You are on page 1of 221

Development of a Sizing Equation for a Multi-Stage Choke Valve

Trim

by
Andrew Grace

Submitted for the award of Doctor of Philosophy at the


University of Limerick
February 2011

Supervisor:
Dr. Patrick Frawley
Department of Mechanical and Aeronautical Engineering
University of Limerick

i
CONFIDENTIALITY RESTRICTION

A confidentiality restriction has been placed on this thesis at the request of Cameron Flow
Control Ireland. This restriction limits the distribution of this report to Cameron Flow Control
Ireland and internal / external examiners. This restriction will remain in place until the year
ending 2016

ii
Abstract

Choke valves are used to regulate the pressure from natural gas reservoirs. Traditional choke
valves control the reservoir pressure drop using a single variable orifice. At choked flow
conditions a shock wave forms in the vena contracta in the valve creating high downstream
velocities and shock cell turbulence interaction. The high velocities increase erosion when
there are sand particles in the gas. Shock cell turbulence interaction is a highly efficient noise
generation mechanism and can create noise far in excess of industrial limits.

Multistage (MS) technology is used to reduce high pressure in stages within a valve. Instead
of a single port the MS valve uses a flow path of sequential restrictions and expansions. The
segmenting of the pressure drop eliminates the presence of shock waves. This reduces the
velocity (and hence erosion) and changes the primary noise generation mechanism to less
efficient turbulent shear. A new MS valve geometry was developed by Cameron Flow
Control. As part of this development a flow equation was required to determine the restrictive
area in the valve (referred to as valve coefficient or Cv) which is necessary to control a set of
reservoir conditions. The Cv of the valve is a function of its internal geometry and is
complicated by the nature of the MS flow path.

A mathematical model of flow through the MS path was constructed based on a series of
sequential thick walled orifice plates. The choked flow conditions for the single stage
geometries were investigated and linked to the gas expansion factor (Gy) and critical pressure
drop ratio (c). Experimental data and theory taken from (Rhode, 1969) was used to estimate
a Cv value for the overall flow path for a series of different pressure differentials. The fluid
properties were modelled using a suitable gas compressibility equation (Peng et al, 1976) and
a Joule Thomson relationship (Bessieres et al, 2006) to account for changes in the expansion
zones. The mathematical model did not produce a choked flow condition.

Experimental tests were conducted using a model of the MS flow path, a mass flow loop and
a Laser Doppler Anemometry (LDA) measurement system. The mass flow rate tests showed
that the rate of change of flow rate reduced significantly at high pressure drops without the
gas becoming choked. The LDA velocity measurements indicated the existence of three flow
phenomena within the MS path. Computational Fluid Dynamics (CFD) was used to
investigate the mechanism that caused the reduction in flow rate. Both the mass flow rate and
LDA velocity measurements were used to benchmark the CFD simulations. Three large
vortices were proven to exist in the restrictive channels and their size and location were
shown to ultimately limit the effective flow area.

A full flow test was conducted on the MS valve to finalise the sizing equation and account for
any upstream or downstream geometrical effects caused by the valve body. At the outlet of
the MS paths, which exited into the same volume, further vortices were seen. These proved to
further reduce the overall flow rate of the valve.

The primary outcomes of this research were the design of an effective MS valve (EU Patent
Number: IB2008053368). As part of this, a new limiting flow mechanism was described and
included as part of the sizing equation. In addition, a specific case where high inlet pressures
created shock waves at the outlet was presented. Furthermore this research detailed the first
use of LDA velocities measurements taken within a MS valve with a full scale design.

This research expands on the state of the art knowledge of valve sizing and design.

iii
DECLARATION

This thesis is presented in partial fulfillment of the requirements for the degree of Doctor of
Philosophy in Engineering. The substance of this thesis is the original work of the author and
due reference and acknowledgement has been made, when necessary, to the work of others.
No part of thesis has previously been accepted for any degree nor has it been submitted for
any other award.

Signed: Date: 14/4/2011

Andrew Grace, C.Eng (Candidate)


Dept. of Mechanical & Aeronautical Engineering
University of Limerick

Signed:_______________ Date:_____________
Dr. Patrick Frawley (Supervisor)
Dept. of Mechanical & Aeronautical Engineering
University of Limerick

iv
ACKNOWLEDGEMENTS

Ultimately this work was possible due to the patience of my wife Jill, who sacrificed many, many,
many weekends so I could spend time learning new things. I know I will spend every weekend
henceforth making it up to her.

I would also like to thank my parents, Greta and Mick, who started this all 30 years ago. Their support
throughout my life gave me all the tools I needed to understand the work I have undertaken. In turn I
can only hope to pass on their knowledge.

Pat Frawley deserves a special mention for his efforts in expanding my horizons. At times it must
have seemed like he was steering a rather large, unwieldy tug boat.

Also all my colleagues in Cameron who have come and gone over the years need mentioning. Some
gave good technical advice, some bought the drinks, others gave direction when none was apparent
and all have some part in this work. In no particular order; Britt Schmidt, Dave Quin, Martin
ODonnell, Declan Elliot, Martin Meehan, Collin Matthews, Stephen Chambers, Eddie McHugh, and
Ray Smyth.

v
Contents

Tables of Contents
Page
Confidentiality Restriction ii
Abstract iii
Declaration iv
Acknowledgements v
Table of Contents vi
List of Figures ix
List of Tables xiii
Nomenclature xv
Abbreviations xx

Section Title Page


Chapter 1- Introduction
1.1 Introduction 1
1.2 Control Choke Location and Function 1
1.3 Control Choke Anatomy 2
1.4 MS Trim Technology 4
1.5 Definitions and Service Conditions 5
1.6 CFD Modelling 7
1.7 LDA Benchmarking and Cv Flow Test 8
1.8 Objectives 8
1.9 Methodology 9

Chapter 2- Empirical Analysis


2.1 Introduction 11
2.2 Ideal Restriction and Quasi One Dimensional Flow 11
Valve Coefficient (Cv) and the Orifice Plate Meter Coefficient
2.3 15
(Cm)
2.4 Single Stage- Gas Expansion Factor, Gy 18
2.5 Multistage Flow Path Model Simplification and Assumptions 22
2.6 First Pass Calculation 25

vi
Contents

2.6.1 MS Path Cd 26
2.6.2 Compressibility Factors 33
2.6.3 Temperature Variation Due to Expansion 36
2.6.4 Calculations and Discussion 40
2.7 Summary 42

Chapter 3- Experimental Investigation and Benchmark Data


3.1 Introduction 44
3.2 LDA Overview 44
3.3 Experimental Analysis 49
3.3.1 LDA Apparatus 49
3.3.2 MS Flow Oath Models 54
3.3.3 Flow Rig Components 60
3.4 LDA Literature Review 61
3.5 Non-Dimensional Analysis 65
3.6 Mass Flow Rate Measurements and Discussion 66
3.7 LDA Velocity Measurements and Discussion 68
3.8 Summary 77

Chapter 4- Computational Fluid Dynamics and Turbulence Model


4.1 Introduction 78
4.2 Turbulence Model 78
4.3 Turbulence Model Application Review 79
SST Turbulence Model Closure Equations and Near Wall
4.4 86
Treatment
4.5 CFX EOS Equations 89
4.6 Grid Independence 89
4.7 Heat Transfer Model 91
4.8 Boundary Conditions and Convergence Criteria 91
4.9 Mass Flow Rate and LDA Benchmark Results 94
4.10 Mesh Independence Results 103
4.11 Mass Rate Discussion 104
4.12 LDA Benchmark Discussion 105

vii
Contents

4.13 Choked Flow Mechanism 111


4.13.1 Choked Flow at Experimental Conditions 112
4.13.2 Choked Flow at Reservoir Conditions 115
4.14 Conclusions 119
4.15 Summary 119

Chapter 5- MS Trim Flow Tests


5.1 Introduction 121
5.2 Flow Test Literature Review 121
5.3 MS Valve Flow Test 132
5.3.1 Test Apparatus 132
5.3.2 Flow Test Procedure 134
5.3.3 Flow Test Results and Conclusions 138
5.4 Summary 142

Chapter 6- Discussion and Conclusions


6.1 Introduction 143
6.2 Summary and Discussion 143
6.3 Conclusions 150
6.4 Original Contributions 153
6.5 Potential for Further Work 154

Chapter 7- References 155


Appendix A 166
Appendix B 178
Appendix C 180
Appendix D 185
Appendix E 195

viii
Contents

List of Figures

Figure No. Title Page


Chapter 1- Introduction
1.1 Photograph of a typical low pressure surface x-tree (Cameron FLC) 3
1.2 Overview of major valve components (Cameron FLC) 3
1.3 Trim details for a single stage valve (Cameron FLC) 4
1.4 Sectional view of a multistage stack trim (Cameron FLC) 4
1.5 FLC two plane multistage flow path with expansion zone (Cameron
5
FLC)
1.6 Required Cv valves plotted on various inherent flow characteristics 7
1.7 Research methodology
10

Chapter 2- Empirical Analysis


2.1 Convergent-divergent nozzle with M varying with area (Anderson
12
2003)
2.2 Convergent-divergent nozzle with equations 2.2 - 2.4 plotted and
13
values at the throat noted (Anderson 2003)
2.3 Thin lip orifice plate with converging and diverging flow stream 14
2.4 Thick walled orifice plate VC location (Perry, 1949) 14
2.5 Barrel shock / under-expanded jet formed below the critical break
15
pressure point (Crist et al, 1966)
2.6 Change in Cm based on varying orifice geometry defined by
17
(Essom, 2007)
2.7 Distance to VC in pipe diameters defined by (North
17
American Manufacturing, 1989)
2.8 Plot of gas expansion factor (Y) versus pressure drop ratio (x) for a
21
number of valves with the limit of Y = 0.667 (IEC 60534-2-1)
2.9 Early Prototype models for the MS path 22
2.10 Multistage Path Geometry with throttling sections and expansion
23
zones
2.11 Simplified multistage flow path with increasing throttling and
24
expansion sections
2.12 Flow chart of theoretical calculations for Cv 26
2.13 Thick walled orifice plate test specimen as used by (Rhode, 1969) 26
2.14 Plot of Discharge Coefficient Versus Velocity Head for = 1.05,
28
1.6, 2.0, 2.83 and 4.0
2.15 Plot of Cd versus , for = 4.0, up to the choked flow point 30
2.16 Plot of Cd versus , for = 1.0, up to the choked flow point 31

ix
Contents

2.17 Nelson-Obert compressibility plot based on 25 different pure gases,


34
reproduced and simplified for clarity from (Cengel et al, 2004)
2.18 Curves for gas mixture GC1L (92% methane) plotted for both PR
38
and SRK EOS (Kortekass, 1997)
2.19 Temperature variations for all natural gas mixtures against pressure
38
drop for a fixed upstream pressure of 1,000 bar (Kortekass, 1997)
2.20 Comparison of JTIC obtained for; experimental, Monte Carlo and
40
other EOS Models (Bessieres et al, 2006)
2.21 Mass flow rate versus pressure drop ratio, , using mathematical
41
analysis

Chapter 3- Experimental Investigation and Benchmark Data


3.1 Light intensity scattered from a small particle, (Durst et al, 1976) 45
3.2 Burst signal for a single particle in a LDA probe volume with 45
Gaussian intensity pedestal and Doppler signal (Drain, 1980)
3.3 Dantec Fiber-Flow system with Spectra-Physics 4W Argon-Ion 50
laser
3.4 Dantec BSA-F70 LDA Signal Processer 50
3.5 Fringe pattern for two component velocity measurement formed by 51
blue and green light (Drain, 1980)
3.6 LDA Apparatus with Test Model In-Situ 51
3.7 Optical Layout and Resultant Beam Angle (Dantec Dynamics) 52
3.8 Details of Probe Volume (Dantec Dynamics) 53
3.9 Multi-plane MS Flow Path 55
3.10 Trigonometric relations for refractive index and probe volume shift 55
3.11 Refraction of Laser and resultant probe volume location 56
3.12 Test Model A- MS Path Lower Section 57
3.13 Test Model B- MS Path Upper Section 57
3.14 Test Model Assembly 58
3.15 Test Model Assembly continued 58
3.16 TMA Throttling Sections and Measurement Locations 59
3.17 TMB Throttling Sections and Measurement Locations 59
3.18 Valve Flow Rig (IEC 60534-2-3) 60
3.19 Calibration Data for Annubar 61
3.20 Optical Layout of Ruggerini four stroke piston with optical access 62
(Amato, 1990)
3.21 Variation in tangential velocity, RMS and Turbulence Intensity at 63
measurement point 15mm for varying crank angle (Amato, 1990)
3.22 Variation in velocity RMS across cylinder radius (encompassing the 64
5 measurement points) (Amato, 1990)

x
Contents

3.23 Mass Flow Rate Versus Pressure Drop Ratio for theoretical model 68
and experimental results
3.24 TMA-1 Resultant Velocities Versus Distance Along Restriction 69
3.25 TMA-2 Resultant Velocities Versus Distance Along Restriction 70
3.26 TMA-3 Resultant Velocities Versus Distance Along Restriction 70
3.27 TMB-1 Resultant Velocities Versus Distance Along Restriction 72

3.28 TMB-2 Resultant Velocities Versus Distance Along Restriction 73


3.29 TMB-3 Resultant Velocities Versus Distance Along Restriction 73
3.30 TMB-4 Resultant Velocities Versus Distance Along Restriction 74
3.31 Turbulence Intensity values along the restrictive channels in TMA 75
3.32 Turbulence Intensity values along the restrictive channels in TMB 76

Chapter 4- Computational Fluid Dynamics and Turbulence Model


4.1 Diffuser Geometry under investigation (El-Behery, 2009) 80
4.2 Turbulent kinetic energy (k) versus non-dimensional distance from
81
the diffuser chamber, (El-Behery, 2005)
4.3 Axial velocity profile for RANS models and Buice-Eaton
81
experimental data, (El-Behery, 2005)
4.4 Impact zone in the expansion chamber for the MS path 83
4.5 Simplified impingement model for cool turbulent jet impacting on a
83
hot surface, (Celic, 2006)
4.6 k- turbulence model versus experimental data (Celic, 2006) 83
4.7 v2-f turbulence model versus experimental data downstream of the
84
impingement zone, (Celic, 2006)
4.8 Comparison of turbulences models and LDV data, (Dalbello et al,
85
2005)
4.9 Comparison of SST model for three different mesh gradients,
85
(Dalbello et al, 2005)
4.10 Velocity gradient location for mesh independence validation 90
4.11 Comparison of LDA Sim 1 and LDA rig mass flow rate versus 95
4.12 MS path with LDA measurement plane 96
4.13 TMA-1 LDA and CFD Mean Velocity Comparison, x = 0.751 97
4.14 TMA-1 LDA and CFD Mean Velocity Comparison, x = 0.803 97
4.15 TMA-2 LDA and CFD Mean Velocity Comparison, x = 0.751 97

xi
Contents

4.16 TMA-2 LDA and CFD Mean Velocity Comparison, x = 0.803 97


4.17 TMA-3 LDA and CFD Mean Velocity Comparison, x = 0.751 98
4.18 TMA-3 LDA and CFD Mean Velocity Comparison, x = 0.803 98
4.19 TMB-1 LDA and CFD Mean Velocity Comparison, x = 0.714 98
4.20 TMB-1 LDA and CFD Mean Velocity Comparison, x = 0.75 98
4.21 TMB-2 LDA and CFD Mean Velocity Comparison, x = 0.714 99
4.22 TMB-2 LDA and CFD Mean Velocity Comparison, x = 0.75 99
4.23 TMB-3 LDA and CFD Mean Velocity Comparison, x = 0.714 99
4.24 TMB-3 LDA and CFD Mean Velocity Comparison, x = 0.75 99
4.25 TMB-4 LDA and CFD Mean Velocity Comparison, x = 0.714 100
4.26 TMB-4 LDA and CFD Mean Velocity Comparison, x = 0.75 100
4.27 Velocities profiles for the standard and refined mesh 103
4.28 Pressure contour plot for TMA and TMB from LDA Sim 1 104
4.29 Flow through a poppet valve with vortices in the outlet region 112
(Bernard et al, 2007)
4.30 Primary vortex structures in TMA at experimental conditions 114
4.31 Small vortex structures in TMB at experimental conditions 114
4.32 JTIC Results from (Bessieres et al, 2006) with additional notation 116
4.33 Temperature contour for inlet conditions of 300K at 47MPa and = 117
0.806
4.34 CFD Mach Number contour at outlet for inlet conditions of 300K 118
at 47MPa and = 0.806
4.35 Interaction of a shock wave and turbulent flow (Zank, 2002) 118

Chapter 5- MS Trim Flow Tests


5.1 Multistage Valve Trim as used in (IEC 60534-5-7) 122
5.2 A Full MS Path Disc Assembly in a Choke Valve Body 123
5.3 Flow Test Apparatus (IEC 60534-2-3) 123
5.4 c Characteristic for a Single Stage Valve 126
5.5 Alternate Flow Test Apparatus (Wu-Shung Fu, 1998) 127
5.6 Reservoir Pressure Reduction(Wu-Shung, 1998) 129
5.7 Wu-Shung Choked Flow Check (Wu-Shung, 1998) 130
5.8 Review of Wu-Shung Cv and Gy Results (Wu-Shung, 1998) 131

xii
Contents

5.9 Two discs compressed together making one valve open position 132
5.10 Piping Diagram for MS Valve Flow Test 133
5.11 MS Valve Installed in the NEL Test Facility 133
5.12 GyCv Versus for Four Valve Positions 138
5.13 A comparison of valve annular area versus MS trim opening 139
5.14 Pressure contour for MS trim in valve assembly 139
5.15 Velocity vector plot through the MS trim assembly with 141
downstream vortices
5.16 Cv versus stem travel inherent characteristic for the MS trim 141

Chapter 6- Discussion and Conclusions


Required Cv plotted on the MS trim inherent characteristic for the 150
6.1
parameters in table 7.1

List of Tables

Table No. Title Page


Chapter 1- Introduction
1.1 Typical composition of natural gas in operational fields 6
1.2 Operating Conditions for an Algeria Surface Gas Well 7

Chapter 2- Empirical Analysis


2.1 Rate of change of throttling width and expansion diameter 23
2.2 Values of Cd for various values of x, at inlet Mach number = 0.07 32

Chapter 3- Experimental Investigation and Benchmark Data


3.1 Probe Volume Dimensions 53
3.2 Test Conditions for TMA 69
3.3 Test Conditions for TMB 72

Chapter 4- Computational Fluid Dynamics and Turbulence Model

xiii
Contents

Addition variables for the general transport equation for the SST 88
4.1
turbulence model
4.2 SST Turbulence Model constants 88

4.3 Summary of Boundary Conditions for CFD Simulations 92

4.4 Mean Velocity Calculation from LDA Sim 1 at TMA-1a 96

4.5 LDA Versus CFD Turbulence Intensity Values for TMA 101

4.6 LDA Versus CFD Turbulence Intensity Values for TMB 102

4.7 LDA and CFD Comparison For TMA RUN 1 and RUN 2 106

4.8 LDA and CFD Comparison For TMA RUN 3 and RUN 4 107

4.9 LDA and CFD Comparison For TMB RUN 1 and RUN 2 110

4.10 LDA and CFD Comparison For TMB RUN 3 111

Chapter 5- MS Trim Flow Tests


5.1 Flow Test Measurements 135

5.2 Comparison of CFD and Experimental Mass Flow Rates 136

5.3 Test (B) Gy Characterization for Four Valve Positions 137

Chapter 6- Discussion and Conclusions


6.1 Sizing Results for Table 1.1 150

xiv
Contents

Nomenclature

Letters

Symbol Description Unit


A1 Upstream area in a restriction m2
A2 Smallest area in a restriction (throat or port area) m2
CA Peng Robinson Cubic Coefficient -
Ca Peng Robinson molecular interaction factor -
CB Peng Robinson Cubic Coefficient -
Cb Peng Robinson molecular interaction factor -
Cd Coefficient of Discharge -
CDk Constant for k- turbulence model -
Cm Meter coefficient -
CMS MS Trim mass flow rate correction factor -
Ck Karman Constant -
Cp Heat capacity at constant pressure J/K

CT Constant to Account for Dimensional Analysis -


Cv Valve coefficient -
C Coefficient of thermal expansion -

C Constant for Darcy-Weisbach equation -


C1 Closure Constant for the k- Turbulence Model -
C2 Closure Constant for the k- Turbulence Model -
D Diameter of pipe for Darcy-Weisbach equation m
DL Beam diameter m
E Beam separation m
En Expansion zone reference number, n = 1 to 6 -
F Focal length m
FL Pressure recovery factor -
F1 Blending function for k- turbulence model -

xv
Contents

F2 Blending function for k- turbulence model -

F Specific heat ratio factor (the ratio of the ratio of specific heats -
of a fluid to that of air)
Gy Gas expansion factor -
IT Turbulence Intensity %
KL Kolmogorov Length Scale m
KT Kolmogorov Time Scale s

KV Kolmogorov Velocity Scale m/s


L Length of pipe for Darcy-Weisbach equation m
Lf Distance between fringes for LDA m
Lh Duct Height m
Ls Length Scale for Turbulence m
L1 Angle between source, particle position 1 and particle position 2 degrees
for LDA
L2 Angle between observer, particle position 1 and particle position degrees
2 for LDA
M Mach number -
NF Number of fringes -
Nu Nusselt number- the ratio of convective to conductive heat -
transfer across a boundary
P Pressure kPa
Po Static Pressure kPa
P1 Upstream pressure in a restriction kPa
P2 Downstream pressure in a restriction kPa

Pc Critical Pressure kPa

Pr Reduced Pressure kPa


Pn Prandtl Number -
PnT Turbulent Prandtl Number -
Pvc Pressure at the vena contracta kPa

xvi
Contents


 k- turbulence model constant -

Q Volumetric flow rate m3/hr

U Resultant Velocity m/s


Ua Speed of sound m/sec
UVC Velocity in the VC m/sec
Ux Velocity in the x direction (for LDA) m/sec
Uy Velocity in the y direction (for LDA) m/sec
R Universal Gas Constant JK-1mol-1
Re Reynolds Number -
Rn Restriction reference number , n = 1 to 7 -

Rp Turbulent Production Term -


S Distance along MS path restriction m
Sij Rate of Deformation s-1
Su Momentum source term (Body Forces) N/m3

ST Energy source term (External Energy Sources) J/m3


Sv Momentum source term (Body Forces) -
Sw Momentum source term (Body Forces) -
S Transport equation source term -
T Temperature K
Tc Critical Temperature K

To Static Temperature K
Tr Reduced temperature K
V Velocity vector -
Vm Volume m3
VH Velocity head ratio -
W SST turbulence model constants -

W SST turbulence model constants -


Z Compressibility Factor -
Z Soave compressibility function -

xvii
Contents

Z Peng Robinson acentric factor -

c Velocity of a wave m/sec


d1 Upstream diameter in a restriction m
d2 Smallest diameter in a restriction (throat or port) m
dp Diameter of LDA seeding particle m

f Frequency Hz
fr Darcy-Weisbach friction factor -

f Change in frequency Hz
h slope of a line -
k Turbulent kinetic energy J
m Mass flow rate kg/hr
mp Mass of a LDA seed particle kg
mw Molar mass kg

n Refractive index -
n Number of wavelength arriving at an observer per unit time -
u x component of velocity m/sec
up Velocity of a LDA sees particle m/sec
urms RMS of mean flow velocity m/sec
uv Viscous Velocity m/sec
 Arithmetic mean of velocity m/sec

u Turbulence Intensity -
u+ Non-dimensional velocity term -
tp Response time of a LDA seed particle s
ts Time Scale for Turbulence s
tc Time scale for Convective Turbulent Scale s
td Time scale for Diffusive Turbulent Scale s
v y component of velocity m/sec

xviii
Contents

vs Velocity Scale for Turbulence s


v Kinematic viscosity Pa.s
vT Turbulent Kinematic viscosity Pa.s
w z component of velocity m/sec

Greek Symbols

Symbol Description Unit


Velocity of approach (d 2 / d1) -
Transport equation diffusion term -
Ratio of Specific Heats (Cpressure / Cvolume)
x LDA Probe Volume x axis length m
y LDA Probe Volume y axis length m
z LDA Probe Volume z axis length m
Rate of turbulent dissipation (energy per unit mass) m2/s3
Ratio of restriction wall thickness to port diameter -

Wavelength m
Wavelength of scattered light m
Viscosity Pa.s
T Turbulent Viscosity Pa.s
JT Joule Thomson Coefficient -
1 Upstream density kg/m2

p Density of a LDA seed particle kg/m2


k SST turbulence model constant -
u Standard deviation of a LDA mean velocity measurement -
T Constant for turbulence model -
Constant for k- turbulence model -
Constant for k- turbulence model -

xix
Contents

 
-
Pressure drop ratio 


c Critical pressure drop ratio -


n Pressure drop ratio per stage, where n is the stage -
Transport equation property term -
v Kinematic viscosity m2/s

vT Kinematic turbulent or eddy viscosity m2/s


Specific rate of turbulent dissipation (energy per unit mass) m2/s3kg

Abbreviations

Notation Meaning

CFD Computational Fluid Dynamics

DNS Direct Numerical Simulation

EOS Equation of State

FLC Cameron Flow Control

JT Joule Thomson

JTIC Joule Thomson Inversion Curve

LDA Laser Doppler Anemometry

LES Large Eddy Simulation

MS Multi-Stage

NS Navier Stokes

PR Peng Robinson

SNR Signal to Noise Ratio

SRK Soave Redlich Kwong

SST Shear Stress Transport

VC Vena Contracta

xx
Chapter 1 Introduction

CHAPTER 1

INTRODUCTION

1.1 Introduction

This chapter is an introduction to the function and anatomy of a control choke valve. This
includes an overview of current single stage technology and the application of multistage
(MS) technology. In addition the methodology and chapter objectives for this thesis are
outlined.

The work undertaken in this thesis details the development of a flow equation for a MS
control choke valve. The MS control choke valve (EU Patent Number: IB2008053368, see
appendix A) was developed by the Flow Control (FLC) division of Cameron International.
The flow equation is used to calculate a non-dimensional valve coefficient (Cv) based on a set
of reservoir parameters. This equation is being used on a daily basis to design MS valves for
oil and gas production fields. The research was conducted under the auspices of the John
Holland Research Centre (JHRC).

1.2 Control Choke Location and Function

A control choke valve (usually shortened to choke) is a specific type of control valve used in
heavy industries like oil and gas production. It is the primary control point in the production
system and is located on the production block, known as the Christmas tree (x-tree). The x-
tree is connected to the well bore that is drilled through the upper layer of rock capping an oil
and gas reservoir. A number of gate valves are located in the well bore and on the x-tree to
give various shut-off points. However, the first variable restrictive area in the system is the

1
Chapter 1 Introduction

choke. Figure 1.1 details a typical x-tree layout with the choke position highlighted. The
research presented herein is concerned explicitly with natural gas production.

As the first variable restriction in the system, the choke controls the flow rate from the well.
In addition it must throttle the reservoir pressure during start-up. At start-up the pressure drop
ratio (), across the choke (given by equation (1.1)) can be in excess of 0.6. For gas flow, a
pressure drop ratio greater than 0.6 will typically result in a choked flow condition. This
choked condition is detailed by (Anderson, 2001) with reference to an ideal restriction (quasi-
one dimensional analysis). The correct design of a choke and its controlling characteristic is
vital in ensuring a correctly regulated reservoir pressure and controlled production rate, as
presented by (Moore, 1976). Choke selection is termed sizing and equates the chokes open
restrictive area to flow rate, based on internal geometry and operating conditions. An
incorrectly sized choke will either provide no controllability at high pressure drops, which
can cause damage to the reservoir structure, or it will restrict production capacity at higher
flow rates.

 


(1.1)

1.3 Choke Anatomy

The choke consists of a pressure containing body, a variable area control section (referred to
as the trim) and an actuator as detailed in figure 1.2. The trim design is dependent on
application but generally consists of a plug moving linearly in a cage which has flow ports
that are uncovered to regulate flow. The cage design takes the operational pressure drop in a
single stage across the flow ports. The size and location of these ports dictate the inherent
characteristic of the trim as they are uncovered by the plug (as defined by IEC 60534-2-4).
For any specific set of operating conditions the total area open for the ports is determined
through sizing.

2
Chapter 1 Introduction

Figure 1.1 Photograph of a typical low Figure 1.2 Overview of major valve
pressure surface x-tree (Cameron FLC) components (Cameron FLC)

Flow through the choke is via the inlet flange and then into an annular volume around the
cage which the flow envelopes. The flow is throttled in the cage ports which are diametrically
opposed so that the fluid jets impinge on each other in the centre in a highly turbulent zone,
as detailed in figure 1.3. The impingement of the flow jets is the primary energy
transformation process in pressure reduction, generating both heat and noise as presented by
(Peters, 1987). In the valve outlet there is pressure recovery and alignment of the flow occurs
further downstream in the piping.

In addition to single stage cage trims, there are MS stack trims. Instead of a ported cage, the
MS stack typically has a flow path machined into a disc. The pressure drop is taken across the
path in stages, based on its geometry. The individual discs are compressed in the valve body
or unitised together to form the stack. Control is still produced by a plug actuating linearly in
the trim to uncovered flow paths as required. A section view of the FLC MS stack is given in
figure 1.4.

3
Chapter 1 Introduction

Figure 1.3 Trim details for a single stage Figure 1.4 Sectional view of a MS
valve (Cameron FLC) stack trim (Cameron FLC)

1.4 MS Trim Technology

Due to their geological formation, natural gas reservoirs tend towards high internal pressures
(up to 172MPa and typically between 34-68MPa). Over the life of the well the pressure tends
to deplete but full bore pressure can exist for many years depending on the total reservoir size
(Bai et al, 2010). Transportation pipelines downstream of the production block operate at
much lower pressures than the x-tree. As the primary pressure transition point between the
reservoir and transportation line, the choke will typically be operating in a choked flow
condition for extended periods. The choking mechanism in a ported trim design is due to the
formation of a shock wave in the vena contracta (VC). At high pressure drops the shock can
extend further downstream of the restriction and interact with the high turbulence zone in the
centre of the cage as detailed by (Ribner, 1997) and (Zank, 2002). The shock cell turbulence
interaction is the primary mechanism for noise generation in the valve. The resultant noise
usually exceeds industrial safety limits (Atmaca et al, 2005).

Noise reduction can be achieved by the use of a MS trim where the pressure drop is balanced
over a number of throttling sections. At each throttling section the gas velocity is reduced and
the primary noise generation mechanism becomes turbulent shear. When transmission losses

4
Chapter 1 Introduction

through the valve body and downstream piping are taken into account, the effective noise is
unlikely to exceed 85 dBA.

The sizing of MS trims is not governed by any industry standard and the assumptions given
for single stage analysis do not hold. The research presented here investigated the parameters
required to size a specific MS geometry (designed by FLC) for an industrial gas application.
Figure 1.5 details a single flow path from the FLC MS trim, with flow direction as shown.
The path is machined on two levels with gas expansion occurring in the cylindrical chambers
followed by throttling in the rectangular sections. Both expansion chambers and throttling
sections increase in volume and cross sectional area respectively as the flow path progresses.
Chapter 2 details these rates of change and the affect on gas density.

Figure 1.5 FLC two plane MS flow path with expansion zone (Cameron FLC)

1.5 Definitions and Service Conditions

The primary variable in valve sizing is referred to as the valve coefficient (Cv), as given in
IEC 60534-1. Cv relates the effective flow rate through a restriction (valve open area) to the
pressure drop across it, taking into account surrounding geometry. A valve trim can be flow
tested to determine its rated Cv but an application must be sized to resolve the required Cv.
Chapter 2 further details the parameters of Cv.

The critical (choked) pressure drop ratio (c) defines the limiting flow rate through the valve
for any rated Cv. Values of may exceed c, but the effective flow through the orifice shall be
limited by c. At this pressure drop ratio a shock wave will have formed in the VC which
limits the velocity (and hence flow rate) of the gas at the smallest cross sectional area of flow.

5
Chapter 1 Introduction

The composition of natural gas is primarily methane with some higher order structures
including ethane, propane, butane, pentane and some contaminants including CO2 and H2S as
detailed in table 1.1. The geological conditions associated with natural gas formation mean
deposits are found at much greater depths than oil. As gas exploration continues, new wells
are at increasingly greater depths. Reservoir pressure is a function of depth and at higher
pressures natural gas tends to pure methane, as detailed in (Paull et al, 2000). For this
research the gas is considered to be 100% methane which is more reflective of developing
fields than current production compositions. Flow testing was conducted with clean air. The
governing equations used a specific heat ratio factor to adjust the air test results to methane
applications (see chapter 2.0 for further details).

Component Chemical notation %


Methane CH4 70-90%
Ethane C2H6
Propane C3H8 0-20%
Butane C4H10
Carbon Dioxide CO2 0-8%
Oxygen O2 0-0.2%
Nitrogen N2 0-5%
Hydrogen sulphide H2S 0-5%
Rare gases A, He, Ne, Xe trace
Table 1.1 Typical composition of natural gas in operational fields

Table 1.2 details a typical field application for a gas reservoir. It is seen in table 1.2 that the
inlet pressure is reservoir dependant while the outlet pressure is governed by the downstream
processing system. The flow rate is regulated by the valve position based on Cv for each of
the pressure ranges. The required Cv calculated from the field data is plotted against the rated
Cv for the trim (see inherent characteristic). The resultant plot shows the controllability and
thus suitability for a given trim versus the application, see figure 1.6 (Bauman, 1998).
Controllability is given by a +/- 10% range on each required Cv but can also depend on the
nature of the trim design.

6
Chapter 1 Introduction

Figure 1.6 Required Cv valves plotted on various inherent flow characteristics (Bauman,
1998)

CASE NAME Well Start-Up Early Regulation Operating Restart from Shut-In

Gas Flow Rate (kg/hr) 42,127 52,136 125,132 38,231

Gas Molecular Weight (Kg/Kmol) 19 19 19 19

T1, Temperature (C) 120 98 65 120

Reservior Pressure, (Bar A) 334 318 285 336

Downstream Pressure, (Bar A) 38 80 120 53

dP(P1-P2), (Bar A) 296 238 165 283

Table 1.2 Operating Conditions for an Algeria Surface Gas Well

1.6 CFD Modelling

MS geometries by nature are complex with no space for alignment or recovery of flow. While
classical theory can be applied to simplified geometries with assumptions, the full analysis of
a MS path requires computational investigation. The primary purpose of CFD will be to
examine the rate of expansion of the gas through the various stages of pressure reduction and
the resultant temperature loss and increase in velocity. As such both the equation of state for
a real gas and Joule Thomson (JT) cooling will be of concern in the model. The affect of both
expansion and temperature will have a direct effect on the sequential stages in the MS path
and hence the overall Cv. Furthermore CFD will be used to determine the critical pressure
drop ratio (c) and the mechanism for choked flow in the restriction.

7
Chapter 1 Introduction

1.7 LDA Benchmarking and Cv Flow Tests

Laser Doppler Anemometry (LDA) is a non invasive technique for measuring point velocities
in fluid systems. The principle is based on the change in frequency of a light source to an
observer, as particles travel between source and observer. A LDA analysis was completed on
a single MS flow path as part of the original development work. The analysis was conducted
primarily as a benchmark for CFD models but also for the comparison of a number of
different conceptual flow paths.

In addition to LDA a flow test was complete on a full valve assembly which contained a
complete MS stack in order to measure the rated Cv versus stem travel. Due to the nature of
the valve and limitations on test equipment there was no suitable governing standard for the
test. A new test procedure was specified which also covered the testing of c.

1.8 Objectives

The correct sizing of a MS choke is vital to the control of a gas reservoir. The determination
of the critical pressure drop ratio (c) and the choking mechanism define the operational
limits of a MS choke. This thesis investigated and defined all the parameters required to
determine the required Cv for a specific application and the resultant controllability based on
rated Cv. Figure 1.7 details the complete research methodology with chapter overview and
results. In addition the following steps outline the thesis format;

Create a theoretical equation for the mass flow rate through a MS path based on
pressure drop using the following;
o Research into quasi-one dimensional flow theory and choked flow conditions
o Investigation of gas expansion equations and critical flow conditions for single
stage restrictions
o Definition of a simplified MS flow model with experimental data
o Review of compressibility equations
o Review of Joules Thomson cooling equations

Conduct experimental analysis on a single MS flow path to investigate internal


velocities and critical pressure drop ratio (c) using;
o LDA to measure point velocities and turbulence intensity
8
Chapter 1 Introduction

o Non-dimensional flow tests for mass flow versus pressure drop ratio ()
o Comparison to theoretical mass flow rate

Develop a computational model for the MS path to investigate the effects of reservoir
conditions (high pressure and temperature) on the choking mechanism and critical
pressure drop ratio (c) using;
o A suitable turbulence model (velocity being of primary interest based on the
LDA benchmark data available)
o Benchmarked CFD models
o Reservoir simulations with high pressure and temperature

Conduct a flow test on a complete valve assembly at nominal conditions in order to


derive a sizing equation for the MS choke that;
o Accounts for the choking mechanism investigated in the single flow path (this
will include the effects of gas expansion, internal cooling and gas
compressibility)
o Accounts for the possible reduced mass flow rate due to the geometry of a full
MS stack trim

1.9 Methodology
Figure 1.7 gives a detailed overview of the chapter layout and sections in relation to the key
results and steps for each segment of research.

9
Chapter 1 Introduction

Figure 1.7 Research methodology

10
Chapter 2 Empirical Analysis

CHAPTER 2

EMPIRICAL ANALYSIS

2.1 Introduction

The objective of this chapter is to derive an expression for the mass flow rate through the MS
path based on the pressure drop across it. Expressions for mass flow rate through single stage
restrictions were investigated, including values for c (which includes the gas expansion
factor, Gy), and the associated choked flow mechanism. The mass flow rate equation required
a simplified model of the MS path which used experimental values to equate vena contracta
(VC) area with path geometry. In addition, an inter-stage temperature reduction expression
(Joule Thomson cooling) and a compressibility equation were required. The mass flow rate
expression was used to plot mass flow versus .

2.2 Ideal and Real Restrictions

(Anderson, 2003) detailed the quasi-one dimensional analysis of an ideal convergent and
divergent restriction. The quasi-one dimensional simplification stated that the properties of
flow including temperature, pressure and velocity are only a function of area and the
properties do not vary across that area. Therefore, relationships between the cross sectional
area and flow properties could be derived. The quasi one-dimensional analysis of a restriction
is an ideal case, as there is no separation of fluid from the walls as the flow converges and
diverges. The area-velocity relationship (see equation (2.1)) was derived using the continuity
of mass and the assumptions of an adiabatic inviscid flow (where A is the variable area along
11
Chapter 2 Empirical Analysis

the restriction). M in equation (2.1) has four physical ranges. For M 0 equation (2.1)
describes an incompressible flow, for 0 M < 1 a decrease in area is associated with an
increase in velocity, for M > 1 an increase in velocity is associated with an increase in area
and for M = 1 the rate of change of area is zero. Applying the possible ranges of M to the
geometry in figure 2.1 it is evident that M is less than 1 in the convergent section but
increases as the area decreases, to the point where the smallest area exists (the throat) where
the maximum value of M is 1 (rate of change of area is zero). Downstream of the throat M
can be greater than 1 for the increasing area. Thus the critical flow condition is the limiting
velocity at the throat of M = 1.

 
  1 (2.1)
 

Figure 2.1 Convergent-divergent nozzle with M varying with area (Anderson 2003)

The complete analysis of a convergent-divergent duct or nozzle requires a numerical solution


but the simplification to quasi-one dimensional flow gives practical relationships that are
used in wind tunnel design and other applications. Based on M = 1 at the nozzle throat an
equation was derived which expressed M as a function of the area ratio (using continuity of
mass, known values at the throat and the stagnation density), see equation (2.2). Assuming
the downstream pressure and temperature in the nozzle were stagnant, relationships for the
pressure and temperature change through the nozzle were derived, see equation (2.3) and
(2.4). The change in M, P/Po and T/To were plotted through the nozzle using equations (2.2),
(2.3) and (2.4), see figure 2.2. From equation (2.3) it was evident that M within the nozzle
was dependent on the ratio of pressures rather than the actual pressure conditions. For air
where the specific heat is 1.4, the pressure ratio across the nozzle required to give a value of
M =1 at the throat was 0.528 and the temperature reduction ratio was 0.833 (the pressure drop

12
Chapter 2 Empirical Analysis

ratio (), see equation (1.1), at this condition was 0.472). Further increases in pressure drop
did not affect the convergent portion of the nozzle and M remained unity at the throat.
Oblique shocks formed at the exit further downstream of the throat. These flow phenomena
did not affect the mass flow rate through the nozzle or any associated sizing equation for a
single stage pressure drop.



     
 



  
1 

  (2.2)



     
!
1  
 (2.3)

"   
"!
1  
 (2.4)

Figure 2.2 Convergent-divergent nozzle with equations (2.2) (2.4) plotted and values at the
throat noted (Anderson 2003)

13
Chapter 2 Empirical Analysis

Choked flow in orifice plates is comparable to the ideal restriction detailed by Anderson. For
a thin lipped orifice, choked flow is achieved by a normal shock forming in the vena
contracta (VC), see figure 2.3. The VC forms due to flow separation from the walls creating a
convergent and divergent flow stream. Section 2.2 details the effect the geometry of the
orifice plate has on the location and size of the VC. Orifice plate geometry is primarily a
function of the ratio of port diameter to inlet diameter. Increasing the thickness of the orifice
plate reduces the effects of upstream geometry. Through a thick walled orifice the flow
streams become parallel and the VC forms at the end of the orifice. Varying the upstream
geometry does not vary the location of the VC (Perry, 1949), see figure 2.4. This will be
shown in experimental analysis presented in section 2.6.1.

Figure 2.3 Thin lip orifice plate with converging and diverging flow stream

Figure 2.4 Thick walled orifice plate VC location (Perry, 1949)

Further increases to pressure drop does not increase the flow rate through the orifice
restriction but it does change the shape of the normal shock. There are two distinct structures
14
Chapter 2 Empirical Analysis

evident when increasing pressure drop which are separated by the break point pressure (see
IEC 60534-8-3). Below the minimum choked flow pressure (c) the normal shock in the VC
extends downstream and a number of reflected shocks and shock cells form further
downstream. Energy is dissipated by shock cell turbulence interaction. Below the break point
pressure the normal shock in the VC has full extended downstream and formed a barrel shock
(or under-expended jet) as described by (Crist et al, 1966), see figure 2.5. The barrel shock is
a standing wave and is the most efficient mechanism for noise generation (Jury, 1988).

Figure 2.5 Barrel shock / under-expanded jet formed below the critical break pressure point
(Crist et al, 1966)

2.3 Valve Coefficient (Cv) and the Orifice Plate Meter Coefficient (Cm)

(Driskell, 1983), first introduced the concept of a valve coefficient (Cv) to quantify the
restrictive area of a valve in terms of the effective flow area. As Cv can be determined for any
valve type it can be used as a parameter for direct comparison between valve styles. Different
valve styles with identical flow areas can have different flow rates for comparable pressure
drops. Right angle valves are more efficient throttling devices as they give higher Cv values
for identical flow areas when compared with inline valves.

15
Chapter 2 Empirical Analysis

Driskel based the calculation of Cv on the orifice plate meter coefficient (Cm). For a
simplified restriction, as shown in figure 2.3, the convergence of the flow streams towards the
orifice creates a reduced downstream flow area known as the VC. By applying Bernoullis
equation, the volumetric flow rate for a liquid across the restriction can be given in terms of
the orifice geometry, see equation (2.5) (Massey, 1998). In equation (2.5) Cd is the coefficient
of discharge, accounting for the difference in ideal flow rate to actual flow rate, see equation
(2.6). For a simple single stage orifice equation (2.5) can be written in terms of mass flow
rate for liquid or gas by introducing the gas expansion factor, see equation (2.7), where Gy =1
for incompressible fluids. The gas expansion factor (Gy), accounts for the change in fluid
density across a restriction, for a compressible fluid.


# $ %& ' ( ) * +2  -. /0 (2.5)
 

(

1234567
$ 18(967
(2.6)


: ;< $ %& ' ( ) * +20   -.  (2.7)
 

(

For an orifice plate, Cm is defined as the product of Cd and the area terms of the restriction
geometry. Equation (2.7) can be reduced for an orifice plate meter to equation (2.8), where
the ratio of restriction diameter to inlet diameter is , see equation (2.9). As stated in section
2.2 is used to determine Cm and the location of the VC using experimental data. Figure 2.6
(Essom, 2007) details the change in Cm for fixed Reynolds number based on an increasing
value of . Figure 2.7 (North American Manufacturing, 1989) defines the distance of the VC
from the restriction based on a number of pipe-diameters. is commonly referred to as the
velocity of approach.

16
Chapter 2 Empirical Analysis

: $1 ;< +20   -.  (2.8)


Where; $1 $ =& > ?)@

A (2.9)


Figure 2.6 Change in Cm based on varying Figure 2.7 Distance to VC in pipe diameters
orifice geometry defined by (Essom, 2007) defined by (North
American Manufacturing, 1989)

Equation (2.8) equates the mass flow rate for a given orifice size based on the pressure drop
across it. A valve sizing equation needs to determine the orifice size or Cv, based on
knowing the application pressure drop and required flow rate. Therefore equation (2.8) can be
rearranged in terms of Cm. As a control valve in its simplest form is just a variable restriction,
Driskell proposed to change Cm to Cv, see equation (2.10). The geometrical affects on flow
rate were included in the general equation by using a valve specific gas expansion factor (Gy)
defined by the critical pressure drop ratio (c). Valves are flow tested (see Chapter 6) at a
series of stems positions to determine the rated Cv. In order to equate required Cv an equation
must be derived that relates pressure and mass flow rate to open area based on c and Gy.
There are a number of expressions for Gy for single stage restrictions.

1
$- B (2.10)
C +D  


17
Chapter 2 Empirical Analysis

2.4 Single Stage- Gas Expansion Factor, Gy

The gas expansion factor (Gy), accounts for the change in fluid density across a restriction,
for a compressible fluid. For an ideal restriction (as detailed in section 2.2) adiabatic
expansion is assumed and the adiabatic gas expansion factor is applicable, see equation 2.11
(Perry, 1999). The flow through most restrictions is not ideal and hence not adiabatic (Kegel,
1998). An empirical equation for Gy is used instead.


 
G G

 H 
 I ? ) 
 G G 
;< E F
J (2.11)
  G G

H 
 ? ) I
G G

(Buckingham, 1933) and (Bean, 1935) defined Gy for a single stage orifice plate with flanged
pressure taps upstream and at the VC point, see equation (2.12), which was subsequently
adopted by ASME. F is the specific heat ratio factor which is the ratio of the specific heat
ratio (Cpressure / Cvolume) of the application fluid to that of air, which is the test fluid. To define
the Gy factor, Buckingham and Bean completed a series of discharge coefficient tests using
water on a fixed orifice at various pressure drops. The tests were repeated with gas on the
same orifice at the exact same pressure drops. The variance between the discharge
coefficients for both tests was the Gy factor. At low pressure drops, gas approaches an
incompressible fluid as there is little change in density through the orifice. Therefore Gy
varies between 1 and its lower limit, defined by c (choked flow condition).

P
1 2
(
G y = 1 0.41 + 0.35 4
F
P1
)

(2.12)

The nearest fit curve given by Buckingham and Bean was disputed in further analysis by
(Kinghorn, 1986). Kinghorn proposed that at high values of Gy (circa Gy = 0.95) that the
Buckingham and Bean coefficients were in error by as much as 0.5%.

18
Chapter 2 Empirical Analysis

More recent research into the ASME approved factor by (Seidl, 2002) confirmed the errors
first indicated by Kinghorn. Various tests were completed at values from 0 0.2 and the
expansion factor calculated as per the ASME method. It is important to note that at the
referenced Gy and values, the flow is laminar in an orifice plate. The least squares fit
derived from the Seidl paper resulted in equation (2.13). Seidl concluded that the errors
identified by Kinghorn have some merit while suggesting that further testing was required to
confirm the new interpretation of data.

P
1 2
(
G y = 1 0.357 + 0.557 4 )
F
P1

(2.13)

As the choke is the primary pressure drop point in a production system the typical pressure
differential tends to be above the range of Seidls testing. Research conducted by
(Cunningham, 1951) showed that the fixed critical limit for sharp edge orifice plates was
inconsistent at higher pressure drops. The conclusion was that with suitable corrections Gy for
non-critical flow could be used for thin edged orifice plates in all cases. Choked flow could
be expected for thick walled orifice plates where the thickness was greater than six times the
orifice diameter.

Additional analysis in the paper showed that the ASME definition of Gy was accurate for
values of P2 down as low as 0.63P1. After this point the discontinuity increased to range from
12% to 40% of actual flow. For outlet pressures lower than P2 = 0.63P1 the Cunningham
corrected ASME equation should be used, see equation (2.14).

P
0.63 2
G y = G0.63 (0.49 + 0.45 4 )
P1
(2.14)
F

There is some continued debate as to the accuracy of the expansion factor currently given by
ASME and even the further clarification by Cunningham. (Kegel, 2002) proposed that in
some cases the theoretical calculations are inadequate for subsonic flow. The Kegel paper

19
Chapter 2 Empirical Analysis

investigated the use of a theoretical expansion factor for two different meters and proposed a
meter specific valve for Gy.

The current equation for Gy used by IEC 60534-2-1, which governs single stage valves, was
based on the original Driskel derivation. In equations (2.12) and (2.13) Gy varies linearly with
the pressure drop (1 - P1 / P2), as the nearest fit to the empirical data. The pressure terms in
equations (2.12) and (2.13) can be replaced with . Therefore Gy is written as an equation of a
line, given by equation (2.15), where h is the slope of the line. As flow rate is directly
proportional to Gy times , see equation (2.16), the rate of change of flow-rate with pressure
drop is given by equation (2.17). As the flow rate chokes, equation (2.18), the limit of
becomes equation (2.19). Taking into account the specific heat ratio factor (F) of the gas
under consideration and the limit of (being c), Gy for a single stage valve is given by
equation (2.20).

;< 1  K (2.15)

1 1 3
m (1 h ) = h 2 2 2
(2.16)

1 1

dm 2 3h 2
= (2.17)
d 2 2

dm
=0 (2.18)
d


1 1 1
2
= 3h 2 ; h = (2.19)
3 c


Gy = 1 (2.20)
3F c

20
Chapter 2 Empirical Analysis

Equation (2.20) yields the limit of Gy = 0.667 as the limiting flow factor for the choked flow
condition in a single stage valve, where is equal to the product of Fc. Values of c vary for
equivalent valve openings based on their internal geometry. This is consistent with
Cunninghams research into thin lipped orifice plates wherein choked flow could not be
achieved experimentally when the plate thickness was sufficiently small (affect on flow rate
was negligible). The limit of the pressure drop ratio (c) for a valve opening needs to be
determined experimentally per IEC 60534-2-1 (see chapter 6).

Gy was plotted against for a number of different valves based on their c values (values
taken from IEC 60534-2-1 Table 2 Typical Values for differential pressure ratio factor c at
full rated flow), see figure 2.8. The various valve internal geometries vary the slope of the
line and are completely described by c. Note that the limit of Gy in all cases is Gy = 0.667.

Figure 2.8 Plot of gas expansion factor (Gy) versus pressure drop ratio () for a number of
valves with the limit of Gy = 0.667 (IEC 60534-2-1)

The mechanism of choked flow for gas in an orifice plate or valve is different to that seen in
the ideal restriction (figure 2.1 versus figure 2.3). Assuming the valve port edge and orifice
plate are equivalent, choked flow occurs due to an under-expanded jet forming outside the

21
Chapter 2 Empirical Analysis

port as the VC travels downstream (Chauveau et al, 2006). The limiting velocity occurs in
this cross-section which increases marginally as increases.

2.5 MS Flow Path Model Simplification and Assumptions

The geometry of the MS path consists of a series of throttling sections with cylindrical
chambers between each section enabling recovery. The design of the MS path initially used a
series of paths in sequence, located all on the same plain. The segmentation and balancing of
a value of = 0.7 required approximately 14 individual restrictive channels. Analysis showed
that the flow remained converged between each channel. Adding chambers for expansion
between the channels enabled partial realignment of the flow and increased the efficiency of
the subsequent restriction. Two prototypes from the concept design stage are shown in figure
2.9. Further investigation showed that elongating the expansion chambers further improved
flow alignment and increased mass flow rate through the path. Increasing the expansion
chamber height lead to the two plain layout.

Figure 2.9 Early Prototype models for the MS path

Figure 2.10 details the growth rate of the throttling and expansion zones with proprietary
dimensions removed. The path was machined in two planes on flat plates with constant depth.
The path under review has seven stages of pressure drop. The rate of change of throttling
width and expansion zone diameter is given in table 2.1. The respective widths and diameters
were optimised through CFD to ensure an equal pressure drop at each stage. It is important to

22
Chapter 2 Empirical Analysis

note the obvious sudden increase in the last two stages of the trim. The significance of this
will be further investigated in Chapter 5.

Figure 2.10 Multistage Path Geometry with throttling sections and expansion zones

Restriction Expansion

# #
% Increase from % Increase from
R1 previous stage E1 previous stage

R2 5.33% E2 7.06%

R3 4.43% E3 6.87%

R4 5.45% E4 7.20%

R5 4.60% E5 6.95%

R6 7.14% E6 6.95%

R7 43.59%

Table 2.1 Rate of change of throttling width and expansion diameter

23
Chapter 2 Empirical Analysis

The MS path was simplified to a series of thick walled orifice plates, with the rate of change
of port area taken from the restriction data given in table 2.1. The geometry of the expansion
zones has been ignored. The full alignment and expansion of the gas and all temperature
changes have been assumed to occur between the thick walled orifices. (Stiles, 1984) stated
that the static temperature change in a single stage orifice occurs downstream of the VC over
a number of pipe lengths. Therefore, it is unlikely that the complete temperature change will
occur within the MS path expansion zones. The effects of ignoring the expansion zone
geometry will need to be investigated further using a CFD simulation. The simplified MS
path model is represented in figure 2.11. The upstream geometry of the restrictions has been
taken as the circular area of the expansion zone, based on the transition between the planes.

Figure 2.11 Simplified multistage flow path with increasing throttling and expansion sections

As stated in section 2.2, the Gy is used to correct for density change at the VC for gas versus
liquid flows. In order to derive a theoretical model for the mass flow rate through the
simplified path a number of assumptions must be made. Firstly the flow must be considered
adiabatic. Assuming the geometry has been correctly designed to balance pressure across the
stages, the velocity should be subsonic at all points. The fluid velocities will still be relatively
high and little heat is likely to transfer to the trim. In addition, the gas expansion will be
assumed to be reversible and hence isentropic. For a single stage expansion the isentropic
assumption holds as most of the work is done downstream of the VC (Stiles, 1984). For the
multistage expansion, it is evident that the process is not of constant entropy and friction will
exist making the process irreversible.

24
Chapter 2 Empirical Analysis

The second assumption is with respect to geometry, VC location and downstream recovery.
The expansion zones between each throttling section were designed to enable recovery of
both pressure and temperature. It must be assumed for this process that full recovery and flow
alignment occurs between each stage. Based on this assumption values for VC diameter were
taken directly from thick wall orifice office experimental results.

2.6 First Pass Calculation

The mass flow rate through the MS path is a function of the Cd for each throttling section, as
well as upstream temperature and pressure. If the MS path were not designed to balance the
pressure drop across each stage, the smallest stage and its Cd value would have the
predominant effect on mass flow rate. For a pressure balanced system the total approximate
Cd value can be determined from the addition of restrictive elements using equation (2.21).
The continuity of mass requires that the mass flow rate through each restriction is the same.
Therefore, the mass flow rate determined per stage using the total Cd from equation (2.21)
and the reducing upstream temperature, pressure and density (determined from the equation
of state for a real gas which includes a compressibility factor) should be equal.

   
  (2.21)
L(M!467 L(!NOPO39 L(!NOPO39
L(!NOPO39Q

This section details the parameters required to determine mass flow rate for the mathematical
model detailed in figure 2.11, including; experimental values of Cd, temperature reduction in
the MS path using the Joule Thomson (JT) equation and the compressibility factor (Z) for the
equation of state of a real gas, as seen in the flow chart given in figure 2.12. The density of a
real gas is determined using the equation of state (EOS) (see equation 2.22), which is
corrected using Z for the real gas condition. The EOS also incorporates the reducing
temperature (see JT); see equation 2.10 for the relationship between Cv and .

1S 
0 TU"
(2.22)

25
Chapter 2 Empirical Analysis

Figure 2.12 Flow chart of theoretical calculations for Cv

2.6.1 MS Path Cd

Due to the historical use of orifice plates for flow metering there is a vast amount of
experimental data available for numerous plate profiles and geometries. The simplification
given in figure 2.11 states that the MS path geometry can be approximated as a series of thick
walled orifice plates. (Rhode, 1969) conducted an extensive series of tests on various sizes of
thick-walled orifices based on a 0.635cm (0.25) inlet, throttling to diameters ranging from
0.322cm (0.127) to 1.49cm (0.59). As detailed in figures 2.6 and 2.7, it is the ratio of
upstream and orifice diameter (, the velocity of approach) that is of most interest in
determining Cd rather than exact sizes. The research was primarily concerned with flow
through thick walled orifices at 90 to ducts used to transport air for cooling turbine fans. This
right angled direction of flow is a good representation of the two plane MS flow path. The
effects on Cd were measured for variances in temperature, pressure, orifice surface finish,
multiple orifice interference, and approach passage geometry. The test specimen of interest is
given in figure 2.13.

Figure 2.13 Thick walled orifice plate test specimen as used by (Rhode, 1969)

26
Chapter 2 Empirical Analysis

There were three primary tests conducted including; 1) constant main duct Mach number, 2)
all flow passing through the orifice (main duct closed), 3) constant pressure differential
across the orifice. For test 1) the inlet Mach number was held constant (as was static
pressure) and the downstream pressure was reduced in small increments until just before
choked flow. Test 2) was calculated exactly the same as test 1) but the main duct was sealed
at the end so all flow went through the orifice (no secondary flow measurement was
required). For test 3) the pressure drop across the orifice was maintained but the inlet Mach
number was varied by heating the air being used for testing.

The primary conclusions of the research that are applicable to the MS path model are the
changes in Cd for varying values (where is the ratio of orifice diameter to wall thickness,
see figure 2.13). Rhode presented results for values of 1 to 4 where values were varied
from 0.1 to 1.0 in 0.1 steps. A comparable non-dimensional methodology was applied to
single stage choke valves to predict Cv for new geometries, see (Grace et al, 2011).

In order to give good correlation with previous work, Rhode used the velocity head ratio (VH)
to describe the variance in Cd. VH in this case is the ratio of outlet to inlet pressure. The
results presented by Rhodes need to be in the form of for comparison with the experimental
data to be present in Chapters 3 and 6. Figures 2.14 (A), (B), (C), (D) and (E) taken from
Rhodes, detail the change in Cd for five different values of ( = 1.05, 1.6, 2.0, 2.83 and 4.0).
In each plot the inlet Mach number has been varied. The relationship of Cd versus VH as
increases is convergent. The convergence of Cd is independent of the Mach number as
increases. The flow structure for the thick walled orifice was shown in figure 2.4. This leads
to two approximations of Cd for the simplified MS path. The inlet and outlet stage have
values of approximately 1.0 while the five internal stages have values of approximately 4.0.
The variance in Cd by Mach number for =1 means the inlet Mach number needs to be
considered for the first stage of the MS path.

Figure 2.15 is the nearest fit for test data given in figure 2.14(E), with VH in the form of for
a thick wall orifice, where 4. For = 1.0 the Mach number must be included and figure
2.16 reproduces the values from figure 2.14(A) in the form of . When a MS path takes a
pressure drop across it, the drop is split evenly across the stages, meaning that each stage has
a unique value. Figures 2.15 and 2.16 can be used to calculate values of Cd for each stage.
Table 2.2 details ten sample values of across the entire MS path, with inter-stage Cd values
(1 n) determined at each stage.

27
Chapter 2 Empirical Analysis

Figure 2.14 (A) Plot of Discharge Coefficient Versus Velocity Head for = 1.05

Figure 2.14 (B) Plot of Discharge Coefficient Versus Velocity Head for = 1.6

28
Chapter 2 Empirical Analysis

Figure 2.14 (C) Plot of Discharge Coefficient Versus Velocity Head for = 2.0

Figure 2.14 (D) Plot of Discharge Coefficient Versus Velocity Head for = 2.83

29
Chapter 2 Empirical Analysis

Figure 2.14 (E) Plot of Discharge Coefficient Versus Velocity Head for = 4.0

Figure 2.15 Plot of Cd versus , for = 4.0, up to the choked flow point

30
Chapter 2 Empirical Analysis

Figure 2.16 Plot of Cd versus , for = 1.0, up to the choked flow point

31
Chapter 2 Empirical Analysis

Table 2.2 Values of Cd for various values of , at inlet Mach number = 0.07, based on results
from (Rhodes, 1969)

In addition to Cd data from the thick plates, it was evident from figure 2.15 that the gas in the
thick plate orifice chokes at a value of 0.558, which is less than that seen for most thin plate
orifices. This can be confirmed by comparing the velocity head values given for M = 0.14
across the three plate thicknesses tested in figures 2.14 (C), (D) and (E). For the thinner
plates, see figure 2.14 (A) and (B), the values of VH get progressively higher. As the Mach
number decreases (due to the increased inlet temperature), the choking point moves further

32
Chapter 2 Empirical Analysis

upstream and c can reach almost 0.65. This is consistent with the research presented by
Cunningham.

2.6.2 Compressibility Factor

The compressibility factor (Z) describes the variance between an ideal and real gas Equation
of State (EOS). For an ideal gas the intermolecular forces are considered negligible and Z =1.
For a real gas at high pressure and / or temperature the effect of intermolecular forces
becomes more noticeable and deviations from ideal gas laws are evident. Equation (2.10)
requires the upstream density to determine the mass flow rate. As the gas expands through the
MS path the compressibility and hence density will continually vary. As the area in the
orifices increases the reduced density should ensure mass continuity. There are many
methods of modelling the interaction of gas molecules which typically use either empirically
determined data, conceptual models of interactions or statistical mechanics.

The simplest EOS use two variables (based on the fluids critical pressure and temperature) to
describe the attractive and repulsive forces between molecules (Van der Waals forces). The
most commonly used EOS include the Soave-Redlich-Kwong (SRK) EOS and the Peng
Robinson (PR) EOS (Peng et al, 1976). Both equations use the Van der Waals relationship
between compressibility and the reduced temperature (Tr) and pressure (Pr), see equations
(2.23) and (2.24). The reduced properties of a fluid are a set of variables normalized by the
fluids state properties at its critical point. Van der Waals principle of corresponding states
dictates that all fluids, when compared at the same reduced temperature and reduced pressure,
have approximately the same compressibility factor and all deviate from ideal gas behavior to
about the same degree see (Xiang, 2005). Van der Waals called this relationship the two-
parameter principle of corresponding states.


V  (2.23)
3

"
WV (2.24)
"3

Pure gases at identical Pr and Tr values will exhibit corresponding compressibility factors. P-
V-T data plots vary for pure gases but when compressibility is plotted against pressure the
resultant graphs exhibit similar curves for fixed temperatures. A generalised graph can be

33
Chapter 2 Empirical Analysis

plotted by using Pr and Tr to normalize the compressibility data. The generalised


compressibility plots use multiple substances, with the Nelson Obert graph, given in figure
2.17 (Cengel et al, 2004); using 25 different gases (the graph has been reproduced and
simplified from the original format for clarity).

The original Van der Waal EOS used estimates of inter-molar forces (based on Pr and Tr) to
predict compressibility. The estimates were derived from the best fit data for the generalised
compressibility graph (see Nelson Obert graph). The Redlich Kwong EOS improved the
inter-molar force equations but they became most practical when Soave added an acentric
factor to take account of the data for the vapour pressure of hydrocarbons, (Soave, 1972). The
acentric factor accounts for the characteristic of a molecular structure and is itself a function
of Pr and Tr.

Figure 2.17 Nelson-Obert compressibility plot based on 25 different pure gases, reproduced
and simplified for clarity from (Cengel et al, 2004)

The PR EOS was based on the work completed by SRK with sponsorship by the Canadian
Energy Board, whose particular interest was in natural gas condensates. Both PR and SRK
display comparable results across a range of pressures and temperatures but the most notable
difference is at the critical point. PR conserves both the attraction term used by SWR and the
acentric factor but presents different fitting parameters to describe these dependencies. In
addition, the pressure correction factor (which is seen as part of the attraction term) was
34
Chapter 2 Empirical Analysis

refined. The PR EOS is the most common oil field equation and while not originally
supported by ANSYS CFX a beta version was available during analysis (see chapter 3 for
further details). In conclusion due to its ease of use and proven industrial application the PR
EOS was used throughout the analysis.

The PR EOS in its explicit form is given in equation (2.25). The coefficients describing
molecular interaction are given by equations (2.26) and (2.27), with Z given by (2.28)
(which includes the acentric factor, see equation (2.29)). Z is the Soave compressibility
function and was used by Soave to modify the Redlich Kwong equation to account for the
vapour pressure data for hydrocarbons. Substituting and simplifying these equations gives a
cubic form for Z, see equation (2.32), where the cubic coefficients are given by equations
(2.30), and (2.31). The cubic equation can be solved directly using the general or the monic
formula of roots. For every cubic equation there must be at least one real root. The
determinant can be checked to see whether imaginary roots exist (determinant = 0) in which
case there will be only one answer, or if multiple roots exist (determinant > 0) in which case
there will be multiple real answers. For each of these real roots, it was seen that only one fell
in the range of compressibility i.e. a non-negative value for the physical limits of temperature
and pressure.

U" L T
 X  X
L6 X[
(2.25)
Y LZ Y Z Y LZ

]._`aU
"3

$\ (2.26)
3

].]aacU"3
$b (2.27)
3


de f1  0.37464  1.54226dm  0.26992dm 1  WV].` o (2.28)

dp  log]V   1 (2.29)

T[ L6 
$ U
"

(2.30)

LZ 
$t (2.31)
U"

d u  1  $t d   $  3$t  2$t d  $ $t  $t  $tu  0 (2.32)

35
Chapter 2 Empirical Analysis

The most complex but comprehensive EOS use the virial theorem to explain the time
averaged kinetic energy in a gas with that of the total potential energy. The virial theorem
provides a general equation relating the average (over time) of the total kinetic energy of a
stable system with that of the total potential energy. Using data from a number of gases a
series of experimental coefficients are used to determine the compressibility of a substance
using Pr and Tr and the kinetic theory of gases. The base form of the virial equation is given
in (Reynolds, 1979) where a series of 27 virial coefficients were used to model various pure
gases. (Hamad et al, 1988) correctly highlighted that of all the EOS, the virial equation is the
only one with a sound theoretical basis. Hamad (1988) aalso stated that the first three
coefficients provide suitable accuracy up to approximately 2/3rd of critical density. In chapter
7 it is advised that further work should be completed in implementing the virial equation for
extreme high pressure applications.

2.6.3 Temperature Variation Due to Expansion

There are various methods for determining temperature change during the expansion of a gas.
For any pure gas or mixture there exists a series of Joule Thomson (JT) coefficients (JT)
which describe the rate of change of temperature with respect to pressure, see equation
(2.33). The density of the gas in the MS path was a function of temperature and pressure. As
the gas expands through the throttling section, the temperature change upstream and in the
VC will help describe the mass flow rate.

v" XY
wx" $e W  1 (2.33)
v Ly

The JT effect was explained by (Roy, 2002) using the inter-molar attractive forces in a gas.
At a specific pressure and temperature the total internal energy is the sum of the kinetic and
potential energy. As a gas expands the distance between the molecules increases as does their
attraction to one another and hence the potential energy increases. For this increase in
potential energy there must be a similar decrease in kinetic energy (assuming constant
enthalpy) and thus as a gas expands cooling generally occurs. Values for JT can be taken
directly from Perrys chemical engineering hand book (Perry, 1999).

36
Chapter 2 Empirical Analysis

It is important to note that using the same Van der Waal theory there must be a point at which
the pre-expansion pressure is high enough to cause repulsion between the molecules. At this
pressure a subsequent expansion would lead to a decrease in overall potential energy and
hence an increase in temperature. The point at which this occurs is called the JT Inversion
Temperature (JTIT). From equation (2.33) it is clear the inversion will occur when CT-1 =0,
as both Vm and Cp are fixed. The inversion temperature is fluid and pressure dependant and
for most applications is sufficiently high that only cooling occurs.

Research into natural gas temperature inversion started in the late 1990s with exploration in
High Pressure High Temperature (HPHT) gas fields in the North Sea, (App, 2009). While JT
inversion is a well defined process there was little information available for natural gas
mixtures at high pressure. (Kortekass et al, 1997) investigated specific field conditions seen
in a number of industry papers using an EOS JT inversion model. As part of this study
Kortekass used both Peng-Robinson (PR) and Soave-Redlich-Kwong (SRK) EOSs to model
a series of isenthalpic expansions. Of primary concern was the variation in the gas inversion
curve for various compositions. A series of possible gas mixtures were investigated as well as
a number of mixture equations. Both PR and SRK were used to solve temperature for
equation (2.34).

v v
W z 0 (2.34)
v" vX

The gas mixtures varied in methane content from 45-92%. Of specific concern to Kortekass,
were heavy gases containing higher order molecules of methane (as high as C20). Figure 2.18
references the inversion area resolved for mixture GC1L (92% methane) using both the SRK
and PR calculations. For pressures inside the temperature axis loop, cooling will occur and
for pressures outside the loop, heating will occur. The fluid mixture GC1L was based on an
actual reservoir composition. Figure 2.19 details the complete series of mixtures based on
temperature affects from an initial pressure state. There was only a slight variation in
temperature increase across the light natural gases. Pure methane was included for reference.

37
Chapter 2 Empirical Analysis

Figure 2.18 Curves for gas mixture GC1L (92% methane) plotted for both PR and SRK EOS
(Kortekass, 1997)

Figure 2.19 Temperature variations for all natural gas mixtures against pressure drop for a
fixed upstream pressure of 1,000 bar (Kortekass, 1997)

38
Chapter 2 Empirical Analysis

Kortekass concluded that temperature increases would occur at the reservoir conditions
identified. These reservoir conditions are likely to be seen in the MS trim and are consistent
with the conditions given in table 1.1. As shown, the fluid composition for higher order and
heavier natural gases had a major effect on temperature variation. Most importantly, the
paper established that pressure more than temperature would affect the inversion point for
production applications. On a practical note it was concluded that temperature increases were
unlikely to exceed 20K.

Kortekass provided insufficient experimental data to confirm the reliability of his EOS and
mixtures models. The referenced industrial papers cite that production compositions are
confidential so no direct comparison could be made. Further investigation by (Bessieres et al,
2006) gave much more reliable inversion curve data with supporting experimentation for pure
methane (99.99% purity). As indicated in section 1.4, recent field data indicated an increased
occurrence of very high methane content wells which lends support to the use of Bessieres
inversion values. Bessieres research was also dedicated to natural gas reservoirs, specifically
for temperatures up to 520K and pressures up to 120 MPa.

For the most accurate determination of JTIS Bessieres noted that cubic EOS (as used by
Koertkass) were mostly inadequate. This conclusion was supported by (Vrabec et al, 2007)
who used more complex EOS including DDMX, SUPERTRAPP and BACKONE. (Parsafar
et al, 2000) further demonstrated the use of complex equations by way of comparison, noting
the consistent variation in each. Bessieres was the only author reviewed who supported
simulated data with experimental results wherein a combined experimental and mathematical
approach was presented. Limited experimental data was previously available as direct
experimental measurement was limited to approximately 30MPa and detection of small
temperature differences was unreliable at the largest differentials. Bessieres used a pressure
controlled scanning calorimeter to overcome these issues and provided a sufficient number of
points on the inversion curve to compare to the mathematical simulation. The Monte Carlo
technique was used to resolve a number of points which solved CT-1=0.

39
Chapter 2 Empirical Analysis

In conclusion the Bessieres values for JT inversion were chosen for inclusion in this research
as the most appropriate based on the comprehensive series of tests and simulations that were
completed. The data presented showed a high level of repeatability and correlation. Figure
2.20 details the relationship between both experimental and Monte Carlo inversion points, as
well as previous referenced techniques. There is a significant variation between Kortkass and
Bessieres. For example, at 50MPa (a typical throttling pressure) the inversion temperature is
approximately 400K on the Kortkass curve while closer to 345K on the Bessieres curve
(with Kortkass the 92% pure methane reference was used for comparison, which showed
little variation to pure methane in Kortkass work). The tables presented by Bessieres are
included in Appendix B for completeness.

2.6.4 Calculations and Discussion

Section 2.6 details all the parameters required to determine the mass flow rate through the
MS path, based on a given pressure drop using equation (2.10). Cd for the entire MS path can
be calculated using equation (2.21) with the Cd for the individual orifices taken from table
2.2. The mass flow rate across each stage can also be determined from the upstream density,
temperature and Cd. The resultant average mass flow rates per stage should be consistent with

40
Chapter 2 Empirical Analysis

the overall mass flow rate. It was evident that as increased across the MS path n did not
come near c values for any single stage data. Therefore the mass flow rate increased to the
limit of the pressure drop ratio, which is unity. The mass flow rate versus has been plotted
in figure 2.21.

Figure 2.21 Mass flow rate versus pressure drop ratio, , using mathematical analysis

In theory the reduced pressure drop across each stage means that there is no formation of a
shock wave in any stage. The data from section 2.6.3 indicated that for high inlet pressures,
below the JTIT, the temperature drop through the MS path decreased the velocity of sound
(as given in equation (2.35)). Therefore at the final restriction where the temperature is lowest
the greatest potential for a shock wave exists. For an inlet pressure of 69.6MPa (the
maximum pressure rating of the valve body) the outlet temperature will be approximately
210K at which the speed of sound in air is 280 m/sec. The mean velocities in the VC at this
location are approaching this value but not sufficient to create a shock wave (at M = 1). The
velocity in the restriction is given by equation (2.36). The density in equation (2.36) can be
determined from the EOS where the temperature and compressibility values are known. The
pressure in the VC can be estimated using the pressure recovery relationship given in
equation (2.37) (where FL = 0.89 for orifice plates). The temperature in the VC can be
estimated using equation (2.38).

41
Chapter 2 Empirical Analysis

{\ +|}W (2.35)

 
   
{XL '2  ~1   (2.36)
  3 D

 

-.  

(2.37)

 
3 
W-. W  (2.38)


Due to the simplification of the mathematical model the investigation of the choking
mechanism within the MS path is limited. Stiles has suggested that the temperature effects
will occur further downstream of a restriction but the complex geometry of the MS path
could localise the temperature changes. In conclusion a combined experimental and CFD
investigation of the MS path is required to validate the mass flow rate of the mathematical
model and further examine the last restriction. The results of both these investigations will be
used to correct figure 2.21 for the model simplifications and choked flow condition.

2.7 Summary

This chapter presented an equation for the mass flow rate of gas through a single stage
restriction. The equation was linked to valve sizing and the primary variable Gy was
investigated for single stage restrictions. It was shown that Gy is defined by c and an
associated choked flow mechanism. A mathematical model for the MS path was created
using existing experimental data. The methodology for creating this model had previously
been applied to single stage choke valves, (Grace et al, 2011). In addition, equations for the
change in temperature and compressibility through the MS path were presented. These were

42
Chapter 2 Empirical Analysis

used to determine the change in mass flow rate versus . It was shown that the conditions of
choked flow did not exist at any of the MS path stages and that theoritically the flow rate
would increase until = 1.

It was proposed that choked flow could occur at values of < 1. As the temperature of the
fluid reduces due to expansion and JT cooling the speed of sound created near critical
conditons or values of M 1. The over simplification of the MS path meant further
investigation into the choking mechanism using experimental and computational tools was
recommended.

43
Chapter 3 Experimental Investigation and Benchmark Data

CHAPTER 3

EXPERIMENTAL INVESTIGTION AND BENCHMARK DATA

3.1 Introduction

Laser Doppler Anemometry (LDA) is a technology for fluid flow research which has the
potential of allowing local time resolved measurements of velocity without disturbing the
fluid phenomenon under investigation, (Buchhave, 1979). It is the non-invasive
characteristic of LDA which makes it ideal for turbulent flow measurement versus other
techniques (hot wire anemometry) which can interfere with the parameters being
investigated. This chapter outlines the use of LDA and a flow loop to provide experimental
benchmarking data for the MS flow path. The mass flow rate data from the flow loop was
used to investigate the accuracy of the mathematical model presented in Chapter 2. The LDA
velocity data will be used to verify the suitability of the CFD turbulence model, see Chapters
4 & 5.

3.2 LDA Overview

The LDA method is based on the measurement of the Doppler shift (frequency change) of
light scattered from small particles carried along with a fluid. At a fundamental level a LDA
system must consist of a suitable light source, a means to direct the light into the fluid, a
seeding mechanism to introduce small particles into the fluid, a means of capturing the
scattered light and a method of processing the frequency signal. This section examines the
principles involved in LDA measurement, including suitable test apparatuses. An overview of
the governing equations is given in appendix C. The associated theory was used to construct
the flow loop and test models used for determining the benchmark data.

44
Chapter 3 Experimental Investigation and Benchmark Data

The probe volume created by the intersecting laser beams has a Gaussian distribution for light
intensity. The crossed beams create constructive fringes of high intensity. A single particle
travelling through the probe volume scatters light with a frequency equal to the rate of fringes
crossed per unit time. The amplitude of the signal is proportional to the Gaussian distribution
and dependant on the direction of observation. The Mie theory of scattering light from a
particle states that light scattered in the direction opposite to the incident is much smaller than
in the direction of the incident, see figure 3.1 (Durst et al, 1976). The Gaussian distribution
creates a pedestal for the LDA signal. The resultant LDA burst, the intensity pedestal and
original Doppler signal are given in figure 3.2, (Drain, 1980).

Figure 3.1 Light intensity scattered from a small particle, (Durst et al, 1976)

Figure 3.2 Burst signal for a single particle in a LDA probe volume with Gaussian intensity
pedestal and Doppler signal (Drain, 1980)

45
Chapter 3 Experimental Investigation and Benchmark Data

The ideal signal to noise ratio (SNR) is obtained when the probability that two or more
particles are present in the probe volume is low (the incoherent detection mode). In this case
a large aperture is used to collect a maximum of light scattered from a single particle. As
there is only one particle crossing the probe volume fringes at a time the detector is able to
fully resolve the interference pattern. If multiple particles are present in the probe volume at
any one time the photocurrent contributions add incoherently at the detector surface (the
particles entering the probe volume at different times create destructive interference in the
scattered light). The resultant SNR can be improved by using the coherent detection mode.
By keeping the aperture small, the light falling on it adds coherently and thus the SNR can be
improved by adding more particles.

Consideration must be given to the electrical equipment required to determine a velocity from
the signals produced by the detector. There is no universal signal processor available and
historically most signal processing was completed using spectrum analyzers or simple
counters. The suitability of a specific processor is dependent on the density of particles in the
probe volume. Spectrum analyzers are suited to the continuous constant amplitude signals
associated with multiple particles in the probe volume. For the LDA measurements
completed herein a direct counting method was used. Counting is an economic method that
uses a high frequency timer to count the number of fringes the particle passes through in a
Doppler burst signal.

Counting is initiated by setting the timer to start at the threshold amplitude of the detector
signal. The counter operates at a frequency greater than the Doppler signal, typically in the
region of 200-500 MHz. A Doppler signal cycle count is determined at two different times
and the cycle rate calculated. The timing has to stop before the end of the burst scale. The
probe volume is usually small in such circumstances and can contain approximately 20
fringes with counting applied to the first 10. Anomalies can occur using the counting
technique including; signals from particles that do not travel through the fringe system
completely resulting in insufficient cycles in the burst, additional particles entering the
volume during a timing interval, timing starting in one burst but ending in another and
excessive noise or interference. Measurement validation is controlled by two separate timers
taking separate cycle counts (they should not share a common factor for the timing period). If
two timing periods agree to a specific degree of precision the cycle count is validated. In
addition the threshold amplitude can be used to reject signals of poor quality. However
46
Chapter 3 Experimental Investigation and Benchmark Data

setting this threshold to high may result in insufficient measurements to determine a mean
velocity, conversely low settings can increase noise. The threshold setting can be optimized
by looking at the percentage of measurements validated, where 50% validation typically
represents a good signal.

Counting techniques are subject to velocity biasing. The result of velocity biasing is the
incorrect mean velocity determined from the average particle velocity. When the velocity at a
point is high, the number of particles flowing through the probe volume will increase per unit
time. As the velocity decreases the particle count also reduces and mean velocities
determined from the total number of particles counted tends towards the higher velocity.
Velocity biasing can be reduced by weighting the particle velocities with their transit times
and therefore counter systems need to measure the resident time corresponding to each
measurement as well as the actual frequency.

In order for the LDA apparatus to operate the fluid being measured must have entrained
particles to scatter the light from the probe volume. In liquid flow hydrosols are usually
present in sufficient density to eliminate the requirement for particles to be added. With the
aerodynamic system being used the flow must be seeded. (Adrian, 1977) stated the particles
have more influence on the quality of the signal than any other component of the system. The
signal strength can be improved by 102 to 104 by increasing the particle diameter. Similar
improvements of this nature are difficult to achieve by increasing the laser power or
improvements to the optical system. The velocity being measured by the LDA apparatus is
that of the scattering particle. Therefore the particle needs to travel in the fluid, tracking any
changes of the flow velocity. If the particles are too large their transition through the fringe
pattern will result in an unclear Doppler signal.

A particle in a turbulent fluid is under the force of viscous drag from the relative motion of
the particle and the surrounding fluid. It is the magnitude of this force in relation to the inertia
of the particle that determines the response of the particle to fluctuations in the velocity of the
fluid. The Stokes number is a dimensionless number corresponding to the behavior of a
particle in a fluid. For a Stokes number >>1 a particle will travel in a straight line as the fluid
turns around an obstacle and for a number <<1 the particle will follow the fluid streamlines.
Drain (1980) combined Stokes law with the equation of motion to give equation (3.1), where
up is the particle velocity and mp is the particles mass. If the fluid velocity is consistent the
47
Chapter 3 Experimental Investigation and Benchmark Data

solution to equation (3.1) is given by (3.2). Therefore, after a change in the fluid velocity, the
velocity of the particle will approach a new value in a certain time (tp), given by equation
(3.3). For a particle of diameter 1m, the corresponding time to approach the new velocity
will be 3sec, assuming a relative density of 1 and the viscosity of air to be 1.8x10-5 kg/msec.
For shock wave analysis a time constant of 4sec is considered sufficient.

y
: 6w f   o (3.1)


4
    f 0  o 4y (3.2)

1y

Dy y

(3.3)
y

For a fluctuating flow field the frequency response of the particle becomes important. By
assuming the frequency response of the fluid is sinusoidal equation (3.1) can written as
equation (3.4). The steady state solution of this is given by equation (3.5). For a particle to be
suitable for a fluctuating flow field the variation in up must be equal to the mean velocity to
within 1% at the maximum fluctuation frequency present and therefore the particle radius is
limited by equation (3.6). Applying equation (3.6) to a particle in air, its diameter must be
smaller than 2.7m to follow fluctuations up to 1 kHz (assuming a density of 1000 kg/m3).

y ! . 5  y o


f
(3.4)
 y

f! .5  o
    (3.5)
>_
y
5


\ 0.1 (3.6)
5 Dy

A number of sources that affect the LDA data exist in the system including the presence of
optical noise (particularly associated with operating in backscatter mode, where the signal is
about 1/100th the level of forward scatter), light reflecting from neighboring surfaces, the

48
Chapter 3 Experimental Investigation and Benchmark Data

quantity of seeding and the deposits of seeding fluid on the test section window. Reflected
light can be reduced by applying non-reflective black coatings to the flow path surfaces. The
seeding rate can be controlled at the aerosol generator but the inevitable residue build up on
the test window is a result of the flow path geometry. The effects of the above on the quality
of the data can be seen by inspecting the data rate, standard deviation (u), and count
variables in the LDA measurements. A very high data rate, above 10 kHz for example is
indicative of noise. An unreasonably large value of u with a small data rate or small count is
probably a result of a valid measurement but with too few particles to obtain a statistically
meaningful result. Caution should be given to results obtained from a very low count (i.e. less
than 10) because this is unlikely to be sufficient to provide an accurate estimate of the mean
velocity. Large values of anode current (i.e. > 1000 A) suggest that light reflected from
neighboring surfaces is being measured by the photomultiplier (this can be useful for aligning
the system).

3.3 Experimental Apparatus

3.3.1 LDA Apparatus

The velocity measurements were made using a Dantec Dynamics Fiber Flow 2 component
LDA system with a Spectra-Physics 4W Argon-Ion laser as a light source, see figure 3.3.
LDA signal processing was managed by the Dantec BSA-F70 LDA signal processor, see
figure 3.4. The beam emerging from this laser was split using a Bragg cell into separate green
(514.9m wavelength) and blue (488 m wavelength) beams. For the measurement of two
components of velocity it is necessary to form two super imposed sets of fringes. The fringes
may be discriminated by polarization but the use of color is more satisfactory and the
strongest lines are associated with the wavelengths of 514 nm and 488nm. The resultant
fringe pattern in the probe volume is given in figure 3.5, which is obtained by focusing four
parallel beams. The complete LDA apparatus can be seen in figure 3.6.

49
Chapter 3 Experimental Investigation and Benchmark Data

Figure 3.3 Dantec Fiber-Flow system with Spectra-Physics 4W Argon-Ion laser

Figure 3.4 Dantec BSA-F70 LDA Signal Processer

50
Chapter 3 Experimental Investigation and Benchmark Data

Figure 3.5 Fringe pattern for two component velocity measurement formed by blue and green
light (Drain, 1980)

Figure 3.6 LDA Apparatus with Test Model In-Situ

51
Chapter 3 Experimental Investigation and Benchmark Data

The LDA system was operated in backscatter mode with both transmitting and receiving
optics being located in the same probe. A 100 mm focal length lens with beam spacing (E) of
15 mm and beam diameter (DL) of 1.35 mm was used for testing. Figure 3.7 details the
optical arrangement used, wherein L can be determined from F, E and DL as L = 8.57 for a
focal length of 100mm. The probe volume was determined from equations (3.7), (3.8) and
(3.9) where x, y and z are given by figure 3.8, with the fringe separation given by equation
(3.10) and the total fringe numbers given by equation (3.11). The resultant probe volume
dimensions are given in table 3.1.

From figure 3.8 it can be seen that the greatest number of fringes exists at the origin of the
xyz plane. Due to the angle of intersection of the lasers the probe volume is typically
elongated in the z direction (4.33 mm for the 100m focus lens). A particle travelling through
the probe a distance of 2.3mm from the origin would cross fewer fringes than a particle
traveling through the origin and the fringes would be of lower intensity. As stated in section
3.1 the Doppler signal is processed using a timer set to a threshold amplitude. Particles
further from the origin would not generate signals of sufficient amplitude (intensity) or cross
enough fringes, to cross the threshold amplitude or give a meaningful cycle count. The higher
the threshold frequency is set the smaller the effective volume in the z direction becomes.
Based on threshold amplitude and the counter settings the effective z would have been closer
to 2mm.

Figure 3.7 Optical Layout and Resultant Beam Angle (Dantec Dynamics)

52
Chapter 3 Experimental Investigation and Benchmark Data

_
(3.7)
.
[

_
< (3.8)

_
(3.9)
 [


(3.10)
 [


c\ [


(3.11)

Figure 3.8 Details of Probe Volume (Dantec Dynamics)

Wavelength Focal Length x (mm) y (mm) z (mm) LF (microns)


514.9m 100mm 0.324 0.323 4.33 0.344
488m 100mm 0.307 0.306 4.1 0.32
Table 3.1 Probe Volume Dimensions

The LDA probe was located and moved using a 3D traverse table. This allowed positional
accuracy of better than 0.1 mm. The absolute central location (perpendicular to the channel
walls) of the measuring volume was determined by transversing into the wall boundaries and
detecting the shift frequency. With this set as the zero point the transverse was used to move
the probe volume geometrically (half the restriction width). The probe depth was maintained
centrally using the transverse system (corrected for the shift of the probe due to the refractive

53
Chapter 3 Experimental Investigation and Benchmark Data

index of the optical window). Due to the complex geometry of the MS flow path the location
of the probe volume could not be independently verified.

The flow was seeded with a SciTek LS10 Liquid Droplet Seeder using olive oil as the
seeding medium. This produced droplets in the range 0.5 to 3 m diameter. Applying
equation (3.3) and (3.6) (with a density of olive oil as 820 kg/sec) the time for particle
response (tp) for the largest 3m particle was 23sec and the fluctuation response was just
over 1kHz. The seeding medium was injected into the delivery pipe to the test rig
approximately two meters upstream of the test section.

The complex flow pattern within the MS path meant that the seeding fluid impacted on the
optical window as the flow travelled from the restrictive channel to the expansion chamber.
The deposition of olive oil on the optical window varied the refractive index of the window
and scattered the probe volume making measurements impossible. As will be shown in
section 3.4.2 the MS path models did not facilitate removal of the optical window. In addition
the flow rig required that the flow path model be realigned perpendicular to the LDA probe
and transverse system every time the optical window required cleaning. This issue limited the
time period for measurements. Using water as a seeding fluid would have eliminated the need
to disassemble the test model for cleaning and would have facilitated a greater number of
measurements.

3.3.2 MS Flow Path Models

The MS path, as seen in figures 1.5 and 2.7 (repeated below in figure 3.9), is a complex 3-
dimensional series of throttling, square-section ducts and cylindrical expansion chambers.
The primary location for pressure drop is in the restricting sections and it is the velocities
across these sections that are of most concern. For LDA measurements to be conducted a MS
flow path geometry had to be fabricated with optical access. The actual dimensions of the
path were suited to the available probe volume and pressure drop capacity and thus no scaling
has been used. It will become apparent that maintaining dimensional accuracy impeded the
scope for measurement but did facilitate a non-dimensional pressure drop and hence flow
velocities.

54
Chapter 3 Experimental Investigation and Benchmark Data

Figure 3.9 Multi-plane MS Flow Path

The seal between the transparent front plate and the flow channel was achieved using Dow
Corning Type 732 multi-purpose (silicone) sealant. The refractive index of polycarbonate has
a value of n = 1.58. Figure 3.10 details the trigonometric relationships used to determine the
shift in the probe volume, see equations 3.12, 3.13 and 3.14. For brevity the nomenclature
used for these relations is specific to this figure only. The refractive index and window
thickness had the effect of shifting the LDA probe volume by 4.7 mm as detailed in figure
3.11. The refraction of the beams resulted in an additional 3.5mm beam separation in the
polycarbonate of the optical window, see figure 3.11. This additional beam separation limited
the movement of the probe volume perpendicular to the walls of the throttling chamber.

Figure 3.10 Trigonometric relations for refractive index and probe volume shift

55
Chapter 3 Experimental Investigation and Benchmark Data

B t
(3.12)
"[

"[ "
W (3.13)
"



W 1  W 1  (3.14)
[

Figure 3.11 Refraction of Laser and resultant probe volume location

As the MS path weaves between two planes the test model required three separate layers. The
top layer is a simple window for optical access and the two lower layers represent the path.
For ease of use two models were fabricated with the MS path inverted in one. This enabled
the throttling sections on both planes to be analyzed. Test Model A (TMA), seen in figure
3.12 represents the lower section of the MS path. Test Model B (TMB), seen in figure 3.13
represents the upper section of the MS path which includes the inlet.

56
Chapter 3 Experimental Investigation and Benchmark Data

Inverse Path Block Plane of interest For TMA

Figure 3.12 Test Model A- MS Path Lower Section

Inverse Path Block Plane of interest TMB

Figure 3.13 Test Model B- MS Path Upper Section

The inverted flow path was machined into a cylindrical block of aluminum with flanged end
connections. A matt black spray coating was applied to the block to reduce refection. The
plane of interest was machined into a polycarbonate plate of depth equal to that of the MS
path. The test models were assembled using a series of cap-screws to compress the optical
window, plane of interest and block together, see figures 3.14 and 3.15.

57
Chapter 3 Experimental Investigation and Benchmark Data

Figure 3.14 Test Model Assembly

Figure 3.15 Test Model Assembly continued

Due to the large beam separation, caused by the optical window, measurements were made
centrally down the lengths of the throttling sections. TMA consisted of 3 different restrictive
sections referred to as TMA-1, TMA-2 and TMA-3, with measurements taken at locations as
identified in figure 3.16. Measurements in the section are referenced in alphabetical order
from inlet to outlet i.e. TMA-1a is the first point in this first restriction for TMA. TMB
consisted of both throttling sections and expansion chambers. The measurements taken in the
expansion zones were separated out for later reference, so that TMB-1 through to TMB-4
refers to the throttling sections, as seen in figure 3.17.

58
Chapter 3 Experimental Investigation and Benchmark Data

Figure 3.16 TMA Throttling Sections and Measurement Locations

Figure 3.17 TMB Throttling Sections and Measurement Locations

59
Chapter 3 Experimental Investigation and Benchmark Data

3.3.3 Flow Rig Components

The test apparatus was designed in two sections including a flow loop and the LDA system.
The flow loop consisted of a compressor, pressure sensors, flow meter and temperature
sensor. The flow loop was of the same specification as that required to flow test a valve per
IEC60534-2-3, see figure 3.18. Pressure measurements were made using a Scanivalve Digital
Sensor Array (DSA) Type 3217 (having eight channels in the range up to 1.4 bar and eight
channels in the range up to 7 bar). Air was supplied by an Atlas-CopCo compressor which
delivered a cooled supply up to 1 kg / s at 7 bar and 290 K. The mass flow of air to the test
section was measured using an Annubar flow meter. Since this was mounted with a number
of bends and changes of plane upstream of the measuring element, an in-situ calibration was
carried out using the LDA system to measure the exit velocity from a circular jet of 9.5 mm
diameter and relating this to the pressure differential across the Annubar (measured with the
Scanivalve DSA), as seen in figure 3.19. Temperatures were measured using a K-Type
thermocouple connected to a Fluke Type 54 hand held digital readout. The accuracy of this
device is estimated to be 1 K.

Figure 3.18 Valve Flow Rig (IEC 60534-2-3)

60
Chapter 3 Experimental Investigation and Benchmark Data

Figure 3.19 Calibration Data for Annubar

3.4 LDA Literature Review

A review of current literature that is applicable to turbulence model benchmarking has been
completed. Specific emphasis has been paid to measurements conducted in models with
limited optical access. In addition best practice for comparison measurements has been
sought. LDA has been used extensively in turbulence model validation, see (Escudier et al,
2006), (Popescu, 2007), (Talbi et al, 2009) and (Gardin et al, 1999). Point measurements of
velocity through a flow path are determined from particles traveling through the probe
volume. Due to the finite size of the measuring volume the LDA data is not really a point
measurement but is integrated over the measuring volume. For near wall measurements the
finite volume may cause large velocity gradients that present difficulties in accurately
locating walls. Thus near wall measurement will show dependences based on the measuring
volume size and may require a volume correction. The MS path was of sufficient width and
depth to enable direct measurements of the velocities without the need for a scale model.
However transverse path measurements proved difficult, as will be detailed in section 3.3,
and measurements were primarily taken along the centerlines of the restrictive chambers and
within the expansion zones. The redirection of the flow in the MS path resulted in the seeding
fluid depositing on the optical window. The test apparatus did not facilitate removal of this
residual in a timely manner. It was necessary to review the current literature to determine best

61
Chapter 3 Experimental Investigation and Benchmark Data

practice in comparing turbulent simulations with direct measurements, based on the LDA
data available.

(Amato et al, 1990) completed an investigation into turbulence in an internal combustion


engine wherein five variations of the k- model were compared directly to point
measurements within an engine piston. Five different k- models were used to simulate the
turbulent velocities within the reciprocating chamber for speeds up to 1,000rpm for a
compressible water / air mixture. The primary variables for comparison were mean swirl
velocity, turbulence velocity and turbulence length scale. Amato conducted the measurements
because of the limited benchmarking information available for turbulence models, for high
rpm engines. Tests were conducted at a compression ratio of 18:1 and an rpm of 1000. The
combination of compression and piston revolutions meant that a purely optical model was not
feasible and an actual single cylinder Ruggerini four stroke, direct injection diesel engine
with a 100mm bore and a 95mm stroke was used, as seen in figure 3.20. The optical access
was limited to a top view along a bore radius located 4mm below the piston head. A
lubricating ring was added to the moving piston to ensure that it would run without
lubrication which would foul the optical window.

Figure 3.20 Optical Layout of Ruggerini four stroke piston with optical access (Amato, 1990)

62
Chapter 3 Experimental Investigation and Benchmark Data

The instantaneous tangential component was measured during the compression stroke based
on an average of 150 data points at 5 measurement points along the radius (at 6mm, 11mm,
15mm, 19mm and 24mm from the bore wall). The validation percentage was between 60%
and 70% with measurements taken over approximately 20 cycles. The variation in ensemble
velocity, velocity RMS and turbulence intensity were plotted at each point against the crank
angle, see figure 3.21 for details of measurements at 19mm from bore wall. For engine
velocities Reynolds decompression is not directly applicable as the mean velocity is a
function of the piston stroke and changes with time. Therefore an ensemble velocity was used
which accounts for both the motion of the piston (and hence changes in average velocity) and
local turbulent velocities.

Figure 3.21 Variation in tangential velocity, RMS and Turbulence Intensity at measurement
point 15mm for varying crank angle (Amato, 1990)

Amato made direct comparisons with the LDA measurements and the CFD simulations
completed (the 5 variations of the k- model are notation TURB1 to TURB5). Figure 3.22
details the comparison for the 5 LDA measurement points versus point values from the CFD
simulations for tangential velocity and tangential RMS. Amato stated that the computed
velocities underestimated the measurements near cylinder axis and overestimated near the
combustion chamber wall, but that the overall agreement between computation and

63
Chapter 3 Experimental Investigation and Benchmark Data

measurements were acceptable. For the turbulence intensity all models computed have lower
values than measured, except for TURB2 (Wilkins model) but are within acceptable
standards. Amato concluded the investigation by stating with a proper choice of initial
conditions a 3D model can fit with sufficient accuracy the mean swirl velocity evolution
during the compression stroke. The limited test data was used effectively to comment on the
application of a number of turbulence models.

Figure 3.22 Variation in velocity RMS across cylinder radius (encompassing the 5
measurement points) (Amato, 1990)

(Nomura et al, 2005) conducted similar research to Amato using a model engine designed for
optical access (but unable to function at high rpm and compression ratios). Nomura was
concerned with benchmarking a numerical simulation in an engine cylinder using both LDA
and Particle Image Velocimetry (PIV). The numerical model was used to investigate the
turbulence associated with the intake stroke through single or dual inlet valves. The CFD
turbulence intensities were simply calculated from the assumption of homogenous turbulence
and did not compare well to the experimental results.

The validation of any turbulent model relies on the quantity and quality of the benchmark
data. For open flow fields as analyzed by (Strachan, 2005) the access and time duration can
produce comprehensive velocity profiles. (Escudier et al, 2006) detailed a comprehensive
velocity profile of a swirl flow downstream of an orifice plate. This work was solely aimed at
producing a database of measurements for benchmarking simulations. The scope of this

64
Chapter 3 Experimental Investigation and Benchmark Data

research was to derive a flow equation for a MS path. Amato and Nomura showed that
limited benchmark data can be used to query the reliability of a simulation. The quantity of
measurements taken using the current optical model was limited but sufficient to investigate
the turbulence model used in the numerical simulation.

3.5 Non-Dimensional Analysis

The pressure drop across an orifice divided by the inlet pressure gives a non-dimensional
ratio known as the pressure drop ratio (). For a compressible fluid the velocity in the orifice
is directly proportional to this ratio. Therefore, low pressure tests will produce similar
turbulent velocities as those seen at higher pressures. The MS path is designed to operate with
a value up to 0.85 based on reservoir conditions. The flow loop had sufficient pressure
head, for the flow rate through the MS path, to generate values in this range. More
comprehensive LDA measurements could have been made with a scaled flow path but the
resultant turbulent velocities (using the current flow loop) would not have been achievable. A
compressor capable of supplying pressures and volumes to a scaled path was not available.
Further non-dimensional analysis can be conducted on the MS path to define a flow function.

The Darcy-Weisbach equation, given in equation (3.15), relates head loss (pressure loss) to
friction (fr), for a length of pipe. fr is not constant and is a function of the pipe roughness and
velocity. It is typically determined from empirical data including the Moody diagram or
theoretical relations. The Colebrook equation is used to determine fr for fully turbulent flows
in rough walled pipes. If the MS path is rationalized to a restrictive pipe, equation (3.15) can
be simplified to equation (3.16), using equation (2.3), (2.4) and (2.33). Equation (3.16)
illustrates the relevant dimensionless groups and an expression for fr. Of particular interested

!

is the pressure drop ratio and flow function  .


1U"!

 V . . (3.15)


65
Chapter 3 Experimental Investigation and Benchmark Data


Or V

 
!

V 2 1 
   (3.16)
 1U"! M!  

M

3.6 Mass Flow Rate Measurements and Discussion

The Atlas CopCo compressor was a water-cooled, stationary, centrifugal compressor


designed to supply a steady mass flow rate at a maximum supply pressure of approximately
8bar. The control valves upstream and downstream of the MS path determined the pressure
drop across it. Mass flow rate tests were completed by adjusting the upstream throttling valve
until the pressure sensor located upstream of the MS path was at 7 bar. An inlet pressure of
7bar was used based on two factors. Firstly the maximum pressure drop across the system
would be atmosphere at the outlet, which gave maximum values of 0.846. Secondly the
steady state flow condition was confirmed by checking for fluctuations at the flow meter and
upstream pressure sensor. Increases in the inlet pressure up to 8 bar caused minor fluctuations
in the supply mass flow rate which were caused by the compressor. The small increase in
backpressure created by using the upstream control valve to throttle the inlet pressure to 7bar
eliminated the transient artifact.

Mass flow rate measurements were taken at a series of values by adjusting the downstream
control valve. The resultant experimental plot was compared to the theoretical mass flow rate
determined in Chapter 2, see figure 3.23. The experimental data for < 0.6 indicated that the
mass flow rate determined from the mathematical model was approximately 25% greater than
the experimental data. It was evident that the assumptions given in section 2.5 for MS path Cd
values, did not take into account the overall reduction in flow efficiency (mass flow rate per
pressure drop) of the complex MS path. However, the general trend of mass flow rate versus
was apparently accurate. A correction factor of 0.75 to account for the oversimplification
reduced the overall error to within 7%. Equation 2.8 was modified to include the empirical
correction factor (denoted Coefficient of MS Path or CMS) to account for the reduced mass

66
Chapter 3 Experimental Investigation and Benchmark Data

flow rate due to the assumptions given in section 2.5, see equation 3.17. Equation 3.17 is
specific to the MS path under investigation where CMS = 0.25.

: $ ;< +20   -.  (3.17)

During the development of the MS path geometry, multiple configurations were tested for Cv
and checked for pressure balancing across the restrictive sections. The segmenting of the total
pressure drop across the paths was achieved by iteratively increasing the restrictive channel
cross sectional areas until each stage took an equivalent pressure drop. The Cv of the MS
paths without expansion chambers, between the restrictive sections, was less than those that
had expansion chambers. The research showed that the flow remained converged from
restrictive channel to restrictive channel when no expansions chambers were used. When
expansion chambers were introduced between the channels, partial alignment was evident.
The assumptions given in section 2.5 stated that the mathematical model assumed fully
developed flow before each restriction. This was consistent with the experimental data used.
The correction factor of 0.75 accounts for the partial alignment that occurred due to the use of
expansion chambers in the final MS path design reviewed in this thesis.

The largest variation in mass flow rate was at values of > 0.6. At = 0.612 the rate of
change of mass flow rate decreased significantly. Choked flow is defined by no increase in
flow rate. At values of > 0.6 the flow rate did increase continuously up to the capacity of the
flow rig but the increase from point to point was negligible. From figure 3.23 it can be
concluded that the flow rate reaches a limited condition at = 0.612 where further increases
in pressure drop only increase flow rate by less than 1% of the full flow rate scale. This
limited condition is for all practical purposes a choked flow condition. From figure 3.23 and
the research presented in Chapter 2 the choked flow mechanism associated with the limited
flow condition does not appear to be defined by a fixed velocity (shock wave) at any
restriction. A CFD investigation into the properties of the flow is required before further
comment can be made. The LDA velocity and turbulence measurements will give additional
information regarding the flow field and benchmark data for the CFD simulation.

67
Chapter 3 Experimental Investigation and Benchmark Data

Comparison of Theoretical and Experimental Mass Flow Rate


0.12
Experimental Mass Flow Rate

Theoretical Mass Flow Rate


0.1
Corrected Empirical Mass Flow
Mass Flow Rate, kg/sec

Rate from Chapter 2


0.08

0.06

0.04

0.02

0
0 0.2 0.4 0.6 0.8 1 1.2
Pressure Drop Ratio,

Figure 3.23 Mass Flow Rate Versus Pressure Drop Ratio for theoretical model and
experimental results

3.7 LDA Velocity Measurements and Discussion

LDA velocity measurements were completed after the mass flow rate trend was plotted. From
the data presented in section 3.4.4 the velocities associated with > 0.6 were of most interest
in defining the limited flow condition. The data presented is for four tests at two different sets
of conditions, as seen in table 3.2 and table 3.3. Figures 3.24, 3.25 and 3.26 illustrate the
variation of resultant mean velocity, U (given in equation (3.18)) with distance along the flow
channel S (given in equation (3.19)), in TMA-1, TMA-2 and TMA-3, respectively. The
velocity component in the x direction, Ux, of a flow can be calculated using equation (3.20).
Where fD is the measured Doppler frequency and is the wavelength of the laser light.

For all tests each measured data point is obtained from an average of (typically) 1000
samples or (typically) 10 seconds of data, whichever condition is achieved first. The
relatively low measurement count was due to the fouling of the optical window as described
68
Chapter 3 Experimental Investigation and Benchmark Data

in section 3.4. Steady state flow was confirmed using both the flow meter and upstream and
downstream pressure sensors.

{ +{  {< (3.18)

+    (3.19)


{ (3.20)
 [

Test 1 Test 2
Inlet Pressure (bar) 4.3 6.34
Outlet Pressure (bar) 1.07 1.248
Pressure Drop Ratio, () 0.751 0.803
Mass Flow Rate (kg/s) 0.0525 0.0549
Table 3.2 Test Conditions for TMA

Figure 3.24 TMA-1 Resultant Velocities Versus Distance Along Restriction

69
Chapter 3 Experimental Investigation and Benchmark Data

Figure 3.25 TMA-2 Resultant Velocities Versus Distance Along Restriction

Figure 3.26 TMA-3 Resultant Velocities Versus Distance Along Restriction

70
Chapter 3 Experimental Investigation and Benchmark Data

The fourth measurement point (TMA-2d) for TMA-2 Test 1: Run 2 had a validation of 37%
for the x-axis and 51% for the y-axis. The resultant velocity at this point showed a large
discrepancy from Test 1: Run 1 which had a 97% validation on both axes. Due to the low
validation percentage the resultant velocity for TMA-2d obtained from Test 1: Run 2 can be
ignored. The validation percentage for the majority of measurements was above 85% with the
average lying at 92%. The average count before the end of the timing period (where 1000
successive counts was not possible) was 453 (15.3 counts per second) and 253 (8.8 counts per
second) for Ux and Uy respectively. For TMA, the complex three dimensional nature of the
passage geometry made interpretation of the measurements difficult.

Figure 3.25 (TMA-2) depicts the typical peak velocity profile associated with a VC (see
figure 2.4, (Perry, 1949)). From the mean velocity data under consideration it appeared the
flow stream was converging towards the VC and diverging downstream as expected. The
location of the peak velocity was skewed toward the lower end of the chamber. The
expansion zone below this may have some influence on the VC location. Figure 3.24 (TMA-
1) illustrated a similar singular peak velocity located at the start of the restrictive channel.
This was contrary to the expected location of the VC. Some fluid anomaly had to be
occurring further downstream of this to push the primary restriction back down the channel.
Figure 3.26 (TMA-3) portrays a much more complex mean velocity profile through the
restriction chamber. It appeared that two VCs existed in series with a large velocity, followed
by a decrease, followed by another peak velocity. Velocity profiles of this nature are unlikely
to be associated with a simple convergent and divergent flow stream.

A similar series of measurements was taken from TMB using the conditions given in table
3.3. The exact conditions for TMA could not be replicated for TMB due to the limited test
period. Control of the pressure drop was achieved by manual adjustment of a downstream
throttling valve. Three tests were conducted for two pressure conditions. These extra pressure
conditions gave additional comparison points. Figures 3.27 to 3.30 detail the resultant mean
velocities along the throttling sections (TMB-1 to TMB-4).

71
Chapter 3 Experimental Investigation and Benchmark Data

Test 1 Test 2
Inlet Pressure (bar) 4.3 6.78
Outlet Pressure (bar) 1.23 1.69
Pressure Drop Ratio, () 0.714 0.75
Mass Flow Rate (kg/s) 0.0521 0.0525
Table 3.3 Test Conditions for TMB

Figure 3.27 TMB-1 Resultant Velocities Versus Distance Along Restriction

72
Chapter 3 Experimental Investigation and Benchmark Data

TMB-2
120

100

80
Velocity (m/s)

60

Test 1: Run 1
40
Test 1: Run 2
Test 2: Run 1
20

0
30 30.5 31 31.5 32 32.5 33 33.5 34
Vertical Distance from the Datum Position (mm)

Figure 3.28 TMB-2 Resultant Velocities Versus Distance Along Restriction

Figure 3.29 TMB-3 Resultant Velocities Versus Distance Along Restriction

73
Chapter 3 Experimental Investigation and Benchmark Data

Figure 3.30 TMB-4 Resultant Velocities Versus Distance Along Restriction

As seen with TMA-2 there existed a number of measurement discrepancies for TMB-2,
TMB-3 and TMB-4 (see TMB-2e, TMB-3e and TMB-4c, for the largest values). For all
points the measurements from Test 1: Run 1 had larger validation percentages. As validation
percentage measures the variance between counter results, the values obtained for Test 1:
Run 1 have greater statistical accuracy (as a greater number of particles were crossing fringes
without interference) and will be used as the primary reference. The velocity profiles in TMB
are much more uniform than for TMA. Figures 3.27 and 3.28 (TMB-1 and TMB-2) depict
steadily increasing profiles without any distinct peak velocity. Figures 3.29 and 3.30 (TMB-3
and TMB-4) have similar profiles with no well-defined peak velocity or noticeable trend. For
TMB-3 and TMB-4 the velocities tend to fluctuate around the mean through the restrictive
chambers.

The mean velocities determined from the LDA probe are the arithmetic mean () of all
velocities measured. The standard deviation value (u) indicates the variation of the mean.
The root-mean-square mean (RMS- urms) quantifies the magnitude of the average (removing
the negative variables).  and u are related to urms by equation (3.21). Turbulence intensity
(or turbulence level) is the measurement of the relative level of turbulence in a flow and is the

74
Chapter 3 Experimental Investigation and Benchmark Data

ratio  to urms. In comparison u measures the variance of velocity measurements at each


point. A large standard deviation would indicate multiple results of varying quantity, as
opposed to a small standard deviation which would indicate a narrow spread of repeated
measurements of limited quantity. The standard deviation can be used to determine whether
values being compared to measured results are statistically significant.


V1    (3.21)

The standard deviation for the measurements taken is relatively large which is consistent with
a highly turbulent flow field. The turbulence intensities have been plotted through the
restrictive sections for both TMA and TMB, see figure 3.31 and 3.32. The x-axis on figures
3.31 and 3.32 has been non-dimensionalised for distance along the respective restriction
channel. It was evident that the largest turbulence intensities were concurrent with the fluid
phenomena seen in TMA-1 and TMA-3.

Figure 3.31 Turbulence Intensity values along the restrictive channels in TMA

75
Chapter 3 Experimental Investigation and Benchmark Data

Figure 3.32 Turbulence Intensity values along the restrictive channels in TMB

The geometry of the MS path clearly generates intense turbulence by design. The large
difference between  and urms means a considerable quantity of the flow at each measurement
point is in the opposite (negative) direction. The reversing flow must reduce the overall
increase in  and hence mass flow rate as increases. This could also explain the complex
flow structure seen in TMA-3 which is evident in the velocity profile (see figure 3.26). If
increases in produced increased recirculation of flow the rate of increase of  must decrease
and this decrease could be the mechanism for the limited flow condition seen in figure 3.23.
This limited flow mechanism does not preclude the possibility of a shock wave forming in
one of the restrictions at a low temperature (reservoir high inlet pressure) as indicated in
Chapter 2. As discussed in section 3.4.4 a further investigation into the flow field properties
using a CFD investigation were required before any conclusions regarding the limited mass
flow rate could be substantiated. As velocity and turbulence intensity are the primary
properties of concern further research was required to ensure the CFD simulation used a
suitable turbulence model.

76
Chapter 3 Experimental Investigation and Benchmark Data

3.8 Summary

A detailed overview of the LDA measurement method was presented. The critical properties
affecting LDA measurements were investigated. Further study was conducted into previous
experiments completed using both restricted access and fully assessable models. The flow rig
and LDA measurement system, as required to complete the testing stipulated in the
conclusions of Chapter 2, were detailed. The properties of the LDA probe volume and
seeding particles were determined. The test models and measurement points were presented
and notated for further reference. A non-dimensional analysis was completed using the
Darcy-Weisbach equation and a flow function was defined. Tests were completed using the
flow rig to plot mass flow rate versus . The theoretical mass flow rate plot, given in figure
2.21, was compared with the experimental results and discussed. LDA velocity measurements
were completed for values of > 0.6. A correction factor accounting for flow alignment was
defined.

The velocity profiles associated with TMA indicated well-defined VC locations. For TMA-3
there was clear evidence of a complex flow structure. The results for TMB were more
uniform with no apparent regions of convergent / divergent flow streams. A condition of
limiting flow was reached at = 0.612 and this was linked to the high turbulence intensity
values measured using LDA. In conclusion further comments regarding the practical values
of c could not be substantiated without a CFD simulation. The CFD simulation would
require an appropriate turbulence model too correctly replicate the velocities.

77
Chapter 4 Computational Fluid Dynamics and Turbulence Model

CHAPTER 4

COMPUTATIONAL FLUID DYNAMICS AND TURBULENCE MODEL

4.1 Introduction

This chapter details the 3D CFD investigation into c and the resultant choked flow
mechanism. The mass flow rate tests and LDA velocity measurements, completed in Chapter
3, were used to benchmark the CFD simulations. Based on the conclusions of Chapter 2
additional high pressure simulations using reservoir data have been completed. In conclusion
the choking mechanism within the MS path is explained and values for c are determined and
an equation for the gas expansion factor (Gy) is defined. All fluid modelling presented herein
was conducted using ANSYS CFX version 11.0 with Beta Code for PR EOS.

4.2 Turbulence Model Overview

The experimental benchmark data presented in Chapter 3 identified a number of flow


phenomena in the restrictive channels using the mean velocity measurements. It was shown
that the flow within the MS path was highly turbulent. The conclusion of Chapter 3 was that
further investigation using a computational simulation was required. The turbulence model
has a large influence on the mean velocity computed in CFD simulations, as it is used to close
the Navier Stokes equations. Therefore, the selection of a suitable turbulence model is critical
in enabling further investigation into the choked flow mechanism in the MS path. The
benchmark data presented will be used to validate the selection of the turbulence model. An
overview of the Navier Stokes equations with Reynolds decomposition (RANS- Reynolds
Averaged Navier Stokes) and turbulence modeling is given in appendix D.

78
Chapter 4 Computational Fluid Dynamics and Turbulence Model

From the turbulence research completed (see appendix D) it was evident that the current two-
equation turbulence models in use are based on non-physical relationships which are
corrected using statistical relations unrelated to the flow. The aim of this thesis is not to
resolve the problem of turbulence but rather to determine a value for Cv and investigate the
choked flow mechanism in the MS trim. Further mathematical analysis of the turbulent
movements would certainly not resolve the primary aim of this research. Therefore an
empirical study investigating various turbulence model applications was conducted. The
intention of this research was to investigate CFD simulations of geometries similar to the MS
trim and support the suitability of the chosen turbulence model with published literature.

4.3 Turbulence Model Application Review

(Saad, 1999) detailed a number of turbulence models including zero-equation, one-equation


and two-equation representations. The models presented had their own advantages,
disadvantages, limitations and appropriate flow regimes. It was apparent from Saad research
that there is no universal turbulence model. In order to determine a suitable turbulence model
for the computational analysis of the MS path an empirical review of simulations conducted
on similar geometries was completed.

(El-Behery et al, 2009) applied a number of turbulence models to the study of a planer
asymmetric diffuser, shown in figure 4.1. The diffuser was of particular interest as it
represents a simplified version of the MS path, where a small throttling section opens
abruptly to a large volume. In the diffuser there was separation and reattachment of turbulent
shear layers due to adverse pressure gradients. El-Behery selected the diffuser for turbulence
model comparison as both experimental data and a LES were readily available. The RANS
models chosen for the comparison included; standard k- (SKE), Low Reynolds Number k-
(LRNKE), Reynolds Stress Modelling (RSM), standard k- (SKW) and Shear Stress
Turbulence k- (SST). In addition a four equation model called v2-f (V2F) was implemented
via User Defined Scalars and User Defined Functions. The abbreviations El-Behery used are
related to figures 4.2 and 4.3 only.

79
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Figure 4.1 Diffuser Geometry under investigation (El-Behery, 2009)

The V2F model is similar to the SKE model but incorporates some additional near-wall
turbulence anisotropy and non-local pressure strain effects. V2F also uses separate transport
equations to resolve the Reynolds Stresses. The V2F model was first introduced by (Durbin,
1995). All the turbulence models were applied to the diffuser geometry at three different
mesh intensities (fine, coarse, very coarse). Figure 4.2 details the turbulent kinetic energy (k)
profiles for the various turbulence models for a fine grid with both experimental and LES
data included, versus a dimensionless distance (x/H) along the diffuser. As expected the four
equation V2F model performed best, demonstrating reasonable agreement with experimental
and LES data (both experimental and LES data are within 5% of each other). Figure 4.2
shows that the SST and SKW models are relatively poor at estimating k. The near wall
comparisons across all profiles for the SST model were consistently less than the
experimental results. For the mid-stream values k was seen to be overestimated by the SST
model.

80
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Figure 4.2 Turbulent kinetic energy (k) versus non-dimensional distance from the diffuser
chamber, (El-Behery, 2005)

Figure 4.3 Axial mean velocity profile for RANS models and Buice-Eaton experimental data,
(El-Behery, 2005)

Figure 4.3 details the axial velocity profiles through the diffuser for the various turbulence
models. The V2F and SST models remain in good correlation with the experimental data
(taken from Buice-Eaton). As the flow separates downstream the LRNKE, SKE and RSM
models show increasing divergence from experimental data. In conclusion El-Behery stated
that the V2F model was closest to the benchmark data, followed by SKW and SST models.
The highest level of correlation between these models was seen when using the finest grid.
The SKE, RSM and LRNKE models were noted for their poor performance. While the

81
Chapter 4 Computational Fluid Dynamics and Turbulence Model

estimated mean axial velocities demonstrated a high level of correlation for the SST, V2F and
SKW models values for k were far less consistent. All the models presented had similar
computation time except the RSM model, which was approximately 70% longer than the
others.

(Iaccarino, 2001) investigated a similar diffuser geometry with two turbulence models (V2F
and LRNKE) computed using three difference commercial CFD codes (CFX, Fluent and
Star-CD). As seen in El-Behery the LRNKE model showed poor agreement with
experimental results. However it was also shown that the different CFD codes had variations
as high as 20% for the exact same turbulence models. Iaccarino identified the primary causes
for this difference as variations in grid generation and the CFD technique used with the
boundary conditions.

In addition to fluid separation through the MS path there is a change in direction in the
sequential throttling channels. This directional change creates an impact at the expansion
chamber wall, as seen in figure 4.4. (Celic, 2006) researched a cool turbulent jet impacting at
a right angle onto a hot surface, as seen in figure 4.5. This model is a simplified version of a
turbulent stream impacting the blade of a turbine. Of primary importance to Celic was the
heat transfer onto the impact surface. Similarly to El-Behery, Celic used a number of
turbulence models and compared them to experimental data (for this specific model there was
a number of different benchmarking results available and Celic used (Baughn et al, 1989)).
The turbulence models used were all based on Reynolds decomposition and included the
various k- and k- models as well as the V2F model (modified by Celic). In addition,
turbulence limiters we applied to all models to assess their improvements. The application of
the limiters was to reduce the excessive growth of k and vT in the region of the stagnation
point. This stagnation point anomaly is a feature of classical two equation turbulence
models.

82
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Impact Surface

Figure 4.4 Impact zone in the expansion Figure 4.5 Simplified impingement model
chamber for the MS path for cool turbulent jet impacting on a hot
surface, (Celic, 2006)

Celic determined values for both mean stream wise velocity and Nusselt number, Nu (being
the ratio of convective to conductive heat transfer across a boundary), along the impact
surface. The k- model was presented using the Nusselt number, while k- was presented
using the mean velocity. As seen in figure 4.6 the k- model performed well with the limiters
applied. There was a high degree of correlation in the impact zone and a divergent
relationship with the experimental data further downstream. For the MS path it is important
to note that the flow turns through 90 a short distance after impact rather than continuing
along a surface. The k- results not given in detail as Celic notes they exceed the
experimental data by more than 40%. The V2F model correlated best with the benchmark
data with improvements downstream of the impact zone, as seen in figure 4.7.

Figure 4.6 k- turbulence model versus experimental data (Celic, 2006)

83
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Figure 4.7 v2-f turbulence model versus experimental data downstream of the impingement
zone, (Celic, 2006)

(Dalbello et al, 2005) investigated another diffuser example (with similar geometry to figure
4.1) where only STT, k-, Spalart-Allmaras (SA) and Explicit Algebaric Reynolds Stress
(EASM) turbulence models were used. These results were compared directly with Laser
Doppler Velocimetry (LDV) results given by (Obi et al, 1991). Figure 4.8 shows a
comparison of the various turbulence models against the experimental results. Dalbello
concluded that the SST and the EASM models provided the best predictions in comparison to
Obis data. The k- model again showed least correlation with the experimental data and in
addition predicted separation near the top wall of the diffuser which did not occur. The SST
model was computed for 3 different mesh sizes and results for all meshes were consistent
with the experimental data, see figure 4.9. This apparent mesh intensity independence is
explained further in section 4.5.

84
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Figure 4.8 Comparison of turbulences models and LDV data, (Dalbello et al, 2005)

Figure 4.9 Comparison of SST model for three different mesh gradients, (Dalbello et al,
2005)

Based on the research conducted by El-Behery, Iaccarino, Celic and Dalbello it is evident that
both the V2F and SST models are most suitable when considering flow through a diffuser
and impacting against a surface. The SST model uses k- for near wall modelling and thus
Celics conclusion regarding k- should apply to the SST model (this will be further
examined in section 4.5). As the MS path is similar to both the diffuser and impacting jet the

85
Chapter 4 Computational Fluid Dynamics and Turbulence Model

use of either of these turbulence models would be acceptable. (Durbin, 1995) presented the
original set of equations for V2F, but these are incomplete and cannot be used directly with
UDF. A commercial code is currently available from Cascade Technology (for FLUENT) but
at the time of research had not yet been released for CFX. Therefore the SST turbulence
model was used and its suitability is supported throughout the literature reviewed. In further
research it is recommended that a commercially available V2F code is compared with the
SST results being presented in chapter 5. As seen in the literature all turbulence models were
benchmarked against experimental data in order to assess their suitability. It is clear from the
research presented that the selection of any particular turbulence model would have a large
effect on simulation velocities. Section 4.5 details the STT model and adds the remaining
variables to table D.1 (see appendix D) to form the closure equations.

4.4 SST Turbulence Model Closure Equations and Near Wall Treatment

Turbulent flows are significantly affected by the presence of walls, where the viscosity-
affected regions create large gradients in the solution variables. (Anderson, 2005) detailed the
boundary layer work completed by Prandtl. For flow near a surface there generally exist three
regions including the no-slip condition at the surface, a boundary layer where the effects of
friction are dominant and a fully turbulent region. Within the boundary layer a reduced set of
NS equations can be applied, see (Sharma, 1986) for a derivation. Due to the large gradient in
flow parameters, the solution of these boundary layer equations requires a suitably dense
mesh. As such a number of wall functions have been developed as part of the turbulence
models to reduce the computational time required. Early wall functions were based on the
log-law velocity profile in the boundary layer previously described. The Wilcox model (k-)
uses the pressure gradient to reduce grid dependence.

(Salim, 2009) investigated a strategy for meshing and near wall treatment based on the non-
dimensional wall distance y+. Three y+ values are chosen, where y+ < 5 is in the viscous sub-
layer, 5 < y+ < 30 is in the blending region and y+ > 30 to 60 is in the fully turbulence zone.
Salim investigated two models of flow including flow over a flat plate and flow past a square
obstruction using various turbulence models (including k-, k- and SST). For each model
three different mesh configurations were presented based on the three values of y+ (2.5, 17.5
86
Chapter 4 Computational Fluid Dynamics and Turbulence Model

and 32.5). It was shown that the mesh configurations for the standard two equation turbulence
models performed better against the experimental data for skin friction, with increasing mesh
near the wall. However, for the SST model the results were consistent regardless of the mesh
configuration.

The development of the SST model was driven by the need to predict separation in
compressible flows with high pressure gradients. The k- model is substantially more
accurate than k- in the near wall layers and has, therefore, been successful for flows with
moderate adverse pressure gradients, but fails for flows with pressure induced separation. In
addition the -equation shows a strong sensitivity to the values of in the free-stream
outside the boundary layer. The free-stream sensitivity has largely prevented the -equation
from replacing the -equation as the standard scale-equation in turbulence modeling, despite
its superior performance in the near wall region. (Mentor, 1993) proposed a blending function
to take advantage of both k- and k- without the requirement for user interaction. The
primary additional complication in the model is the requirement to compute a wall distance
(an input into the blending function).

(Mentor, 2003) noted that the SST model benefited from the underlying robustness of both k-
and k-, particularly the Wilcox near wall formulation. The closure equations for the SST
turbulence model are given in addition to table D.1 (see table 4.1 below), with the blending
function (F1) given by equation (4.1). F1 is equal to zero away from the wall (resulting in the
general transport equation with the turbulence variables in table 4.1 reducing to the k-
model), while closer to the wall it changes to one. The turbulent eddy viscosity is given by
equation (4.2), where a second blending function is used (F2). As seen in (Celic, 2006) a
production limiter (Pk) is used to prevent the build up of turbulence in stagnation regions. The
constants for the model are determined from a blend of the k- and k- model constants, as
given in table 4.2.

87
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Equation S

Turbulence (k) k w   w  
? 0m  
0

Turbulence
w  p w  ? 0m  e 0 
(/)
0  21  0p
Table 4.1 Addition variables for the general transport equation for the SST turbulence model

_
 `]]- _D

 tanh : =: ,
, @ (4.1)
p< < p L <

  p
Where $p : 20p p , 10 ]

[ 
(4.2)
[ ,



 `]]-
Where  tanh ~=: , @
< <

W W1 W2 W1 W2 K1 1 k2 2

0.09 3/40 0.0828 5/9 0.44 0.85 0.5 1 0.856

Table 4.2 SST Turbulence Model constants

The dependence of the SST model on the mesh near the wall has been investigated by both
(Menter, 2003) and (Kim et al, 2005). As detailed by Menter the SST model uses a near wall
treatment which automatically shifts from the standard low-Re formulation to wall functions,
based on the grid spacing of the near-wall cell. Menter presented simulations completed using
three different grids (y~ 0.2, y ~ 9 and y ~ 100), where the computed wall shear stresses

88
Chapter 4 Computational Fluid Dynamics and Turbulence Model

varied by less than 2%. (Kim et al, 2005) presented a similar investigation comparing the
various turbulence models ability to predict near wall friction which supports the conclusions
of both Menter and Salim regarding the SST model.

4.5 CFX EOS Equations

At the time of analysis CFX used Redlich Kwong EOS and a version of the Vukalovich virial
equation- the latter only being available through user tables. As stated in chapter 2 section
2.6.1 PR EOS is the preferred real gas model. This was available in beta form and was
implemented through a patch distributed by CFX. The current release of CFX has PR EOS
integrated as a standard option.

4.6 Grid Independence

This section details the mesh used for all simulations completed and the parameters for
considering the independence of this mesh. As cited in section 4.2 the accuracy of a
turbulence model is a function of the mesh intensity. For k- models it was shown that near
wall accuracy increased in line with the node count (El-Behery, 2009). El-Behery used three
definitions of mesh namely; fine, coarse and very coarse. These definitions were based on
node count but did not define location intensity or biasing. (Salim, 2009) presented mesh
strategies for near wall treatment using the non-dimensional wall difference. It was seen that
the various strategies had little effect on the SST model. The effect of mesh intensity on the
SST model was further explained by (Menter, 2003) and (Kim et al, 2005).
The MS path was meshed using a combination of inflations and point controls. Inflation was
used along the throttling channels. The inflation was controlled using layer quantity and
expansion rate, with a specific body spacing set for each section. The individual body
spacings for each section gave optimum control over the changing geometry per stage. The
overall spacing was derived from the LDA transverse system accuracy as the model would be
compared to point velocities. The LDA transverse had a positional accuracy of +/- 0.1mm,
with a probe volume of 0.324mm x 2mm in the x-plane and 0.307mm x 2mm on the y-plane.
To ensure multiple nodes were clustered around the measurement point the body spacing was
set to 0.05, with an expansion rate of 1.1 and 30 layers and control point was located at each
89
Chapter 4 Computational Fluid Dynamics and Turbulence Model

LDA measurement point. The control points at the LDA measurement coordinates created a
number of errors in the transition to the inflation region. This was resolved by manually
adjusting the number of layers in the inflation adjacent to the specific flow channels where
these errors occurred. These settings were optimized visually until it was seen that the mesh
in the centre of the flow had sufficient nodes to enable a reasonable comparison with the
LDA point measurements.

The resultant mesh produced 1,102,332 nodes and converged in 8 hours and 32 minutes (per
the criteria given in section 5.2) using an Intel Core2 Duo CPU running at 2.66GHz with 2.00
GB of RAM. Mesh independence was confirmed using three separate parameters. The first
parameter used was a velocity profile taken across TMA-1 (and subsequently through a LDA
measurement point (see figure 4.10)). The second and third parameters were the resultant
average pressure at the outlet, and the total number of nodes in the LDA measurement
volume at various points in the path. These two parameters were considered to ensure the
mesh density did not unduly bias the average pressures and velocities being compared i.e.
increasing nodes did not affect average.

Figure 4.10 Velocity gradient location for mesh independence validation

A second, finer mesh was derived from the initial mesh by reducing the body spacing and
increasing the inflation layers. This resulted in a mesh of 1,901,231 nodes, which converged
in 20 hours and 18 minutes. Coarser meshes were not considered as the convergence time of
the original mesh was acceptable. The effect of mesh intensity on independence will be
detailed further in the results section.

90
Chapter 4 Computational Fluid Dynamics and Turbulence Model

4.7 Heat Transfer Model

The heat transfer model is used to predict the temperature throughout the flow. Heat transfer
by conduction, convection, and (where appropriate) turbulent mixing and viscous work are
included. The three model options that were available in CFX included Isothermal, wherein
only temperature dependant fluid properties are modelled, Thermal Energy which models the
transport of enthalpy through the domain and Total Energy. The Total Energy model takes
into account the transport of enthalpy but also includes the effects of mean kinetic energy.
The Total Energy model should be used when fluid Mach number exceeds 0.2, which was the
case for the MS path investigated.

The selection of the Total Energy heat transfer model enables the modelling of compressible
flow in CFX when used in conjunction with a real gas. The inlet boundary conditions need to
be set prior to the solver run. Based on the LDA results the inlet boundary was set to subsonic
(see velocities in TMB-1, figure 3.27). Supersonic conditions are generally more difficult to
resolve with the possibility of oblique shocks impinging on boundaries. For the compressible
condition the inlet boundary must be set with a pressure. If the boundary pressure is not set
CFX assigns a zero relative pressure to node 1 and during the course of the solution the
absolute pressure may turn negative. The negative absolute pressure would result in a
negative density which would crash the solution (apart from being physically inconsistent).

4.8 Boundary Conditions and Convergence Criteria

Over or under specified boundary conditions can result in nonphysical solutions or


convergence failure for numerical models. CFX Help detailed a series of recommended
conditions that produce the most robust models. Robustness, in this case, relates to the
boundary conditions which are most likely to generate a physical solution and hence a stable
convergence. The optimum constraint for convergence is the definition of inlet mass flow rate
(or velocity) and outlet static pressure (the inlet total pressure is the result of the prediction).
Due to the requirements of the Total Energy model and compressibility, pressure must be
given at the inlet. This means the second most robust model had to be used which utilizes
inlet total pressure and mass flow rate at the outlet (in this case the velocity at the inlet and
static pressure at the outlet are part of the solution).

91
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Constraining the model by defining inlet total pressure / outlet static pressure or inlet static
pressure / outlet static pressure leads to the most unreliable solution. The total pressure / static
pressure condition is very sensitive to the initial guess (mass flow is the part of the solution)
and for the static pressure / static pressure condition the total pressure boundary is
unconditionally unstable when the fluid flows out of the domain where the total pressure is
specified.

For the benchmark testing completed the total pressure at inlet and mass flow rate at outlet
were available (from Chapter 3). For reservoir conditions the mass flow rate is unknown but
as detailed in Chapter 2 (see figure 2.1), a fixed value of ensured similar mass flow rates for
higher inlet pressure and temperature conditions. Therefore simulations using reservoir data
had to be constrained by mass flow at the outlet and the outlet static pressure was checked to
confirm consistent values. Table 4.3 summarizes the LDA velocity benchmark series of
CFD simulations with boundary conditions conducted for this research.

A series of mass flow rate simulations were also completed where the outlet mass flow rate
was reduced in 15 even steps for a fixed inlet pressure (6.34bar) and the resultant outlet
pressure was recorded. These simulations were concurrent with the mass flow rate tests
completed on the MS path.

Outlet Total
Inlet Condition
Simulation Condition Temperature Reference dP
(Total Pressure),
Reference (Mass Flow K Ratio,
bar
Rate), kg/s
LDA Sim 290
4.3 0.0525 0.751
1
LDA Benchmark

LDA Sim 290


6.34 0.0572 0.803
2
LDA Sim 290
4.3 0.0521 0.714
3
LDA Sim 290
6.78 0.0525 0.750
4
Table 4.3 Summary of Boundary Conditions for CFD Simulations

92
Chapter 4 Computational Fluid Dynamics and Turbulence Model

The boundary conditions for the wall of the MS path were set to no-slip for the near wall
velocity condition (default of zero) and smooth wall for the wall roughness condition
(required by the turbulence model). The wall heat transfer model was set to adiabatic such
that there was no heat transfer across the wall.

The timescale for steady state problems is automatically determined by the CFX-Solver,
unless otherwise specified. It is based on the specified boundary conditions, initial guesses
and the geometry of the domain. The automatic setting uses a false timestep as a means of
under-relaxing the transport equations as they iterate towards a final solution. The number of
iterations for a solution is generally dependant on the initial guess. CFX states that typically
50-100 iterations are required for a robust, steady state model. The CFX simulations
completed were robust (using Total Pressure inlet and mass flow at outlet) and, based on the
LDA benchmarking, were steady state (see section 3.4). Both separation and possible
transonic flow were of concern when modeling the MS path. Initial simulations were
completed using the automatic timestep. The resulting convergence was reviewed to ensure
no noticeable fluctuations occurred which would be indicative of the automatic timestep
being too large.

A number of preliminary meshes were used to test the mesh density (with smaller node
counts than the standard mesh presented in section 5.2) and they converged within 50
iterations. When the node count was increased in stages, the convergence of the solution
started to oscillate. It became apparent that the denser mesh was recognizing transient
behavior within the flow path. This transient nature was internal to the path as opposed to the
overall steady state behavior at the boundaries. Transient flow of this nature is generally
associated with high turbulence where both large flow separation and transonic phenomena
occur. Simulations where the boundary conditions are steady state and there is an internal
transient condition are referred to as borderline transient.

A borderline transient solution can be resolved using a steady state model with a defined time
scale. Alternatively a transient simulation can be used with a defined time step. For the steady
state solution a time scale indicative of the fluid transit time is generally appropriate. In this
case if the time scale is too small transience within the flow field starts to be resolved and the
solution begins to oscillate. For transient simulations the oscillation period inherent in the
steady state solution trends the transient behavior itself. Convergence of the the final MS path
93
Chapter 4 Computational Fluid Dynamics and Turbulence Model

mesh was completed by initiating the steady state model (using the standard mesh) and then
noting the period of the resultant oscillation. The solution was changed to transient and the
time step was varied using various factors of the steady state period.

The steady state solution oscillated at a time scale of 3.044x10-5, with a resultant period of
3.44x10-3. The reduction factor for the transient simulation was 1/200 which gave a time step
of 1.7x10-7. It will be shown that this time step is approximately 1/100 of the time scale of the
largest fluctuation in the MS path.

The convergence criteria for the model were a combination of the maximum iterations and
the definition of the residual. The residual, as detailed in Chapter 4, is the measurement of the
local imbalance in the transport equations. A solution can be set to stop based on the
maximum number of iterations or the residual value. CFX uses the normalized residual with
either a maximum or RMS value option. A RMS residual value of 0.0001 or lower is a very
conservative convergence criterion, suitable for use with geometrically sensitive problems
like the MS path. Balanced convergence is a combination of RMS residual as well as the size
of the MAX residual and the overall flow balances (overall conservation). The MS path
simulations were run using the RMS residual with the target set to 0.0001.

4.9 Mass Flow Rate and LDA Benchmark Results

Two distinct benchmarks were available for comparison with the CFD simulations. The first
was the variation in mass flow rate versus and the second was the LDA mean velocity
measurements throughout the restrictions.

For the mass flow rate comparison the LDA-Sim2 model was used. The data was collected by
fixing the inlet pressure, and decreasing the outlet mass flow rate from 0.0572 kg/sec to 0.015
kg/sec in 15 equal steps. The resultant outlet pressure was used to determine and this was
plotted against mass flow rate. Figure 4.11 compares the MS path mass flow rate tests
completed in Chapter 3 versus the simulation mass flow rate and corrected theoretical results.
At the critical flow condition ( > 0.8) the CFD model started to become unstable and was
very sensitive to the initial pressure estimates (downstream pressure initial guesses were used
to stabilize the model). Figure 4.11 validated the accuracy of the CFD simulation based on
the consistency between the mass flow rate and upper limit of (c). Mass flow rates in
94
Chapter 4 Computational Fluid Dynamics and Turbulence Model

excess of 0.057 were used to test if lower values of outlet pressure were possible. The
resultant instability demonstrated by the solution crashing proved the non-physical state of
increased flow rates.

Figure 4.11 Comparison of LDA Sim 1, experimental and theoretical mass flow rates versus

The LDA mean velocity and turbulent kinetic energy (k) measurements were used to validate
the turbulence model used in the CFD simulation, based on a comparison of the mean
velocities in the restrictive sections of the MS path. CFD simulations were completed per the
boundary conditions given in table 4.3. The velocities and k values at each node were
exported to an excel file using coordinate references. A cluster of nodes was taken around
each LDA reference point. Figure 4.12 depicts the LDA probe volume in the MS flow path.
The node cluster was taken from the CFD results by filtering the coordinates using the LDA
measurement and surrounding points that fitted into the probe volume. This included the
additional offset created by transverse of 0.1mm. For each filtered measurement the mean
velocity in the coordinate plane was determined from the arithmetic mean of the all the
resultant velocities at the nodes. The resultant velocity for the measurement point was
calculated per equation 3.18. Table 4.4 is a sample calculation from LDA Sim 1 for LDA
measurement point TMA-1a. The simulation mean velocities were plotted against the LDA

95
Chapter 4 Computational Fluid Dynamics and Turbulence Model

mean velocity, see figures 4.13 to 4.18 for TMA and figures 4.19 to 4.26 for TMB. Values
for turbulence intensity (IT) were calculated using the results for k, taken in the same
coordinate plane as the velocity results and equation 4.3. The CFD results, including
comparisons to the LDA experimental values, are given in tables 4.5 and 4.6.


" > (4.3)
u

Figure 4.12 MS path with LDA measurement plane

Z Axis Velocity Velocity Resultant


X Axis (mm) Y Axis (mm)
(mm) u, (m/s) v, (m/s) Velocity
0.76 0.17 11.00 7.01 24.94 25.91
-0.11 0.23 10.50 7.20 25.34 26.34
0.10 -0.30 9.90 7.95 28.93 30.00
0.95 0.16 9.20 9.20 26.69 28.23
0.00 0.00 8.70 Reference Centre Point
0.87 -0.96 8.20 8.90 30.53 31.80
0.70 0.11 7.40 7.89 28.20 29.28
-0.72 0.56 7.10 7.62 30.15 31.10
0.08 -0.85 6.80 6.90 34.53 35.21
Component Velocities (m/s) 7.83 28.66
LDA RUN 1 Comonent Velocities (m/s) 8.29 32.06
Mean Resultant
TMA-1a Velocity (m/s)
29.57

Table 4.4 Mean Velocity Calculation from LDA Sim 1 at TMA-1a

96
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Figure 4.13 TMA-1 LDA and CFD Mean Figure 4.14 TMA-1 LDA and CFD Mean
Velocity Comparison, x = 0.751 Velocity Comparison, x = 0.803

Figure 4.16 TMA-2 LDA and CFD Mean Velocity


Figure 4.15 TMA-2 LDA and CFD Mean
Comparison, x = 0.803
Velocity Comparison, x = 0.751

97
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Figure 4.17 TMA-3 LDA and CFD Mean Figure 4.18 TMA-3 LDA and CFD Mean Velocity
Velocity Comparison, x = 0.751 Comparison, x = 0.803

Figure 4.19 TMB-1 LDA and CFD Mean Velocity Figure 4.20 TMB-1 LDA and CFD Mean Velocity
Comparison, x = 0.714 Comparison, x = 0.75

98
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Figure 4.21 TMB-2 LDA and CFD Mean Velocity Figure 4.22 TMB-2 LDA and CFD Mean Velocity
Comparison, x = 0.714 Comparison, x = 0.75

Figure 4.23 TMB-3 LDA and CFD Mean Figure 4.24 TMB-3 LDA and CFD Mean Velocity
Velocity Comparison, x = 0.714 Comparison, x = 0.75

99
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Figure 4.25 TMB-4 LDA and CFD Mean Figure 4.26 TMB-4 LDA and CFD Mean Velocity
Velocity Comparison, x = 0.714 Comparison, x = 0.75

100
Chapter 4 Computational Fluid Dynamics and Turbulence Model

TMA
RUN 1 RUN 2 Aver CFD Sim 1 Error
TMA-1a 130% 154% 142% 102% 28%
TMA-1b 39% 58% 49% 46% 5%
TMA-1c 57% 84% 70% 62% 12%
TMA-1d 49% 75% 62% 54% 13%
TMA-1e 46% 87% 67% 56% 16%
TMA-1f 90% 113% 102% 78% 23%
TMA-2a 169% 171% 170% 123% 28%
TMA-2b 96% 105% 100% 68% 32%
59% 68% 64% 56% 12%
TEST 1

TMA-2c
TMA-2d 35% 117% 76% 67% 12%
TMA-2e 43% 49% 46% 39% 16%
TMA-2f 59% 71% 65% 58% 11%
TMA-3a 91% 105% 98% 78% 20%
TMA-3b 104% 94% 99% 81% 18%
TMA-3c 133% 139% 136% 87% 36%
TMA-3d 81% 91% 86% 78% 10%
TMA-3e 73% 86% 80% 71% 11%
TMA-3f 69% 85% 77% 64% 17%
TMA-3g 74% 88% 81% 72% 11%
RUN 1 RUN 2 Aver CFD Sim 2 Error
TMA-1a 148% 156% 152% 122% 20%
TMA-1b 39% 40% 40% 24% 39%
TMA-1c 53% 54% 54% 41% 24%
TMA-1d 62% 57% 59% 51% 14%
TMA-1e 54% 51% 53% 47% 11%
TMA-1f 86% 116% 101% 92% 9%
TMA-2a 182% 184% 183% 47% 74%
TMA-2b 94% 89% 91% 80% 12%
TMA-2c 50% 57% 53% 44% 17%
TEST 2

TMA-2d 33% 42% 37% 21% 44%


TMA-2e 47% 49% 48% 42% 12%
TMA-2f 66% 62% 64% 58% 10%
TMA-3a 110% 111% 110% 98% 11%
TMA-3b 111% 86% 98% 87% 12%
TMA-3c 160% 139% 149% 122% 18%
TMA-3d 85% 82% 84% 75% 10%
TMA-3e 67% 73% 70% 58% 17%
TMA-3f 77% 69% 73% 62% 15%
TMA-3g 76% 66% 71% 61% 14%

Table 4.5 LDA versus CFD Turbulence Intensity values for TMA

101
Chapter 4 Computational Fluid Dynamics and Turbulence Model

TMB
RUN 1 RUN 2 Aver CFD Sim 3 Error
TMB-1a 48% 60% 54% 42% 22%
TMB-1b 75% 78% 76% 62% 19%
TMB-1c 47% 22% 35% 29% 16%
TMB-1d 38% 48% 43% 40% 7%
TMB-2a 34% 49% 42% 36% 13%
TMB-2b 25% 34% 29% 22% 25%
TMB-2c 28% 30% 29% 23% 20%
TMB-2d 43% 26% 35% 27% 22%
TMB-2e 33% 62% 47% 41% 13%
TEST 1

TMB-3a 42% 52% 47% 42% 11%


TMB-3b 22% 24% 23% 21% 9%
TMB-3c 49% 53% 51% 48% 6%
TMB-3d 30% 11% 20% 24% -18%
TMB-3e 36% 52% 44% 51% -16%
TMB-4a 51% 66% 58% 45% 23%
TMB-4b 58% 67% 62% 48% 23%
TMB-4c 62% 87% 75% 54% 28%
TMB-4d 54% 71% 63% 44% 30%
TMB-4e 47% 64% 56% 38% 32%
TMB-4f 59% 74% 67% 40% 40%
RUN 1 Aver CFD Sim 4 Error
TMB-1a 43% - 43% 51% -21%
TMB-1b 76% - 76% 63% 17%
TMB-1c 39% - 39% 36% 8%
TMB-1d 48% - 48% 49% -3%
TMB-2a 37% - 37% 44% -17%
TMB-2b 31% - 31% 31% -1%
TMB-2c 25% - 25% 25% 1%
TMB-2d 27% - 27% 31% -14%
TMB-2e 32% - 32% 41% -31%
TEST 2

TMB-3a 48% - 48% 46% 4%


TMB-3b 17% - 17% 30% -74%
TMB-3c 29% - 29% 52% -78%
TMB-3d 26% - 26% 31% -18%
TMB-3e 12% - 12% 60% -396%
TMB-4a 45% - 45% 55% -22%
TMB-4b 98% - 98% 55% 44%
TMB-4c 134% - 134% 60% 55%
TMB-4d 60% - 60% 44% 26%
TMB-4e 70% - 70% 43% 39%
TMB-4f 66% - 66% 41% 37%
Table 4.6 LDA versus CFD Turbulence Intensity values for TMB

102
Chapter 4 Computational Fluid Dynamics and Turbulence Model

4.10 Mesh Independence Results

Before the results given in section 5.5 are discussed further the independence of the solution
from the mesh needs to be examined. Per the guidelines given in section 5.2, the velocity
profiles across the second restriction and through TMA-1b were plotted in figure 4.27. Both
the standard and refined meshes gave consistent results for the mean flow in the profile
selected. The refined mesh clearly indentified a high peak velocity at 2mm from the channel
wall. In addition, a number of velocity spikes were seen in the middle of the flow profile. The
percentage variation in mean velocity measurements and consistency in overall profile did
not support the use of the refined mesh.

Figure 4.27 CFD velocity profiles for the standard and refined mesh

The number of nodes residing in the measurement planes as given by section 5.5, increased
by approximately 40% with the refined mesh. This can be seen in figure 4.27 where the
number of nodes around the midpoint (6mm from the wall, in the measurement volume)
intensifies. Figure 4.27 depicts a number of velocities spikes in this region for the increased
mesh. However, the overall effect across the entire measurement plane was not substantial as
the increased number of nodes averages the spikes more than the standard mesh would. The
average outlet pressures, in relation to mass flow rate, were consistent with the experimental
data (see figure 4.11). This is further detailed in the discussion in section 4.11.

103
Chapter 4 Computational Fluid Dynamics and Turbulence Model

4.11 Mass Flow Rate Discussion

Figure 4.11 supports the conclusion that the CFD simulations completed accurately predicted
the downstream pressure (per the robust boundary conditions stipulated in section 5.4). The
stability of the model at higher outlet mass flow rates did not adequately model the continued
but negligible increase in the mass flow rate for > 0.778. The pressure profile at the outlet is
relatively flat, as seen in the pressure contour given in figure 4.28 (taken from LDA Sim 1).
The largest low pressure zones occurred in the expansion chambers of TMB and the
restriction channels in TMA. As the velocity and pressure are linked by the turbulence model
the correlation between mass flow rate and outlet pressure supports the case of the SST
model.

Figure 4.28 CFD pressure contour plot for TMA and TMB from LDA Sim 1

104
Chapter 4 Computational Fluid Dynamics and Turbulence Model

4.12 LDA Benchmark Discussion

Section 3.4.5 indicated that there were a number of fluid phenomena that produced two
distinct restrictions and peak velocities (see TMA-1 and TMA-3) in the MS path. The
resultant velocities determined from the CFD simulations were consistent in trending these
phenomena for TMA, see figure 4.13 to 4.18. Tables 4.7 and 4.8 compare the average results
of the two LDA measurement runs, completed for each boundary condition, with the CFD
resultant velocity. As detailed in section 5.5, the CFD resultant velocity is the average of all
the resultant velocities at the nodes in the LDA probe volume. For reference the number of
nodes for each CFD measurement has been included. The percentage error between the LDA
and CFD results was less than 12%, except where noted in table 4.7 (highlighted in yellow).

For table 4.7 there were five points at which the error exceeded 12%. The first point (TMA-
1f) was over an expansion zone where a large velocity gradient existed. The large error at this
point was a function of the number of nodes used in the CFD resultant velocity calculation.
The next point, TMA-2d, had a very low validation rate for RUN 2 (as noted in section
3.4.5). A direct comparison of the CFD resultant velocity with the data from RUN 1 reduced
the error to within 2%. The final three points (TMA-3a, b, c) were all clustered near a known
flow phenomena and an expansion zone. Both these effects created large turbulence
intensities and velocity gradients. As indication is section 3.4.5 the particle counts for RUN 1
and RUN 2 were lower than that for RUN 3 and RUN 4.

Considering the same three points in table 4.8 where the LDA particle count increased, the
error was greatly reduced (except for TMA-3b). The only other point of interest, in table 4.8,
was TMA-2a. As no error existed in RUN1 and RUN2 for the same point the most probably
explanation for the error, was an error is small values. Taking the entire series of results
together the error values are consistent with the acceptable errors used to validate turbulence
models by (El-Behery et al, 2009), (Celic, 2006), (Dalbello et al, 2005) and (Durbin, 1995).

105
Chapter 4 Computational Fluid Dynamics and Turbulence Model

TMA Sim 1

Experimental Data From Chapter 3 From Chapter 5


LDA Point LDA Resultant LDA Resultant LDA Resultant CFD Resultant LDA Versus
CFD Node
Measurement Velocity Run 1 Velocity Run 2 Average Velocity Average Velocity CFD % Notes
Count
Reference (m/s) (m/s) (m/s) (m/s) Variance
TMA-1a 33.11 34.40 33.76 8 29.57 12%
TMA-1b 95.88 85.19 90.54 7 88.23 3%
TMA-1c 86.02 60.39 73.20 9 71.23 3%
TMA-1d 65.76 53.34 59.55 8 58.12 2%
TMA-1e 56.57 52.85 54.71 8 48.3 12%
High Turbulence Intensity Over
TMA-1f 32.78 38.98 35.88 6 26.32 27% Expansion Zone

TMA-2a 12.62 15.60 14.11 8 14.89 -6%


TMA-2b 37.94 41.90 39.92 9 35.48 11%
TMA-2c 68.82 70.78 69.80 8 75.6 -8%
Run 2 had a validation Rate of
TMA-2d 90.25 48.40 69.33 6 98.52 -42% 37%, within 2% of Run 1

TMA-2e 70.73 70.91 70.82 6 68.9 3%


TMA-2f 51.09 48.41 49.75 7 44.23 11%
TMA-3a 52.39 45.61 49.00 4 36.47 26%
High Turbulence Intensity with
TMA-3b 49.15 53.85 51.50 5 40.12 22% Large Velocity Gradients and
Low CFD Node Counts
TMA-3c 36.17 37.04 36.61 3 48.45 -32%
TMA-3d 53.11 57.09 55.10 7 56.89 -3%
TMA-3e 69.81 64.18 67.00 6 75.23 -12%
TMA-3f 66.41 65.23 65.82 8 70.78 -8%
TMA-3g 60.70 56.42 58.56 6 65.28 -11%

Table 4.7 LDA and CFD Comparison For TMA RUN 1 and RUN 2

106
Chapter 4 Computational Fluid Dynamics and Turbulence Model

TMA Sim 2

Experimental Data From Chapter 3 CFD From Chapter 5

LDA LDA LDA CFD


LDA Point CFD LDA Versus
Resultant Resultant Resultant Resultant
Measurement Node CFD % Notes
Velocity Run Velocity Run Average Average
Reference Count Variance
1 (m/s) 2 (m/s) Velocity (m/s) Velocity (m/s)

TMA-1a 9.92 10.67 10.29 8 8.5 17%


TMA-1b 99.14 96.69 97.92 7 106.59 -9%
TMA-1c 88.78 87.19 87.98 9 96.53 -10%
TMA-1d 60.68 61.89 61.28 8 63.59 -4%
TMA-1e 58.74 60.61 59.68 8 61.23 -3%
TMA-1f 36.96 31.71 34.34 6 34.23 0%
Small Value Error- Low CFD
TMA-2a 8.02 7.33 7.67 8 12.63 -65% Node Count

TMA-2b 42.87 44.62 43.74 9 44.69 -2%


TMA-2c 83.56 76.35 79.96 8 94.56 -18%
TMA-2d 89.79 89.91 89.85 6 98.57 -10%
TMA-2e 71.97 70.32 71.15 6 80.23 -13%
TMA-2f 53.16 54.65 53.90 7 58.75 -9%
TMA-3a 46.90 47.36 47.13 4 44.99 5%
High Turbulence Intensity- CFD
TMA-3b 50.99 63.79 57.39 5 36.59 36% Low Node Count

TMA-3c 29.28 37.72 33.50 3 33.26 1%


TMA-3d 59.45 66.24 62.84 7 63.4 -1%
TMA-3e 81.77 77.29 79.53 6 85.2 -7%
TMA-3f 71.04 79.35 75.19 8 76.3 -1%
TMA-3g 63.58 71.09 67.34 6 64.5 4%

Table 4.8 LDA and CFD Comparison For TMA RUN 3 and RUN 4

The turbulence intensity (IT) is the measure of the mean velocity versus the magnitude of the
variation. Section 3.7 has already highlighted that values for IT in the MS path are high. As IT
describes the fluctuating component of the velocity it was expected that the experimental and
CFD results would have a similar relationship to that of the mean velocity results. As seen by
(El-Behery, 2005) and (Nomura, 2005) this correlation is not always apparent. Nomura
(2005) commented that using k, from the CFD simulation, to determine IT resulted in a flat
distribution across the channel being measured which resulted in larger errors near the
channels walls. This was seen in figure 4.2 where the SST model consistently underestimated
IT near the diffuser wall (and over estimated it in the centre of the diffuser). In the same area
the experimental and computational mean velocities had a much greater correlation.

107
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Table 4.5 shows that the computational values of IT are on average approximately 20% in
error with some points exceeding 30%. While the quantitive accuracy of the data comparison
is relatively poor the qualitative explanation is consistent with the physics of the flow. For
TMA-1 the results tend to map the inverse of the mean velocity around the flow anomaly.
The highest values of IT are apparent either side of the peak mean velocity. Physically this
would be consistent with the edge of a vortex where the shear layer at a point is subject to
much higher fluctuations in time as the vortex grows and declines. The flow stream enters
TMA-1 from the upstream expansion chamber, wherein some re-alignment has occurred.
Downstream of this the re-aligned flow, which is approximately parallel to the channel wall,
shears against the recirculating flow of the vortex. Nearer the centre of the vortex the
fluctuation is more intense but less likely to degrade as much over time and hence the
resultant magnitude is reduced.

For the CFD mean velocities to be more consistent with the experimental results than the IT
values are, the CFD simulation must be tracking the flow fluctuations accurately but under
estimating the fluctuating peaks. This is consistent with the velocity profile seen in the mesh
independence test, see figure 4.27. Increasing the node count resolved higher values of the
fluctuations without affecting the mean result. Thus the average of the fluctuation is
consistent with the experimental results but the magnitude of the measurement is
underestimated. The underestimation could also be a function of internal mathematical
viscosity used to resolve the residuals in the time step given. The additional mathematical
viscosity added to the solution in order to dampen the equation response and create stability
would also dampen the magnitude of the velocities seen and thus reduce the computational
values of IT. This would occur as part of the relaxation technique used to solve the
compressible flow, wherein the mathematical viscosity is a by-product of the solution
equation. The mean of the fluctuation (mean velocity) would not be greatly affected. The
same mechanism is commonly seen in shock capturing wherein the location of the shock is
displaced from its physical location as the CFD solution adds mathematical viscosity to
enable the large fluctuation in the properties across the shock to be resolved.

The IT results for TMA-3 do not follow this logic, as IT is at points overestimated. From
section 3.7 it is apparent that the LDA count in this channel was less than the others. The
impact of particles into the optical window along the path meant that the total count was
reduced in the final section and the measurements were taken over the full 10 second time
period. This means the correlation between computational and experimental results has less
108
Chapter 4 Computational Fluid Dynamics and Turbulence Model

statistical significance in this section than the others. While the mean velocities consistently
correspond in this channel additional measurements could shift the LDA mean velocities.

Section 3.4.5 did not highlight any particular anomalies in the restrictive channels of TMB.
Tables 4.9 and 4.10 detail the comparison of LDA and CFD results, from figures 4.19 to 4.26.
The major errors of interest are highlighted in yellow. For table 4.9 the first two errors can be
explained by the low LDA validation rates. For TMB-1c, the LDA error positively effects the
resultant velocity by bringing it closer to the CFD resultant velocity. In the case of TMB-2a,
the comparison is negatively affected by the low validation rate LDA measurement. TMB
had only three runs completed, at two different boundary conditions, and table 4.10 highlights
3 more errors of interest. The first two were a result of having no second run for reference in
areas of high turbulent intensity. TMB-4f had a particularly low node count for the CFD
resultant velocity. As with TMA, the CFD model with the SST turbulence model was
accepted as accurately predicting the conditions for TMB.

The IT results for TMB TEST 1 displayed similar underestimation as TMA but without the
same peak fluctuations. Of most concern are the results for TMB TEST 2 where only one
LDA pass was made. The values of IT using the boundary conditions for LDA Sim 4 should
be similar to LDA Sim 3 based on the slight change in boundary conditions and both sets of
CFD data are consistent. The major difference is that two sets of LDA measurements were
not available for TEST 2. This indicates that the measurement period of 10 seconds is directly
affecting the magnitude of the experimental results. It is the last flow channel with the reduce
LDA count where this affect is most apparent. At this point the largest error for the mean
velocities was also apparent (-17% and -36%).

In addition to the resultant velocities the component velocities, u and v, were available for
comparison. The resultants were chosen as they give a better overall description of the flow
phenomena under consideration. The component velocities were comparable with the LDA
component measurements. At some points the averaging of u and v positively and negatively
affected the simulation and LDA comparison. For reference the entire series of measurements
for LDA Sim 1 are included in Appendix E.

109
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Experimental Data From Chapter 3 CFD From Chapter 5

LDA LDA LDA CFD


LDA Point CFD LDA Versus
Resultant Resultant Resultant Resultant
Measurement Node CFD % Notes
Velocity Run Velocity Run Average Average
Reference Count Variance
1 (m/s) 2 (m/s) Velocity (m/s) Velocity (m/s)

TMB-1a 60.47 59.39 59.93 6 52.36 -14%


TMB-1b 54.67 52.44 53.56 8 58.9 9%
TMB-1 RUN 2 Low Validation
TMB-1c 68.68 80.38 74.53 9 74.59 0% Rate. Higher % for RUN 1

TMB-1d 59.04 57.82 58.43 7 53.23 -10%


RUN 1 Higher Validation Rate.
TMB-2a 87.74 89.46 88.60 6 75.23 -18% Changes error to 16%

TMB-2b 83.56 93.39 88.47 8 78.52 -13%


TMB-2c 88.33 88.85 88.59 7 80.23 -10%
TMB-2d 92.87 95.63 94.25 6 107.8 13%
TMB-2e 106.20 80.08 93.14 6 98.4 5%
TMB-3a 102.15 85.40 93.78 7 95.42 2%
TMB-3b 122.44 114.13 118.28 8 124.9 5%
TMB-3c 91.26 83.37 87.31 9 80.4 -9%
TMB-3d 107.76 121.42 114.59 7 112.78 -2%
TMB-3e 103.50 84.19 93.84 6 95.47 2%
TMB-4a 91.82 83.92 87.87 6 98.25 11%
TMB-4b 61.41 59.33 60.37 7 55.23 -9%
TMB-4c 84.22 69.81 77.01 8 90.58 15%
TMB-4d 94.72 77.30 86.01 9 98.63 13%
TMB-4e 91.77 88.95 90.36 8 85.74 -5%
TMB-4f 56.91 68.33 62.62 3 50.47 -24% Low CFD Node Count

Table 4.9 LDA and CFD Comparison For TMB RUN 1 and RUN 2

110
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Experimental Data From Chapter 3 CFD From Chapter 5

LDA LDA LDA CFD


LDA Point CFD LDA Versus
Resultant Resultant Resultant Resultant
Measurement Node CFD % Notes
Velocity Run Velocity Run Average Average
Reference Count Variance
1 (m/s) 2 (m/s) Velocity (m/s) Velocity (m/s)

TMB-1a 60.60 60.60 6 55.89 -8%


TMB-1b 78.96 78.96 8 85.51 8%
High due to no averaging with a
TMB-1c 101.84 101.84 9 122.5 17% second RUN

TMB-1d 99.70 99.70 7 88.47 -13%


TMB-2a 82.72 82.72 6 76.25 -8%
No Second Run for these Conditions
TMB-2b 88.31 88.31 8 94.58 7%
High due to no averaging with a
TMB-2c 96.87 96.87 7 118.5 18% second RUN

TMB-2d 90.14 90.14 6 80.54 -12%


TMB-2e 74.27 74.27 6 69.3 -7%
TMB-3a 83.91 83.91 7 77.5 -8%
TMB-3b 132.86 132.86 8 122.3 -9%
TMB-3c 116.97 116.97 9 105.3 -11%
TMB-3d 114.76 114.76 7 101.5 -13%
TMB-3e 118.36 118.36 6 108.69 -9%
TMB-4a 53.30 53.30 6 60.58 12%
TMB-4b 36.10 36.10 7 40.12 10%
TMB-4c 79.86 79.86 8 88.24 9%
TMB-4d 78.07 78.07 9 80.3 3%
TMB-4e 72.97 72.97 8 62.3 -17%
TMB-4f 75.85 75.85 3 55.8 -36% Low CFD Node Count

Table 4.10 LDA and CFD Comparison For TMB RUN 3

4.13 Choked Flow Mechanism

There were two distinct investigations into the choked flow mechanism required. Choked
flow was apparent at the low inlet pressures (7bar) used during the experimental testing. As
discussed in the conclusions of Chapter 2 it was also shown that at high inlet pressures the
reduced temperatures could create conditions suitable for shock waves at the outlet stage
(conditions that could not be investigated experimentally). Therefore the choked flow
investigation was broken into two sections. Section 5.9.1 details the choked flow mechanism
at experimental conditions while section 5.9.2 used the reservoir conditions.

111
Chapter 4 Computational Fluid Dynamics and Turbulence Model

4.13.1 Choked Flow at Experimental Conditions

As shown in Chapter 2 the limitation of flow rate in a single stage restriction is created by the
geometrical convergence of the flow streams. In the MS path the complex geometry has no
distinctive throttling point and is designed to reduce pressure across a series of channels.
Within each channel there must exist a mechanism for dissipating the potential energy into
kinetic energy for pressure reduction to occur. (Bernad et al, 2007) investigated the mean
flow rate through a complex restriction, in this case a poppet valve. Bernard introduced a
model for quantitatively describing the vortex flow. Figure 4.29 depicts the poppet valve
geometry and the vortex analysis completed. The restriction in the valve existed between the
seat and the poppet but the three downstream vortices defined the flow rate through the
outlet. As part of the investigation Bernard suggested various changes to the geometry to
reduce the primary vortex (V1 in figure 4.29) and increase the overall flow rate per unit
pressure drop. Additional research presented by (Christensen et al, 2001), (Hendricks et al,
2004) and (Tamburrino et al, 1999) detailed how large scale vorticity dominated the
properties of a flow field for various applications where the primary variables of
consideration were velocity, pressure and temperature respectively.

Figure 4.29 Vortices in poppet valve outlet (Bernad et al, 2007)

112
Chapter 4 Computational Fluid Dynamics and Turbulence Model

(Tennekes et al, 1972) detailed the relationship between vortices and turbulent flow. All
turbulent flow is characterised by high levels of fluctuating vorticity. Turbulent flow is
defined by both diffusive and dissipative scales. The coefficient of eddy diffusion is defined
by the integral length scale, while the dissipative scales are defined by the energy transfer
rates (which equal the energy dissipation per unit mass). The extraction of energy from the
mean flow at large scales becomes balanced by the dissipation of energy at the smaller scales.
The transfer of energy between the scales is facilitated by vortices and vortex stretching.
Therefore the pressure reduction through the MS path restrictive channels was understood
using the dissipative nature of turbulence. This was concurrent with the high turbulence
intensities seen in Chapter 3. In order to dissipate high values of (or potential energy) the
velocities within the vortex structure must be large and continuously fluctuating.

Figures 4.30 and 4.31 depict the velocities within the MS path using a vector plot. The MS
path has been presented according to the layout of the LDA models TMA and TMB. For the
restrictive channels in TMB it is seen that no large scale recirculation exists. This is
consistent with the LDA benchmarking presented where no definitive peak velocity was seen
along the restrictive channels in TMB. Small vortices are seen in the transition from the
expansion chamber to the restrictive channel but these are not the predominant restriction in
the flow. In this path vortices occur in the expansion chambers which do not represent
minimum flow areas. The three restrictive channels in TMA are dominated by a number of
large vortices. In TMA-1 (see figure 4.30) a vortex with a length scale of approximately 5mm
exists in the channel whose width is just over 10mm. The fluctuation of the velocity within
this vortex had a time scale of approximately 9.47x10-5 sec.

113
Chapter 4 Computational Fluid Dynamics and Turbulence Model

Figure 4.30 CFD primary vortex structures in TMA at experimental conditions

Figure 4.31 CFD small vortex structures in TMB at experimental conditions

114
Chapter 4 Computational Fluid Dynamics and Turbulence Model

The next restriction of interest is TMA-3. Similarly to TMA-1 it is defined by a single large
vortex located centrally within the channel. At the outlet another large vortex exists but this is
associated with the expansion zone. The vortex in TMA-3 is similar in size and had a time
scale concurrent with the vortex seen in TMA-1. The velocities either side of the restriction
are much greater. The increased velocities are consistent with the overall pressure drop across
this channel which is greater than that for TMA-1.

It was concluded from the size and flow direction that these two vortices were dictating the
effective flow area within these two channels. The length scales of both vortices account for
over 50% of the width of the path on the plane shown. Further sizing of the vortices could be
completed by taking multiple planes at varying depths within the restrictions. The actual
widths of the channel increase through the MS path and this was done to balance the pressure
drop. From table 2.1 it is seen that the rate of increase of TMA-3 is far greater than TMA-1
(7.14% for TMA-3 and 5.33% for TMA-1). This is because the vortex in TMA-3 is greater in
overall area than the vortex in TMA-1. At the choked flow condition the mass flow rate
through the MS path was defined by the vortex structure at TMA-3 and the resultant VC
formed.

The velocity profile given in figure 4.27 was taken at the edge of the vortex in TMA-2 and is
consistent with the velocity increase away from the centre of the vortex. In this case the
beginning of the VC is clearly located in an area between 1.5mm and 3mm from the wall.

4.13.2 Choked Flow at Reservoir Conditions

Figure 2.18 (reproduced below as figure 4.32, with addition notation) was given as the most
accurate determination of the JTIC. Initial conditions of temperature and pressure which
reside within the loop on the y-axis are classified by cooling on pressure reduction. For
values outside of the loop a temperature increase is apparent. As stated by Bessieres a number
of EOS give widely varying results for the JTIC. The simulations completed used CFX PR
EOS which was shown to perform poorly versus the Monte Carlo technique and experimental
data. The PR JTIC given by both Kortekas (see figure 2.16) and Bessieres were in agreement.
Both figures show the PR JTIC predicting temperature drops for high pressure inlet
conditions where a temperature increase is apparent. The JTIC and EOS within CFX are
115
Chapter 4 Computational Fluid Dynamics and Turbulence Model

inherently linked and Bessieres did not offer an alternative means of determining Z within his
research. Therefore, there is no facility for readily updating the JTIC within CFX without
affecting or redefining the EOS model.

Figure 4.32 JTIC Results from (Bessieres et al, 2006) with additional notation

Both the PR and Monte Carlo JTIC agree within certain ranges. An increase in temperature
through the first stage of the pressure reduction in the MS path would ultimately increase the
outlet temperature. The research present here was most interested in quantifying the effects of
a low outlet temperature on the choked flow mechanism within the MS path. Therefore,
simulations were completed for temperature decreasing boundary conditions that existed
within the JTIC loop. From figure 4.32 it is seen that the optimum condition for temperature
decrease is 300K at 47MPa. Further increases in pressure require higher inlet temperatures to
produce cooling. The rate of change on the inlet temperature requirement is far greater than
the rate of change of cooling. The boundary condition of 47MPa and 300K is also within the
PR JTIC. At these inlet conditions a series of simulations was completed using increasing
values of . The simulations of most interest were where > 0.6 and the outlet stage
temperature decreased below 220K.

Figure 4.33 details the temperature change through the MS path for = 0.806. The outlet
stage was of most interest and it was apparent that a significant temperature drop occurred in
this location. Figure 4.28 showed that the pressure drop through the MS path was evenly
distributed. If the velocities in the MS path were sufficiently low, the temperature change
would be expected to occur in place with the pressure drop. As indicated by Stiles, the

116
Chapter 4 Computational Fluid Dynamics and Turbulence Model

resultant temperature changes actually occurred further downstream due to the relatively high
velocities. The extremely low temperature outlet condition in the last stage gave low values
for ua (approximately 276m/s based on pure methane).

Figure 4.33 CFD temperature contour for inlet conditions of 300K at 47MPa and = 0.806

Figure 4.34 details the Mach number contours through the last stage of the MS path (TMB-
4). It was evident that the local velocities exceeded Mach 1, based on the local speed of
sound calculated. Downstream of a shock, as seen in figure 2.1, the velocities should continue
to increase. This is not apparent in figure 4.34 and the temperature contour in figure 4.33
showed a temperature recovery downstream of the last stage. The Mach number contour of
the last stage, given in figure 4.34, does not indicate a definitive location for a shockwave
(typified by a large variation in flow properties across a membrane). The problem with
defining the location of shock waves in turbulent flow was detailed by (Zank, 2002) in figure
4.35. In a highly turbulent flow field the shock wave becomes highly distorted by the
repeated chaotic interaction of upstream turbulence with the shock. The problem in defining
the shock wave is in the computation of the mean velocity, position and variance of the
quantities in the presence of upstream turbulence. In addition the upstream turbulence

117
Chapter 4 Computational Fluid Dynamics and Turbulence Model

characteristics and their subsequent decay as it is dissipated into the downstream flow are
required.

Figure 4.34 CFD Mach Number contour at outlet for inlet conditions of 300K at 47MPa and
= 0.806

Figure 4.35 Interaction of a shock wave and turbulent flow (Zank, 2002)

118
Chapter 4 Computational Fluid Dynamics and Turbulence Model

The research presented here is limited to defining the choked flow mechanism for the MS
path. It was evident that a complex shock wave existed in the outlet of the MS path at high
inlet pressure and low temperature conditions. However, this shock wave occurred
simultaneously with the high upstream turbulence conditions which limited flow rate. It was
not conclusive that the outlet shock formed a limiting velocity within the MS path under the
conditions investigated. It has been concluded that the large vortex structures did form that
limiting flow area and that velocities at these available areas were comparable to the overall
flow rate.

4.14 Conclusion

In conclusion the mass flow rate through the MS path was defined by large vortices occurring
in the restrictive section of TMA-1 and TMA-3. Distorted shock waves occur under certain
operating conditions but they were not the pre-dominant restrictive phenomena. Gy was
solely defined by c (given as 0.812) and the rated Cv value for the MS path can be
determined from the experimental data given in Chapter 3 and equation (2.10). The MS path
is part of a much larger assembly referred to as an MS trim which is located in a choke valve.
The choke valve itself has an internal annular area which directs the bore fluid into the MS
path where it is separated into individual MS paths. A choke sizing equation must incorporate
all the elements of the valve assembly in order to form an equation for sizing a valve with a
MS trim. Further flow tests had to be conducted on a full valve assembly to ensure c values
and mass flow rate results were applicable to the full valve assembly.

4.15 Summary

The boundary conditions and solver criteria for the MS path simulations were presented. A
series of parameters for investigating mesh independence were also stated. The results of the
simulations completed were shown to be comparable to both the experimental mass flow rate
and velocity measurements made. Based on the benchmark CFD model further investigations
were made into the choked flow mechanism at two conditions; namely experimental and
reservoir. It was shown that at experimental conditions that choked flow occurred due to

119
Chapter 4 Computational Fluid Dynamics and Turbulence Model

large vortices in TMA-1 and TMA-3. At increased inlet pressures (reservoir conditions)
downstream temperatures decreased and created the parameters for a shock to occur. The
nature and location of this shock was difficult to examine based on the turbulent flow field.
This shock occurred concurrently with the large vortices and it was concluded that no
preference for limiting velocity occurred with the shock. The pre-dominant phenomena of
limiting flow in all cases were the vortices previously seen. As the MS path is part of a full
valve assembly it was concluded that further testing would be required before a final sizing
equation could be defined.

120
Chapter 5 MS Trim Flow Tests

CHAPTER 5

MS TRIM FLOW TESTS

5.1 Introduction

The MS path is part of a much larger assembly called a disc stack trim, as described in
Chapter 1. The opening of the MS paths is controlled by an internal linearly actuated plug.
The Cv equation used to size the entire valve is a function of this collection of paths and the
upstream geometry. A flow test of an entire valve assembly had to be completed to ensure the
Cv equation for the single MS path relates to the full valve. The affects on mass flow rate
through the combined system of MS paths were compared to mass flow through the single
path analysed. The inherent characteristic, as described in section 1.3, was plotted and related
to the overall rate of change of mass flow.

5.2 Flow Test Literature Review

At the time the literature reviewed was completed there was no standard specifically
designated for the flow testing of MS valves. The literature review presented here details two
existing flow test procedures and highlights concerns regarding accuracy and suitability for
using them in the testing of the MS valve. In conclusion, a new procedure is presented and
applied (see section 5.3).

Flow tests of meters and measurement systems have historically been defined by international
standards to ensure consistency in the transfer of valuable fluids. An international test
standard for valve Cv was first published in 1985 (ISA S75.1). Previously valve
manufacturers published proprietary equations and values for Cv. The ISA standard used a

121
Chapter 5 MS Trim Flow Tests

combined Cv flow test and equation first published by (Driskel, 1984). This method was
combined with an IEC standard to form (ANSI/ISA-75.01.01 (IEC 60534-2-1 Mod)-2007).
Both these standards were specific to single stage trims. In Driskels original work, separate
conditions were detailed for profiled flow ports (where the port was not sharp edged but had
curved or shaped edges), these conditions did not appear in the current standards. Due to the
individual nature of a MS path there was no international standard for flow testing or valve
sizing. IEC 60534-5-7 described a series of equations that could be used for MS valves but
these were limited to a very specific geometry and only 5 stages. The IEC 60534-5-7
geometry is detailed in figure 5.1.

Figure 5.1 Multistage valve trim as used in IEC 60534-5-7

Figure 5.1 details a flow path that is created by a series of internal cages. The port of each
cage is offset to the other such that flow must travel along a channel before exiting the
sequential port. This type of arrangement is somewhat common in clean service industries,
but does not represent the MS path reviewed (the MS path was designed specifically for high
pressure, high temperature service with entrained particles- a specialized service for oil and
gas production). In order to determine a Cv value for the full valve assembly, as detailed in
figure 5.2, it was necessary to review the current single stage flow test procedure (IEC
60534-2-3) to understand the limitations for multistage testing.

122
Chapter 5 MS Trim Flow Tests

Figure 5.2 A full MS trim assembly in a choke valve body

The test rig detailed in IEC 60534-2-1, as given in figure 5.3, consisted of the test valve
(specimen), upstream and downstream throttling valves, pressure sensors at either side of the
test valve, a flow meter and a temperature sensor. The standard required that a number of
pipe length distances were used upstream and downstream of each component of the rig. For
example, a run of 18 straight pipe diameters was required before the first pressure sensor as
well as a further two downstream of the test valve, before the downstream pressure sensor.
Additionally, the pipe diameter was to be equal to the end-connections of the test valve.

Figure 5.3 Flow Test Apparatus (IEC 60534-2-3)

The IEC test arrangement described could cause inaccuracies in the determination of Cv
values. When the test valve represented a high resistance to flow in the line, slight errors in
pressure calculation had negligible effects on Cv. However, when a valve was fully opened

123
Chapter 5 MS Trim Flow Tests

the effect of small pressure errors increased. The error in pressure measurement had two
sources- the pressure sensors and pipe losses. The reading error in the sensors could be
accounted for and their calibration requirement was governed by the standard. The pressure
loss through the recommended straight pipe runs could not be accounted for. Driskell (1984)
stated that since a control valve and the process pipe are seldom the same sizes, it is not
even feasible to compensate for this situation by subtracting these pipe lengths in
calculation. The IEC standard stated that the accuracy of the tests were only within 5% for
Cv
2
< 0.0465 , where d 1 was the internal diameter of the upstream pipe. By increasing the
d1
overall pressure drop across the valve, the pressure loss in the pipe had a decreasing affect
(not a linear relationship as the pressure loss is a function of flow rate and pipe surface
Cv
finish). For the MS valve assembly was substantially less than 0.0465, as the MS path is
d12

highly restrictive in relation to the line size, even at the fully open position.

The IEC flow test procedure used for compressible fluids is given below in points A to E
below.

A. Install test valve in flow rig as determined by standard


B. Flow tests need to be conducted at predefined intervals / positions along the
valve stem travel. At each interval / position steps C to E are repeated. At each
position complete 3 separate tests at 3 different pressures. To ensure
incompressible flow the pressure ratio (x = P/P1) must be less than 0.02.
C. The following data is to be recorded at each test
C.1) valve travel
C.2) inlet pressure
C.3) pressure differential (outlet pressure)
C.4) fluid inlet temperature
C.5) volumetric flow rate
C.6) barometric pressure

124
Chapter 5 MS Trim Flow Tests

D. Based on the incompressible flow approximation of the above data can be


used to calculate Cv for each of the three pressure differentials, using equation
(5.1).

1
CV = Q (5.1)
P1 P2

E. If the Cv values calculated at each differential pressure are not within 4% of


each other three additional pressure tests are required to check for flow
stability. The final Cv value at the valve travel point is the arithmetic mean of
the 3 values.

As detailed in section 2.3, Gy for a single stage valve trim was based on a linear interpretation
of test data. This linear equation was used for all single stage trims, therefore, using a low
pressure drop for an incompressible approximation in a gas test was acceptable. In some
cases valve manufacturers simply used values of Cv from water tests and the linear gas
expansion equation for the sizing equation. While not explicitly prohibited by the IEC
standards, Driskels original work noted that a Cv duality could occur within complex valve
geometries where both gas and water tests were not comparable. The dual Cv theory is
contrary to orifice plate standards which traditionally used a single discharge coefficient for
both gas and liquid flows. The orifice plate equation was based on experiments conducted on
plates with a fully developed flow profile with negligible swirl and no flow fluctuations, (Z.
Husain, 1995). It was proposed by (Z. Husain, 1995) that deviations from this standard flow
arrangement introduced uncertainty in orifice plate calculations based on changes to the
discharge coefficient. The effects of profile distortion, swirl flow and flow fluctuations at an
orifice would have the same effect on a valve port. These effects would most likely occur in
valves where the internal geometry promotes them. Driskell warned that valves with
smoothly contoured orifice passages may have a Cv for liquid that is different from the gas
Cv. Due to the geometrical nature of the MS path the assumption of a single Cv value for
both liquid and gas was questionable. In addition the characterization of Gy as linear required
confirmation across the range of .

125
Chapter 5 MS Trim Flow Tests

For the experimental determination of c the valve flow must be choked. In a control valve of
10.16cm (4) nominal size an immense amount of gas was required to reach choked flow.
IEC 60534-2-3 advocated that if direct values of c could not be determined then results could
be extrapolated based on the approximate linear relationship between Gy and . With the
valve in the fully opened position and the inlet pressure at the highest achievable for the flow
rig the downstream pressure was to be varied. With five back pressures in a wide a range as
possible the flow rate, temperature, pressure drop and test fluid molecular weight were to be
recorded. Using equation (5.2) values of the product GyCv could be plotted against .

1
;< $- (5.2)
+D 

A typical plot, given in figure 5.4, was used to extrapolate c at Gy= 0.667. If any of the test
points deviated from the linear estimate by more than 5% the valve was seen to be exhibiting
anomalous behaviour and a full test for c should be completed (Driskel, 1983). If anomalous
behaviour was evident it was recommended that a full plot of the relationship for Gy versus
be completed. It will be shown in section 5.2.2 that the full relationship for the MS trim
exhibits varying Gy values based on trim plug position.

Gy Cv = Experimental
Point

0.667*GyCv
= pressure ratio

Figure 5.4 c characteristic for a single stage valve

In addition to the IEC industrial standard, (Wu-Shung, 1998) proposed a new method for
determining the valve coefficient using data acquisition techniques. The approach was
specific to gas applications, so it was only applicable to the calculation of Cv and c. The
technique enabled a blow-down system to be used where a constant pressure drop was not
required and thus the total gas reservoir could be reduced. As a result of the data acquisition

126
Chapter 5 MS Trim Flow Tests

frequency, the method could also be applied to valves under transient flow conditions. In
addition, the method eliminated the need for a flow meter which had benefits over the
standard IEC technique, as well as transient flow control valves, where transient flow meters
were not available. The proposed test rig, as seen in figure 5.5, used high frequency data
acquisition sensors. Unlike the IEC rig, the flow meter was excluded and additional
temperature and pressure sensors were placed in the reservoir. A PLC was used to ensure
synchronisation between valve openings and data acquisition.

Figure 5.5 Alternate flow test apparatus (Wu-Shung Fu, 1998)

The primary advantage of the Wu-Shung method was the economic use of high pressure gas.
By removing the requirement for a fixed pressure drop, the reservoir pressure charge could be
maintained longer. The primary cost in high pressure testing is reservoir charging. This was
an important aspect in the flow testing conducted on the MS valve. As indicated previously
the standard IEC test did not account for Gy at high pressure drop but used a linear
approximation, based on low pressure test data.

The Wu-Shung test procedure used a mathematical method to determine the flow rate for the
test. Assuming adiabatic flow, (based on the short time to fully release the test gas from the
reservoir) the relationship between temperature and pressure, as well as mass and pressure,
was defined for the reservoir, using the EOS for ideal gases. Mass flow was determined from
the EOS by differentiating with respect to time yielding equation (5.3)

127
Chapter 5 MS Trim Flow Tests

1
dm 1 mr ,t =i Pr ,t dPr ,t
mt = r ,t = (5.3)
dt Pr ,t P dt
r , t =1
Where r subscript is for reservoir, t is subscript for time and i is for initial conditions.

According to (Saad, 1993) mass flow rate could be expressed in terms of upstream static
pressure, the stagnation temperature, Mach number and cross-sectional area, see equation
(5.4) (also reference Chapter 2 section 2.2 equations (2.2), (2.3), (2.4)).

Pu ,t 1 2
mt = AM u ,t 1 + M u ,t (5.4)
Tr ,t R 2

Where subscript u represents upstream conditions of test section

Experimentation defined the pressure conditions with respect to time such that equation (5.3)
was solved for mass flow rate (mass flow rate related to rate of change of reservoir pressure
and the initial conditions). Equation (5.4) was then solved for Mach number where the
upstream conditions were determined from the same experimentation information as equation
(5.3). The static temperature at any time t was established for upstream test section conditions
(denoted subscript u) per equation (2.4) (repeated below as equation (5.5) for reference and
with subscripts).

Tr ,t
Tu ,t = (5.5)
1 2
1+ M u ,t
2

The same method was applied downstream of the test piece where the subscript was denoted
d.

Wu-Shung proposed that Cv and a new dimensionless mass flow rate factor Gt (given in
equation (5.6)) should be used to characterise the flow in a valve. Gt was determined either
upstream or downstream of the valve based on the temperature as identified in equation (5.6)
and the experimental results from the pressure sensors at any time t.

128
Chapter 5 MS Trim Flow Tests

m RTu ,t
Gt = (5.6)
APu ,t

The experimental procedure to determine Cv and Gt was similar to the IEC standard except
that the pressure sensors accumulate data on the pressure reduction in the reservoir as well as
upstream and downstream of the test section at a sampling rate of 500 Hz. The resultant curve
depicting pressure drop in the reservoir is given in figure 5.6. Wu-Shung proposed a least-
squares-fit for the pressure in the reservoir, by slightly modifying a governing equation
derived by Saad. The errors between experimental calculations of reservoir pressure and the
mathematical fit were less than 6%. The initial section of test data was further inspected to
check for choked flow. A pressure sample across the test section was taken from time equal
to 0.4 seconds and time equal to 0.9 seconds. It proved that there was no increase in flow rate.
Thus the reservoir volume and pressure drop was sufficient to give a good prediction of c
from choked flow to low pressure drop, see figure 5.7.

Figure 5.6 Reservoir pressure reduction (Wu-Shung, 1998)

129
Chapter 5 MS Trim Flow Tests

Figure 5.7 Wu-shung choked flow check (Wu-Shung, 1998)

Wu-Shung determined both the Cv and c factors from plotting the product of the expansion
factor Gy and Cv versus the pressure drop data, see figure 5.8. The Cv value for each stem
opening position was determined at the intersection of the y axis and the resultant plot. At
this point the Gy value was unity and the pressure drop was zero. The c valve was
determined at Gy = 0.667, as seen in figure 5.8. It is interesting to note that Wu-Shung found
errors up to 10% between the experimental values in figure 5.8 and the nearest linear fit. The
IEC standard stated that such large variances indicated flow anomalies and that the standard
did not cover such circumstances.

130
Chapter 5 MS Trim Flow Tests

Figure 5.8 Review of Wu-Shung Cv and Gy results (Wu-Shung, 1998)

The test results presented by Wu-Shung Fu concluded that the IEC calculated value of Cv
(56.3- using equation (5.1)) was 16% less than the Cv extrapolation identified in figure 5.8.
However, Wu-Shung did not highlight the IEC recommendation for accuracy in the
linearization of Gy values. The values at the lower end of c were the most inconsistent and
based on the reservoir reaching near zero pressure were probably taken at the smallest time
interval. Before Wu-Shungs conclusion can be justified, a number of low pressure drop tests
in the region of c < 0.1 should be conducted for an extended duration. IEC found greatest
consistency in their results at small valves of c and hence used them to extrapolate the critical
value. Wu-Shung found the opposite using his method.

In conclusion, for flow testing the MS valve the use of the IEC method presented a number of
issues. For the MS trim the main trend of interest was the variation in expansion factor with
pressure drop ratio for fixed Cv positions. This required multiple tests at each valve open
position to determine the change in Gy. The Wu-Shung method had practical advantages over
multiple fixed pressure drop tests as all the varying data could be computed during one test.
However it was evident that results to date had not been consistent with IEC requirements.
Therefore a modified version of the IEC standard was used for flow testing, as detailed in
section 5.3.

131
Chapter 5 MS Trim Flow Tests

5.3 MS Valve Flow Test


5.3.1 Test Apparatus

The valve opening positions in the MS trim assembly equated to one section of fully opened
flow paths or one disc, see figure 5.9. To characterise Gy at each position a series of
increasing pressure drop tests were completed. For a single stage trim it was shown that
whatever the valve position the resultant Gy characteristic could be linearized. As stated
previously the limit of Gy for a single stage is 0.667. The National Engineering Laboratory
(NEL) in East Kilbride Scotland had a facility capable of testing to pressure drop ratios of
approximately 0.8. As values above this level were seen to increase flow rates negligibly,
during MS path testing, the limits of the facility were deemed acceptable. A line diagram
schematic of the choke valve installed in the gas flow facility is given in figure 5.10. The
choke body was installed horizontally in the 100mm test line with both inlet and outlet
flanges also horizontal. Figure 5.11 is a view of the choke installed in the facility. The flow
rig used air compressors to discharge dry air to outside storage vessels (70 bar, 12m3
capacity) that were connected to pressure regulating valves inside the laboratory. These
supplied the ambient temperature gas at a pre-set pressure to a sonic flow nozzle. The gas
mass flow rate was determined by measuring the local nozzle pressure and temperature. After
the nozzle the gas entered the 100mm test line pipework and then went into the choke before
passing through the back-pressure control valve (used to adjust the choke downstream
pressure) and exhausting to atmosphere. The test pipe-work was set up according to IEC
guidelines with a maximum working pressure of 1400kPa.

Figure 5.9 Two discs compressed together making one valve open position

132
Chapter 5 MS Trim Flow Tests

Figure 5.10 Piping Diagram for MS Valve Flow Test

Figure 5.11 MS Valve Installed in the NEL Test Facility

The test pressures were measured using static (at inlet and outlet) and differential electronic
pressure transmitters. The differential pressure sensors used ranged between 100 and 500
kPa. For measurements above a 500kPa differential, the inlet and outlet static pressure
readings were used. Below the 500kPa range the difference between the static and differential
pressure sensors was less than 50kPa. The gas temperature at the choke inlet was recorded
using a Platinum Resistance Thermometer located ten diameters upstream of the choke inlet.

The uncertainty on the gas flow measurement was 0.5% of reading with flows > 0.3kg/s and
below this the uncertainty was 0.7% The pressure measurement sensors were calibrated

133
Chapter 5 MS Trim Flow Tests

prior to the tests and had an uncertainty of measurement of 0.2% of reading. Gas
temperature has an uncertainty of 0.2%C.

5.3.2 Flow Test Procedure


Two separate series of tests were conducted. The first test (A) characterised the Cv across the
full valve travel. The actual Cv value at each valve travel point could not be calculated until
Gy was determined. The procedure for test (A) recorded all the variables required by the IEC
standard as noted in section 5.1 C. The second test (B) was used to characterise Gy for
various stem positions. The operating conditions for test (A) and (B) required separate setups.
Test (A) was run first in order to commission the flow rig and to verify flow stability. The Gy
equation from test (B) was combined with test (A) results and equation (5.9) to give
experimental Cv values.

1
$- B (5.7)
C + D

Test (A) was accomplished by charging the external reservoir via the compressors and
discharging the gas into the test lines (with pressure drop controlled by upstream and
downstream valves). As part of the pre-test calculations the total mass flow rate was required
for each stem position. The depletion of the reservoir was determined for each test and the
remaining pressure charge used for additional tests. The charging of the reservoir took
considerable time and test (A) was designed to ensure multiple test points across the entire
valve travel with only one reservoir charge. Preliminary CFD simulations were conducted at
each valve travel point to determine the mass flow rate. The predicted mass flow rates were
used to establish the reservoir charging sequence and ultimately the test costs. As detailed in
section 5.3 the mass flow rate simulations were completed inversely with P2 determined by
stepping the outlet mass flow rate.

Table 5.1 details the results for the experimental variables, as stipulated in section 5.1 C.
Once the flow tests were completed the preliminary CFD models used to determine the
reservoir charge were updated with the actual test pressures / temperatures / flow rates and re-
run for completeness. Table 5.2 details the values determined from the CFD simulations
versus the experimental mass flow rate and the resultant variance.

134
Chapter 5 MS Trim Flow Tests

Table 5.1 Flow Test Measurements

135
Chapter 5 MS Trim Flow Tests

Table 5.2 Comparison of CFD and Experimental Mass Flow Rates

Test (B) was used to describe Gy for a number of stem positions at various values up to 0.8,
in approximately 0.1 steps. At each step the flow rate was recorded and the GyCv value
determined using equation (5.2). At large values of stem positioning (large valve openings)
and high values of , the flow rig was operating at maximum capacity. Test (B) was
completed at a total of four stem positions equating to the lower range of the trim. Each series
of tests at the various positions required the reservoir to be recharged and thus testing was
limited to four positions. For valve openings greater than the fourth position the flow rate at
= 0.8 did not provide sufficient time for the test to reach a stable condition before the
upstream pressure decreased. Results for all tests completed are given in table 5.3. The
resultant GyCv value was graphed against , as seen in figure 5.12.

136
Chapter 5 MS Trim Flow Tests

Table 5.3 Test (B) Y Characterization for Four Valve Positions

137
Chapter 5 MS Trim Flow Tests

Figure 5.12 GyCv Versus for Four Valve Positions

5.3.3 Flow Test Results and Conclusions

As tends towards zero the flow approaches an incompressible approximation and Gy 1.


From figure 5.12 at = 0, Cv is equal to 5.81, 9.48, 13.48 and 17.43 at the four test positions
respectively (see figure 5.8 for technique). From positions 2-4 the increase in Cv for stem
position is equivalent, with a Cv of approximately 4.0 per stage. It was evident that the
upstream annular area in the valve had little effect on the total Cv of each disc in the MS trim.
The relative area of the valve annulus was very large compared with the MS port opening, as
seen in figure 5.13 (a magnified view of figure 5.2). At increased trim openings more of the
valve annular area affects the inlet of the trim discs. The MS path is primarily a complex
geometrical restriction. It was evident that the external flow pattern had no significant effect.
The affects of the valve annular area on inlet pressure are seen in a contour plot of pressure
from full valve CFD simulations, given in figure 5.14. The pressure contour changes
dramatically in the y direction as the 3-D vortex structure narrows at the base. Figure 4.28,
taken 4mm either side of the central plane, shows that both TMA-1 and TMA-3 act as
transitional channels for the pressure drops. TMB-1, TMB-2 and TMB-3 have relatively
steady pressure drops across them whereas TMA-1 and TMA-3 naturally take a higher

138
Chapter 5 MS Trim Flow Tests

pressure drop. This is associated with the increased velocities and dissipation of kinetic
energy in the vortex structure. The contour in figure 5.14 is taken 2mm from the down from
the top of TMA and shows the variance in the pressure profile as the plane moves in the y
direction. There were no low pressure zones in the annulus or significant vortices upstream of
the MS trim inlet. The pressure drop was evenly balanced across the paths, as assumed in
Chapter 2 and shown in Chapter 4.

Figure 5.13 A comparison of valve annular area versus MS trim opening

Figure 5.14 Pressure contour for MS trim in valve assembly

139
Chapter 5 MS Trim Flow Tests

The Cv associated with the first stage was approximately 5.81. The valve assembly was
designed to shut-off flow against an angled seat. The plug actuates linearly within the disc
stack with a tolerance on the disc internal diameter and the plug external diameter. Once the
plug has been unseated flow can leak by the tolerance. The leak-by is constant and not
additive and thus only seen at the first position. This accounts for the larger Cv seen at the
first opening.

The mass flow rate per stage, ignoring the first stage which includes flow-by, was
approximately 0.259 kg/sec for the second disc and reduces to 0.2381 kg/sec for the third set.
At the same values the single MS path tested had a mass flow rate of 0.0512 kg/sec. There
were 6 flow paths per disc in the valve assembly tested which, assuming no upstream or
downstream flow affects, produced a combined mass flow rate of 0.306 kg/sec. The full trim
assembly evidently reduced the overall mass flow rate by approximately 18%-28%. As seen
in figure 5.14 there were no low pressure zones in the annular area of the valve so the
reduction in overall flow rate occurred downstream of the path outlets. Figure 5.15 details the
downstream velocity using a vector plot. As seen in figure 5.14 there were no large upstream
vortices or low pressure regions. Downstream of the outlet there were a number of significant
vortices and as the flow paths opened the flow complexity increased. The reduced mass flow
rate through the third disc was indicative of the increasing vortex structures and their affect
on the overall flow field. The variation in flow rate through the discs was ultimately
accounted for by the Cv versus stem travel plot, given in figure 5.16.

140
Chapter 5 MS Trim Flow Tests

Figure 5.15 Velocity vector plot through the MS trim assembly with downstream vortices

Figure 5.16 Cv versus stem travel inherent characteristic for the MS valve

141
Chapter 5 MS Trim Flow Tests

5.4 Summary

This chapter detailed the current standards and industrial procedures for the flow testing of
control valves. A new procedure was proposed and implemented in the testing of the MS
trim. The resultant mass flow rates were compared to CFD simulations and were seen to be
comparable. The product of GyCv was plotted against and the value of c confirmed, and the
valve inherent characteristic determined. It was shown that the mass flow rate per disc was
18-28% less than the extrapolated mass flow from a single path. CFD was used to show that
this was a result of the large downstream vortices and turbulence impeding flow. In addition,
pressure contour plots showed the variance in the pressure drop across the channel as a
function of y. As the number of discs openings increased the rate of change of mass flow rate
decreased. This relationship was shown by the inherent characteristic.

142
Chapter 6 Discussion and Conclusions

CHAPTER 6

DISCUSSION AND CONCLUSIONS

6.1 Introduction

The primary aim of this research was to determine the rated Cv of the MS trim so that a valve
sizing equation could be derived. In determining the valve size for an application, the limiting
pressure drop ratio (c) is required. The flow field mechanism that limits the flow rate and
effective pressure drop through the MS trim was investigated in order to establish c. This
chapter will review the research, discuss overall observations and outline conclusions from
the work. In addition, possible topics for further work are suggested.

6.2 Summary and Discussion

The control choke is the primary pressure control device in the production system for an oil
and gas reservoir. In the case of a gas well, the reservoir pressure can range from 69MPa to
138MPa. The processing equipment and transportation pipeline is typically rated for much
lower operating pressures. Historically the control choke was a variable area, single stage
restriction. The pressure drop from the reservoir to the downstream systems was taken across
a single orifice. For gas wells, at large pressure differentials, the flow becomes choked due to
the formation of a shock wave in the vena contracta (VC). The shock wave generates a
number of undesirable side-effects including; high levels of noise caused by the interaction
between turbulent flow streams and shock waves, flow induced vibration and high velocities
within the valve. Gas wells can contain sand and increased levels of velocity amplify erosion
within the valve.

143
Chapter 6 Discussion and Conclusions

A MS trim was designed by Cameron FLC for high pressure gas reservoir applications with
the following characteristics;

Large pressure drops are balanced across multiple stages, such that the pressure drop
ratio () at each stage is insufficient to form a shock wave.
The choke valve is sized based on the area required to control the flow rate for a given
pressure drop across it (using the non-dimensional valve coefficient- Cv).
Multiple conditions exist during the life of a reservoir with a range of Cv values called
the controlling range.

The primary aim of this research was to determine the Cv value for the new geometry, the
associated value of c, and to use these to size the new valve for industrial applications. The
first step was to create an empirical model of the MS path using;

A correction factor for the expansion of the gas (Gy) through the multiple paths of the
MS trim (see (Anderson, 2003), (Driskell, 1983), (Buckingham, 1933), (Bean, 1935),
(Kinghorn, 1986), (Seidl, 2002) and (Kegel, 2002))
Experimental data to replicate the unique path of the MS trim. Data from thick walled
orifice plates, commonly referred to as restriction plates was used, see (Rhodes,
1969). This data was non-dimensionalised using the ratio of orifice wall thickness to
diameter ().
An Equation of State (EOS) which considered the effects of the compressibility of gas
under high pressure conditions. It was shown that the Peng Robinson equation was the
most suitable (Peng et al, 1976).
A Joule Thomson (JT) coefficient that described the cooling of the gas on expansion,
see (App, 2009), (Kortekass et al, 1997), (Bessieres et al, 2006), (Vrabec et al, 2007)
and (Parsafar et al, 2000). The cooling affected each stage differently. It was shown
that at very high inlet pressures heating would occur before cooling. Also that cooling
would reduce the speed of sound and decrease the velocity required to form a shock
wave.

144
Chapter 6 Discussion and Conclusions

The flow rate through the MS path was calculated with the assumption of full pressure
recovery and alignment between each consecutive restriction. The change in mass flow rate
was plotted against , see figure 2.21. It was shown that the conditions of choked flow did not
exist at any of the MS path stages and that theoritically the flow rate would increase until =
1. It was proposed that choked flow would occur at values of < 1. As the temperature of the
fluid reduces due to expansion and JT cooling, the speed of sound creates near critical
conditons or values of M 1. The over simplification of the MS path meant further
investigation into the choking mechanism using experimental and computational tools was
recommended.

The experimental research into the MS path flow dynamics was completed using a flow loop
and a Laser Doppler Anemometry (LDA) apparatus. A number of difficulties were
encountered in completing the mean velocity and IT measurements. The complex geometry of
the MS path meant the LDA seeding impacted and fouled the optical window. This limited
the total time available for measurement runs and ultimately reduced the total number of
particles available for counting in the final two restrictive channels. The LDA runs were
limited to either 1,000 counts or 10 second of measurement. In addition the wide of the
restrictive channels and the refractive index of the optical window meant measurements could
only be taken centrally down the channels. The experimental research included;

Plotting the mass flow rate versus and comparing it to the empirical model. It was
shown that rate of change of flow rate decreased rapidly at 0.6. The empirical
model and experimental data was consistent with 25% of measurement of each other
below this point, see figure 3.23. A correction factor was employed to reduce the
overall error to within 7%.
A literature review, which was used to examine the properties of the LDA apparatus
used and confirm the methodology of measurements in a confined space, see (Amato
et al, 1990).
Plots of the mean velocities along the restrictive channels for different values of , see
figures 3.24 to 3.30.
Plots of the turbulence intensity (IT) along the restrictive channels for different values
of , see figures 3.31 and 3.32.

145
Chapter 6 Discussion and Conclusions

The LDA measurements completed indicated the potential locations for VCs. The second
pressure drop stage (TMA-2) depicted the typical peak velocity profile associated with a VC
near the orifice opening (Perry, 1949), see figure 2.4 and 3.25. For the first restriction the
location of the peak velocity was skewed toward the lower end of the chamber, see figure
3.24. The final stage had a much more complex mean velocity profile. It appeared that two
VCs existed in series, with a large velocity followed by a decrease, followed by another peak
velocity, see figure 3.26. Velocity profiles of this nature are unlikely to be associated with a
simple convergent and divergent flow stream. The associated values of IT measured were
very high which was consistent with a very turbulent flow exhibiting large areas of
recirculation / fluctuation. The velocity profiles in the lower plain were much more uniform
than for the upper plain. In conclusion further comment regarding the practical values of c
could not be substantiated without a Computational Fluid Dynamics (CFD) simulation.

The values of mean velocity in the CFD model would be a function of the turbulence model
chosen. As the benchmark data (see chapter 3) used mean velocity and mass flow rate
measurements the turbulence model needed to be determined prior to any simulations. The
selection of the turbulence model was completed by;

Reviewing the three primary turbulence movements including; deterministic,


structural and statistical. This including work by; Boussinesq (1877), Reynolds
(1894), Prandtl (1925), Taylor (1935) and Kolmogorov (1941), see appendix D.
Reviewing the derivation of a two equation turbulence model (k-). This review
incorporated the Boussinesq theory of turbulence viscosity (vT), supported by the
Prandtl mixing length theory.
Noting that the experimental relationships in the k- equations still described the
original hypothesises of the statistical movement which are not universally applicable.
In conclusion an empirical study investigating various turbulence model applications,
with similar geometries to the MS path, was completed. This included work by; (El-
Behery et al, 2009), (Celic, 2006), (Dalbello et al, 2005) and (Durbin, 1995).

146
Chapter 6 Discussion and Conclusions

It was shown that the V2F and SST models accurately predicted the velocities within the
geometries reviewed. As the V2F model was unavailable at the time of research the SST
model was selected as the appropriate turbulence model for use.

The first steps in completing the CFD simulations conducted included;

Validating the grid density and structure used by comparing a velocity profile taken
across the first restriction at two different grid densities, see figures 4.10 and 4.27. In
addition, a cluster of nodes was taken at each measurement point which represented
the LDA probe volume. The number of nodes within the coordinate volume was a
function of mesh density. The grid was shown to be independent when an increase in
node density did not significantly affect the resultant velocity of the node cluster. The
same premise was applied to the average outlet pressure.
Selecting a suitable Heat Transfer model that modelled compressible behaviour. The
Total Energy heat transfer model was chosen because it was the only model that
would account for the transport of enthalpy but also includes the effects of mean
kinetic energy. Using this model meant that the inlet pressure boundary needed to be
specified. Using the outlet pressure to constrain the model would have produced an
unstable convergence. The mass flow rate at the outlet was used to create a very
robust model and the outlet pressure was then compared to the experimental data as a
resultant.

Once CFD model was defined the following simulations were completed;

Fifteen mass flow rate versus measurements were generated to compare to the
experimental mass flow rate tests from the flow loop, see figure 4.11. For < 0.8 the
simulation and experimental curves were within 5% of each other. At > 0.8 the CFD
simulations became less stable and very sensitive to the initial pressure estimates.
A series of simulations using the experimental benchmark data boundary conditions
were completed. The resultant velocities, calculated from the node clusters taken at
the LDA measurement coordinates, were plotted against the experimental velocity
measurements, see figures 4.13 to 4.26. It was shown that the measurements
compared were for the most part were within 12% error of measurement. This was

147
Chapter 6 Discussion and Conclusions

consistent with the acceptable criteria presented in the CFD turbulence model
literature review, see section 4.3.
A high pressure (47MPa) and low temperature (300K) simulation was completed to
investigate the possibility of a shock wave at the outlet, due to the reduced speed of
sound caused by the gas cooling.

As indicated by the LDA results, there was obviously a complicated flow field in the second
and sixth restrictive channels. The velocity vector plots from the benchmark simulations were
reviewed and three large vortices were evident (one in the second restriction and two in the
sixth), see figure 4.30. Research was presented regarding vortices and restricted flow wherein
(Zhan et al, 2009), (Siewert et al, 2004) and (Bernad et al, 2007) showed reduced flow rates
due to vortex structures.

(Tennekes et al, 1972) stated that the transfer of energy between the scales is facilitated by
vortices and vortex stretching. The expansion chambers coupled with the restrictive channels
create areas of flow separation upstream of the vortices. When the flow streams merge with
the re-circulating vortices further separation is seen. This explanation was consistent with the
mass flow rate plots generated. At increasing values of the vortex structure increases in size
and the effective flow area reduces. The vortex size reached a maximum at 0.612. The
reduced effective flow area only allows a relatively small increase in total flow rate as
increases up to 0.83.

The high pressure simulation was completed using an inlet pressure of 47MPa and a low
temperature of 300K, see figure 4.33. The temperature at the outlet produced values of the
speed of sound as low as 276 m/s. The Mach number at the outlet was shown to exceed 1, see
figure 4.34. The problem with defining the location of shock waves in turbulent flow was
detailed by (Zank, 2002). In a highly turbulent flow field the shock wave becomes highly
distorted by the repeated chaotic interaction of upstream turbulence with the shock. The
problem in defining the shock wave is in the computation of the mean velocity, position and
variance of the quantities in the presence of upstream turbulence. While it was evident that
the conditions for a shock wave did exist it occurred simultaneously with the vortices (at the
values of used).

148
Chapter 6 Discussion and Conclusions

It was not conclusive that the outlet shock formed a limiting velocity within the MS path
under the conditions investigated. It was concluded that the large vortex structures did form
that limiting flow area and that velocities at these available areas are comparable to the
overall flow rate.

The MS path is part of a full MS trim with the paths machined into a series of discs that are
compressed together, see figure 5.9. Before an equation for the full valve can be finalised a
flow test of the full assembly needed to be completed. This testing included;

Researching the current test standards and developing a new procedure based on the
original IEC equations, with a focus on completing the tests at high pressure and
determining the values of Cv at Gy = 0.
Conducting test at NEL using a gas blow-down flow loop which pressurized a
reservoir and controlled the pressure drop across the test section using upstream and
downstream throttling valves. Lower pressure tests were conducted across the full
stem travel to determine the inherent trim characteristic. Higher pressure tests were
conducted at the first 4 openings to characterise Gy.
Completing CFD simulations of an entire valve assembly to compare the simulation
and experimental mass flow rates.

In conclusion it was seen that;

The Cv per trim disc did not change as the valve opened to its maximum setting. This
proved that the upstream geometry had little effect on the inlet flow characteristic to
an individual path, see figure 5.12.
The Cv calculated for each stage was less than the multiple of the Cv values calculated
for the individual MS paths. This proved that the mass flow rate through the MS trim
was further limited. The CFD simulations used to determine the mass flow rate for the
flow tests were used to identify further vortices inside the MS trim outlet. These
vortices further restricted the mass flow rate at the outlet channels, see figure 5.15.

149
Chapter 6 Discussion and Conclusions

The flow testing on the full valve completed the research required to conduct sizing on a full
valve assembly. The rated Cv per stage was determined experimentally and the inherent
characteristic of the valve plotted (see figure 5.16). Required values of Cv are calculated
using equation 2.10, which utilize the variables c and Gy

6.3 Conclusions

Table 1.2 detailed a number of typical application flow rates and pressure drops that a MS
valve would be required to control. The research completed gives sufficient data to calculate
equation 2.10 and plot the required Cv on the rated Cv of the inherent characteristic of the
valve, see figure 6.1 and table 6.1 below. The equations size the valve in terms of
determining the number of discs required to control the range of operating requirements. By
increasing the valve size additional discs can be added to cover the application requirements.

Application "Required Cv " Plot on "Rated Cv" Inherent Characteristic of


50
MS Valve

45

40
Valve Coefficient, Cv

35 Controlling Range
30

25

20 Application Required Cv Point

15 Inherent Flow Characteristic


From Experimental Data
10 Given in Table 6.2

Valve Position (mm)


Figure 6.1 Required Cv versus plotted on the MS valve inherent characteristic for the
parameters in table 6.1

150
Chapter 6 Discussion and Conclusions

CASE NAME Well Start-Up Early Regulation Operating Restart from Shut-In

Gas Flow Rate (kg/hr) 42,127.00 52,136.00 125,132.00 38,231.00

Gas Molecular Weight (Kg/Kmol) 19.10 19.10 19.10 19.10

T1, Temperature (C) 120.00 98.00 65.00 120.00

Reservior Pressure, (Bar A) 334.40 318.00 285.00 335.70

Downstream Pressure, (Bar A) 38.00 80.20 120.00 53.00

dP(P1-P2), (Bar A) 296.40 237.80 165.00 282.70

(Pressure Drop Ratio) 0.78 0.75 0.58 0.78

F (Specific Heat Ratio Factor) 0.94 0.94 0.94 0.94

Gy (Gas Expansion Factor) 0.67 0.68 0.75 0.67

Flow Condition Choked Normal Normal Choked

Density (inlet) Kg/m3 162.48 153.82 154.56 162.48

Density (outlet) Kg/m3 27.30 62.67 98.81 39.33

Compressibility (Z) 1.01 1.07 1.05 1.01

Choked dP (Bar A) 261.73 237.80 165.00 262.75

Required Cv 7.46 10.36 26.94 6.93

Stem Travel (mm) 5.98 8.23 28.20 4.12

7.62cm 7.62cm 7.62cm 7.62cm


Valve Model
(3 Nom) (3 Nom) (3 Nom) (3 Nom)

Table 6.1 Sizing results for table 1.1

In addition to the sizing of the valve the following conclusions can be made;

An empirical model of flow was developed from experimental data using non-
dimensional values representing geometry. This was shown to be applicable to the
velocity of approach factor () for orifice plates and then extended to values of . The
empirical model proved to be accurate (within 25%) for values of < 0.6 when
compared with experimental data. The assumptions of convergent and divergent flows
in each restriction were shown to oversimplify the flow dynamics. A correction factor

151
Chapter 6 Discussion and Conclusions

of 0.75 reduced the overall error to within 7%. The correction factor accounted for the
partial flow alignment between each restriction created by the expansion chambers. In
addition, the JT and compressibility equations used accurately predicted the presence
of shock waves in the last stage of the MS path for specific inlet conditions. The
methodology of the empirical model could be applied to new geometries, using the
non-dimensional value and an applicable correction factor.
The mass flow rate experimental data proved that the MS path had a limiting flow
condition, inconsistent with the classical single stage pressure drops researched.
Figure 3.23 proved that the flow rate was not limited by a single anomaly in the
smallest cross section of flow, because the flow rate increased up to the limit of the
test rig ( 0.83).
The LDA measurements within the restrictive channels successfully identified the
location of a number of flow anomalies. The experimental data showed structures that
produced multiple peak velocities where one was expected. In addition, it indicated
the location of some peak velocities away from their predicted position. The standard
deviation (u) of the experimental data was used to show the high level of turbulence
intensity within the path, see figures 3.30 and 3.31. This was later linked to the
requirements of the flow structures to dissipate energy from large to small scales, in
order to change potential energy (pressure across the stage) to kinetic energy.
The two-equation turbulence models were shown to not be universally applicable. The
equations used theories with insufficient experimental support and correction factors
to account for properties unrelated to the actual flow. A suitable turbulence model was
selected by researching benchmarked CFD work completed on geometries similar to
the MS path. The SST turbulence model chosen using this methodology was shown to
accurately model the resultant mean velocities in the flow field when benchmarked
against the LDA data (see tables 5.3 to 5.6).
The CFD model, which incorporated the SST turbulence model, the Total Energy
thermal model and used the pressure inlet / mass flow rate outlet boundary conditions,
was shown to accurately predict mean velocity and outlet pressure. Once it was
benchmarked against the experimental mean velocity data, it was used to describe the
flow phenomena identified in the LDA measurements. The simulations were used to
prove the existence of two primary vortices. The vortices were seen to limit flow as

152
Chapter 6 Discussion and Conclusions

they increased in size but not to completely choke the VC. This model explained the
experimental mass flow rate versus plot.
Values of IT from CFD where shown to have relatively low quantitive accuracy in
comparison to the experimental data. However the qualitative information was
consistent with the physics of the flow explaining the large upstream shear
fluctuations in front of and after the vortices.
A CFD simulation completed using a specific high pressure (47MPa) and low
temperature (300K) inlet condition, was used to show the presence of shock waves in
the last restriction. This proved the JT cooling equations used for the empirical model
equated the correct conditions in the last restriction. Due to shock cell turbulence
interaction the exact location and form of the shock could not be determined. The
conditions that did generate the shock wave were also shown to create the vortices
that did ultimately limit the flow.
The flow testing completed on the full assembly proved that the mass flow rate for a
collection of MS paths was further reduced. This reduced the overall rated Cv for the
valve assembly. Using data from the flow test the inherent characteristic for the trim
was determined by plotting the rated Cv against the stem travel. Equation 2.10 was
used to calculate the required Cv and this was subsequently plotted on the inherent
characteristic to give the controlling range of the valve. CFD simulations completed
on the full assembly confirmed that the reduction in Cv occurred due to increased
turbulent flow at the outlet, rather than the inlet, of the MS trim.

6.4 Original Contributions

The original contributions made to the area of valve design and sizing include;

The design of the MS trim was completed by the author as preliminary work for this
research and is filed under the patent number IB2008053368. This is an entirely new
design with key innovative features including; large channel areas facilitating particle
entrained flow, inter-stage recovery which increases overall Cv for the flow path and
the use of LDA in the measurement of velocities in MS systems.
The MS valve is currently being used in multiple industrial applications in various
geographical locations including; Saudi Arabia, Thailand, Australia, Oman and

153
Chapter 6 Discussion and Conclusions

Mexico. The valves were sized for the varying application using the research
completed in this paper, per the methodology shown in section 6.3.
The measurement of point velocities within a MS path. In all the literature reviews
completed measurements within a complex MS path were derived from pressure
readings or ignored completely. The use of LDA, including the description of the
apparatus, test models and flow rig are original to the research of MS valves.
The determination of Cv, Gy and c for a new flow geometry.
The finding and correlation of limiting flow rate with complex structures in
restrictive channels.
The identification of highly distorted shock waves at the outlet stage of a MS path for
a specific high pressure condition.
The development of a predictive empirical model for a MS geometry.

6.5 Potential for Further Work

The research undertaken in this project gives rise for several opportunities for further work
including;

The measurements made within the MS path where limited by the LDA system
available. Further measurements using different optical layouts could facilitate more
measurements in the transverse direction closer to the walls.
The V2F turbulence model was unavailable with the CFX CFD package at the time of
research. For completeness simulations using the V2F model should be compared
with the SST simulations.
The MS trim reduces erosion in control valves by controlling velocity. The geometry
of the MS path is radically different from that of a single stage trim, as are the impact
angles of the fluid. The study of erosion was outside the scope of this thesis but the
prediction of erosion rates and their affect on the trim Cv would assist in practical
applications.
The MS trim reduces noise in control valves by eliminating shock cell turbulence
interaction which is the primary noise generating mechanism. Actual measurements
of noise are greatly affected by transmission losses through valve bodies and
attenuation through downstream piping. Research into the effective noise versus
would assist in practical applications.

154
Chapter 7 References

CHAPTER 7

REFERENCES

Adrian, R.J., Orloff, K.L., (1977), Laser anemometer Signals: Visibility


Characteristics and Application to Particle Sizing, Applied Optics 16, pages 667-684

Amato U., Bertoli, C., Corcione, F.E., Petrillo F., Valentino G. (1990), Turbulence
Models Validation by LDA in an Internal Combustion Engine, International
Symposium COMODIA 90: 411-418, 1990

Anderson, J. D., (1995), Computational Fluid Dynamics: The Basics with


Applications, 1st Ed., McGraw-Hill

Anderson J.D, (2001) Fundamentals of Aerodynamics, 3 rd ed, Singapore: McGraw-


Hill

Anderson J.D, (2003) Modern Compressible Flow with Historical Perspective, 3rd ed,
McGraw-Hill

Anderson, J. D., (2005), Ludwig Prandtls Boundary Layer, Physics Today,


December 2005, American Institute of Physics

155
Chapter 7 References

App. J.F., (2009), Fields Cases: Nonisothermal Behaviour Due to Joule-Thomson


and transient Fluid Expansion / Compression Effects, presented at 2009 SPE Annual
Technical Conferene and Exhition, New Orelans, Louisiana, USA, October 2009

Atmaca1, E., Peker1, I., Altin, A., (2005), Industrial Noise and Its Effects on Humans,
Polish Journal of Environmental Studies Vol. 14, No 6 (2005), 721-726

Baumann, H., (1998), Control Valve Primer: A Users Guide, ISA Publishers

Bean, H.S. Values of Discharge Coefficients of Square-Edged Orifices. American


Gas Association Monthly, July, 1935

Bessie`res D., Randzio S. L., Pin eiro M. M., Lafitte Th, Daridon J-L., (2006), A
Combined Pressure-controlled Scanning Calorimetry and Monte Carlo Determination
of the Joule-Thomson Inversion Curve. Application to Methane, J. Phys. Chem. B
2006, 110, 5659-5664

Bernad, S., Susan-Resiga, R., Muntean, S., Anton, I., (2007), Cavitation Phenomena
In Hydraulic Valves Numerical Modelling, Proceedings Of The Romanian Academy,
Series A, Volume 8, Number 2/2007

Bia, Yong, Bia, Qiang, (2010) Subsea Engineering Handbook, Elsevier

Boussinesq (1877), J., Essai sur la theorie des eaux courantes, Mem. pres. par div.
savant `a lAcad. Sci. 23, 1680, 1877.

Buchhave, P., (1979), The Measurement of Turbulence With The Burst-Type Laser
Doppler Anemometer- Errors And Correction Methods, Thesis submitted to the
Faculty of the Graduate School of State University of New York (Buffalo) 1979.

Buckingham, E. Notes on the Orifice Meter: The Expansion Factor For Gases.
Research Paper No. 459, Bureau of Standards Journal of Research, Vol. 9, July 1932

156
Chapter 7 References

Celi, A., Chevalier G., Negulescu C. A., (2006), Comparison of Modern Turbulence
Models For Aero-Thermal Applications, European Conference For Aerospace
Sciences (Eucass)

Cengel, Y., Turner R., Cimbala, J., (2004). Fundamentals of Thermal-Fluid


Sciences, 2nd Ed, McCraw Hill, Appendix A, figure A-28.

Chauveau, C., Davidenko D., Sarh B., Gkalp I., Avrashkov V., Fabre C., (2006)
PIV measurements in an under expanded hot free jet, presented at 13th Int. Symp on
Appl. Laser Techniques to Fluid Mechanics, Lisbon, Portugal,

Christensen, K. T., Adrian R. J., (2002) The Velocity and Acceleration Signatures of
Small-scale Vortices in Turbulent Channel Flow, Journal of Turbulence, 3, 1-28,
2002

Cunningham, R.G. (1951), Orifice Meters with Supercritical Compressible Flow"


Trans. ASME, Vol. 73, pp. 625-638

Dalbello, T., Dippold, V., Georgiadis, N., (2005), Computational Study of Separating
Flow in a Planar Subsonic Diffuser, NASA/TM2005-213894

Durbin, P.A., (1995), Separated Flow Computations with the k-e-v2 Model, AIAA
JOURNAL Vol. 33, No. 4, April 1995

Durbin, P.A., (1995), Turbulence modeling for non-equilibrium flow, Center for
Turbulence Research Annual Research Briefs 1995

Drain, L.E., (1980), The Laser Doppler Technique, A John Wiley & Sons Limited,
1980

Driskell, Les, (1983), Control Valve Selection And Sizing, 1 st ed., North Carolina,
Publishers Creative Services Inc.

157
Chapter 7 References

Driskell, Les, (1984) Sizing Theory and Applications, in ISA Handbook of Control
Valves, Hutchison, J.W., North Carolina, 180-190

El-Behery, S.M., Hamed, M. H., (2009) A Comparative Study of Turbulence


Models Performance for Turbulent Flow in a Planar Asymmetric Diffuser, World
Academy of Science, Engineering and Technology 53 2009

Escudier, M., Nickson, K., Poole R., (2006), Intensely swirling turbulent pipe flow
downstream of an orifice: the influence of an outlet contraction, 13th Int. Symp. on
Appl. Laser Techniques to Fluid Mechanics, Lisbon, Portugal, June 26-29, 2006 #
1031

Essom Company Limited, (2007), Determination of Flow Rate by Orifice Plate

Gardin, P., Brunet, M., Domgin, J.F., Periceous, K., (1999), An Experimental and
Numerical Study of Turbulence in a Tundish container, Second International
Conference on CFD in the Minerials and Process Industries, CSIRO, 6-8 December
1999

Grace. A., Frawley, P., (2011), Experimental Parametric Equation For The
Prediction Of Valve Coefficient (Cv) For Choke Valve Trims, International Journal
of Pressure Vessels and Piping, Reference: IPVP3085, DOI information:
10.1016/j.ijpvp.2010.11.002

Hamad E.Z., Mansoori G. A., (1988) Statistical Mechanical Test of Mhemhs Model
of the interaction of third viral coefficients, in Fluid Phase Equilibria, Elsevier
Science Publisher, Netherlands, 205-212

158
Chapter 7 References

Hendricks, R.C., Shouse, D.T. , Roquemore, W.M., (2004), Experimental and


Computational Study of Trapped Vortex Combustor Sector Rig With Tri-Pass
Diffuser, NASA/TM2004-212507

Husain, Z., (1995), Industry paper Theoretical Uncertainty Of Orifice Flow


Measurement Zaki D. Husain, PhD Daniel Flow Products, Inc. P. O. Box 19097
Houston, Texas 77224 1995 Daniel Flow Products, Inc.

Iaccarino, G., (2001), Predictions of a Turbulent Separated Flow Using Commercial


CFD Codes, Journal of Fluids Engineering Copyright , ASME, December 2001,
Vol. 123

IEC 60534-8-3, Aerodynamic Noise in Control Valves (2000), Switzerland,


International Electro-technical Commission

IEC 60534-2-4, Industrial-process control valves Part 2-4: Flow capacity Inherent
flow characteristics and rangeability, (2009), Switzerland, International Electro-
technical Commission

IEC 60534-1, Control valve terminology and general considerations (2004),


Switzerland, International Electro-technical Commission

IEC 60534-2-1 Mod: Flow Equations for Sizing Control Valves, Switzerland,
International Electro-technical Commission

IEC 60534-2-3: Flow Capacity Test Procedures, Switzerland, International Electro-


technical Commission

Jiyuan, T., Yeoh G. H., Liu, C., (2008), Computational Fluid Dynamics: A Practical
Approach, 1 st Ed., Butterworth-Heinemann (Elsevier)

159
Chapter 7 References

Kegel, T., (2002) Compressible flow effects in subsonic venturis, in Fluid Flow
Measurement 3rd International Symposium, Colorado Engineering

Kim, J. Ghajar, A.J., Tang, C., Foutch G.L., (2003), Comparison of Near-Wall
Treatment Methods for High Reynolds Number Backward-Facing Step Flow,
International Journal of Computational Fluid Dynamics, Vol. 19, No. 7, October
2005, 493500

Kinghorn, F.C. The Expansibility Correction for Orifice Plates: EEC Data. Paper
5.2, Presented at the International Conference on Flow Measurement in the Mid 80s,
National Engineering Laboratory, East Kilbride, Glasgow, Scotland, June 9-12, 1986.

Kolmogorov, A. N., (1941), The local structure of turbulence in incompressible


viscous fluid for very large Reynolds number, Dokl. Acad. Nauk. SSSR 30, 913,
1941; On degeneration (decay) of isotropic turbulence in an incompressible viscous
liquid, Dokl. Acad. Nauk. SSSR 31, 538540, 1941; Dissipation of energy in locally
isotropic turbulence, Dokl. Acad. Nauk. SSSR 32,1618, 1941.

Kortekaas W.G., Peters C.J., de Swaan Arons J., (1997), Joule-Thomson expansion
of high-pressure-high-temperature gas condensates, Fluid Phase Equilibria 139
(1997) 205-218, ELSEVIER

Massey B., (1998) Mechanics of Fluids, 7th ed, UK: StanleyThornes

McDonough, J. M., (2007), Introductory Lectures On Turbulence Physics,


Mathematics And Modeling, Departments Of Mechanical Engineering And
Mathematics University Of Kentucky

Menter, F.R., (1993), Zonal Two-Equation k-w Turbulence Model for Aerodynamic
Flows, AIAA Paper 1993-2906, 1993.

160
Chapter 7 References

Menter, F. R., Kuntz, M., Langtry, R., (2003), Ten Years of Industrial Experience
with the SST Turbulence Model, Turbulence, Heat and Mass Transfer 4, 2003 Begell
House, Inc.

Moore, Ralph L. (1976), Flow Characteristics of Valves, in ISA Handbook of Control


Valves, Hutchison, J.W., North Carolina (165-179)

Nomura, T.,, Takahashi, Y., Ishima, T., Obokata, T., (2008), LDA and PIV
Measurements and Numerical Simulation on In-Cylinder Flow Under Steady State
Flow Condition, Department of Mechanical System Engineering, Gunma University
in association with Honda R&D Co., Ltd., SAE International, Paper Number-SP-2178
(2008)

North American Maufacturing Co., (1989), Bulletin 8695, Orifice Plates and Flange
Specifications

Paull, C.K., Matsumoto, R., Wallace, P.J., and Dillon, W.P. , (2000), Occurrence,
Structure, And Composition Of Natural Gas Hydrate Recovered From The Blake
Ridge, Northwest Atlantic, Proceedings of the Ocean Drilling Program, Scientific
Results, Vol. 164

Parsafar G. A., Noparast E., Keshavarzi E., (2000), Comparison of the Equations of
State from Joule-Thomson Coefficient, J. Sci. I. R. Iran, Vol. 11, No. 2, Spring 2000

Peng, D., Robinson, D., A New Two-Constant Equation of State, Ind. Eng. Chem.
Fundamen., 1976, 15 (1), pp 5964

Peters, J, (1987), Reliable Hydrodynamic Energy Dissipation In Subsea Chokes


Valve, Subsea Choke Valve Seminar, East Kilbride, Glasgow, 24 Feb 1987

161
Chapter 7 References

Perry R.H., (1999), Perrys Chemical Engineers Handbook, McGraw Hill, p 2-132
to 2-135.

Perry, J.A., (1949), Critical Flow Through sharp Edged Orifices, TRANS ASME,
vol. 71, no. 7, p 757-764

Popescu, F., Panait, T., (2007), Numerical Modelling and Experimental Validation
of a Turbulent Separated Reattached Flow, International Journal of Mathematics and
Computers In Simulation, Issue 1, Volume 1, 2007

Prandtl, L., (1925). Bericht uber Untersuchungen zur ausgebildeten Turbulenz, Zs.
agnew. Math. Mech. 5, 136139, 1925.

Reynolds, O., (1894), On the dynamical theory of turbulent incompressible viscous


fluids and the determination of the criterion, Phil. Trans. R. Soc. London A 186,
123161, 1894.

Reynolds, W.C., (1979), Thermodynamic Properties in SI units, Department of


Mechanical Engineering Stanford University Press, page 158

Rhode, J.E., Hadley R. T., Metger G. W., (1969), Discharge Coefficients for Thick
Plate Orifices with approach Flow Perpendicular and Inclined to the Orifice Plate,
NASA TN D-5467

Ribner, H.S., (1997), Shock Turbulence Interaction and the Generation of Noise,
Report 1233 National Advisory Committee for Aeronautics, 682-701.

Roy B. N., (2002), Fundamentals of Classical and Statistical Thermodynamics, John


Wiley & Sons

Saad (1993), M.A., Compressible Fluid Flow, 2nd ed., Ch. 3, PrenticeHall, NJ, 1993.

162
Chapter 7 References

Saad, T., (1999), Turbulence Modeling for Beginners, University of Tennessee


Space Institute

Salim, M., Cheah, S.C., (2009), Wall y+ Strategy for Dealing with Wall-bounded
Turbulent Flows, Proceedings of the International Multi-Conference of Engineers
and Computer Scientists 2009 Vol II, IMECS 2009, March 18 - 20, 2009, Hong Kong

Seidl, W., (2002) The orifice expansion correction factor for a 50mm line size at
various diameter ratios, in Fluid Flow Measurement 3 rd International Symposium,
Colorado Engineering Experiment Station Inc, Colorado, USA, 1-11

Sharma, M., Carey G.F., (1986), Turbulent Boundary-Layer Analysis Using Finite
Elements, International journals for numerical Methods in Fluids, VOL. 6, 769-787
(1986)

Siewert, P., Griebel, P., Schren, R., (2004), Measurements Of A Turbulent Flow In
A Generic Gas Turbine Combustor By Means Of 2d Particle Image Velocimetry,
Paul Scherrer Institute, 2004

Soave, G., (1972). "Equilibrium Constants from a Modified Redlich-Kwong Equation


of State", Chem. Eng. Sci., 27:1197-1203.

Stiles G.F. (1984) Cavitation and Flashing Considerations, in ISA Handbook of


Control Valves, Hutchison, J.W., North Carolina, 206-220

Strachan, R., Knowles, K., Lawson, N., (2005), Comparisons between CFD and
experimental results for a simplified car model in wall proximity, Department of
Aerospace, Power and Sensors Cranfield University

163
Chapter 7 References

Talbi, K., Nemouchi, Z., Donnot, A., Belghar, N., (2009), An Experimental Study
and a Numerical Simulation of the Turbulent Flow under the Vortex Finder of a
Cyclone Separator, Journal of Applied Fluid Mechanics, Vol. 4, No. 1, 69-75, 2011.

Tamburrino, A., Gulliver, J., (1999), Large Flow Structures in a Turbulent Open
Channel Flow, Journal of Hydraulic Research, VOL. 37, 1999, No.3

Tennekes, H., Lumley, J.L., (1972), A First Course in Turbulence, The MIT Press

Vrabec J., Ashish A., Kumar Hasse K.H., (2007), Joule-Thomson inversion curves of
mixtures by molecular simulation in comparison to advanced equations of state:
natural gas as an example, 15 April 2007, Elsevier

Wilcox, D. C., (1993), Turbulence Modeling for CFD, DCW Industries, Inc., La
Canada, CA,.

Wu-Shung Fu, (1998) A concise method for determining a valve flow coefficient of a
valve under compressible gas flow, in Experimental Thermal and Fluid Science 18,
Elsevier Science Publisher, Netherlands, 307-313

Xiang, Hong Wei, (2005), The Corresponding-States Principle And Its Practice,
Elsevier, Section 3.1

Zank, G. P., Ye Zhou, Matthaeus, W. H., Rice, W. K. M., (2002), The Interaction of
Turbulence with Shock Waves: A Basic Model, Phys. Fluids 14, 3766 (2002)

164
Chapter 7 References

Zanoun, E.-S., Durst, F., (2003), Evaluating The Law Of The Wall In Two-
Dimensional Fully Developed Turbulent Channel Flows, Physics Of Fluids Volume
15, Number 10 October 2003

Zhang, N., Chato, D., McQuillen, J., Motil, B., Chao, D., (2009), CFD simulation of
Pressure Drops in Liquid Acquisition Device Channel with Sub-Cooled Oxygen,
World Academy of Science, Engineering and Technology, 58 2009

165
Appendices

Appendix A:
Multi-Stage Trim Patent

EU Patent Number: IB2008053368

US Patent number: 2010/0186835 A1

166
Appendices

167
Appendices

168
Appendices

169
Appendices

170
Appendices

171
Appendices

172
Appendices

173
Appendices

174
Appendices

175
Appendices

176
Appendices

177
Appendices

Appendix B:
Extract of the experimental data of the Volume Expansion Coefficient, from Bessie`res
(2006), A Combined Pressure-controlled Scanning Calorimetry and Monte Carlo
Determination of the Joule-Thomson Inversion Curve. Application to Methane, J. Phys.
Chem. B 2006, 110, 5659-5664

178
Appendices

179
Appendices

Appendix C:
LDA Overview

180
Appendices

(Drain, 1980) described the simple shift in frequency from a source to an observer by
illustrating the change in optical path length. If a particle, at time (t), moves between a source
and observer, at time t+t the particle will have moved a distance and two optical path
lengths will be seen, see figure C.1. If n is the number of wavelengths arriving at the
observer per unit time, it is clear that the extra number of wavelengths arriving at the
observer as the particle travels a distance, is the change in frequency (f), given by equation
(C.1).

Figure C.1 Optical path diagram for single particle scattered light


 
(C.1)

The rate of change of n is the change in optical path length in unit time. If the distance
travelled by the particle is infinitesimally small and and are the wavelengths before and
after scattering, the change in path can be determined from the velocity vector V and the
initial angle of scatter (L1 and L2), see equation (C.2). As the velocity of a wave does not
change before and after scattering (or when travelling through different medium) equation
(C.2) can be reduced using equation (C.3), to equation (C.4), given by Drain (1980).

v.[ v.[



(C.2)

(C.3)

.[ .[

 .
 .
(C.4)

181
Appendices

The above example demonstrates the measurement of a shift in scattered light frequency but
does not give a means of determining the actual velocity of the particle. The wave theory of
light states that electromagnetic radiation obeys the properties of superposition and can thus
generate constructive and destructive interference. Two monochromatic light sources when
crossed will produce a series of constructive fringes (with opposing destructive fringes),
referred to as a probe volume. Laser light has properties most suited to Doppler frequency
measurement. A laser beam is both monochromatic and coherent such that the frequency is in
a known range. The intensity distribution across the narrow part of the beam (referred to as
the waist) is Gaussian. The selection of a laser source varies the distance between the
constructive fringes. A simple laser source and optical system suited to Doppler frequency
measurement is given in figure C.2.

LDA systems use multiple optical layouts to direct the probe volume and gather the scattered
light. The nature of the apparatus is often dictated by the fluid path geometry and
accessibility. The particular layout of the MS path meant that optical access was only
available from one side. The LDA apparatus used operated in back scatter, which meant the
scattered light was observed in the same plane as the laser, as seen in figure C.2. Forward
scatter systems also exist wherein the scattered light is observed on the opposite side to the
laser source. The different optical layouts affect the characteristics of the scattered light and
the signal to noise ratio (SNR), detailed later in this section.

Figure C.2 Typical back scatter LDA arrangement (Drain, 1980)

182
Appendices

A single source laser can be split into two separate beams using a Bragg cell. A probe volume
is formed by crossing the two beams along their coherent length. With Gaussian laser beams
the probe volume is an ellipsoid and is described using the half-axis equations. The resultant
interference fringes are shown geometrically in figure C.3. An equation will be derived that
equates frequency shift with particle velocity. Red and green have been used to notate the
crossed laser beams for clarity but in applications these beams will be of the same frequency
(and thus color). The geometry of the interference fringes are further described by figure C.4.
If the probe volume is directed into a seeded flow the particles will scatter light as they travel
through the fringes. If the distance between the fringes can be equated to the shift in
frequency at the observer than the velocity of the particle through the probe volume is known.

Figure C.3 Probe volume generated by two crossing laser beams

Figure C.4 Interference fringe geometry

183
Appendices

Figure C.5 Geometrical diagram of laser beam and fringe separation

Investigating the triangle described in figure C.5 (highlighted and enlarged from figure C.4),
the distance between the fringes (Lf) is given in terms of the laser wavelength, see equation
(C.5). If a particle travels through the fringes an extra wave of motion appears at the
observer, as it scatters from one plane to the next. Therefore the shift in frequency for a
particle travelling at an angle L to the fringes is equal to the number of fringes crossed in
unit time; see equation (C.6). The equation for Doppler shift and velocity can also be
described using relativistic transformations.

[ [
|&| 2L (C.5)
 

. . [
(C.6)
P 

184
Appendices

Appendix D:
Turbulence Overview

185
Appendices

The most common and economical methods of resolving turbulent flow apply Reynolds
Decomposition to the fluctuating properties in the governing equations. Reynolds proposed
that the property at any point in a turbulent region is the sum of the average value and a
fluctuating value. The turbulent fluctuating flow properties can be defined in the form of
equation (D.1) and inserted into the governing equations. The transport equation (see (D.2))
is a generic form of PDE that is used to describe the Reynolds Averaged Naiver Stokes
(RANS) equations (using the variables in table D.1) for a turbulent flow. The transport
equation describes transport phenomena such as heat transfer, mass transfer, momentum
transfer and energy transfer. Equation (D.2) contains four terms which explain the properties
influencing transport. From left to right these include; a transient term, a convection term, a
diffusion term and a source term. Respectively each of these terms controls the accumulation
of transport phenomena, the transport due to the velocity field, transport due to the
phenomenas own gradients and other sources which either increase or decrease the
phenomena. The variables given in table D.1 describe the transport of mass, moment and
energy.

These equations can be further reduced for compressible flow (Euler equations) where the
dissipative transport phenomena of viscosity, mass diffusion and thermal conductivity are
ignored. The inclusion of additional fluctuating terms in the governing equations necessitates
the definition of additional relationships (equations), such that a closed set can be resolved.
The closure equations are referred to as the turbulence model, as they define how the
fluctuating components are determined. For the MS trim a RANS turbulence approach has
been taken as the most economical solution based on the primary interest being the average
values in the MS trim. The theory behind the closure equations is presented in details in
section 4.3.


     K :  (D.1)

 - 
      (D.2)
< < <

186
Appendices

Equation S
Continuity 1 0 0
1
Momentum (x) u  "   
0
1
Momentum (y) v  "   -
0
1
Momentum (z) w  "  
0
"
Energy T  ST
 "

Table D.1. Transport equation variables for continuity, momentum and energy equations

The most notable difficulty with the investigation into turbulence was the lack of an accepted
definition. In order to understand the different turbulence theories and their inaccuracies /
limitations the three current movements that have formulated definitions were investigated.
(McDonough, 2007) detailed the three opposing movements using the major historical
contributions, see figure D.1. This research summarized the work of the three movements and
their contributions to turbulence modeling.

Figure D.1 Historical inputs for the three turbulence movements; deterministic, statistical and
structural (McDonough, 2007)

187
Appendices

The statistical movement finds its roots in research completed by Reynolds (1894). Reynolds
defined turbulence as entirely random and hence unsuited to direct analysis. His conclusion
was that the average change in the parameters could be measured and statistics used to
determine the fluctuating component in time. A considerable amount of current research has
been influenced by this approach and focuses on statistical tools for determining the
fluctuating components. The structural movement is primarily concerned with the form,
shape and change of the various fluid configurations associated with turbulence. Turbulence
is dominated by multiple scales of fluid motion which theoretically are linked. Structural
theory concentrates on the non-dimensional relationships between the scales and endeavors to
form relationships which predict the evolution of these scales.

The most modern turbulence research focuses on the deterministic nature of the NS
equations. Contrary to Reynolds belief of a completely random nature of turbulence, it has
been seen mathematically and by experimentation that the fluid motion is more chaotic than
random. Chaotic motion and random motion are distinctly different. Chaotic motion can be
predicted but only for a short period. It is also directly affected by and very sensitive to its
starting conditions. As stated previously (section 4.2) it is generally accepted that the NS
equations can completely resolve all turbulent motion should the correct scale (mesh) be
used. If the NS equations are deterministic it follows that turbulent motion cannot be random
and statistical analysis is not justified. However, current turbulence modelling is strongly
affected by the theories and nomenclature of historical statistical analysis. Statistical analysis
simplifies the mathematical complexity and hence reduces the computational power required
to resolve a turbulent field.

The founding theories of statistical analysis that contribute most to modern turbulence models
were formulated by; Boussinesq (1877), Reynolds (1894), Prandtl (1925), Taylor (1935) and
Kolmogorov (1941). The Boussinesq theory endeavoured to relate turbulent shear stress to
the mean flow strain rate. This was based on Newtons law of viscosity which states that
shear stress is proportional to strain rate, with viscosity being the constant of proportionality.
For laminar flow Newtons theory can be applied, as given in equation (D.3). In this case the
viscosity () is a property of the fluid. Boussinesq applied the same theory to a turbulent
flow, where turbulent viscosity (T), as given in equation (D.4), replaced the fluid viscosity.
In this case T is a function of the flow and not the fluid. There is little experimental data that
supports this conclusion but its speculation has been applied or alluded to in most commonly
used turbulence models.

188
Appendices


w (D.3)
<

 w "   -
 (D.4)
<

Reynolds experiment of injecting a coloured dye into a flow stream in a pipe with smooth,
transparent walls demonstrated two of the primary properties of turbulence. Firstly, Reynolds
identified that the ratio of inertia to viscous forces was related to the motion of the fluid
(Reynolds number, Re). At values of Re < 2,000 it was seen that the colored dye mixed very
slowly when compared with the average fluid velocity. In this case mixing was via molecular
diffusion only. As Re increased to 2000 < Re < 2300 the colored dye became wavy but
mixing was still as a result of molecular diffusion. Further increases in Re caused an erratic
change in direction with significant local mixing. This demonstrated that turbulence was a
mathematical bifurcation and occurred after only a few transitions. Ultimately this
experiment showed that turbulence had a diffusive nature bounded at its lower end by
molecular diffusion. In addition, determinists would later relate the sensitivity of the
bifurcation to the starting conditions as indicative of a chaotic nature (changes in the wall
roughness greatly affects the value of Re for the turbulent condition).

The nature of the diffusive length scale can be seen in the NS equations. In the moment
equation there is a convective (or advective) term. This term defines the macroscopic
movement of the fluid. There is also a second order movement term whose constant is
viscosity. The viscosity is a property of the fluid and does not depend on the flow situation. If
the viscosity term were large in comparison to the inertia forces it is evident that the second
order moment term and its constant of viscosity would predominate and diffusion would be
by molecular diffusion. If viscosity were small in comparison to the inertia forces the
convection term would dominate and diffusion would be by turbulent motion.

Prandtl furthered Boussinesq turbulent viscosity theory by formulating a method of


predicting vT (also known as the coefficient of eddy kinematic viscosity). Prandtl introduced
the mixing length theory based on the correlation between turbulent eddies and the molecules
of a gas. The kinetic theory, used to predict viscosity in gases, requires the establishment of a
length and velocity scale associated with the transportation of momentum. In the case of the
kinetic theory the length scale equates to the distance between molecules before collision.

189
Appendices

The velocity scale is resolved from an average over the Maxwellian distribution for a fixed
volume of gas in a thermodynamic equilibrium. Prandtl proposed that turbulence had a
similar characteristic and that the eddies were bundles of fluid parcels randomly moving
around. These bundles would collide similarly to the molecules in the kinetic theory and
would thus need a length (mixing length) and velocity scale. In conclusion, a number of
experimental coefficients used to determine the mixing length were correlated for various
flows. The predictions of turbulence proved acceptable for 2-D free shear flows and boundary
layers. This added significant support to the Bousinesq hypothesis, even though it still
maintained the unproven proposition that vT was representative of the flow and not the fluid.

Taylor (1935) was the first researcher to use sophisticated formal statistical methods. He
introduced the Taylor hypothesis which provided a means of converting temporal data into
spatial data. Measurements were made at a single point or at a number of points without
having to make the measurements simultaneously. Essentially if a measurement was made at
a point in the flow where the velocity was varying in time (u(t)), and another measurement
was made a time later a distance x from the first point (where the measurement probe was
moving with velocity U), then the velocity at that subsequent point would be u(x/U).

As identified by Reynolds early experiments, turbulence contains multiple scales of length


and time. Tennekes (1972) identified the four main scales namely; the large scale, the integral
scale, the Taylor micro-scale (inertial sub-range) and the Kolmogorov scale (dissipation
scale). The large scale length is defined by the fluid domain and its time by the average fluid
velocity. This scale is known as the convective scale as it represents the highest level of
diffusion through the domain (tc = Ls / U). As stated previously if diffusion is also related to
kinematic viscosity a molecular diffusion scale can be obtained (kinematic viscosity units are
v = distance2/time, thus td = Ls2/v). The ratio of both these scales i.e. molecular diffusion
versus convection is equal to Re. This is equivalent to the explanation given previously
regarding the NS equations where turbulence is dominated by the convective scale and
laminar flow by the molecular scale. The Kolmogorov scales are the smallest associated with
turbulence and by definition are associated with molecular dissipation and viscosity.
Kinematic viscosity generalized units are distance2/time2 and those for energy dissipation are
distance2/time3. Therefore a length scale (KL) can be formed by dividing v3 by with an
associated time (KT) and velocity (KV) scale given by equations (D.5), (D.6) and (D.7).

190
Appendices


-Q )
 (D.5)


-

"  (D.6)


X ) (D.7)

The k- turbulence model is a two-equation model that utilizes the turbulent kinetic energy
(k) and the turbulent kinetic energy dissipation rate () to establish values of velocity and
length scale. Using the Prandtl mixing length theory, k and can be used to calculate vT. The
k- model is a closed model as it predicts all the terms necessary to understand vT without
appeal to experimental results. An equation for k is derived from a total kinetic energy and a
mean flow kinetic energy equation. These are formed from the product of the NS equations
and the velocity vector. Kinetic energy in both equations is given per unit mass as half the
square of the velocity term. (Tennekes et al, 1972) derived k from the subtraction of the time
averaged total and mean kinetic energy equations after the application of Reynolds
decomposition, see equation (D.8). The original tensor notation has been maintained for
consistency. This is not the only expression for k, and Wilcox (1993) obtained a similar
expression, see equation (D.9). Both equation (D.8) and (D.9) are essentially the same with
corresponding terms, except for a diffusive kinetic energy term. This term does not appear in
the Tennekes review due to the cancellation of strain rate from the original NS equation. The
Wilcox version is that typically used in actual RANS modelling while the Tennekes form is
accompanied by additional notes on the physical meaning of the terms

    
  
      2  
 
   2 (D.8)



  

O   
Where 
  and   O

O

  
     
       
  (D.9)


191
Appendices


Where 2

Whether equation (D.8) or (D.9) is used to determine k, additional equations are required to
model the constituent terms. It is therefore necessary to have an understanding of the physical
meaning of each term in the definition of k. From equation (D.8) it is clear that the term on
the far left is the advection of turbulent kinetic energy. Moving from left to right the next
term is pressure work done due solely to turbulence, the next term is the transport of turbulent
kinetic energy by turbulent fluctuations and the third term on the right hand side is transport
by viscous stresses. Tennekes proposed definitions for the final two terms on the left, namely;
turbulence production (Rp) and the rate at which fluctuating viscous stress (vsij) performs
deformation work against the fluctuating strain rate (sij). Both definitions are subject to
contrary physical explanations but are consistent with current k- models. Tennekes stated
that Rp often holds. The definition of each term in equation (D.8) has been summarized in
figure D.2.

Figure D.2 Summary of terms of Tennekes Turbulent Kinetic Energy Equation

Returning to the Wilcox form of k, as used in RANS modelling, it is evident that three terms
must be modelled, excluding the modelling of . In addition an equation for vT must be

192
Appendices

derived. The turbulent production term is modelled using the Bousinnesq hypothesis in the
form of equation (D.10). Turbulent viscosity has now been introduced to equation (D.9).
There has been much effort committed to modeling the pressure work term based on
additional modifications to the Boussinesq hypothesis (see (Wilcox, 1993)) but little physical
justification has been presented in support. A simpler approach is used in the standard k-
model and the pressure work term and velocity triple correlation term are combined to form
the diffusion of kinetic energy term, as given in equation (D.11). McDonough (2004) stated
that equation (D.11) has been supported with minimal experimental data and debatable
physical relationships.

 
  2 "

 u (D.10)

   
  - 
  M (D.11)


Where k is a closure constant

The final term in the Wilcox equation is . The derivation of itself is complex and
introduces a further six higher order expressions which must be modelled. Essentially is
approximated from the fact that dissipation is locally isotropic, meaning that small-scale
behaviour is independent of direction, which can be expressed in equation (D.12) below.
Definitions of have already been given as part of the Tennekes and Wilcox determination of
k (see equation (D.9)). The six higher order terms are correlated statistically using many of
the techniques and data incorporated in Taylors original work. The use of statistics to model
the remaining terms is not connected with the actual physics of flow but rather the original
PDEs. Equation (D.12) is the most widely used form of in RANS modelling.

 
O
-M
  $    $  =  @ (D.12)
 

193
Appendices

Equations (D.8) to (D.12) represent a complete set of closed equations that are sufficient to
determine the velocity and length scales needed to resolve vT. Using dimensional analysis it is
evident that k~(Ls/ts)2 and ~Ls2/ts3. Therefore the velocity scale can be obtained from k-1/2,
the time scale from k/~ts and it follows that k-3/2/~Ls. Turbulence viscosity (vT) is obtained
from vT~Ls2/ts as given in equation (D.13), where CT is a constant to account for the purely
dimensional analysis. The k- model cannot be directly applied to a flow field due to
boundary conditions. As does not tend towards zero at a wall, the Law of the Wall must be
utilized to model this boundary.

" $" (D.13)


194
Appendices

Appendix E:
Complete tables for CFD simulation results for LDA Sim 1, with LDA benchmark data for
TMA RUN 1.

195
Appendix A

TMA-1

Z Axis Velocity Velocity Resultant


X Axis (mm) Y Axis (mm)
(mm) u, (m/s) v, (m/s) Velocity
0.76 0.17 11.00 7.01 24.94 25.91
-0.11 0.23 10.50 7.20 25.34 26.34
0.10 -0.30 9.90 7.95 28.93 30.00
0.95 0.16 9.20 9.20 26.69 28.23
0.00 0.00 8.70 Reference Centre Point
0.87 -0.96 8.20 8.90 30.53 31.80
0.70 0.11 7.40 7.89 28.20 29.28
-0.72 0.56 7.10 7.62 30.15 31.10
0.08 -0.85 6.80 6.90 34.53 35.21
Component Velocities (m/s) 7.83 28.66
LDA RUN 1 Comonent Velocities (m/s) 8.29 32.06
Mean Resultant
TMA-1a Velocity (m/s)
29.57

9.23 1.49 10.20 84.56 7.91 84.93


10.27 2.19 9.50 83.89 7.22 84.20
9.49 2.30 8.80 81.45 8.12 81.85
10.15 2.66 8.50 99.34 7.23 99.60
10.00 2.00 8.70 Reference Centre Point
9.53 2.91 7.80 101.30 7.46 101.57
10.40 2.42 7.40 86.81 7.23 87.11
10.15 2.51 7.10 80.43 7.68 80.80
Component Velocities (m/s) 88.25 7.55
LDA RUN 1 Comonent Velocities (m/s) 94.94 13.42
Mean Resultant
TMA-1b Velocity (m/s)
88.23

19.28 3.80 10.40 64.56 15.40 66.37


20.42 4.24 10.20 80.22 11.23 81.00
20.59 3.57 9.80 81.45 14.55 82.74
21.00 3.01 9.20 81.34 6.56 81.60
19.60 4.47 8.80 78.56 13.23 79.67
20.00 3.50 8.70 Reference Centre Point
20.29 2.86 7.80 68.89 14.21 70.34
20.17 3.95 7.70 71.45 13.86 72.78
19.97 3.85 7.40 64.56 12.28 65.72
20.97 3.92 6.90 56.34 13.11 57.85
Component Velocities (m/s) 71.93 12.71
LDA RUN 1 Comonent Velocities (m/s) 83.88 19.05
Mean Resultant
TMA-1c Velocity (m/s)
71.23

196
Appendix A

29.88 5.99 10.20 58.57 9.67 59.36


30.14 5.96 9.80 54.21 9.43 55.02
30.48 4.27 9.50 55.62 8.49 56.26
30.74 5.33 9.20 62.34 9.34 63.04
30.00 5.00 8.70 Reference Centre Point
29.11 5.46 8.50 55.42 10.76 56.45
30.97 5.71 8.20 57.89 7.21 58.34
29.64 5.43 8.00 56.31 7.23 56.77
30.77 5.14 7.80 58.51 9.12 59.22
Component Velocities (m/s) 57.36 8.91
LDA RUN 1 Comonent Velocities (m/s) 64.72 11.65
Mean Resultant
TMA-1d Velocity (m/s)
58.12

40.72 5.71 9.80 42.34 2.12 42.39


40.87 6.78 9.50 52.92 2.67 52.99
39.44 7.44 9.20 52.12 1.89 52.15
40.37 6.09 8.80 44.32 2.01 44.37
40.00 6.50 8.70 Reference Centre Point
40.47 6.58 8.00 49.67 2.44 49.73
39.08 7.16 7.80 49.90 2.67 49.97
40.41 6.74 7.40 47.58 3.80 47.73
40.18 7.16 6.90 48.88 4.20 49.06
Component Velocities (m/s) 48.47 2.73
LDA RUN 1 Comonent Velocities (m/s) 56.42 4.12
Mean Resultant
TMA-1e Velocity (m/s)
48.3

50.17 8.54 10.80 26.45 1.98 26.52401


50.82 8.47 10.50 25.89 2.12 25.97665
49.47 8.80 10.20 26.12 2.21 26.21333
50.00 8.00 8.70 Reference Centre Point
50.24 7.40 8.50 27.89 2.78 28.02821
50.72 8.37 7.10 25.56 1.67 25.6145
50.73 8.13 6.90 26.66 1.54 26.70444
Component Velocities (m/s) 26.43 2.05
LDA RUN 1 Comonent Velocities (m/s) 32.65 2.92
Mean Resultant
TMA-1f Velocity (m/s)
26.32

197
Appendix A

TMA-2

Velocity Velocity Resultant


X Axis (mm) Y Axis (mm) Z Axis (mm)
u, (m/s) v, (m/s) Velocity
-0.74 25.82 11.1 8.34 11.23 13.98816
-0.38 25.76 10.5 8.89 12.89 15.65836
0.15 26.40 10.2 7.98 12.23 14.60319
-0.52 25.95 9.2 7.47 12.88 14.88944
0.00 25.50 8.7 Reference Centre Point
0.98 24.55 7.7 8.52 11.56 14.3605
0.74 25.15 7.4 8.88 12.99 15.73514
-0.11 25.19 7.1 7.35 12.21 14.25155
-0.87 25.22 6.9 7.25 11.99 14.01152
Component Velocities (m/s) 8.085 12.2475
LDA Comonent Velocities (m/s) 7.62 10.06
Mean Resultant
TMA-2a Velocity (m/s)
14.89

10.80 27.87 10.8 35.35 3.34 35.51


9.18 26.68 9.8 33.89 3.21 34.04
10.68 27.65 9.2 32.46 3.45 32.64
10.44 27.37 8.8 31.22 3.25 31.39
10.00 27.00 8.7 Reference Centre Point
10.08 27.90 8.2 36.89 3.11 37.02
10.31 27.61 8 37.88 3.26 38.02
9.84 27.08 7.7 37.45 3.78 37.64
10.90 27.40 7.4 37.13 3.43 37.29
9.97 27.34 6.9 37.18 3.21 37.32
Component Velocities (m/s) 35.49 3.34 35.65
LDA Comonent Velocities (m/s) 37.49 5.81 37.94
Mean Resultant
TMA-2b Velocity (m/s)
35.48

20.45 27.08 10.50 70.87 5.23 71.06


20.21 28.58 10.40 78.58 5.67 78.78
19.01 28.10 9.50 75.98 6.24 76.24
20.71 27.25 8.80 68.89 6.21 69.17
20.00 28.00 8.7 Reference Centre Point
19.91 28.17 8.5 71.82 6.15 72.08
20.95 28.20 8.2 78.53 5.98 78.76
20.09 27.58 8 79.27 5.89 79.49
20.24 28.03 7.8 78.42 6.05 78.65
Component Velocities (m/s) 75.30 5.93
LDA Comonent Velocities (m/s) 68.67 4.59
Mean Resultant
TMA-2c Velocity (m/s)
75.6

198
Appendix A

29.90 30.73 11.10 101.45 3.21 101.50


28.20 30.25 10.80 95.85 3.87 95.93
29.89 29.79 9.20 96.24 3.72 96.31
29.85 30.37 8.80 97.52 3.45 97.58
30.00 29.00 8.7 Reference Centre Point
30.45 29.05 7.1 98.35 4.21 98.44
29.42 28.21 6.9 99.21 4.23 99.30
Component Velocities (m/s) 98.10 3.78
LDA Comonent Velocities (m/s) 90.20 2.89
Mean Resultant
TMA-2d Velocity (m/s)
98.52

40.34 30.58 9.80 67.64 8.56 68.18


39.85 30.80 9.50 66.37 8.34 66.89
40.81 30.80 8.80 67.93 8.73 68.49
40.00 30.00 8.7 Reference Centre Point
40.15 30.81 8.20 69.24 8.58 69.77
39.70 30.76 7.80 67.52 8.92 68.11
39.91 30.94 7.70 68.78 8.59 69.31
Component Velocities (m/s) 67.91 8.62
LDA Comonent Velocities (m/s) 70.14 9.15
Mean Resultant
TMA-2e Velocity (m/s)
68.9

43.11 30.37 10.20 44.78 2.34 44.84


42.31 30.75 9.80 43.64 1.28 43.66
43.80 30.67 9.50 43.62 1.67 43.65
43.96 30.66 8.80 44.52 1.82 44.56
43.00 30.30 8.7 Reference Centre Point
43.22 31.25 8.20 43.38 1.59 43.41
43.82 29.94 8.00 44.68 1.64 44.71
43.45 29.51 7.80 43.23 1.78 43.27
Component Velocities (m/s) 43.98 1.73
LDA Comonent Velocities (m/s) 51.05 2.10
Mean Resultant
TMA-2f Velocity (m/s)
44.23

199
Appendix A

TMA-3

Velocity Velocity Resultant


X Axis (mm) Y Axis (mm) Z Axis (mm)
u, (m/s) v, (m/s) Velocity
2.850 51.214 10.400 38.870 15.670 41.910
1.061 51.056 10.200 37.230 16.560 40.747
2.191 51.153 9.800 38.670 15.340 41.601
1.319 49.793 9.500 39.560 9.670 40.725
2.00 50.50 8.70 Reference Centre Point
2.32 50.46 7.40 18.32 8.42 20.16
Component Velocities (m/s) 34.53 13.13
LDA Comonent Velocities (m/s) 49.46 17.27
Mean Resultant
TMA-3a Velocity (m/s)
36.47

5.63 51.32 9.80 40.23 25.67 47.72


4.54 52.82 9.50 41.44 24.31 48.04
5.00 52.00 8.70 Reference Centre Point
5.15 52.52 8.50 30.63 24.83 39.43
4.20 52.92 7.40 23.56 24.93 34.30
5.98 51.96 7.10 21.94 24.26 32.71
Component Velocities (m/s) 31.56 24.8
LDA Comonent Velocities (m/s) 39.4 29.39
Mean Resultant
TMA-3b Velocity (m/s)
40.12

10.06 53.65 9.20 35.58 10.34 37.05


9.09 54.60 8.80 41.29 9.64 42.40
10.00 54.00 8.70 Reference Centre Point
10.19 54.02 6.90 61.94 21.57 65.59
Component Velocities (m/s) 46.27 13.85 48.30
LDA Comonent Velocities (m/s) 34.39 11.22 36.17
Mean Resultant
TMA-3c Velocity (m/s)
48.45

200
Appendix A

14.62 56.11 10.70 57.37 6.27 57.71


14.53 56.14 10.50 57.27 6.94 57.69
15.39 55.86 9.80 58.72 5.94 59.02
15.00 56.00 8.70 Reference Centre Point
15.71 56.23 8.50 55.42 5.59 55.70
15.97 56.25 8.20 56.38 5.55 56.65
14.65 56.81 7.40 55.97 6.42 56.34
15.29 55.40 6.90 55.37 6.78 55.78
Component Velocities (m/s) 56.64 6.21
LDA Comonent Velocities (m/s) 52.61 7.24
Mean Resultant
TMA-3d Velocity (m/s)
56.89

20.98 58.85 11.10 74.77 5.36 74.96


19.06 57.08 9.80 76.93 5.89 77.16
19.78 58.14 9.50 78.35 5.62 78.55
20.00 58.00 8.70 Reference Centre Point
19.24 58.19 7.80 79.36 5.74 79.57
19.98 57.58 7.70 72.36 5.28 72.55
19.52 58.13 7.10 71.46 5.82 71.70
Component Velocities (m/s) 75.54 5.62
LDA Comonent Velocities (m/s) 69.41 7.44
Mean Resultant
TMA-3e Velocity (m/s)
75.23

24.76 60.23 9.80 69.94 6.70 70.26


24.90 60.75 9.50 68.48 6.23 68.76
25.98 59.39 9.20 69.39 6.46 69.69
24.11 59.72 8.80 70.12 6.88 70.46
25.00 60.00 8.70 Reference Centre Point
24.09 60.73 8.50 73.82 5.92 74.06
25.56 60.99 7.80 71.39 5.42 71.60
24.21 59.89 7.70 69.83 5.92 70.08
25.90 60.48 7.40 71.83 4.52 71.97
Component Velocities (m/s) 70.51 5.84
LDA Comonent Velocities (m/s) 65.96 7.70
Mean Resultant
TMA-3f Velocity (m/s)
70.78

30.83638685 61.71358454 11.1 63.98 3.40 64.07


30.43395828 61.66977425 10.4 64.31 3.41 64.40
30.06226907 62.01783802 10.2 66.93 2.98 67.00
30.83309734 62.13708792 9.5 65.27 3.68 65.37
30 62 8.7 Reference Centre Point
30.78551012 62.60031233 8.5 66.67 2.91 66.73
30.86001833 61.79757235 8.2 64.52 3.62 64.62
Component Velocities (m/s) 65.28 3.33
LDA Comonent Velocities (m/s) 60.62 3.20
Mean Resultant
TMA-3g Velocity (m/s)
65.28

201

You might also like