You are on page 1of 10

Chemical Engineering Science 71 (2012) 264273

Contents lists available at SciVerse ScienceDirect

Chemical Engineering Science


journal homepage: www.elsevier.com/locate/ces

Experimental validation of ke RANS-CFD on a high-pressure


homogenizer valve
a,n
a

ardh

Andreas Hakansson
, Laszlo Fuchs b, Fredrik Innings c, Johan Revstedt b, Christian Trag
,
a

Bjorn
Bergenstahl
a

Lund University, Department of Food Technology, Engineering and Nutrition, P.O. Box 124, SE-221 00 Lund, Sweden
Lund University, Department of Energy Sciences, P.O. Box 118, SE-221 00 Lund, Sweden
c
Tetra Pak Processing Systems, Lund, Ruben Rausings Gata, SE-221 86 Lund, Sweden
b

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 26 September 2011
Received in revised form
21 December 2011
Accepted 22 December 2011
Available online 3 January 2012

Since the emulsication in the High-Pressure Homogenizer (HPH) is controlled by hydrodynamic


forces, the turbulent ow eld in the valve region is of signicant interest. Computational Fluid
Dynamics (CFD) simulations have been used for this in many studies. However, there are reasons to
question if the utilized turbulence models, with their inherent assumptions and simplications, could
accurately describe the inuential aspects of the ow.
This study compares CFD simulations using the methods from previous studies with experimental
measurements in a model HPH valve. The results show that the region upstream of the gap can be
described accurately regardless of turbulence model and that the gap region can be captured by using
one of the more rened k e models. None of the studied turbulence models were able to describe the
details of the highly turbulent region downstream of the gap. The obtained results are also discussed in
relation to generalizability and limitations in using CFD simulations for understanding the emulsication in the HPH.
& 2011 Elsevier Ltd. All rights reserved.

Keywords:
Emulsion
Fluid mechanics
Turbulence
Computational Fluid Dynamics
Homogenisation
Hydrodynamics

1. Introduction
The emulsication in a High Pressure Homogenizer (HPH) can
be seen as a combination of three simultaneous processes;
fragmentation of drops, recoalescence of insufciently covered

droplets and adsorption of emulsiers (Walstra, 1993; Hakansson


et al., 2009). All these processes are highly dependent on the ow
eld in general, and on the turbulence characteristics in particular. Computational Fluid Dynamics (CFD) was used early in order
to describe the ow eld in the homogenizer, starting with
Kleinig and Middelberg (1996, 1997) and Stevenson and Chen
(1997). Since then, a large number of studies containing CFD on
HPHs of various geometry have been published, for example,

Miller et al. (2002), Floury et al. (2004), Kohler


et al. (2007),

Steiner et al. (2006), Raikar et al. (2009), Hakansson


et al. (2010)
and Casoli et al. (2010). The studies differ in treatment of a large
number of factors such as dimensionality (two or three), grid size,
wall treatment and turbulence modeling; see Table 1 for an
overview of the assumptions made in the aforementioned papers.
The CFD results have been used for different purposes, e.g. gaining

Corresponding author. Tel.: 46 46 222 9670; fax: 46 46 222 4620.

E-mail address: andreas.hakansson@food.lth.se (A. Hakansson).

0009-2509/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2011.12.039

an overview of the ow (Stevenson and Chen, 1997), investigating

possible zones of cavitation (Floury et al., 2004; Hakansson


et al.,
2010), used as basis for predicting drop movement in the valve
(Casoli et al., 2010) or predicting strength of fragmenting hydrodynamic forces (Kleinig and Middelberg, 1997; Steiner et al.,
2006; Casoli et al., 2010).
Of the studies in Table 1, Kleinig and Middelberg (1996)
assume laminar ow since only the inlet chamber, upstream of
the gap, is modeled (see Fig. 1). The remaining studies in Table 1
model turbulence with the Reynolds Averaged NavierStokes
(RANS) equations using turbulence models based on the Boussinesq hypothesis; more specically with different types of k  e
turbulence models. RANS-CFD has the clear advantages of being
low in computational cost and widely available in commercial
implementations. The disadvantage is that the assumptions from
the Boussinesq hypothesis and formulations of source terms for
the k and (especially) e equations can sometimes give rise to
non-physical results. Ideally, the turbulence should therefore be
described by more accurate techniques; such as by Large Eddy
Simulations (LES) or even Direct Numerical Simulation (DNS).
However, these techniques are very computationally demanding
and are disregarded in this study, since to our knowledge, no
successful CFD using LES or DNS on homogenizer valves has been
published. A more constructive approach can therefore be asking

A. H
akansson et al. / Chemical Engineering Science 71 (2012) 264273

265

Table 1
Some previous CFD-models of High Pressure Homogenizer valves.
Reference

Modeled geometry Turbulence model

Geometry

Cells

Kleinig and Middelberg (1996)


Kleinig and Middelberg (1997)
Stevenson and Chen (1997)
Floury et al. (2004)
Miller et al. (2002)
Kelly and Muske (2004)
Steiner et al. (2006)

Kohler
et al. (2007)
Raikar et al. (2009)
Casoli et al. (2010)

Hakansson
et al. (2010)

Inlet chamber
Outlet chamber
Complete valve
Complete valve
Complete valve
Complete valve
In-house valve
In-house valve
Complete valve
Complete valve
Complete valve

2D
2D
2D
2D
2D
2D
3D
2D
2D axisymmetric
2D axisymmetrical
2D

5852
5015
20 000
350 000
65 000
65 000
850 000

Laminar
High-Re k e
Standard k  e
RNG k e
Standard k  e, RNG k  e, Realizable ke
Realizable k e
Standard k  e
RNG k e
Standard k  e
Standard k  e
Standard k  e

Phases

Wall treatment

1
Laminar
1
WF
1
WF
1
LReF
a
1
a
1
1
LReF
a
a
1
a
a
1
350 000 1 Cavitation model LReF
278 000 1
WF

WFWall functions, LReFLow Reynolds number Formulation.


a

Information not clear or missing in the publication.

in previously published studies. Furthermore, to discuss the


effects on utilization CFD models for describing emulsication
in the HPH valve.
Geometrical differences (such as inlet chamber rounding, gap
length, impact ring distance, etc.) between homogenizers of
different scales and between manufacturers are often substantial.
ardh

The studies by Innings and Trag


(2007) and Hakansson
et al.
(2011) investigate a design with a rather short gap (10 gap
heights) and long impact ring distance (  60 gap heights) which
is a common design for production scale homogenizers for
emulsication. Numerical investigations will be limited to this
design based on availability of experimental results, whereas
generalization to other geometries is treated briey in Section 4.7.
Since the experimental measurements available (Innings and
ardh,

Trag
2007; Hakansson
et al., 2011) were performed without
disperse phase and/or cavitation, this study is restricted to one
phase HPH ow only.

Fig. 1. Geometry of the computational domain. The position of ve lines, IIV,


used for analyzing the results, can also be seen in the gure. All measures are
given in gap height units.

what aspects of the ow in the homogenizer could be described


by the common RANS-CFD simulations.
In light of the large number of RANS-CFD performed previously (Table 1) and the theoretical limitations of the utilized
models, experimental validation of the calculated ow elds are
needed. Some authors have made comparison with secondary
factors such as pressure drop to semi-empirical correlations
(Kleinig and Middelberg, 1996; Stevenson and Chen, 1997;
Floury et al., 2004) or comparison to measured ow rate and
pressure drops (Casoli et al., 2010). However, none of the studies
have been validated directly against measured velocity elds.
The large scale structures of the one phase ow eld in a HPH
ardh

scale model was rst measured by Innings and Trag


(2007).
These measurements show the formation of a turbulent jet
downstream of the gap and the creation of recirculation zones
on either side of the jet. Later, the same experimental model was

investigated by Hakansson
et al. (2011), obtaining high resolution
measurements of the turbulence in the outlet chamber as well as
detailed information about the mean velocity in inlet chamber
and gap.
The aim of this paper is to investigate to what extent the
turbulent ow eld in a HPH valve can be predicted using
RANS-CFD techniques combined with the turbulence models used

2. Experimental measurements
Mean velocities and mean uctuations were measured in a scale
model of a high pressure homogenizer using 2D Particle Image

Velocimetry (PIV) as described elsewhere (Hakansson


et al., 2011).

The reader is referred to the original publication (Hakansson


et al.,
2011) for a more comprehensive discussion on experimental set-up
and conditions, this section will only introduce the general concepts
of the measurements. A scale model of a high pressure homogenizer

was constructed (Innings and Tragardh,


2007). The model was
scaled on gap Reynolds number, Reg
Reg

U g hrC

mC

gap size to Kolmogorov scale, h/Z, and dimensionless laminar gap


boundary thickness in the gap using a at plate approximation
s
d99
nC L
2
5:0
2
h
h Ug
All experiments were carried out in a one phase ow at room
temperature with gap height, h4.8 mm, mean gap velocity,
Ug 5.6 m/s corresponding to Reg 27 000. The scaling thus
ensures turbulence levels, length-scales and relative boundarylayer thickness comparable to those of a production scale HPH
ardh,

(Innings and Trag


2007; Hakansson
et al., 2011).
Production of turbulent kinetic energy was estimated from the
measured instantaneous velocity elds by calculation from

266

A. H
akansson et al. / Chemical Engineering Science 71 (2012) 264273

Fig. 2. 2D Mesh (Mesh B) containing 800 000 cells. Inserts show zoomed in images of the region close to the gap inlet and outlet.

denition
P k /ui uj S

@/U i S
@xj

Reynolds stresses and gradients were obtained as described

earlier (Hakansson
et al., 2011).

3. Computational methods and models


3.1. Computational Fluid Dynamics (CFD)
Instantaneous ow velocity and pressure can be obtained by
solving mass and momentum conservation laws, i.e. the Navier
Stokes equation. A one phase incompressible, isothermal, continuous ow with Newtonian viscosity is assumed.
3.2. Turbulence modeling
The turbulence models used in this study are based on the
RANS approach where the instantaneous velocity U is divided in
two; a mean velocity /US and a uctuation u. Substitution of this
division in the NavierStokes equation yields the RANS equations.
These will contain an extra term including the Reynolds stresses,
/uiujS. Closure can be obtained by utilization of the Boussinesq
hypothesis where the uctuations are assumed to be described by
an efcient turbulent viscosity and the mean velocity gradients:


2
@/U i S @/U j S
/ui uj S kdij nT
4

3
@xj
@xi
In the Standard ke turbulence model (Launder and Spalding,
1974) the turbulent viscosity, nT, is modeled by
2

nT C m

where Cm 0.09 is a model constant.


Production of turbulent kinetic energy is modeled using
P k 2nT Sij Sij

where
Sij



1 @/U i S @/U j S

:
2
@xj
@xi

A large number of rened k  e turbulence models exist in the


literature. In this study, the focus is on the models used in

previous studies of HPH valves (see Table 1): RNG ke (Yakhot


and Orszag., 1986) and Realizable ke (Shih et al., 1995). The
major difference between the Standard ke model and the RNG
ke model is a modication to the e equation in the latter;
reducing turbulent kinetic energy for high strain rates by increasing the loss of turbulent kinetic dissipation rate. The Realizable
ke model contains a e equation with source terms differing
completely from the Standard model. Also the constant Cm in
Eq. (5) is substituted with a function, implying a reduction in
turbulent kinetic energy for high levels of strain (or rotation).
These changes are supposed to increase the ability of the RNG ke
and Realizable ke models to describe ows with straining,
rotation, curved stream lines and separation which are all likely
to occur in the homogenizer valve. However, the improved
models still contain a number of simplications and modeling
assumptions that can be questioned, e.g. isotropy, a simplistic
representation of the production rate of turbulent kinetic energy
(cf. Eqs. (3) and (6)), the Boussinesq hypothesis and representing
all scales of turbulence with only two variables.
3.3. Boundary conditions
A constant velocity magnitude was specied at the inlet (see
Fig. 1). Turbulent variables were set by specifying a turbulent
length scale of 75 mm (equal to the length of the inlet boundary)
and constant turbulence intensity. If not otherwise stated, the
turbulence intensity was set to 2%. A constant pressure boundary
condition was set on the outlet (see Fig. 1).
3.4. Wall boundary conditions
Wall treatment is an essential part of modeling a turbulent
ow. Different CFD studies of the homogenizer valve have used
different wall treatments: wall functions (Launder and Spalding,
1974) or a low Reynolds number formulation (LReF) approach
(Launder and Sharma, 1974) (see Table 1). Wall functions require
the rst mesh node to be positioned relatively far from the wall
(y 430300) whereas LReF require a very high resolution close
to walls (y o5). The ow prole in proximity to the gap walls
are of interest for studying the HPH turbulence, this would
suggest the low Reynolds number approach. However, ow close
to the remaining walls is less interesting and using high resolution on these would be computationally expensive. Therefore the
majority of simulations have been carried out using Enhanced
Wall Treatment (EWT), combining the two-layer near wall model
with wall functions, resulting in walls with high resolution being

A. H
akansson et al. / Chemical Engineering Science 71 (2012) 264273

described as using LReF and less rened walls being described as


when using wall functions. Further details on the EWT can be
elsewhere (ANSYS, 2009).
Calculations with LReF and mesh rened in order to ensure
y o4 on all walls for the converged solution, were also carried
out for one case for comparison.
3.5. Mesh
Unstructured hexagonal meshes were created over the 2D
geometry seen in Fig. 1 using Workbench 12 (ANSYS, Canonsburg,
PA, USA) in two sizes. The meshes were manually rened in the
areas of high spatial velocity gradients (in and around the gap)
and close to gap walls. The nal number of cells was approximately 450 000 (Mesh A) and 800 000 (Mesh B), respectively.
Fig. 2 displays mesh B, utilized in a majority of the simulations.
Insets of the mesh near the gap inlet and outlet can also be seen in
the gure.
A third mesh, mesh C, was created with wall renement
y o4 for utilization with the LReF wall treatment. This resulted
in a mesh with 850 000 cells and approximately the same average
cell distance as mesh A in regions far from the walls.
3.6. Numerical methods and discretization
The SIMPLE scheme (Patankar and Spalding, 1972) was used
for pressurevelocity coupling. Second order Upwind spatial
discretization was used throughout the study.
3.7. Convergence
Three factors were used in order to test for convergence. First
the calculations were performed until all residuals were levelled
out, i.e. further iterations did not signicantly affect their value.
Secondly, a less than 2% difference in the integral of turbulent
kinetic energy over line IV (in Fig. 1) was required over iterations.
After the criteria above had been satised, a further 4000 iterations were calculated and three proles of mean velocity and
turbulent kinetic energy differing by 2000 iteration steps were
compared. The solutions were seen as converged if there was no
signicant difference between the three proles.
3.8. Implementation
All CFD simulations were performed using FLUENT 12.0
(ANSYS, Canonsburg, PA, USA) with single precision. Turbulence
models and wall treatments were used as implemented.

267

4. Results and discussion


4.1. Mesh sensitivity
The converged proles of turbulent kinetic energy, k, on three
lines (I, III and IV in Fig. 1) can be seen in Fig. 3 for two mesh sizes;
mesh A (450 000 cells) and mesh B (800 000 cells). Differences
between meshes are small upstream (Fig. 3A), inside (Fig. 3B) and
downstream (Fig. 3C) of the gap exit. The largest differences in k
can be seen inside the gap and is below 3%. Thus, the solution is
considered mesh independent and all further simulations are
performed with the resolution of mesh B.

4.2. Wall treatment and boundary conditions


The previous studies (Table 1), handle turbulence modeling in
the near wall region using different approaches. Standard Wall
functions were not possible to combine with a sufciently ne
mesh in the gap to obtain a mesh independent solution in the
present geometry and were therefore disregarded in this study.
Simulations using Enhanced Wall Treatment (EWT) and Low
Reynolds number Formulations (LReF) are compared in Fig. 3.
The EWT and LReF meshes were constructed with the same
resolution far from walls whereas the LReF mesh (mesh C) has a
higher resolution close to walls and therefore also more nodes. As
seen in Fig. 3, the LReF predicts slightly higher turbulence levels
than the EWT does at all positions investigated. However, there
are no signicant differences in the qualitative proles. Furthermore, since the differences between the studied wall treatments
are small in comparisons to experimental and computational
results (see Sections 4.34.5), no clear advantage with either
method can be seen at this stage and all further simulations use
the EWT which is less computational expensive.
The simulations seen in Fig. 3 use a constant turbulent
intensity of 2% on the inlet boundary. Simulations were also
performed using 10% turbulent intensity for comparison. The
higher inlet intensity gives an increase in turbulent kinetic energy
upstream of the gap. In this region, production and dissipation of
turbulent kinetic energy is low and thus the level is expected to
be largely inuenced by the inlet boundary condition. In and
beyond the gap, no difference could be seen when altering inlet
turbulence intensity in this interval. This is reasonable since the
turbulent kinetic energy transported from the inlet, is insignificant compared to the high local production and dissipation
downstream of the gap entrance. The proceeding simulations
were carried out using 2% turbulence intensity on the inlet.

Fig. 3. Mesh sensitivity and wall treatment. Comparison of turbulent kinetic energy, k, over three positions: (A) line I in the inlet chamber, (B) line III in the gap and (C) line
IV in the outlet chamber (lines dened in Fig. 1). Simulations performed with Enhanced Wall treatment (EWT) on meshes A and B and Low Reynolds number formulation
(LReF) on mesh C.

268

A. H
akansson et al. / Chemical Engineering Science 71 (2012) 264273

Fig. 4. (A) Calculated velocity prole in the gap (RNG k e turbulence model). The velocity magnitudes are scaled with gap velocity, Ug. (B) Elongation rate, Ey, versus
distance downstream of gap entrance versus distance from the gap entrance on the dashed line in (A).

Fig. 5. (A) Contours of velocity magnitude in the gap inlet region from CFD with RNG k  e turbulence model. Velocity magnitudes have been scaled with mean gap
velocity, Ug. (B) Boundary layer thickness, d70, at the right wall (A) versus downstream distance from gap entrance.

4.3. The inlet chamber


The inlet chamber constitutes the narrowing ow channel
upstream of the gap (see Fig. 1). Contours of velocity magnitudes
from CFD simulations using the RNG ke turbulence model can be
seen in Fig. 4A. The ow accelerates into the gap with an
asymmetrical prole due to the geometry. Both the acceleration
and asymmetry in velocity is qualitatively well described by the

CFD-model (cf. Hakansson


et al., 2011). No principal differences in
the predicted inlet chamber ow could be seen between the
different turbulence models except very close to the gap (see
further discussion in Section 4.4).
With regards to the emulsication in the homogenizer, and the
drops moving through the valve, the inlet chamber gives rise to an
elongational ow, exerting an elongational force on the drops. The
elongational rate
Ey

@U y
@Y

was obtained from numerical differentiation of the measured

ow eld of Hakansson
et al. (2011) and is shown in Fig. 4B versus
upstream distance from the gap entrance on the dashed line in
Fig. 4A. Elongation increases up to a maximum of 0.49Ug/h at a
position close to the gap inlet, which is in good agreement with
previous studies in the same experimental model (Innings and
ardh,

Trag
2007).

Elongation rates calculated from CFD with the three different


turbulence models have also been inserted in Fig. 4B. All models
describe the elongation rate rather well, levels and position of
maximal elongation are captured whereas the slope of the
elongation rate is somewhat steeper in the simulations. Fig. 4B
shows some differences between turbulence models; the largest
deviation of maximum elongation is with the Standard k  e
which can be interpreted in terms of its inability to handle the
separation zone in the gap (see Section 4.4).
Turbulence levels are low in the inlet chamber, both from
measurements and simulations. A quantitative comparison is
hindered by the large dependence on assumed turbulence intensity in the inlet.
4.4. The gap
Contours of velocity magnitude in the region close to the gap
entrance from CFD simulations using the RNG k  e model can be
seen in Fig. 5A. As the ow enters the gap, a boundary-layer
develops close to the right wall in Fig. 5A. This was also shown

experimentally by Hakansson
et al. (2011, Fig. 3A). The development of a boundary-layer zone with separation is well known
from previous experiments for this type of geometries (Phipps,
1974) and has been discussed in relation to cavitation (Phipps,

1974; Hakansson
et al., 2010). It was suggested that the high local
velocity and thus low static pressure close to this zone can induce

A. H
akansson et al. / Chemical Engineering Science 71 (2012) 264273

269

Fig. 6. Velocity magnitude (A) and turbulent kinetic energy, k (B) for line II (see Fig. 1). Comparison of experimentally obtained values with simulations using three
different turbulence models.

Fig. 7. Contours of turbulent kinetic energy in the early part of the gap from CFD with three different turbulence models: Standard k  e (A), RNG k  e (B) and Realizable
k  e (C).

cavitation locally. Cavitation visualization by Hakansson


et al.
(2010) offer some support for cavitation starting in this general
region of the gap for a similar geometry.
The boundary-layer thickness is also expected to have an
important effect on the single phase ow by effectively reducing
the gap height creating a vena contracta effect; as seen in Fig. 5A,
a low velocity region is created close to the wall. The experimentally measured effective reduction in gap height, dened as the
proportion of the gap with mean velocity magnitude below 70% of
the maximum velocity, d70/h, can be seen in Fig. 5B. The
experimentally obtained boundary layer thickness increases with
distance from the entrance and reaches a maximum of approximately 10% of the gap height 0.6h into the gap. Direct comparison
of separation and reversed ow was not possible using the
experimental model. Comparison to the CFD simulations suggest
a very narrow separation zone (  0.01h thick) which is below the
resolution of the measurements in the gap.
Simulations with RNG k e and Realizable k e turbulence
models describe the boundary layer thickness rather well; however,
the position of maximum thickness occurs somewhat later in
simulations compared to experiments. The standard k e model
underestimates the boundary layer thickness (see discussion below).

The mean velocity magnitude and turbulent kinetic energy


over a line (II in Fig. 1) 0.2h downstream of the gap entrance can
be seen in Fig. 6. Fig. 6A compares velocity magnitude from PIV
experiments with CFD using three different turbulence models.
The characteristic velocity maxima close to the wall and at
velocity prole seen in the experiments is predicted by the RNG
and Realizable k  e whereas the Standard k  e predicts a rather
different prole with maximum velocity close to the opposite
wall. This difference is even more pronounced when looking at
the turbulent kinetic energy in Fig. 6B. From experiments,
turbulence kinetic energy is high in front of the boundary-layer
zone, have a much smaller local maximum close to the opposite
wall and very low levels of turbulence in the center of the gap.
This behavior is also captured by the CFD with RNG and Realizable k  e. The Standard k e model, on the other hand, overestimates the turbulence in the gap and fails altogether to give a
reasonable prole of turbulent kinetic energy. Contour plots of
turbulent kinetic energy from CFD with the different models can
be seen in Fig. 7. The RNG (Fig. 7B) and Realizable k e (Fig. 7C)
models predict a high production of turbulence close to the wall
and the boundary-layer zone, which is physically reasonable. The
Standard k e model (Fig. 7A) on the other hand, predicts

270

A. H
akansson et al. / Chemical Engineering Science 71 (2012) 264273

Fig. 8. The jet created in the outlet chamber at the jet exit: (A) velocity vectors from PIV experiments (Hakansson
et al., 2011) compared with CFD simulations using
Standard k e (B), RNG k  e (C) and Realizable k e (D) turbulence model. All velocities have been scaled with the mean gap velocity, Ug.

production in a much wider area giving high turbulence in a


much larger proportion of the gap.
Over prediction of turbulent kinetic energy due to separation
and high strain zones is a well known problem for the Standard
k  e model. (Pope, 2000). The more physical behavior seen when
using the RNG and Realizable k  e models can be explained from
the modications reducing turbulent kinetic energy for high
strain rates as discussed above in Section 3.2.
A number of studies have used RANS-CFD in order to discuss
the turbulent fragmenting force in the homogenizer gap (e.g.

Casoli et al., 2010; Kohler


et al., 2007; Raikar et al., 2009; Steiner
et al., 2006). The results in this study show there is a risk of
overestimating the turbulence, and hence also the fragmentation,
in the gap by using the Standard k e turbulence model on the
HPH valve. As a comparison, the area averaged dissipation rate of
turbulent kinetic energy, is 1.8 (1.9) times larger with the
Standard k  e compared to the RNG (Realizable k  e) model.
4.5. The outlet chamber
The outlet chamber constitutes the large volume downstream
of the gap exit (see Fig. 1). Fig. 8A shows the mean velocity eld in
a region close to the gap exit obtained from PIV experiments

(Hakansson
et al., 2011). The ow exits the gap creating a
turbulent jet. The jet attaches to the right wall roughly 10 gap
heights downstream in the chamber. A small recirculation zone is
created between the jet and the right wall. Earlier experiments
using a larger eld of view also show how a second, larger and
slow moving, recirculation region is formed on the left side of the
ardh,

jet. (Innings and Trag


2007) Fig. 8BD show velocity

contours from CFD simulations with different turbulence models.


All models give a qualitatively reasonable representation of the
ow showing a jet exiting the gap, bending and adhering to the
wall further downstream. However, the velocity prole differs
between simulations, and between experiments and simulations.
For example, the attachment distance is quite severely underestimated by the Standard k  e model whereas simulations with
RNG k  e and Realizable k  e models predicts attachment further
downstream of the gap.
The early part of the outlet chamber, with jet ow past a
recirculation zone followed by wall attachment, resembles the
well studied test case of a Backwards Facing Step (BFS).
Standard k  e is known to underestimate the re-attachment
point of a BFS with up to 20% (Nellasamy, 1987) and RNG k  e and
Realizable k e have been shown to produce somewhat larger
attachment distances (Kim et al., 2005).
Fig. 9A shows turbulent kinetic energy obtained with PIV in
parts of the outlet chamber. The values were obtained using a
large eld of view, yielding a low resolution, and are consequently
only used for a qualitative comparison of the turbulent eld.
Maximum levels of turbulent kinetic energy can be seen in the
shear layers of the jet. The right hand side shear layer have
substantially higher turbulent kinetic energy which could be
explained by the larger velocity gradients created when the jet
interacts with the small and rapidly moving recirculation zone
close to the right wall. Fig. 9BD show contours of turbulent
kinetic energy in the outlet chamber using CFD with different
turbulence models. The simulations agree with experiments in
predicting higher levels of turbulence in the shear layer close to
the right wall, however, the spatial distribution of turbulent

A. H
akansson et al. / Chemical Engineering Science 71 (2012) 264273

271

Fig. 9. Turbulent kinetic energy in the turbulent jet exiting the outlet chamber: (A) PIV, (B) Standard k e, (C) RNG k e and (D) Realizable k  e. Note the different scales for
k from PIV and CFD. Stream lines have been inserted for BD.

Fig. 10. Production of turbulent kinetic energy on line IV (dened in Fig. 1).

kinetic in and before the attachment point differs substantially.


The CFD simulations predict high turbulent kinetic energy in a
rather large region in the end of the recirculation zone, just before
attachment. In the simulations, this region extends almost all the
way to the wall, whereas experiments (Fig. 9A) show high levels
in the shear layers only. The extent of the high turbulent kinetic
energy region differs between models with the largest region seen
for the Standard k  e model and the smallest for the Realizable
k  e model.
For the BFS, Nellasamy (1987) argues that the under prediction
of attachment distance is due to an over estimation of the
turbulent kinetic energy in the recirculation zone via an increase
in turbulent viscosity, nt. This could be a part of the explanation
why Standard k  e underestimates the attachment in the outlet
chamber. However, in the HPH outlet chamber, both RNG k e
and Realizable k e overestimate the turbulent kinetic energy in
the recirculation zone without grossly over predicting the attachment length, although, it could be seen that the models predicting

longer attachment distances also predicts a smaller and less


intense high k zone.
Due to the high degree of nonlinearity in the RANS equation, it
is not obvious what causes the observed discrepancy between
models and experiments in this specic case. Some further insight
can be obtained by investigation of terms in the budget equation
of turbulent kinetic energy. Fig. 10 shows the production of
turbulent kinetic energy, Pk, estimated from PIV experiments
(Eq. (3)). The raw data is rather noisy, and has been ltered using
a LOWESS lter (second order robust method with a 5% lter
window) in order to highlight the general trends. Production is
high in the shear layers and low in the central parts of the jet. The
production is also somewhat higher in the right shear layer which
is reasonable because of the curvature of the jet. The modeled
production (Eq. (6)), using the different turbulence models, has
also been inserted in Fig. 10. The models overestimate the
production in the left shear peak and under estimate the height
of the right peak. Furthermore, the maximum level of production
in the left peak increases in order Standard k  e, RNG k e and
Realizable k  e which is also the order of the size of the erroneous
high k zone seen in Fig. 9. Thus, it seems as an inability to
estimate production in the early part of the jet is, at least partly,
responsible for the observed deviation. RANS models are well
known for failing to handle ows with curved stream lines and
separation. As seen for the gap in Section 4.4, the RNG k  e and
Realizable k  e give more physical results than the Standard k  e,
however, in contrast to the gap, the improvement by using these
models are not enough for producing accurate ow elds in the
outlet chamber. This can be understood in the differences
between the ow in the two regions. In the gap, the Standard
k e model most likely fails because of not being able to handle
high strain rates. This is signicantly improved in the RNG and
Realizable k  e models. However, even these more advanced
models still contain questionable assumptions, e.g. the model
for production of turbulent kinetic energy (Eq. (6)) that cannot

272

A. H
akansson et al. / Chemical Engineering Science 71 (2012) 264273

capture the negative terms of Eq. (3) which could be signicant


for curved ows (Pope, 2000) as the jet seen in Figs. 8 and 9.
In addition to production, overestimation of the turbulent
transport in the transverse direction could be another source of
the buildup of k in the end of the recirculation zone. Testing of
this hypothesis is unfortunately not possible using the available
data since it requires the calculation of the divergence of a third
order Reynolds stress tensor which is highly sensitive to noise.
In summary, the utilized turbulence models were not able to
capture the details of the ow eld in the outlet chamber. The
experimental comparisons suggest that this may be partly due to
problems in modeling the strong production in the early parts of the
bending jet. Further studies are needed in order to fully understand
which modeling steps in the utilized models are violated in the
present case and what level of complexity that would be needed in
order to accurately predict the outlet chamber ow.
The level and distribution of turbulence over the outlet
chamber is important when using CFD data in order to predict
where turbulent fragmentation takes place (e.g. Casoli et al., 2010;

Hakansson
et al., 2009). This study shows that the qualitative
velocity prole and turbulence as a function of gap distance could
be estimated approximately, however, the exact position of the
jet is highly sensitive to the choice of turbulence model and
the turbulence is signicantly overestimated in the part closest to
the small recirculation zone at the right wall using all three models.
This difference can lead to large errors in substantial parts of the
outlet chambers when describing the effect of drops from turbulent
forces on emulsion drops with the studied RANS models.

Complications for modeling includes possible laminarturbulent


transition, separation, re-attachment and boundary layer growth,
all of which can be considered problematic for the simplistic
turbulence models treated in this study (Pope, 2000). The predictive power of the models for cases differing widely from the
investigated one, such as the very long gap length of Phipps
(1985) could not be determined from this validation, and predictions are likely to decrease in quality when more of the aforementioned complications are present.
The outlet chamber ow will independently of geometry, assuming turbulent ow, be a turbulent jet exiting into a conned volume.
This will result in wall interaction with separation, re-attachment
and possibly impingement. Prediction of these ows is known to be
complicated using the common turbulence models (and more so for
the Standard k e model than Realizable and RNG k e) (Nellasamy,
1987) which is also conrmed in the present study (see Section 4.5).
Thus, the problems seen with simulating the ow in the outlet
chamber is likely to be a general issue for gaps with turbulent ow.
There has also been indirect experimental evidence for the creation
of a laminar outlet chamber jets for some geometries (Kleinig and
Middelberg, 1997). A separate validations would be required for this
case, however, predictions are likely to improve signicantly based
on the indications in Section 4.5, where an inability to describe
transport and production of turbulence seem to be an important
factor for the observed deviations.

5. Conclusions
4.6. Dimensionality
The experimental investigations were conducted in a scaleup model (see Section 2). The models presented thus far are two
dimensional, describing the middle plane where PIV data was
obtained. The walls in the experimental model are 7.8 gap
heights above and below the middle plane. An interesting
question is to what extent the walls in the experimental model
have any signicant effect on the ow in the middle plane where
measurements were made. This could give rise to an apparent
difference between measurements and simulations. A 3D simulation was therefore performed over the full cubical experimental geometry with RNG k  e turbulence models and either
symmetry or wall boundary conditions on the upper and lower
walls. Differences in turbulent kinetic energy were less than 1%
on all investigated positions indicating that the walls in the
experimental model have no signicant inuence on the ow in
the center plane.
4.7. Generality
This study is focused on validating the models for a specic
geometry, through scaling it applies directly to a production scale
homogenizer with a short gap and long impact ring distance
ardh,

(Innings and Trag


2007). An interesting question is to what
extent the results on the predictive power of the models in the
different regions apply to a larger set of HPH geometries.
The inlet chamber can in a more general case be described as a
narrowing region with high strain rates. Excluding the gap entrance,
the ow is laminar or show low turbulence intensities. Based on the
discussion in Section 3.1 and deviations found with the Standard k e
in the gap due to high strain, this model is not recommended for a
more general geometry that could give rise to higher strain in the
inlet chamber than in the investigated geometry, e.g. due to a rounded
or otherwise modied inlet seat wall.
Gap ow has been described as either laminar or turbulent
depending on geometry (e.g. Kiefer, 1977; Phipps, 1985).

The ability of some RANS-CFD models to describe the one


phase ow eld of a (short gap, large impact ring distance) HPH
valve were studied by comparison to experimental measurements. The turbulence models were chosen since they have been
used in previous CFD studies of HPH ow.
The inlet chamber could be well described by the simulations
regardless of turbulence model. The qualitative ow pattern was
highly similar between experiments and simulations. Elongation
rates were also described accurately by all models. Thus the effect
on emulsion drops in the inlet chamber (at least at very low volume
fractions of oil) could be determined by using the CFD simulations.
The gap ow velocity and turbulent kinetic energy is described
with good accuracy by the RNG k  e and Realizable k e whereas
the Standard k  e signicantly overestimates both the turbulent
kinetic energy and the mean area averaged dissipation of turbulent kinetic energy. The boundary-layer thickness in the inlet
of the gap can also be described accurately by using RNG k  e and
Realizable k e turbulence models. The velocity proles in
the early part of the gap are distorted with the Standard k  e
model. The inability of the Standard k e to describe the gap ow
is most likely due to the high strain and boundary-layers in
the early parts of the gap. Thus, when using the other two
turbulence models the effect on drops in the gap (for low volume
fractions of oil and without cavitation) could be captured by CFD
simulations.
In the outlet chamber, all turbulence models are reasonable at
a crude qualitative level. The details of the velocity prole, such as
point of wall attachment differs between models and between
simulations and experiments. Turbulent kinetic energy is not well
described by either model. Comparison with experimental data
suggests an inability of the models to predict the production of
turbulent kinetic energy in the early part of the bending jet as one
of the reasons for the observed differences. Thus, CFD in combination with the investigated RANS k e turbulence models is not
suitable for obtaining the detailed effect on drops in the outlet
chamber.

A. H
akansson et al. / Chemical Engineering Science 71 (2012) 264273

Nomenclature
Symbols
/US
Cm
d
Ey
h
k
L
r
Reg
t
u
Ug
X,Y
x,y
y

dij
dp

e
mC
rC
uC
uT
Sij
Pk

mean velocity vector, m/s


constant in Eq. (5), dimensionless
distance downstream gap entrance (see Fig. 4A), m
elongation rate, s  1
gap height, m
turbulent kinetic energy, m2/s2
gap length, m
distance from gap center to wall, m
Reynolds number based on gap dimensions (Eq. (1)),
dimensionless
time, s
uctuation velocity vector, m/s
mean gap velocity, m/s
coordinates (see Fig. 1), m
coordinates (see Fig. 1), m
wall friction distance from wall, dimensionless
Kronecker delta, 1 if (ij), 0 otherwise, dimensionless
boundary layer thickness in the gap. First wall distance
where the velocity magnitude is equal to or above p % of
the maximum velocity, m
dissipation rate of turbulent kinetic energy, m2/s3
dynamic viscosity of uid, Pa s
uid density, kg/m3
kinematic viscosity of uid, m2/s
turbulent viscosity, m2/s
mean rate of strain, s  1
production of turbulent kinetic energy, m2 s  3

Abbreviations
BFS
CFD
DNS
EWT
HPH
LES
LReF
PIV
RANS

Backwards Facing Step


Computational Fluid Dynamics
Direct Numerical Simulation
Enhanced Wall Treatment
High-Pressure Homogenizer
Large Eddy Simulation
Low Reynolds number Formulation
Particle Image Velocimetry
Reynolds Averaged NavierStokes

Acknowledgments
This study was nanced by the Swedish Research Council (VR).
References
ANSYS, 2009. Theory GuideFLUENT 12.0. ANSYS, Canonsburg, PA.
Casoli, P., Vacca, A., Berta, G., 2010. A numerical procedure for predicting the
performance of high pressure homogenizing valves. Simul. Model. Pract.
Theory 18, 125138.

273

Floury, J., Bellettre, J., Legrand, J., Desrumaux, A., 2004. Analysis of a new type of
high pressure homogeniser. A study of the ow pattern. Chem. Eng. Sci. 59,
843853.

ardh,

Hakansson,
A., Trag
T., Bergenstahl,
B., 2009. Dynamic simulation of emulsion
formation in a high pressure homogenizer. Chem. Eng. Sci. 64, 29152925.

ardh,

Hakansson,
A., Fuchs, L., Innings, F., Revstedt, J., Bergenstahl,
B., Trag
C., 2010.
Visual observations and acoustic measurements of cavitation in an experimental model of a high-pressure homogenizer. J. Food Eng. 100, 504513.

ardh,

Hakansson,
A., Fuchs, L., Innings, F., Revstedt, J., Trag
C., Bergenstahl,
B., 2011.
High resolution experimental measurement of turbulent ow eld in a high
pressure homogenizer model and its implications on turbulent drop fragmentation. Chem. Eng. Sci. 66, 17901801.
ardh,

Innings, F., Trag


C., 2007. Analysis of the ow eld in a high-pressure
homogenizer. Exp. Therm. Fluid Sci. 32, 345354.

Kiefer, P., 1977. Der Einuss von Scherkraften


auf die Tropfchenzerkleinerung
l-in-Wasser-Emulsionen in Hochdruckhomogenbeim Homogenisieren von O

isierdusen.
Doctoral Thesis, Karlsrhue Technische Hochschule, Karlshue.
Kelly, W.J., Muske, K., 2004. Optimal operation of high-pressure homogenization
for intracellular product recovery. Bioprocess. Biosyst. Eng. 27, 2537.
Kim, J.-Y., Ghajar, A.J., Tang, C., Foutch, G.L., 2005. Comparison of near-wall
treatment methods for high Reynolds number backward-facing step ows.
Int. J. Comput. Fluid Dyn. 19, 493500.
Kleinig, A.R., Middelberg, A.P.J., 1996. The correlation of cell disruption with
homogenizer valve pressure gradient determined by computational uid
dynamics. Chem. Eng. Sci. 51, 51035110.
Kleinig, A., Middelberg, P., 1997. Numerical and experimental study of a homogenizer impinging jet. AIChE J. 43, 11001107.

Kohler,
K., Aguilar, F., Hensel, A., Schubert, K., Schubert, H., Schuchmann, H.P.,
2007. Design of a microstructured system for homogenization of dairy
products with high fat content. Chem. Eng. Technol. 30, 15901595.
Launder, B.E., Sharma, B.I., 1974. Application of the energy-dissipation model of
turbulence to the calculation of ow near a spinning disc. Lett. Heat Mass
Transfer 1, 131137.
Launder, B.E., Spalding, D.B., 1974. The numerical computation of turbulent ows.
Comput. Method Appl. Mech. Eng. 3, 269289.
Miller, J., Rogowski, M., Kelley, W., 2002. Using a CFD model to understand the
uid dynamics promoting E. coli breakage in a high-pressure homogenizer.
Biotechnol. Prog. 18, 10601067.
Nellasamy, M., 1987. Turbulence models and their applications to the prediction of
internal ows: a review. Comput. Fluids 15, 151194.
Phipps, L.W., 1974. Cavitation and separated ow in a simple homogenizing valve
and their inuence on the break-up of fat globules in milk. J. Dairy Res. 41,
18.
Phipps, L.W., 1985. The High Pressure Dairy Homogenizer. The National Institute
for Research in Dairying, Reading.
Pope, S.P., 2000. Turbulent Flows. Cambridge University Press, Cambridge.
Patankar, S.V., Spalding, D.B., 1972. A calculation procedure for heat, mass and
momentum transfer in three-dimensional parabolic ows. Int. J. Heat Mass
Transfer 15, 17871806.
Raikar, N.B., Bhatia, S.R., Malone, M.F., Henson, M.A., 2009. Experimental studies
and population balance equation models for breakage prediction of emulsion
drop size distributions. Chem. Eng. Sci. 64, 24332447.
Shih, T.-H., Liou, W.W., Shabbir, A., Yang, Z., Zhu, J., 1995. A new k e eddyviscosity model for high Reynolds number turbulent ows. Comput. Fluids 24,
227238.
Steiner, H., Teppner, R., Brenn, G., Vankova, N., Tcholakova, S., Denkov, N., 2006.
Numerical simulation and experimental study of emulsication in a narrowgap homogenizer. Chem. Eng. Sci. 61, 58415855.
Stevenson, M.J., Chen, X.D., 1997. Visualization of the ow patterns in a highpressure homogenizing valve using a CFD package. J. Food Eng. 33, 151165.
Walstra, P., 1993. Principles of emulsion formation. Chem. Eng. Sci. 48, 333349.
Yakhot, V., Orszag, S.A., 1986. Renormalization group analysis of turbulence: I.
Basic Theory. J. Sci. Comput. 1, 351.

You might also like