-spherical solid particles of
doi:10.1017/S0022112084000586
regular and irregular shape. Powder Technology, 182(3),
342. doi:10.1016/j.powtec.2007.06.001 Lutsko, J. (1997). Approximate Solution of the Enskog
Equation Far from Equilibrium. Physical Review Letters,
Louge, M. Y., Mastorakos, E., & Jenkins, J. T. (1991).
78, 243. doi:10.1103/PhysRevLett.78.243
The role of particle collisions in pneumatic transport.
S0022112091003427
451
Lutsko, J. (2001). Model for the Atomic Scale Structure Ma, J., Ge, W., Xiong, Q., Wang, J., & Li, J. (2009).
of the Homogeneous Cooling State of Granular Fluids. Direct numerical simulation of particle clustering in
Physical Review E: Statistical, Nonlinear, and Soft Matter gas-solid flow with macro-scale particle method. Chemi-
Physics, 63, 061211. doi:10.1103/PhysRevE.63.061211 cal Engineering Science, 64(1), 43–51. doi:10.1016/j.
ces.2008.09.005
Lutsko, J. (2002). Atomic-Scale Structure of Hard-Core
Fluids Under Shear Flow. Physical Review E: Statisti- Mabrouk, R., Chaouki, J., & Guy, C. (2007). Effective
cal, Nonlinear, and Soft Matter Physics, 66, 051109. drag coefficient investigation in the acceleration zone of
doi:10.1103/PhysRevE.66.051109 an upward gas-solid flow. Chemical Engineering Science,
62, 318–327. doi:10.1016/j.ces.2006.08.055
Lutsko, J. (2004). Kinetic Theory and Hydrodynamics of
Dense, Reacting Fluids Far From Equilibrium. The Journal Mahdi, F. A., Dudley, A. S., & Sundaresan, S. (2002).
of Chemical Physics, 120, 6325. doi:10.1063/1.1648012 The effect of static electrification on gas-solid flows
in vertical risers. Industrial & Engineering Chemistry
Lutsko, J. (2004). Rheology of Dense Polydisperse
Research, 41(25), 6224–6234. doi:10.1021/ie010982w
Granular Fluids Under Shear. Physical Review E: Sta-
tistical, Nonlinear, and Soft Matter Physics, 70, 061101. Manyele, S. V., Pärssinen, J. H., & Zhu, J.-X. (2002).
doi:10.1103/PhysRevE.70.061101 Characterizing particle aggregates in a high-density and
high-flux CFB riser. Chemical Engineering Journal, 88,
Lutsko, J. (2005). Transport Properties for Dense Dissi-
151–161. doi:10.1016/S1385-8947(01)00299-6
pative Hard Spheres for Arbitrary Energy Loss Models.
Physical Review E: Statistical, Nonlinear, and Soft Matter Marchisio, D. L., Pikturna, J. T., Fox, R. O., Vigil, R. D.,
Physics, 72, 021306. doi:10.1103/PhysRevE.72.021306 & Barresi, A. (2003a). Quadrature method of moments
for population-balance equations. AIChE Journal. Ameri-
Lutsko, J. (2006). Chapman-Enskog expansion about
nonequilibrium states with application to the sheared
doi:10.1002/aic.690490517
granular fluid. Physical Review E: Statistical, Nonlin-
ear, and Soft Matter Physics, 73, 021302. doi:10.1103/ Martin, T. W., Huntley, J. M., & Wildman, R. D. (2006).
PhysRevE.73.021302 Hydrodynamic model for a vibrofluidized granular bed.
Ma, D., & Ahmadi, G. (1988). A kinetic model for rapid
S0022112005004866
granular flows of nearly elastic particles including inter-
stitial fluid effects. Powder Technology, 56, 191–207. Martin, T. W. (2005). Capturing gas and particle motion
doi:10.1016/0032-5910(88)80030-5 in an idealised gas-granular flow. Powder Technology,
155, 175–180. doi:10.1016/j.powtec.2005.05.043
Ma, J., & Ge, W. (2008). Is standard symmetric formulation
always better for smoothed particle hydrodynamics? Com- Martin, M. P., Turlier, P., Wild, G., & Bernard, J. R.
puters & Mathematics with Applications (Oxford, Eng- (1992). Gas and solid behavior in cracking circulating
land), 55, 1503–1513. doi:10.1016/j.camwa.2007.08.010 fluidized beds. Powder Technology, 70(3), 249–258.
doi:10.1016/0032-5910(92)80060-A
Ma, J., Ge, W., Wang, X., Wang, J., & Li, J. (2006).
High-resolution simulation of gas-solid suspension us- Marzocchella, A., Zijerveld, R. C., & Schouten, J. C.
ing macro-scale particle methods. Chemical Engineering (1997). Chaotic behavior of gas-solids flow in the riser
Science, 61, 7096–7106. doi:10.1016/j.ces.2006.07.042 of a laboratory-scale circulating fluidized bed. AIChE
Journal. American Institute of Chemical Engineers, 43,
1458. doi:10.1002/aic.690430609
452
Mathiesen, V., Solberg, T., & Hjertager, B. H. (2000). Mehrani, P., Bi, H. T., & Grace, J. R. (2007). Electrical
An experimental and computational study of multiphase behaviour of different fines added to beds of glass beads
flow behavior in a circulating fluidized bed. International and polyethylene particles. Journal of Electrostatics,
Journal of Multiphase Flow, 26, 387–419. doi:10.1016/ 65(1), 1–10. doi:10.1016/j.elstat.2006.05.002
S0301-9322(99)00027-0
Meier, H. F., & Mori, M. (1998). Gas-solid flow in cy-
Matsen, J. M. (1982). Mechanisms of choking and entrain- clones: The Eulerian-Eulerian approach. Computers &
ment. Powder Technology, 32, 21–33. doi:10.1016/0032- Chemical Engineering, 22(1), S641–S644. doi:10.1016/
5910(82)85003-1 S0098-1354(98)00114-8
McGraw, R. (1997). Description of aerosol dy- Meynard, F. (1997). An analysis of the energy minimi-
namics by the quadrature method of moments. zation multi-scalemodel for circulating fluidized beds.
Aerosol Science and Technology, 27(2), 255–265. Unpublished.
doi:10.1080/02786829708965471
Mickley, H. S., & Fairbanks, D. F. (1955). Mechanism
Mckeen, T., & Pugsley, T. (2003). Simulation and experi- of heat transfer to fluidized beds. A.I. Ch.E. Journal,
mental validation of a freely bubbling bed of FCC catalyst. 1(3), 374–384.
Powder Technology, 129(1-3), 139–152. doi:10.1016/
Minier, J.-P., & Peirano, E. (2001). The PDF approach to
S0032-5910(02)00294-2
turbulent polydispersed two-phase flows. Physics Reports,
McKenna, T. F., & Soares, J. B. P. (2001). Single particle 352, 1. doi:10.1016/S0370-1573(01)00011-4
modeling for olefin polymerization on supported catalysts:
Mittal, R., & Iaccarino, G. (2005). Immersed boundary
A review and proposals for future developments. Chemical
methods. Annual Review of Fluid Mechanics, 37, 239–261.
Engineering Science, 56(13), 3931–3941. doi:10.1016/
S0009-2509(01)00069-0
Mohd-Yusof, J. (1996). Interaction of massive particles
McKenna, T. F., Spitz, R., & Cokljat, D. (1999). Heat
with turbulence. Cornell University.
transfer from catalysts with computational fluid dynamics.
AIChE Journal. American Institute of Chemical Engineers, Molerus, O. (1982). Interpretation of Geldart’s type A,
45, 2392–2410. doi:10.1002/aic.690451113 B, C and D powders by taking into account interpar-
ticle cohesion forces. Powder Technology, 33, 81–87.
McLennan, J. A. (1989). Introduction to Nonequilib-
doi:10.1016/0032-5910(82)85041-9
rium Statistical Mechanics. Upper Saddle River, NJ:
Prentice-Hall. Monaghan, J. J. (1992). Smoothed particle hydrodynam-
ics. Annual Review of Astronomy and Astrophysics, 30,
McNamara, S., & Young, W. (1996). Dynamics of a Freely
543–574. doi:10.1146/annurev.aa.30.090192.002551
Evolving, Two-dimensional Granular Medium. Physi-
cal Review E: Statistical Physics, Plasmas, Fluids, and Monazam, E. R., Shadle, L. J., & Lawson, L. O. (2001).
Related Interdisciplinary Topics, 53, 5089. doi:10.1103/ A transient method for determination of saturation carry-
PhysRevE.53.5089 ing capacity. Powder Technology, 121, 205. doi:10.1016/
S0032-5910(01)00354-0
Mehrani, P., Bi, H. T., & Grace, J. R. (2005). Electro-
static charge generation in gas-solids fluidized beds. Monazam, E. R., & Shadle, L. J. (2004). A transient method
Journal of Electrostatics, 63, 165–173. doi:10.1016/j. for characterizing flow regimes in a circulating fluid
elstat.2004.10.003 bed. Powder Technology, 139(1), 89–97. doi:10.1016/j.
powtec.2003.10.007
453
Montanero, J., Garzó, V., Alam, M., & Luding, S. (2006). Mountain, J. R., Mazumder, M. K., Sims, R. A., Wankum,
Rheology of Two and Three Dimensional Granular Mix- D. L., Chasser, T., & Pettit, P. H. Jr. (2001). Triboelectric
tures Under Uniform Shear Flow: Enskog Kinetic Theory charging of polymer powders in fluidization and transport
Versus Molecular Dynamics Simulation. Granular Matter, processes. IEEE Transactions on Industry Applications,
8, 103. doi:10.1007/s10035-006-0001-7 37(3), 778–784. doi:10.1109/28.924759
Montanero, J., & Santos, A. (1996). Monte Carlo Simula- Mudde, R. F., & Simonin, O. (1999). Two- and three-
tion Method for the Enskog Equation. Physical Review E: dimensional simulations of a bubble plume using a
Statistical Physics, Plasmas, Fluids, and Related Interdis- two-fluid model. Chemical Engineering Science, 54(21),
ciplinary Topics, 54, 438. doi:10.1103/PhysRevE.54.438 5061–5069. doi:10.1016/S0009-2509(99)00234-1
Montanero, J. M., & Garzó, V. (2002). Monte Carlo Mueller, P., & Reh, L. (1993). Particle drag and pressure
simulation of the homogeneous cooling state for a granu- drop in accelerated gas-solid flow. In A. A. Avidan (Ed.),
lar mixture. Granular Matter, 4, 17–24. doi:10.1007/ Preprint Volume for Circulating Fluidized Beds IV (pp.
s10035-001-0097-8 193-198). Somerset: AIChE.
Montanero, J. M., Santos, A., & Garzó, V. (2007). First- Muzzio, F. J., Shinbrot, T., & Glasser, B. J. (2002).
order Chapman-Enskog velocity distribution function Powder technology in the pharmaceutical industry: the
in a granular gas. Physica A, 376, 75–93. doi:10.1016/j. need to catch up fast. Powder Technology, 124(1-2), 1–7.
physa.2006.10.080 doi:10.1016/S0032-5910(01)00482-X
Montanero, J., Santos, A., & Garzó, V. (2005). DSMC Nakamure, K., & Capes, C. E. (1973). Vertical pneumatic
Evaluation of the Navier-Stokes Shear Viscosity of a conveying: a theoretical study of uniform and annular
Granular Fluid. In Capitelli, M. (Ed.), Rarefied Gas particle flow models. Canadian Journal of Chemical
Dynamics 24, AIP Conference Proceedings 762, 797. Engineering, 51, 39–46. doi:10.1002/cjce.5450510107
Moran, J. C., & Glicksman, L. R. (2003). Experimental Neri, A., & Gidaspow, D. (2000). Riser hydrodynamics:
and numerical studies on the gas flow surrounding a single simulation using kinetic theory. AIChE Journal. American
cluster applied to a circulating fluidized bed. Chemical Institute of Chemical Engineers, 46, 52. doi:10.1002/
Engineering Science, 58(9), 1879–1886. doi:10.1016/ aic.690460108
S0009-2509(02)00684-X
Neri, A., Ongaro, T. E., Macedonio, G., & Gidaspow, D.
Morris, J. P., Fox, P. J., & Zhu, Y. (1997). Modeling low (2003). Multiparticle simulation of collapsing volcanic
Reynolds number incompressible flows using SPH. Jour- pyroclastic flow. Journal of Geophysical Research, 108,
nal of Computational Physics, 136, 214–226. doi:10.1006/ 2202..doi:10.1029/2001JB000508
jcph.1997.5776
Noskowicz, S. H., Bar-Lev, O., Serero, D., & Goldhirsch,
Morton, B. R., Taylor, G. F. R. S., & Turner, J. S. (1956). I. (2007). Computer-aided kinetic theory and granular
Turbulent gravitational convection from maintained and gases. Europhysics Letters, 79, 60001. doi:10.1209/0295-
instantaneous sources. Proceedings of the Royal Society of 5075/79/60001
London. Series A, 234, 1–23. doi:10.1098/rspa.1956.0011
Noymer, P. D., & Glicksman, L. R. (1998). Cluster and
Mostoufi, N., & Chaouki, J. (1999). Prediction of ef- particle-convective heat transfer at the wall of a circu-
fective drag coefficient in fluidized beds. Chemical lating fluidized bed. International Journal of Heat and
Engineering Science, 54, 851–858. doi:10.1016/S0009- Mass Transfer, 41(1), 147–158. doi:10.1016/S0017-
2509(98)00290-5 9310(97)00065-3
454
Noymer, P. D., & Glicksman, L. R. (1999). Near-wall O’Rourke, P. J. (1981). Collective drop effects on vapor-
hydrodynamics in a scale-model circulating fluidized izing liquid sprays. Ph.D. Thesis, Princeton University,
bed. International Journal of Heat and Mass Transfer, Princeton, NJ.
42(8), 1389–1403. doi:10.1016/S0017-9310(98)00238-5
Osher, S., & Fedkiw, R. P. (2001). Level set methods: an
Noymer, P. D., & Glicksman, L. R. (2000). Descent overview and some recent results. Journal of Computa-
velocities of particle clusters at the wall of a circulating tional Physics, 169, 463. doi:10.1006/jcph.2000.6636
fluidized bed. Chemical Engineering Science, 55(22),
Ottino, J. M., & Khakhar, D. V. (2000). Mixing and seg-
5283–5289. doi:10.1016/S0009-2509(00)00171-8
regation of granular materials. Annual Review of Fluid
O’Rourke, P., Brackbill, J., & Larrouturou, B. (1993). On Mechanics, 32, 55–91. doi:10.1146/annurev.fluid.32.1.55
Particle-Grid Interpolation and Calculating Chemistry
Pai, M. G., & Subramaniam, S. (2009). A comprehensive
in Particle-in-Cell Methods. Journal of Computational
probability density function formalism for multiphase
Physics, 109(1), 37–52. doi:10.1006/jcph.1993.1197
flows. Journal of Fluid Mechanics.
Oberhuber, J. M., Herzog, M., Graf, H.-F., & Schwanke,
Paisley, M. A., Farris, G., Slack, W., & Irving, J. (1997).
K. (1998). Volcanic plume simulation on large scales.
Commercial Development of Battelle/FERCO Biomass
Journal of Volcanology and Geothermal Research, 87,
Gasification Process: Initial Operation of the McNeil
29–53. doi:10.1016/S0377-0273(98)00099-7
Gasifier. In R.P. Overend, E. Chornet (Eds.), Making a
O’Brien, T. J., & Syamlal, M. (1993). Particle Cluster Business from Biomass in Energy, Environment, Chemi-
Effects in the Numerical Simulation of a Circulating Flu- cals, Fibers and Materials: Proceedings of the Third
idized Bed. In Avidan, A. A. (Ed.), Circulating Fluidized Biomass Conference in Americas (pp. 579-592) Oxford:
Beds IV (pp. 345–350). Somerset, PA. Elsevier Science, Oxford.
Oey, R. S., Mudde, R. F., & Van den Akker, H. E. A. (2003). Pan, T. W., Joseph, D. D., Bai, R., Glowinski, R., & Sarin,
Numerical simulations of an oscillating internal-loop air- V. (2002). Fluidization of 1204 spheres: simulation and
lift reactor. Canadian Journal of Chemical Engineering, experiment. Journal of Fluid Mechanics, 451, 169–191.
81(3-4), 684–691. doi:10.1002/cjce.5450810347 doi:10.1017/S0022112001006474
Ogawa, S., Umemura, A., & Oshima, N., J. (1980). On the Paolotti, D., Cattuto, C., Marconi, U. M. B., & Puglisi, A.
equations of fully fluidized granular materials. [ZAMP]. (2003). Dynamical properties of vibrofluidized granular
Applied Math. and Physics, 31(4), 483–493. doi:10.1007/ mixtures. Granular Matter, 5(2), 75–83. doi:10.1007/
BF01590859 s10035-003-0133-y
Oliveira, P. J., & Issa, R. I. (1994). On the numerical treat- Passalacqua, A., Galvin, J. E., Vedula, P., Hrenya, C. M.,
ment of interphase forces in two-phase flow. Numerical & Fox, R. O. (2010). A quadrature-based kinetic model
Methods in Multiphase Flows, 185, 131–140. for dilute non-isothermal granular flows. Communication
in Computational Physics.
O’Rourke, P. J., Brackbill, J. U., & Larrouturou, B. (1993).
On particle grid interpolation and calculating chemistry Patankar, S. (1980). Numerical Heat Transfer and Fluid
in particle-in-cell methods. Journal of Computational Flow. New York: Hemisphere Publising Corporation.
Physics, 109, 37–52. doi:10.1006/jcph.1993.1197
Pell, M., & Jordan, S. P. (1988). Effects of fines and
O’Rourke, P. J., Zhao, P., & Snider, D. (2009). A Model velocity on fluid bed reactor performance. AIChE Symp.
for Collisional Exchange in Gas/Liquid/Solid Fluidized Ser., 84(262), 68-73.
Beds. Chemical Engineering Science, 64, 1784–1797.
doi:10.1016/j.ces.2008.12.014
455
Perot, B. (2000). Conservation properties of unstructured Pritchett, J., Blake, T., & Garg, S. (1978). A Numberical
staggered mesh schemes. Journal of Computational Phys- Model of Gas Fluidized Beds. Fluidization: Application
ics, 159, 58–89. doi:10.1006/jcph.2000.6424 to Coal Conversion Processes: AIChE Symp. Ser. 74
(176) 134-148.
Perry, R. H., & Green, D. W. (2007). Perry’s Chemical
Engineers’ Handbook. New York: Mcgraw-Hill. Pugsley, T. S., Lapointe, D., Hirschberg, B., & Werther,
J. (1997). Exit effects in circulating fluidized bed ris-
Perthame, B. (1992). Second-order Boltzmann schemes
ers. Canadian Journal of Chemical Engineering, 75(6),
for compressible Euler equations in one and two space
1001–1010. doi:10.1002/cjce.5450750602
dimensions. SIAM Journal on Numerical Analysis, 29(1),
1–19. doi:10.1137/0729001 Pugsley, T. S., & Malcus, S. (1997). Partial Oxidation of
Methane in a Circulating Fluidized-Bed Catalytic Reactor.
Peskin, C. S. (1982). The Fluid-Dynamics of Heart-
Industrial & Engineering Chemistry Research, 36(11),
Valves - Experimental, Theoretical, and Computational
4567–4571. doi:10.1021/ie970088y
Methods. Annual Review of Fluid Mechanics, 14, 235–259.
doi:10.1146/annurev.fl.14.010182.001315 Qi, X., Tao, Z., & Huang, W. (2005). Experimental study
of solids holdups inside particle clusters in CFB risers.
Peyret, R., & Taylor, T. (1985). Computational Method
Journal of Sichuan University, 37(5), 46.
for Fluid Flows. Berlin: Springer.
Rahaman, M. F., Naser, J., & Witt, P. J. (2003). An unequal
Pigford, R.L. & Baron, T. (1965). Hydrodynamic stability
temperature kinetic theory: description of granular flow
of a fluidized bed. I&EC Fundamentals, 4, 81-87.
with multiple particle classes. Powder Technology, 138,
Pita, J., & Sundaresan, S. (1991). Gas-solid flow in verti- 82–92. doi:10.1016/j.powtec.2003.08.050
cal tubes. AIChE Journal. American Institute of Chemical
Rai, M. M., Gatski, T. B., & Erlebacher, G. (1995). Direct
simulation of spatially evolving compressible turbulent
Pita, J., & Sundaresan, S. (1993). Developing flow of boundary layers. Paper presented at the AIAA 950583,
a gas-particle mixture in a vertical riser. AIChE Jour- 33rd Aerospace Sciences Meeting and Exhibit, Reno, NV.
nal. American Institute of Chemical Engineers, 39(4),
Rall, R. (May 1, 2001). Apparatus for Contacting of
541–552. doi:10.1002/aic.690390402
Gases and Solids in Fluidized Beds. US Patent 6,224,833.
Plumpe, J. G., Zhu, C., & Soo, S. L. (1993). Measurement
Ramanathan, S., & Koch, D. L. (2009). An efficient
of fluctuations in motion of particles in a dense gas—solid
direct simulation Monte Carlo method for low Mach
suspension in vertical pipe flow. Powder Technology,
number noncontinuum gas flows based on the Bhatnagar-
77(2), 209–214. doi:10.1016/0032-5910(93)80057-H
Gross-Krook model. Physics of Fluids, 21, 033103.
Pomraning, E., & Rutland, C. J. (2001). A Dynamic doi:10.1063/1.3081562
One-Equation Non-Viscosity LES Model. AIAA Journal,
Ramkrishna, D. (2000). Population balances: theory and
25502.
applications to particulate systems in engineering. San
Pope, S. B. (2000). Turbulent Flows. Cambridge, UK: Diego, CA: Academic Press.
Cambridge University Press.
Rashidi, M., Hetsroni, G., & Banerjee, S. (1999). Particle-
Pöschel, T., & Luding, S. (Eds.). (2001). Granular Gases. turbulence interaction in a boundary layer. International
New York: Springer. doi:10.1007/3-540-44506-4 Journal of Heat and Mass Transfer, 42, 935–949.
Pöschel, T., & Luding, S. (Eds.). (2003). Granular Gas

Dynamics. New York: Springer.
456
Rauwoens, P., Vieerndeels, J., & Merci, B. (2007). A Rhodes, M. J., Mineo, H., & Hirama, T. (1992). Par-
solution for the odd-even decoupling problem in pressure- ticle motion at the wall of a circulating fluidized bed.
correction algorithms for variable density flows. Journal Powder Technology, 70(3), 207–214. doi:10.1016/0032-
of Computational Physics, 227, 79–99. doi:10.1016/j. 5910(92)80055-2
jcp.2007.07.010
Rhodes, M. J., Sollaart, M., & Wang, X. S. (1998). Struc-
Ray, W. H., & Villa, C. M. (2000). Nonlinear dynamics ture of the Dense-Dilute Interface in Fast Fluidization.
found in polymerization processes – a review. Chemical In Fan, L.-S., & Knowlton, T. M. (Eds.), Fluidization
Engineering Science, 55, 275–290. doi:10.1016/S0009- IX (pp. 141–148). New York: Engineering Foundation.
2509(99)00323-1
Rice, R. B. (2009). (Manuscript submitted for publica-
Reed, T. M., & Gubbins, K. E. (1973). Applied Statistical tion). Clustering in rapid granular flows of binary and
Mechanics. New York: McGraw-Hill. continuous particle size distributions. Physical Review
E: Statistical, Nonlinear, and Soft Matter Physics.
Reh, L. (1971). Fluid bed combustion in Processing,
Environmental Protection and Energy Supply. Chemical Richardson, J. F., & Zaki, W. N. (1954). Sedimentation
Engineering Progress, 67(2), 63–67. and fluidisation: Part I. Transactions of the Institution of
Chemical Engineers, 32, 35–53.
Reh, L., & Li, J. (1991). Measurement of voidage in
fluidized beds by optical probe. In Basu, P., Horio, M., & Risk, M. A. (1993). Mathematical modeling of densely
Hasatani, M. (Eds.), Circulating Fluidized Beds Technol- loaded, particle laden turbulent flows. Atomization and
ogy III (p. 105). Oxford, UK: Pergamon Press. Sprays, 3, 1–27.
Rericha, E. C., Bizon, C., Shattuck, M. D., & Swinney, H. Rivault, P., Nguyen, C., Laguerie, C., Bernard, J. R., &
L. (2002). Shocks in supersonic sand. Physical Review Let- Aquitaine, E. (1995). Countercurrent stripping dense
ters, 88(1), 014302. doi:10.1103/PhysRevLett.88.014302 circulating beds effect of the baffles. In Large, J. F., &
Laguerie, C. (Eds.), Fluidization VIII (pp. 491–499). New
Résibois, P., & De Leener, M. (1977). Classical Kinetic
Theory of Fluids. New York: Wiley Interscience.
Roache, P. J. (1998). Verification and Validation in Compu-
Resnick, W., & White, R. R. (1949). Mass Transfer in
tational Science and Engineering (p. 446). Albuquerque,
Systems of Gas and Fluidized Solids. Chemical Engineer-
NM: Hermosa Publishers.
ing Progress, 45(6), 377–389.
Rosato, A., Prinze, F., Standburg, K. J., & Swendsen, R.
Revel, J., Gatumel, C., Dodds, J. A., & Taillet, J. (2003).
(1987). Why Brazil nuts are on top: size segregation of
Generation of static electricity during fluidization of
particulate matter by shaking. Physical Review Letters,
polyethylene and its elimination by air ionization.
58, 1038–1040. doi:10.1103/PhysRevLett.58.1038
Powder Technology, 135–136, 192–200. doi:10.1016/j.
powtec.2003.08.015 Samuelsberg, A., & Hjertager, B. J. H. (1996). Compu-
tational modeling of gas/particle flow in a riser. AIChE
Rhie, C., & Chow, W. (1983). Numerical study of the
turbulent flow past an airfoil with trailing edge separation.
1536–1546. doi:10.1002/aic.690420605
AIAA Journal, 1(21), 1527–1532.
Santos, A., Garzó, V., & Dufty, J. (2004). Inherent Rheol-
Rhodes, M. J. (1989). The upward flow of gas/solids
ogy of a Granular Fluid in Uniform Shear Flow. Physical
suspensions: Part 2: A practical quantitative flow regime
diagram for the upward flow of gas/solid suspensions.
69, 061303. doi:10.1103/PhysRevE.69.061303
Chemical Engineering Research & Design, 67, 30–37.
457
Santos, A., Yuste, S., & López de Haro, M. (2002). Con- Senior, R. C., & Grace, J. R. (1998). Formation of particle
tact values of the radial distribution functions of additive streamers in the wall region of circulating fluidized bed
hard-sphere mixtures in d dimensions: A new proposal. risers. In Fan, L.-S., & Knowlton, T. M. (Eds.), Fluidization
The Journal of Chemical Physics, 117(12), 5785–5793. IX (pp. 165–172). New York: Engineering Foundation.
doi:10.1063/1.1502247
Senior, R. C., Smalley, C. G., & Gbordzoe, E. (1998).
Santos, A., & Montanero, J. (2008). The Second and Third Hardware modifications to overcome common operat-
Sonine Coefficients of a Freely Cooling Granular Gas ing problems in FCC catalyst strippers. In Fan, L. S., &
Revisited. arXiv:0812.3022. Knowlton, T. M. (Eds.), Fluidization IX (pp. 725–732).
New York: Engineering Foundation.
Savage, S. B., & Sayed, M. (1984). Stresses developed by
dry cohesionless granular materials sheared in an annular Serero, D., Goldhirsch, I., Noskowicz, S. H., & Tan, M.-L.
shear cell. Journal of Fluid Mechanics, 142, 391–430. (2006). Hydrodynamics of granular gases and granular
doi:10.1017/S0022112084001166 gas mixtures. Journal of Fluid Mechanics, 554, 237–258.
doi:10.1017/S0022112006009281
Schiller, L., & Naumann, A. (1933). Fundamental calcula-
tions in gravitational processing. Zeitschrift Des Vereines Settle, M. (1978). Volcanic eruption clouds and the
Deutscher Ingenieure, 77, 318–320. thermal power output of explosive eruptions. Journal
of Volcanology and Geothermal Research, 3, 309–324.
Schiller, L., & Naumann, A. (1935). A drag coefficient
doi:10.1016/0377-0273(78)90041-0
correlation. V.D.I. Zeitung, 77, 318–320.
Seville, J. P. K., Simons, S. R. J., Broadbent, C. J., Martin,
Schiller, L., & Naumann, A. Z. (1933). A Drag Coefficient
T. W., Parker, D. J., & Beynon, T. D. (1995). Particle
Correlation. Z. Ver. Deutsch Ing., 77, 318–320.
velocities in gas-fluidized beds. In Fluidization VIII (p.
Schmeal, W. R., & Street, J. R. (1971). Polymerization 319). New York: Engineering Foundation.
in expanding catalyst particles. AIChE Journal. Ameri-
Seville, J. P. K., Simons, S. R. J., Broadbent, C. J., Parker,
D. J., & Beynon, T. D. (1994). Particle velocities in gas
doi:10.1002/aic.690170526
fluidized beds. In 1st International Particle Technical
Schneider, G., & Raw, M. (1987). Control volume Forum of AIChE (p. 493). Denver.
finitie-element method for heat transfer and fluid flow
Seville, J. P. K., Tuzun, U., & Clift, R. (1997). Process-
using collocated variables - 1. Computational proce-
ing of Particulate Solids. New York: Blackie Academic.
dure. Numerical Heat Transfer Part A, 11, 363–390.
doi:10.1080/10407788708913560 Shaffer, F. (2008). Personal communication. National
Energy Technology Laboratory, Pittsburgh, PA.
Schröter, M., Ulrich, S., Kreft, J., Swift, J. B., & Swinney,
H. L. (2006). Mechanisms in the size segregation of a Sharma, A. K., Tuzla, K., & Matsen, J. (2000). Parametric
binary granular mixture. Physical Review E: Statistical, effects of particle size and gas velocity on cluster charac-
Nonlinear, and Soft Matter Physics, 74(011307). teristics in fast fluidized beds. Powder Technology, 111,
114. doi:10.1016/S0032-5910(00)00247-3
Segrè, P. N., Liu, F., & Umbanhowar, P. (2001). An effec-
tive gravitational temperature for sedimentation. Nature, Sharma, A. K., Matsen, J., Tuzla, K., et al. (2001). A cor-
409, 594. doi:10.1038/35054518 relation for solid fraction in clusters in fast fluidized beds.
In M. Kwauk, J. Li, & W.C. Yang (Eds.), Proceedings of
Sela, N., & Goldhirsch, I. (1998). Hydrodynamic equa-
the 10th Engineering Foundation Conference on Fluidi-
tions for rapid flows of smooth inelastic spheres, to
zation, (pp. 301). New York: Engineering Foundation.
Burnett order. Journal of Fluid Mechanics, 361, 41–74.
doi:10.1017/S0022112098008660
458
Shi, D., Nicolai, R., & Reh, L. (1998). Wall-to-bed heat Song, H. (2004). Bounds on operating conditions leading
transfer in circulating fluidized beds. Chemical Engi- to melting during olefin polymerization. Doctoral dissera-
neering and Processing, 37, 287. doi:10.1016/S0255- tion. Houston, TX: University of Houston.
2701(98)00039-7
Song, H., & Luss, D. (2004). Bounds on operating
Simonin, O. (1991). Prediction of the dispersed phase conditions leading to melting during olefin polymeriza-
turbulence in particle-laden jets. In 4th International tion. Industrial & Engineering Chemistry Research, 43,
Symposium on Gas-Solid Flows (pp. 197 – 206, Vol. 270–282. doi:10.1021/ie020981j
121), ASME-FED.
Soong, C. H., Tuzla, K., & Chen, J. C. (1995). Experimental
Simonin, O. (1996). Continuum modeling of dispersed determination of cluster size and velocity in circulating
two-phase flows. Combustion and turbulence in two- fluidized bed. In Large, J. F., & Laguerie, C. (Eds.), Flu-
phase flows. Von Karman Institute of Fluid Dynamics, idization (p. 219). New York: Engineering Foundation.
VKI Lecture Series, 1-47, Rhodes-St-Geneses, Belgium.
Soong, C. H., Tuzla, K., & Chen, J. C. (1994). Identifi-
Sinclair, J., & Jackson, R. (1989). Gas-particle flow in cation of particle clusters in circulating fluidized bed. In
a vertical pipe with particle-particle interactions. AIChE AA Avidan (Ed.), Proceeding of the Fourth International
Journal. American Institute of Chemical Engineers, 35(9), Conference on Circulating fluidized beds, (pp. 615). New
1473–1486. doi:10.1002/aic.690350908 York: Pergamon Press.
Sjogren, L. (1980). Kinetic Theory of Current Fluctua- Sorensen, J. P., & Stewart, W. E. (1974). Computation
tions in Simple Classical Fluids. Physical Review A 22, of forced-convection in slow flow through ducts and
2866. ibid 2883. packed-beds. 2. velocity profile in a simple cubic array of
spheres. Chemical Engineering Science, 29(3), 819–825.
Smagorinsky, J. (1963). General circulation experiments
doi:10.1016/0009-2509(74)80200-9
with the primitive equations, part I: the basic experiment.
Monthly Weather Review, 91, 99–164. doi:10.1175/1520- Spalding, D. B. (1980). Numerical computation of multi-
0493(1963)091<0099:GCEWTP>2.3.CO;2 phase fluid flow and heat transfer. In Taylor, C., & Morgan,
K. (Eds.), Recent advances in numerical methods in fluids
Snider, D. M. (2001). An incompressible three-dimen-
(pp. 139–168). Swansea, UK: Pineridge Press Limited.
sional multiphase particle-in-cell model for dense par-
ticle flows. Journal of Computational Physics, 170(2), Sparks, R. S. J. (1986). The dimension and dynamics of
523–549. doi:10.1006/jcph.2001.6747 volcanic eruption columns. Bulletin of Volcanology, 48,
3–15. doi:10.1007/BF01073509
Snider, D. M., O’Rourke, P. J., & Andrews, M. J. (1998).
Sediment flow in inclined vessels calculated using a Sparks, R. S. J., Bursik, M. I., Carey, S. N., Gilbert, J.
multiphase particle-in-cell model for dense particle S., Glaze, L. Z., Sigurdson, H., & Woods, A. W. (1997).
flows. International Journal of Multiphase Flow, 24(8), Volcanic Plumes (p. 574). Hoboken, NJ: John Wiley.
1359–1382. doi:10.1016/S0301-9322(98)00030-5
Squires, K. D., & Eaton, J. K. (1991). Preferential con-
Snider, D. M. (2008). Ozone Decomposition. Paper centration of particle by turbulence. Physics of Fluids,
presented at AIChE Annual Meeting, Philadelphia, 3, 1169. doi:10.1063/1.858045
Pennsylvania.
Stewart, P. S. B., & Davidson, J. F. (1967). Slug Flow
Snyder, L. J., & Stewart, W. E. (1966). Velocity and in Fluidized Beds. Powder Technology, 1, 61–66.
pressure profiles for newtonian creeping flow in regular doi:10.1016/0032-5910(67)80014-7
packed beds of spheres. A.I. Ch.E.J., 12(1), 167.
459
Stromberg, L. (1983). Operational modes for fluidized Syamlal, M., & O’Brien, T. (1989). Computer simulation
beds. Studsvik Report E4-79/83. of bubbles in a fluidized bed. AIChE Symp. Ser. 85(27)
22-31.
Struchtrup, H. (2005). Macroscopic Transport Equations
for Rarified Gas Flows. Berlin: Springer-Verlag. Syamlal, M., Rogers, W., & O’Brien, T. J. (1993), MFIX
documentation: theory guide. (Tech. Rep. DOE/METC-
Strumendo, M., & Canu, P. (2002). Method of moments
94/1004, DE9400087). Morgantown, WV: US Department
for the dilute granular flow of inelastic spheres - art. no.
of Energy, Office of Fossil Energy, Morgantown Energy
041304. Physical Review E: Statistical, Nonlinear, and
Technology Center.
Soft Matter Physics, 6604, 1304.
Takeda, H., Miyama, S. M., & Sekiya, M. (1994). Nu-
Strumendo, M., & Arastoopour, H. (2008). Solution of PBE
merical simulation of viscous flow by smoothed particle
by MOM in finite size domains. Chemical Engineering Sci-
hydrodynamics. Progress of Theoretical Physics, 92,
ence, 63(10), 2625–2640. doi:10.1016/j.ces.2008.02.010
939–960. doi:10.1143/PTP.92.939
Subbarao, D., & Gambhir, S. (2002). Gas Particle Mass
Takeuchi, H., Hirama, T., Chiba, T., Biswas, J., & Leung,
Transfer in Risers. 7th International Conference on Cir-
L. S. (1986). A quantitative definition and flow regime
culating Fluidized Beds, Niagra Falls, Ontario, Canada.
diagram for fast fluidization. Powder Technology, 47(2),
Subramaniam, S. (2001). Statistical modeling of sprays 195–199. doi:10.1016/0032-5910(86)80115-2
using the droplet distribution function. Physics of Fluids,
Tammers, R. F., Shaw, D. F., Reinman, K. J., & Melfi, G.
13(3), 624–642. doi:10.1063/1.1344893
(March 2, 1993). Tangential Solids Separation Transfer
Sun, B. (1996). Simulations of Gas-Liquid and Gas-Solid Tunnel. US Patent 5,190,650.
Two Phase Flows. Doctoral Dissertation, Illinois Institute
Tanaka, T., Yonemura, S., & Tsuji, Y. (1995). Effects of
of Technology, Chicago.
particle properties on the structure of clusters. ASMD
Sundaresan, S. (2001). Some outstanding questions in Publications FED, 228, 297–302.
handling of cohesionless particles. Powder Technology,
Taneda, S. (1957). Negative magnus effect. Republic
115, 2–7. doi:10.1016/S0032-5910(00)00423-X
Research Institution of Mechanics, 5, 123–128.
Suzuki, Y. J., Koyaguchi, T., Ogawa, M., & Hachisu, I.
Thomas, P. J. (1992). On the influence of the Basset history
(2005). A numerical study of turbulent mixing in erup-
force on the motion of a particle through a fluid. Phys-
tion clouds using a three-dimensional fluid dynamics
ics of Fluids A, 4(9), 2090–2093. doi:10.1063/1.858379
model. Journal of Geophysical Research, 110, B08201..
doi:10.1029/2004JB003460 Thompson, P. A. (1972). Compressible-fluid dynamics.
New York: McGraw Hill.
Syamlal, M. (1994). MFIX documentation. In User’s
manual, (pp. 87). Washington, DC: U.S. Dept. of Energy, Todesco, M., Neri, A., Eposti Ongaro, T., Papale, P.,
DOE/METC-95/1013, DE95000031. Macedonio, G., Santacroce, R., & Longo, A. (2002).
Pyroclastic flow hazard assessment at Vesuvius (Italy)
Syamlal, M. (1998). MFIX documentation. In Numeri-
by using numerical modeling. I. Large-scale dynam-
cal Technique, (pp. 80). Washington, DC: U.S. Dept. of
ics. Bulletin of Volcanology, 64, 155–177..doi:10.1007/
Energy. DOE/MC/31346-5824, DE98002029.
s00445-001-0189-7
Syamlal, M., & O’Brien, T. (April 1987) Derivation of
a drag coefficient from velocity-voidage correction. US
Dept. of Energy, Office of Fossil Energy, National Energy
Technology Laboratory, Morgantown, West Virginia,
460
Tomiyama, A., Sou, A., Zun, I., Kanami, N., & Sakaguchi, Tuzla, K., Sharma, A. K., & Chen, J. C. (1998). Transient
T. (1995). Effects of Eötvös number and dimensionless dynamics of solid concentration in downer fluidized
liquid volumetric flux on lateral motion of a bubble in bed. Powder Technology, 100, 166. doi:10.1016/S0032-
a laminar duct flow. In Serizawa, A., Fukano, T., & Ba- 5910(98)00137-5
taille, J. (Eds.), Advances in multiphase flow (pp. 3–15).
Unverdi, S. O., & Tryggvason, G. A. (1992). Front-
Amsterdam: Elsevier.
tracking method for viscous, incompressible multi-fluid
Toor, H. L., & Machello, J. M. (1958). Film-Penetration flows. Journal of Computational Physics, 100, 25–37.
Model For Mass And Heat Transfer. AIChE Journal. doi:10.1016/0021-9991(92)90307-K
American Institute of Chemical Engineers, 4(1), 97–101.
US DOE, Energy Information Administration. (2006).
doi:10.1002/aic.690040118
Retrieved from http://tonto.eia.doe.gov/dnav/pet/hist/
Tran-Cong, S., Gay, M., & Michaelides, E. E. (2004). Drag mcrccus2A.htm
coefficients of irregularly shaped particles. Powder Tech-
Valentine, G. A., & Wohletz, K. H. (1989). Numerical
nology, 139(1), 21. doi:10.1016/j.powtec.2003.10.002
models of Plinian eruption columns and pyroclastic
Travis, J. R., Harlow, F. H., & Amsden, A. A. (1976). flows. Journal of Geophysical Research, 94, 1867–1887.
Numerical calculation of two-phase flows. Nuclear Sci- doi:10.1029/JB094iB02p01867
ence and Engineering, 61, 1–10.
Valentine, G. A., Wohletz, K. H., & Kieffer, S. W. (1991).
Tryggvason, G., Bunner, B., & Esameeli, A. (2001). A Sources of unsteady column dynamics in pyroclastic
front-tracking method for the computations of multiphase flow eruptions. Journal of Geophysical Research, 93,
flow. Journal of Computational Physics, 169, 708–759. 21887–21892. doi:10.1029/91JB02151
doi:10.1006/jcph.2001.6726
Valentine, G. A. (1998). Eruption column physics. In
Tsao, H.-K., & Koch, D. L. (1995). Simple shear flows of Freundt, A., & Rosi, M. (Eds.), From Magma to Tephra:
dilute gas-solid suspensions. Journal of Fluid Mechanics, Modeling physical processes of explosive volcanic erup-
296, 211–245. doi:10.1017/S0022112095002114 tions (pp. 91–138). New York: Elsevier Science.
Tsuji, Y., Kawaguchi, T., & Tanaka, T. (1993). Discrete van Beijeren, H., & Ernst, M. H. (1973). The Modified
particle simulation of two-dimensional fluidized bed. Enskog Equation. Physica A 68, 437. Ibid, 70, 225.
Powder Technology, 77(1), 79–87. doi:10.1016/0032-
van Beijeren, H., & Ernst, M. H. (1979). Kinetic Theory
5910(93)85010-7
of Hard Spheres. Journal of Statistical Physics, 21, 125.
Tsuji, Y., Oshima, T., & Morikawa, Y. (1985). Numerical doi:10.1007/BF01008695
simulations of pneumatic conveying in a horizontal pipe.
van Breugel, J., Stein, J. J. M., & de Vries, R. (1969).
Kona, 3, 38–44.
Isokinetic sampling in a dense gas-solids stream. Proceed-
Tsuo, Y. P., & Gidaspow, D. (1990). Computation of ings - Institution of Mechanical Engineers, 184, 18–23.
flow patterns in circulating fluidized beds. AIChE Jour-
van der Hoef, M., Beetstra, R., & Kuipers, J. (2005).
nal. American Institute of Chemical Engineers, 36(6),
Lattice Boltzmann simulations of low Reynolds number
885–896. doi:10.1002/aic.690360610
flow past mono- and bidisperse arrays of spheres: results
Tuchbreiter, A., Marquardt, J., Kappler, B., Honerkamp, for the permeability and drag force. Journal of Fluid
J., Kristen, M. O., & Muhlhaupt, R. (2003). High-output Mechanics, 528, 233. doi:10.1017/S0022112004003295
polymer screening: exploiting combinatorial chemistry
and data mining tools in catalyst and polymer develop-
ment. Macromolecular Rapid Communications, 24,
47–62. doi:10.1002/marc.200390014
461
van der Hoef, M., van Sint Annaland, M., & Kuipers, Vanka, S. (1985). Block-implicit calculation of steady tur-
J. (2004). Computational Fuid dynamics for dense gas- bulent recirculating flows. International Journal of Heat
solid Fuidized beds: a multi-scale modeling strategy. and Mass Transfer, 28, 2093–2103. doi:10.1016/0017-
Chemical Engineering Science, 59, 5157. doi:10.1016/j. 9310(85)90103-6
ces.2004.07.013
Veera, U. P., Weickert, G., & Agarwal, U. S. (2002).
Van der Hoef, M. A., Van Sint Annaland, M., Deen, N. Modeling monomer transport by convection during ole-
G., & Kuipers, J. A. M. (2008). Numerical simulation of fin polymerization. AIChE Journal. American Institute
dense gas-solid fluidized beds: A multiscale modeling of Chemical Engineers, 48, 1062–1070. doi:10.1002/
strategy. Annual Review of Fluid Mechanics, 40, 47–70. aic.690480515
Venderbosch, R. H., Prins, W., & van Swaaij, W. P. M.
van Noije, T. P. C., & Ernst, M. H. (1998). Velocity Dis- (1999). Mass Transfer and Influence of the local catalyst
tributions in Homogeneous Granular Fluids: the Free and activity on the conversion in a Riser reactor. Cana-
the Heated Case. Granular Matter, 1, 57. doi:10.1007/ dian Journal of Chemical Engineering, 77, 262–274.
s100350050009 doi:10.1002/cjce.5450770211
van Noije, T. P. C., & Ernst, M. H. (2001). Kinetic Theory Verzicco, R., Mohd-Yusof, J., Orlandi, P., & Haworth,
of Granular Gases. In Pöschel, T., & Luding, S. (Eds.), D. (2000). Large eddy simulation in complex geometric
Granular Gases. New York: Springer. doi:10.1007/3- configurations using boundary body forces. AIAA Journal,
540-44506-4_1 38(3), 427–433. doi:10.2514/2.1001
van Swaaij, W. (1990). Chemical reactors. In Davidson, J., Vijay, G. N., & Reddy, B. V. (2005). Effect of dilute
& Clift, R. (Eds.), Fluidization. London: Academic Press. and dense phase operating conditions on bed-to-wall
heat transfer mechanism in a circulating fluidized bed
van Wachem, B. G. M., Schouten, J. C., van den Bleek,
combustor. International Journal of Heat and Mass
C. M., Krishna, R., & Sinclair, J. L. (2001). Comparative
Transfer, 48, 3276–3283. doi:10.1016/j.ijheatmasstrans-
analysis of CFD models of dense gas-solid systems. AIChE
fer.2005.03.013
1035–1051. doi:10.1002/aic.690470510 Vincenti, W. G., & Kruger, C. H. Jr. (1975). Introduc-
tion to physical gas dynamics. Huntington, NY: Krieger
van Wachem, B. G. M., Schouten, J. C., van den Bleek, C.
Publishing Company.
M., Krishna, R., & Sinclair, J. L. (2001). CFD modeling
of gas-fluidized beds with a bimodal particle mixture. Viollet, P. L., Simonin, O., Olive, J., & Minier, J. P. (1992).
AIChE Journal. American Institute of Chemical Engineers, Modeling turbulent two-phase flows in industrial equip-
47(6), 1292–1302. doi:10.1002/aic.690470607 ments. In Hirsch, C. (Ed.), Computational Methods in
Applied Sciences. New York: Elsevier science.
Van Wachem, B. G. M., Schouten, J. C., Krishna, R., &
van den Bleek, C. M. (1991). Validation of the Eulerian Vladimir, B. S., Vladimir, A. Z., & Valerii, A. K. (1996).
simulated dynamic behaviour of gas-solid fluidized Investigation of kinetics of ethylene polymerization
beds. Chemical Engineering Science, 54, 2141–2149. support titanium-magnesium catalyst of various compo-
doi:10.1016/S0009-2509(98)00303-0 sition. Macromolecular Chemistry and Physics, 197(5),
1615–1631. doi:10.1002/macp.1996.021970504
van Zoonen, D. (1962). Measurement of diffusional
phenomena and velocity profiles in a vertical riser. In
Proceeding of the Symposium on the Interaction between
Fluids and Particles, London, (pp. 64-71).
462
Vreman, B., Geurts, B. J., Deen, N. G., Kuipers, J. A. M., Wesseling, P. (2000). Principles of Computational Fluid
& Kuerten, J. G. M. (2009). Two- and Four-Way Coupled Dynamics. Berlin: Springer.
Euler–Lagrangian Large-Eddy Simulation of Turbulent
Wesseling, P., Segal, A., Kassels, C., & Bijl, H. (1998).
Particle-Laden Channel Flow. Flow, Turbulence and Com-
Computing flows on general two-dimensional nonsmooth
bustion, 82(1), 47–71. doi:10.1007/s10494-008-9173-z
staggered grids. Journal of Engineering Mathematics, 34,
Wang, J., & Ge, W. (2005). Collisional particle-phase 21–44. doi:10.1023/A:1004341115180
pressure in particle-fluid flows at high particle inertia.
Wildman, R. D., & Parker, D. J. (2002). Coexistence of
Physics of Fluids, 17, 128103. doi:10.1063/1.2145757
two granular temperatures in binary vibrofluidized beds.
Wang, W., & Li, J. (2007). Simulation of gas-solid two- Physical Review Letters, 88(6), 064301. doi:10.1103/
phase flow by a multi-scale CFD approach-extension of the PhysRevLett.88.064301
EMMS model to the sub-grid level. Chemical Engineer-
Wilhelm, R. H., & Kwauk, M. (1948). Fluidization of
ing Science, 62, 208–231. doi:10.1016/j.ces.2006.08.017
solid particles. Chemical Engineering Progress, 44(3),
Wang, W., Lu, B., Zhang, N., Shi, Z., & Li, J. (2010). A 201–218.
review of multiscale CFD for gas-solid CFB modeling.
Williams, F. A. (1985). Combustion Theory. 2nd. Reading,
International Journal of Multiphase Flow..doi:10.1016/j.
MA: Addison-Wesley.
ijmultiphaseflow.2009.01.008
Williams, K., & Cocco, R. (2007). Validation of Bar-
Wang, J., van der Hoef, M. A., & Kuipers, J. A. M. (2009).
racuda CPFD Dense-Phase Particle Physics Models
Why the two-fluid model fails to predict the bed expan-
Using Unique Data From a Large Deep-Bed Experiment.
sion characteristics of Geldart A particles in gas-fluidized
Paper presented at the AIChE Annual Meeting, Salt Lake
beds: A tentative answer. Chemical Engineering Science,
City, Utah.
64, 622–625. doi:10.1016/j.ces.2008.09.028
Willits, J. T., & Arnarson, B. O. (1999). Kinetic theory of
Wassgren, C. R., Cordova, J. A., Zenit, R., & Karion,
a binary mixture of nearly elastic disks. Physics of Fluids,
A. (2003). Dilute granular flow around an immersed
11(10), 3116–3122. doi:10.1063/1.870169
cylinder. Physics of Fluids, 15(11), 3318–3330.
doi:10.1063/1.1608937 Wilson, L., Sparks, R. S. J., Huang, T. C., & Watkins,
N. D. (1978). The control of volcanic column heights by
Wegerer, D., & Lomas, D. (Sept. 19, 1995). FCC Feed
eruption energetic and dynamics. Journal of Geophysical
Contacting with Catalyst Recycle Reactor. US Patent
Research, 83, 1829–1836. doi:10.1029/JB083iB04p01829
5,451,313.
Wohletz, K. H., McGetchin, T. R., Standford, M. T.
Weinstein, H., Graff, R. F., Meller, M., & Shao, M. J.
II, & Jones, E. M. (1984). Hydrodynamics aspects of
(1983). The Influence of the Imposed Pressure Drop
caldera-forming eruptions: Numerical models. Journal
Across a Fast Fluidized Bed. In Kunii, K., & Toei, R.
of Geophysical Research, 89, 8269–8286. doi:10.1029/
(Eds.), Fluidization IV (pp. 299–306). New York: Engi-
JB089iB10p08269
neering Foundation.
Wolny, A., & Ka Zmierczak, W. (1989). Triboelectrifi-
Wen, C. Y., & Yu, Y. H. (1966). Mechanics of fluidization.
cation in fluidized bed of polystyrene. Chemical Engi-
Chem. Eng. Progr. Symp., 62, 100-110.
neering Science, 44(11), 2607–2610. doi:10.1016/0009-
Wenneker, I., Segal, A., & Wesseling, P. (2003). Conserva- 2509(89)85204-2
tion properties of a new unstructured staggered scheme.
Computers & Fluids, 32, 139–147. doi:10.1016/S0045-
7930(01)00094-9
463
Wood, A. M. (2005). Preferential concentration of particles Xu, M., Ge, W., & Li, J. (2007). A discrete particle
in homogeneous and isotropic turbulence. International model for particle-fluid flow with considerations of sub-
Journal of Multiphase Flow, 31, 1220. doi:10.1016/j. grid structures. Chemical Engineering Science, 62(8),
ijmultiphaseflow.2005.07.001 2302–2308. doi:10.1016/j.ces.2006.12.008
Wormsbecker, M., & Pugsley, T. (2008). The influence Xu, K. (2008). A generalized Bhatnagar-Gross-Krook
of moisture on the fluidization behaviour of porous phar- model for nonequilibrium flows. Physics of Fluids, 20,
maceutical granule. Chemical Engineering Science, 63, 026101. doi:10.1063/1.2837174
4063–4069. doi:10.1016/j.ces.2008.05.023
Xu, Y., & Subramanian, S. (2006). A multiscale model for
Wright, D. L., McGraw, R., & Rosner, D. E. (2001). Bi- dilute turbulent gas–particle flows based on the equili-
variate extension of the quadrature method of moments bration of energy concept. Physics of Fluids, 18, 21–38.
for modeling simultaneous coagulation and sintering doi:10.1063/1.2180289
particle populations. Journal of Colloid and Interface
Xu, Y. (2008). Modeling and direct numerical simulation
Science, 236(2), 242–251. doi:10.1006/jcis.2000.7409
of particle laden turbulent flows. Iowa State University,
Wu, L. R., Grace, J. R., & Lim, C. J. (1990). A model for Ames, IO.
heat transfer in circulating fluidized beds. Chemical Engi-
Yan, A., Manyele, S. V., Parssinen, J. H., et al. (2002).
neering Science, 45(12), 3389–3398. doi:10.1016/0009-
The interdependence of micro and macro flow structures
2509(90)87144-H
under a high-flux flow. In Proceedings of 7th Interna-
Wylie, J. J., & Koch, D. L. (2000). Particle clustering due tional Circulating Fluidized Beds Conference, (pp. 357).
to hydrodynamic interactions. Physics of Fluids, 12(5), Canadian Society for Chemical Engineering.
964–970. doi:10.1063/1.870351
Yang, J. Z., & Renken, A. A. (2003). Generalized correla-
Xie, H. Y. (1997). Drag coefficient of fluidized particles at tion for equilibrium of forces in liquid-solid fluidized beds.
high Reynolds numbers. Chemical Engineering Science, Chemical Engineering Journal, 92, 7–14. doi:10.1016/
52(17), 3051–3052. doi:10.1016/S0009-2509(97)00058-4 S1385-8947(02)00084-0
Xie, T., McAuley, K. B., Hsu, J. C. C., & Bacon, J. W. Yang, N., Wang, W., Ge, W., & Li, J. (2003). CFD
(1994). Gas phase ethylene polymerization: Production simulation of concurrent-up gas-solid flow in circulating
processes, polymer properties, and reactor modeling. In- fluidized beds with structure-dependent drag coefficient.
dustrial & Engineering Chemistry Research, 33, 449–479. Chemical Engineering Journal, 96, 71–80. doi:10.1016/j.
doi:10.1021/ie00027a001 cej.2003.08.006
Xu, H., Louge, M., & Reeves, A. (2003). Solutions of Yang, N., Wang, W., Ge, W., Wang, L., & Li, J. (2004).
the kinetic theory for bounded collisional granular flows. Simulation of heterogeneous structure in a circulating
Continuum Mechanics and Thermodynamics, 2003, fluidized-bed riser by combining the two-fluid model with
321–349. doi:10.1007/s00161-003-0116-6 the EMMS approach. Industrial & Engineering Chemistry
Research, 43(18), 5548–5561. doi:10.1021/ie049773c
Xu, B. H., & Yu, A. B. (1997). Numerical simulation
of the gas-solid flow in a fluidized bed by combining Yang, N., Ge, W., Niu, G., Yang, C., & Li, J. (2005).
discrete particle method with computational fluid dynam- Simulation of gas/solid flow behaviors and choking for
ics. Chemical Engineering Science, 52(16), 2785–2809. a CFB riser: the EMMS/CFD approach. In Cen, K. (Ed.),
doi:10.1016/S0009-2509(97)00081-X Circulating Fluidized Bed Technology VIII (pp. 291–298).
Hangzhou, China: World Publishing Corporation.
464
Yang, N., Wang, W., Zhao, H., Ge, W., & Li, J. (2006). Yin, X., & Sundaresan, S. (2009). Drag law for bidisperse
Structure-oriented multi-scale simulation of two-phase gas-solid suspensions containing equally sized spheres.
flows—methodology and application. In Fifth Inter- Industrial & Engineering Chemistry Research, 48(1),
national Conference on CFD in the Process Industries 227–241. doi:10.1021/ie800171p
(CDROM). Melbourne, Australia: CSIRO.
Yip, S. (1979). Renormalized Kinetic Theory of Dense
Yasuna, J., Moyer, H., Elliot, S., & Sinclair, J. (1995). Fluids. Annual Review of Physical Chemistry, 30, 547.
Quantitative predictions of gas-particle flow in a vertical doi:10.1146/annurev.pc.30.100179.002555
pipe with particle-particle interactions. Powder Technol-
Yoon, D. K., & Jenkins, J. T. (2006). The influence of
ogy, 84(1), 23–34. doi:10.1016/0032-5910(94)02971-P
different species’ granular temperature on segregation in
Ye, M., van der Hoef, M. A., & Kuipers, J. A. M. (2004). a binary mixture of dissipative grains. Physics of Fluids,
A numerical study of fluidization behavior of Geldart A 18, 073303. doi:10.1063/1.2219437
particles using a discrete particle model. Powder Technol-
Yuan, E., Letzsch, W., Jackson, G., Evans, J., & Hood,
ogy, 139(2), 129–139. doi:10.1016/j.powtec.2003.10.012
J. (2006, Aug. 10). Riser Termination Device. US Patent
Ye, M., van der Hoef, M. A., & Kuipers, J. A. M. (2005). Application 2006/0177357.
From discrete particle model to a continuous model of
Yusof, J. M. (1996). Interaction of massive particles with
Geldart A particles. Chemical Engineering Research &
turbulence. Ithaca, NY: Cornell University.
Design, 83(7), 833–843. doi:10.1205/cherd.04341
Zamankhan, P. (1995). Kinetic theory of multicomponent
Ye, M., Wang, J., van der Hoef, M. A., & Kuipers, J. A.
mixtures of slightly inelastic spherical particles. Physi-
M. (2008). Two-fluid modeling of Geldart A particles
cal Review E: Statistical Physics, Plasmas, Fluids, and
in gas-fluidized beds. Particuology, 6(6), 540–548.
Related Interdisciplinary Topics, 52(5), 4877–4891.
doi:10.1016/j.partic.2008.07.005
doi:10.1103/PhysRevE.52.4877
Ye, M. (2005). Multi-level modeling of dense gas-solid
Zenz, F. A. (1949). Two-phase fluid-solid flow. Indus-
two-phase flows. PhD Dissertation, University of Twente,
trial & Engineering Chemistry, 41(12), 2801–2806.
Twente, The Netherlands.
doi:10.1021/ie50480a032
Yerushalmi, J., Tuner, D. H., & Squires, A. M. (1976). The
Zhang, D. Z., & VanderHeyden, W. B. (2002). The effects
fast fluidized bed. Industrial & Engineering Chemistry
of mesoscale structures on the macroscopic momentum
Process Design and Development, 15, 47–53. doi:10.1021/
equations for two-phase flows. International Journal
i260057a010
of Multiphase Flow, 28, 805–822. doi:10.1016/S0301-
Yerushalmi, J., & Cankurt, N. T. (1979). Further studies 9322(02)00005-8
of the regimes of fluidization. Powder Technology, 24(2),
Zhang, J., Ge, W., & Li, J. (2005). Simulation of hetero-
187–205. doi:10.1016/0032-5910(79)87036-9
geneous structures and analysis of energy consumption
Yerushalmi, J., & Squires, A. M. (1977). The phenomenon in particle-fluid systems with pseudo-particle model-
of fast fluidization. AIChE Symposium Series, 72, 44-47. ing. Chemical Engineering Science, 60, 3091–3099.
doi:10.1016/j.ces.2004.11.057
Yerushalmi, J., Cankurt, N., Geldart, D., & Liss, B.
(1978). Flow regimes in vertical gas-solid contact systems. Zhang, N., Lu, B., Wang, W., & Li, J. (2008). Virtual
Fluidization: Application to Coal Conversion Processes: experimentation through 3-D full-loop CFD simulation
AIChE Symp. Ser. 74 (176) 1-13. of a CFB. Particuology, 6, 529–539. doi:10.1016/j.
partic.2008.07.013
465
Zhang, Y., & Reese, J. M. (2003). The drag force in Zick, A. A., & Homsy, G. M. (1982). Stokes flow through
two-fluid models of gas-solid flows. Chemical Engi- periodic arrays of spheres. Journal of Fluid Mechanics,
neering Science, 58, 1641–1644. doi:10.1016/S0009- 115, 13–26. doi:10.1017/S0022112082000627
2509(02)00659-0
Zimmermann, S., & Taghipour, F. (2005). CFD model-
Zhao, H., Castle, G. S. P., Inculet, I. I., & Bailey, A. G. ing of the hydrodynamics and reaction kinetics of FCC
(2000). Bipolar charging in polydisperse polymer powders fluidized-bed reactors. Industrial & Engineering Chem-
in industrial processes. In Proceedings of the IEEE-IAS istry Research, 44, 9818–9827. doi:10.1021/ie050490+
Annual Meeting, Rome, October 2000, (pp. 835–841).
Zou, B., Li, H., & Xia, Y. (1994). Cluster structure in a
Zheng, Y., & Struchtrup, H. (2005). Ellipsoidal statistical circulating fluidized bed. Powder Technology, 78, 173.
Bhatnagar-Gross-Krook model with velocity-dependent doi:10.1016/0032-5910(93)02786-A
collision frequency. Physics of Fluids, 17, 127103.
Zuber, N. (1964). On the dispersed two-phase flow in the
doi:10.1063/1.2140710
laminar flow regime. Chemical Engineering Science, 19,
Zhu, H. P., Zhou, Z. Y., Yang, R. Y., & Yu, A. B. (2007). 897. doi:10.1016/0009-2509(64)85067-3
Discrete particle simulation of particulate systems:
Zwart, P. (1999). The integrated space-time finite volume
theoretical developments. Chemical Engineering Sci-
method. PhD thesis, University of Waterloo, Canada.
ence, 62(13), 3378–3396. doi:10.1016/j.ces.2006.12.089
Zhu, H. P., Zhou, Z. Y., Yang, R. Y., & Yu, A. B. (2008).

Discrete particle simulation of particulate systems: a
review of major applications and findings. Chemical
Engineering Science, 63(23), 5728–5570. doi:10.1016/j.
ces.2008.08.006
466
467
About the Contributors
Sreekanth Pannala is a senior research staff member in the Computer Science and Mathematics
Division at Oak Ridge National Laboratory. He received his M.S. (1994) and Ph.D. (2000) in Aerospace
Engineering specializing in the area of computational combustion of two-phase flows from Georgia
Tech. His expertise is primarily in the area of developing parallel algorithms and models for heteroge-
neous chemically reacting flows from device to micro scale. He received Federal Laboratory Technology
Transfer award in 2006 and R&D 100 award in 2007 for his contribution to the development of MFIX
(http://mfix.netl.doe.gov), an open-source multiphase flow simulation suite. He has served as a principal
investigator on various DOE computational science projects and has over 75 conference and technical
publications in various areas of computational science and engineering.
Madhava Syamlal is Focus Area Leader of Computational and Basic Sciences at National Energy
Technology Laboratory, where he is responsible for using computational science to accelerate the devel-
opment of advanced energy systems. He earned a B.Tech in chemical engineering from IT-BHU (1977)
and an M.S. (1981) and a Ph.D. (1985) in chemical engineering from Illinois Institute of Technology. He
has led the development of the open source multiphase flow code MFIX and the integration CFD and
process simulation, both of which have won R&D 100 awards. In 2009 he received the PTF Fluidization
Process Recognition award from the American Institute of Chemical Engineers.
Thomas J. O’Brien has been a Research Scientist at the NETL since 1979. He obtained a B.S.
in chemistry from the College of St. Thomas (St. Paul, MN) in 1962, a Ph.D. in Physical/Theoretical
Chemistry from the University of Wisconsin (Madison) in 1968, has held postdoctoral positions at The
Queen’s University (Belfast, Northern Ireland) and The Johns Hopkins University, and has been a Na-
tional Research Council Associate (U. S. Army Ballistics Research Laboratory). He was an Assistant
Professor in the Chemistry Department at Texas Tech University (Lubbock, TX) from 1969-79.Most of
Dr. O’Brien’s research efforts at NETL have been in developing mechanistic mathematical models for
coal conversion processes.
***
Aparna Baskaran was born in Chennai, India on May 12 1979. She obtained her Ph.D from the
University of Florida in May 2006 working with James W. Dufty on the statistical mechanics of fluidized
granular systems. She is a postdoctoral researcher at Syracuse University working on active biomateri-
als. She will start as a tenure track assistant professor in physics at Brandeis University in June 2010.
Sébastien Dartevelle is a Los Alamos National Laboratory technical staff member in geophys-
ics. He obtained his Master in geo-sciences at the University of Brussels in 1993 and his Ph.D. in
geophysical-Computational Fluid Dynamic at Michigan Technological University in 2003. He worked
on multiphase code development and V&V for the licensing application of the Yucca Mountain Project,
Nevada (underground nuclear waste repository) of the Office of Civilian Radioactive Waste Manage-
ment. His main interests are geophysical multiphase turbulence phenomenology in the atmosphere and
volcanic granular flows.
James W. Dufty is currently Professor of Physics (Emeritus) at the University of Florida, where he
has been involved in teaching and research since 1970. During that time he also served as Associate
Dean for Research and American Institute of Physics Fellow at the U. S. State Department. His general
area of research is non equilibrium statistical mechanics of classical and quantum systems, described in
more than 220 refereed publications. Current research emphases include the theory of granular matter,
strongly coupled Coulomb systems, and warm dense matter. This research includes multiple interna-
tional collaborators, and is supported by the National Science Foundation and the Department of Energy.
Rodney O. Fox has made many ground-breaking contributions to the field of reactive flow model-
ing. Fox spearheaded many fundamental advances in the development of novel computational fluid
dynamics (CFD) models to overcome specific scientific challenges faced in engineering applications.
Fox pioneered the use of in situ tabulation (ISAT) for efficiently handling complex chemistry in detailed
multiphase reactor models, and developed powerful solution techniques (DQMOM) for treating distribu-
tion functions (particle size, bubble size, etc.) required for CFD models of single and multiphase reactors.
The impact of Fox’s work extends far beyond chemical engineering and touches every technological
area dealing with turbulent flow and chemical reactions (e.g., combustion, atmospheric science, nuclear
fuel processing, etc.). His monograph, Computational Models for Turbulent Reacting Flows, offers an
authoritative treatment of the field.
Christine Hrenya is a tenured Associate Professor in the Department of Chemical and Biological
Engineering at the University of Colorado. Her interests lie in the field of multiphase and solids flows,
using a combination of theory, simulations, and experiments. To date, this research program has resulted
in 75 invited lectures, with funding from the U.S. National Science Foundation, U.S. Department of
Energy, U.S. National Aeronautics and Space Administration, and American Chemical Society. Recent
recognitions include being named a 2009-10 University of Colorado Emerging Leaders Fellow, the 2008
University of Colorado Provost’s Achievement Award, and the 2006 ACS Progress/Dreyfus Lecture-
ship Award. In 2005, she chaired the Patten Centennial Scientific Workshop: The Next Millennium in
Chemical Engineering, the findings of which were published in a series of articles in Chemical Engi-
neering Education (2006). Prof. Hrenya served as Chair of the 2006 Gordon Conference on Granular
Flow (Oxford, U.K.) and as co-Director of a 2001-2004 GAANN program in Micro- and Nano-Particle
Technology.
Michael Muhle received a BS in Chemistry and Mathematics from Creighton University and gradu-
ated in1969. He received a MS in Chemical Engineering from Iowa State University in 1971 and a
PhD in Chemical Engineering from Iowa State University in 1974. Following his graduate degrees he
has been continuously employed at several companies, and most recently at ExxonMobil Corporation,
468
where he has over 30 years of service. He has worked in R&D, commercial manufacturing, patent and
licensing of technology and is currently assigned in a joint venture technology and licensing company
that was formed between ExxonMobil and Dow Chemical called Univation Technologies. His academic
interests include polymerization reaction modeling, computational fluid dynamics, organometallic
metallocene polymerization catalysis, advanced signal processing, advanced experimental design and
multivariate statistical analysis.
Alberto Passalacqua is a post-doctoral research associate at Iowa State University, and obtained
his Ph.D. at Politecnico di Torino, Italy (2008), on the CFD simulation of gas-particle flows applied to
fluidized systems, in particular bubbling fluidized bed reactors and risers. His research interests include
computational fluid dynamics applied to multiphase flows, direct numerical simulations and large eddy
simulation, with particular attention to the development, the implementation and the validation of CFD
methods for the simulation of granular and fluid-particle flows. His current research activity is focused
on the development and implementation of quadrature-based moment methods for the solution of the
kinetic equation to describe the dispersed phase in fluid-particle flows.
Todd Pugsley received his BSc in Chemical Engineering from the University of New Brunswick
in 1990 and his MSc and PhD in Chemical Engineering from the University of Calgary in 1992 and
1995, respectively. He is known internationally for his research in the field of fluidized bed technology,
particularly for the application of electrical and X-ray tomography and radioactive particle tracking
techniques for imaging the gas and solids flow patterns inside lab and pilot scale test rigs. His group
was also the first to obtain realistic CFD predictions of fluidized beds containing Geldart A powders.
Professor Pugsley has co-chaired the 11th International Fluidization Conference in British Columbia,
Canada in 2007 and the inaugural Particulate Processes in the Pharmaceutical Industry conference in
Montreal in 2005. His current research efforts focus on fluidized bed gasification and CFD modeling
using the multiple-particles-in-cell (MPIC) approach.
Ram G. Rokkam received a B.Tech in Chemical Engineering from Andhra University College of
Engineering, India and graduated in 2003. He received a M.Tech in Chemical Engineering from Indian
Institute of Technology Madras, India. Ram joined Dr. Rodney Fox group at Iowa State University in
January 2007 for his doctoral studies in Chemical Engineering and is presently working on the com-
putational fluid dynamic modeling of gas-solid polymerization fluidized-bed reactors. In particular his
research work focuses on electrostatic modeling and coarse grid simulations of pilot plant and com-
mercial scale reactors. His research interests are in Multiphase flows, Chemical reaction engineering,
CFD modeling.
Prakash Vedula received his BTech (1996) degree from Indian Institute of Technology, Madras,
and both MS (1998) and PhD (2001) degrees from Georgia Institute of Technology, Atlanta (in Aero-
space Engineering). He is currently an assistant professor at the School of Aerospace and Mechanical
Engineering, University of Oklahoma, Norman (from 2006). His research interests include turbulence,
mixing, multiphase flows, numerical methods, reduced order modeling, computational fluid dynamics,
direct numerical simulation, large eddy simulation and optimal prediction. More recently, he also de-
veloped several innovative computational algorithms and multi-purpose codes for model reduction via
quadrature based methods, that have shown promising potential for applications in diverse fields relating
469
to computational (non-equilibrium) kinetic theory, multiphase flows, stochastic nonlinear dynamical

systems, aeroelasticity, uncertainty quantification, nonlinear filtering and control.
Berend van Wachem obtained his PhD in Applied Physics at Delft University of Technology in 2000.
He is currently a senior lecturer at Imperial College London in the UK. His research projects involve
multiphase flow modeling and simulation, ranging from understanding the behaviour of turbulence on
the scale of individual particles, to the large-scale modelling of gas-solid and gas-liquid flows.
470
471
Index
A circulating fluidized beds (CFB) 277, 316, 317,

318, 319, 320, 321, 322, 323, 324, 325,
Archimedes’ principle 131, 132, 133 330, 333, 334, 335, 336, 337, 338, 339,
artificial compressibility 204, 206, 207 340, 346, 348, 349, 351
B clusters 322, 327, 329, 330, 331, 332, 352, 354
coarse grids 4
Basset force 131, 134, 143 collocated grids 204, 205
bed expansion 361, 362, 363, 364, 365, 368 computational fluid dynamics (CFD) 2, 4, 5,
bed hydrodynamics 183 6, 53, 54, 55, 58, 64, 65, 128, 129, 140,
Boltzmann-Enskog kinetic equation 66, 67 143, 146, 147, 154, 157, 166, 172, 176,
Boltzmann kinetic equation 107, 221, 222, 177, 178, 181, 182, 190, 191, 194, 198,
223, 230, 242 199, 205, 206, 217, 316, 333, 334, 337,
Brazil nut problem 103 356, 357, 358, 360, 361, 362, 364, 365,
bubble columns 374 366, 367, 368, 369, 370, 371, 375, 376,
bubble motion 4 377, 379, 380, 388, 389, 392, 393, 394,
bubbling bed model 181 398, 426
bubbling beds 361 conservative buoyancy 132
bubbling fluidized bed 1, 44, 56, 63 continuous fluids 316, 334, 343
buoyancy 130, 131, 132, 133, 137, 138, 166, continuum gas-solids flow models 4
167, 169 convective heat transfer 178
buoyancy-driven plumes 399 coupled solver approach 204
cyclones 317, 319, 339, 352
C
Cartesian grids 204
D
Cartesian meshes 251 deferred correction methods (DCM) 402
CFB, hydrodynamics of 322 dense fluidized systems 181, 182, 183
Chapman-Enskog (CE) expansion 108, 222 diffusion 178, 179, 180, 181, 188, 193, 196,
Chapman-Enskog (CE) perturbative expansion 199
114 direct numerical simulation (DNS) method 3,
chemical looping 318, 320, 321 4, 129, 140, 144, 159, 164, 245, 246,
Christoffel symbols 205 247, 260, 270, 271, 272, 273
circulating fluidized bed (CFB) combustor direct quadrature method of moments (DQ-
103, 124, 125 MOM) 222
discrete element method (DEM) 3, 5, 63, 65,
68, 214, 277, 278
Index
discrete particle method (DPM) 68, 278, 362 fluidized bed reactors 2, 44, 48
discrete quadrature method of moments (DQ- fluidized beds 316, 317, 318, 319, 320, 321,
MOM) 103 322, 323, 324, 327, 333, 335, 338, 340,
drag coefficients 4, 17, 18, 22, 23, 56 341, 342, 343, 345, 347, 348, 349, 350,
drag force 129, 135, 151, 358, 360, 361, 362, 351, 352, 353, 354, 355, 357, 358, 360,
363, 364, 365, 368, 370 361, 368, 370, 371, 372, 374
drag interaction 128 fluidized bed systems 133, 140, 146
drag laws 246, 247, 248, 256, 262, 264, 265, fluidized bed technology 358, 360
266, 267, 268, 269, 270, 272, 273 fluidized catalytic cracking (FCC) riser reactor
drag losses 128 316, 317, 318, 319, 320, 321, 324, 325,
drag models 213 327, 328, 329, 332, 341, 342, 343, 346,
DuPont process 320, 321 349, 352, 353, 356
dynamic drag 132 fluxes 66, 67, 68, 73, 74, 75, 84, 85, 89
functions method of moments (FC-MOM) 222
E
G
electrostatics 375, 393, 395
elutriation 374, 377, 379, 380, 381 gas-fluidized beds 68, 98
Enksog assumption 107 gasification technology 2
Enskog equation 105, 107, 113, 114 gas-particle flows 221, 222, 223, 224, 225,
entrainment 374, 375, 377, 379, 380, 393 229, 234, 240, 241, 242
entrainment rates 323 gas-phase processes 374
epoxy resins 374 gas phases 128, 129, 131, 133, 134, 137, 138,
Eulerian approach 360 139, 140, 142, 143, 144, 145, 147, 149,
Eulerian equation 222, 230 150, 151, 152, 153, 154, 156, 157, 158,
Eulerian-Eulerian equations 215 159, 160, 162, 163, 164, 165, 166, 167,
Eulerian-Eulerian model 3, 221, 222, 226, 238, 168, 169, 170, 171, 172, 173, 174, 175,
239, 242 176, 177, 399, 401, 402, 403, 404, 405,
Eulerian-Eulerian multi-fluid model 373 408, 414, 415, 416, 417, 422
Eulerian flow statistics 245, 246, 249, 250 gas-solids drag 1, 3, 4, 16, 19, 21, 22
Eulerian fluid solver 221 gas-solids flows 1, 2, 5, 50, 66, 68, 69, 70, 71,
Eulerian grid 277, 278, 284, 287, 288, 289, 82, 83, 89, 102, 104, 119, 128, 129, 149,
290, 291, 292, 298 159, 204, 219, 245, 246, 247, 249, 270,
Eulerian-Lagrangian methods 277 271, 272, 273, 277, 306, 309, 312, 373
Eulerian model 362 gas-solids hydrodynamic behavior 245
Eulerian reference frame 277 gas-solids reacting flows 1, 6, 50
Eulerian statistics 245 gas-solids suspension flow 246
Euler-Lagrange simulations 221, 226, 242 gas-solids systems 68, 69, 100
Geldart A powders 357, 359, 360, 361, 362,
F 363, 365, 370, 371
fictitious speed of sound 207 Geldart B powders 359, 360, 362
flow velocity 66, 76, 81, 101 Geldart classification of powders 359
fluid catalytic cracking (FCC) 2, 3, 4, 59, 359, Geldart C powders 360
361, 371, 372 Geldart D powders 360
fluidization 373, 394 geophysical multiphase flow with interphase
fluidized bed hydrodynamics 357 exchange (GMFIX) 401, 402, 407, 410,
fluidized bed processes 358 412, 414, 415, 416, 417, 418, 420, 426
472
Index
geo-validation 398, 418, 426 kinetic theory 67, 68, 69, 70, 72, 73, 76, 77,
granular energy balances 110 78, 79, 80, 88, 89, 90, 94, 97, 102, 103,
granular flows 2, 56, 221, 242, 243 119, 123, 124, 125, 126, 127
granular gases 222 Knudsen numbers (Kn) 108, 114, 221, 222,
granular material 66 225, 226, 230, 238, 239, 242
granular phase 66, 67, 68, 69, 72, 76, 77, 78,
79, 80, 81, 82, 84, 88, 89, 90, 91, 92, 93, L
94, 96, 99, 101 Lagrangian advection 277
granular stresses 1, 26 Lagrangian computational-particle model 277
granular temperature 66, 101, 316, 331, 335, Lagrangian flow statistics 246
336, 338, 348 Lagrangian models 214
granular turbulent kinetic energy 331 Lagrangian parcels 277
grid-to-particle interpolation method 279 Lagrangian reference frame 277
Gunn’s model 181, 182 Lagrangian statistics 245
large eddy simulation (LES) approach 401,
H
403, 404, 405, 426
heat transfer coefficients 182, 183, 199 laser Doppler velocimetry 331
high density polyethylene (HDPE) 374 lattice-Boltzmann calculations 278
homogeneous cooling states (HCS) 107, 108, lattice-Boltzmann equations 3
127 lattice-Boltzmann method (LBM) 3, 4, 19, 21
homogeneous suspension flow 247 Liouville density 247
hydrodynamic equations 66, 67, 68, 69, 76, 78, liquid-phase processes 373, 374
81, 86, 90, 93 loop reactors 374
hydrodynamic fields 66, 67, 68, 70, 71, 72, 74, low density polyethylene (LDPE) 373
75, 76, 77, 78, 81, 82, 83, 85, 92
hydrodynamics 66, 67, 68, 69, 70, 76, 77, 78, M
82, 85, 88, 89, 90, 93, 94 Magnus lift force 131, 134
maleic anhydride 320, 321, 322, 348
I
mass and heat transfer 178
ice Boltzmann method (LBM) 246, 248, 261, mass transfer 178, 179, 180, 181, 182, 183,
262, 264, 265, 266, 267, 268, 270, 271 184, 186, 187, 188, 189, 190, 193, 195,
immersed boundary methods (IBM) 128, 245, 196, 198, 199, 201
246, 251, 252, 254, 255, 256, 259, 260, Maxwellian velocity distribution 222
261, 262, 263, 264, 265, 266, 267, 268, mesoscale phenomena 357
269, 270, 271, 272, 273 metallocene catalyst 374, 376, 385
implicit multiphase formalism 402 MFIX software 1, 6, 33, 44, 46, 50, 63
interfacial drag closure relations 357 microclusters 183, 184, 188, 198
interstitial gas phase 66 mixture velocity 104, 127
molecular dynamics (MD) 108
K momentum transfer 245, 246, 249, 273
kinetic equations 66, 68, 69, 72, 73, 74, 77, 78, monodisperse materials 102, 103, 104, 108,
80, 81, 82, 83, 84, 85, 86, 88, 89, 90, 93, 111, 112, 113, 114, 115, 118, 119, 124
101 multi-phase flow problems 204
multiphase flows 205, 213
473
Index
multiphase flow with interphase exchanges polyolefins 374, 375

(MFIX) 178, 179, 180, 184, 186, 200, polypropylene (PP) 373
363, 366 polystyrene (PS) 374
multiphase particle-in-cell (MP-PIC) numerical polyurethanes (PU) 374
method 277, 278, 279, 280, 284, 286, poly vinyl chloride (PVC) 374
287, 289, 290, 291, 292, 293, 295, 296, population balance equation (PBE) 222, 244
297, 298, 302, 305, 306, 307, 308, 312 pressure-velocity coupling 204, 205, 219
multizone circulating reactors 374 pressure-velocity decoupling 205
pressure-weighted interpolation 205
N
Q
Navier-Stokes continuum 68
Navier-Stokes equations 3, 204, 215, 219, 222, quadrature method of moments (QMOM) 222,
245, 246, 270, 271, 275, 402, 426 223, 226, 232, 234, 235, 236, 237, 238,
Navier-Stokes hydrodynamic equations 66, 68, 239, 240, 241
93
Navier-Stokes limitations 66 R
Navier-Stokes momentum equation 278 reactants 178
Navier-Stokes Partial Differential Equations revised Enskog theory (RET) 105, 106, 113,
401 114
nodal points 208 Reynolds Average Navier Stokes (RANS)
Nusselt number 182, 201 framework 401, 403, 404, 405, 426
Reynolds numbers 129, 130, 134, 135, 136,
O
138, 143, 144, 151, 155, 156, 173, 181,
oxygen carriers 320 182, 201, 246, 247, 248, 255, 256, 259,
260, 262, 264, 265, 266, 267, 268, 269,
P 270, 271, 273, 274
packet theory 183 riser hydrodynamics 318, 324, 327, 333, 338
particle-in-cell (PIC) method 277, 278, 279,
S
280, 284, 286, 287, 289, 290, 291, 292,
293, 295, 296, 297, 298, 302, 305, 306, Saffman force 131, 134
307, 308, 312 segregation 373, 374, 375, 379, 380, 381, 391,
particle-particle-stresses 278 393, 395
particle phase 221, 222, 225, 234, 235, 239, separated particulate multiphase flow 277
241 Sherwood numbers 180, 181, 186, 201, 202
particle viscous-stresses 278 slip velocities 323, 326
particle volume fraction 221 solid-fluid flow regimes 278
phenolics 374 solid phases 66, 69, 80, 83, 88, 91, 92, 129
plinian cloud 399, 401, 415, 418, 419, 420 solids clusters 178, 182, 183, 188, 190, 191,
plug flow reactors 374 196, 198
polyamide (PA) 374 solids flows 102, 103
polydisperse granular flows 104, 105 solids flux 322, 323, 326, 327, 328, 329, 332,
polydisperse materials 103 333
polyethylene 373, 374, 376, 393, 396 solids volume fraction 182
polyethylene terephthalate (PET) 374 source terms 204, 205, 206, 209, 213, 214,
polymers 373, 374, 376, 394 215, 219
474
Index
species densities 66, 67, 70, 76, 85, 86 U

species segregation 102, 103, 104, 116, 117,
119 underexpanded jets 401, 406, 408
species velocity 104, 127 V
standard Enskog theory (SET) 105, 106, 113,
114 velocity field 205, 206, 215
stirred tank reactors 374 virtual mass force 131, 133
Stokes drag force 131 viscous terms 205, 209
Stokes flow 248, 262, 263, 264, 268, 270, 271, voidage profiles 323
273, 276, volcanic clouds 399
Stokes numbers 221, 222, 223, 226, 238, 239, volcanic eruptions 398, 427, 429
242, 247, 248 volcanic explosive phenomenology 398
Stokes numbers, finite 221 volcanic hazards 398, 399
streamers 316, 329, 330, 331, 332, 336, 343, volcanic phenomenology 398, 399, 406
344 volcanology 398, 399, 406
supersonic jets 401, 406, 407, 408, 409, 421,
422, 423 W
surface tension models 213 Winkler gasifier 358
synthesis gas (syngas) 2, 5
Z
T
Ziegler-Natta catalyst 374
thermoplastics 373, 374
thermosets 373, 374
tribo-electrification 375
turbulent dispersion 178, 182, 184, 193, 194
two fluid theory 68, 246, 249, 250, 255, 273
475

Computational Gas-Solids Flows and Reacting Systems - Theory, Methods and Practice (2011)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Computational Gas-Solids Flows and Reacting Systems - Theory, Methods and Practice (2011)

Uploaded by

Copyright:

Available Formats

Computational Gas-Solids

Flows and Reacting

EnginEEring sciEncE rEfErEncE

Published in the United States of America by

Library of Congress Cataloging-in-Publication Data

Computational gas-solids flows and reacting systems : theory, methods and

British Cataloguing in Publication Data

Foreword ............................................................................................................................................ xiv

Preface ................................................................................................................................................ xvi

Compilation of References .............................................................................................................. 430

About the Contributors ................................................................................................................... 467

Index ................................................................................................................................................... 471

Foreword ............................................................................................................................................ xiv

Preface ................................................................................................................................................ xvi

Compilation of References .............................................................................................................. 430

About the Contributors ................................................................................................................... 467

Index ................................................................................................................................................... 471

The primary audience of this book is intended to be:

Multiphase Continuum Formulation for Gas-Solids Reacting Flows

Hydrodynamic Equations from Kinetic Theory: Fundamental Considerations

Kinetic Theory for Granular Materials: Polydispersity

Interfacial Interactions: Drag

Mass and Heat Transfer Modeling

Coupled Solvers for Gas-Solids Flows

Quadrature-Based Moment Methods for Polydisperse Gas-Solids Flows

Direct Numerical Simulation of Gas-Solids Flows Based on the Immersed Boundary

The MP-PIC Method for Dense Particle Flows

Circulating Fluidized Beds

CFD Modeling of Bubbling Fluidized Beds of Geldart A Powders

Computational Modeling of Gas-Solids Fluidized Bed Polymerization Reactors

Validation Approaches to Volcanic Explosive Phenomenology

2. gas-sOlids cOntinuuM MOdel

2.1. Mixture theory

2.2. averaging Methods

and of the solids (considering only one solids-phase—M=1):

Similarly, we can obtain the averaged gas velocity:

and solids velocity as

3.1. equation of state

or the fluid is considered incompressible.

4.1. fluid-solids Momentum transfer

Force balance under terminal settling for single particle:

and for multi-particle

and the Reynolds number of the mth solids phase is given by

To ensure that the function Bnew is continuous, d must be related to c as follows:

4.2. solids-solids Momentum transfer

4.3. fluid-Phase stress tensor

4.4. solids-Phase stress tensor

Figure 4. Schematic of the plastic/transition/viscous granular rheology regime map as a function of

and the rate of strain tensor

where the strain rate tensor can be written as:

and the normal solids pressure is given as:

where ; ; ; , g0 stands for radial

distribution function and e stands for coefficient of restitution.

The collisional dissipation term is given as and the exchange terms

5. cOnservatiOn Of sPecies Mass

5.1. reaction Kinetics

Figure 5. Shrinking core model for coal combustion

where the Schmidt number is defined as

and the ash layer resistance is given by

where, De is an effective ash diffusivity given by (C.Y. Wen, et al., 1982)

and the ratio of core diameter to particle diameter

can be related to the solids mass fraction as

The surface reaction resistance is given by (Desai & Wen, 1978)

From equation 5.3, the other formation rates can be obtained as