Вы находитесь на странице: 1из 202

CFD SIMULATION OF IMMISCIBLE

LIQUID DISPERSIONS

by

Srinath Madhavan

Submitted
in partial fulfillment of the requirements
for the degree of

MASTER OF APPLIED SCIENCES

Major Subject: Chemical Engineering

at

DALHOUSIE UNIVERSITY

Halifax, Nova Scotia July, 2005

© Copyright by Srinath Madhavan, 2005


Dalhousie University
Faculty of Engineering
Department of Chemical Engineering
The undersigned hereby certify that they have examined, and recommend to the Faculty
of Graduate Studies for acceptance, the thesis entitled “CFD Simulation of Immiscible
Liquid Dispersions” by Srinath Madhavan in partial fulfillment of the requirement for the
degree of Master of Applied Sciences.

Dated: ____ _____ ____

Supervisor: ______________________________
Dr. A M. Al Taweel

Co Supervisor: _______________________________
Dr. M. Koksal

Examiners: _______________________________
Dr. J. Chuang

________________________________
Dr. Y. Gupta

________________________________
Dr. T. Dabros

ii
Dalhousie University
Faculty of Engineering

DATE: _______________________

AUTHOR: Srinath Madhavan

TITLE: CFD Simulation of Immiscible Liquid Dispersions

MAJOR SUBJECT: Chemical Engineering

DEGREE: Master of Applied Sciences

CONVOCATION: October, 2005

Permission is herewith granted to Dalhousie University to circulate and to have copied


for non-commercial purposes, at its discretion, the above title upon the request of
individuals or institutions.

_______________________________
Signature of Author

The author reserves other publication rights, and neither the thesis nor extensive extracts
from it may be printed or otherwise reproduced without the author’s written permission.

The author attests that permission has been obtained for the use of any copyrighted
material appearing in the thesis (other than the brief excerpts requiring only proper
acknowledgement in scholarly writing), and that all such use is clearly acknowledged.

iii
TABLE OF CONTENTS
TABLE OF CONTENTS................................................................................................... iv
LIST OF TABLES........................................................................................................... viii
LIST OF FIGURES ........................................................................................................... ix
NOMENCLATURE ......................................................................................................... xii
ACKNOWLEDGEMENTS............................................................................................. xiv
ABSTRACT...................................................................................................................... xv
1 INTRODUCTION ...................................................................................................... 1
1.1 Multiphase Flow ................................................................................................. 1
1.2 Immiscible Liquid Dispersions and their Importance......................................... 3
1.2.1 Dispersed Phase Holdup and its Significance............................................. 4
1.2.2 The Need for Numerical Studies................................................................. 4
1.3 Overview of Computational Modeling and CFD................................................ 5
1.3.1 Computational Modeling ............................................................................ 5
1.3.2 Computational Fluid Dynamics (CFD)....................................................... 6
1.4 Problem Statement .............................................................................................. 7
2 USING CFD TO SIMULATE LIQUID-LIQUID SYSTEMS: PREVIOUS WORK 9
2.1 CFD Simulation of Multi-fluid Flows ................................................................ 9
2.1.1 Direct Numerical Simulations (DNS)....................................................... 10
2.1.2 Dispersed Multi-fluid Flow Modeling ...................................................... 10
2.1.2.1 Volume of Fluid (VOF) ........................................................................ 11
2.1.2.2 Eulerian-Lagrangian (E-L).................................................................... 12
2.1.2.3 Eulerian-Eulerian (E-E) ........................................................................ 12
2.1.2.4 Mixture.................................................................................................. 13
2.1.3 Treatment of Turbulence in Multi-fluid Flows......................................... 14
2.1.3.1 Large Eddy Simulations (LES)............................................................. 14
2.1.3.2 Turbulence Models based on RANS..................................................... 15
2.1.4 Treatment of Interphase Forces in Dispersed Multi-fluid flow ................ 18
2.1.4.1 Drag Force ............................................................................................ 18
2.1.4.2 Lift Force .............................................................................................. 19
2.1.4.3 Virtual/Added Mass Force.................................................................... 20
2.1.4.4 Basset Force .......................................................................................... 21
2.1.4.5 Wall Lubrication Force ......................................................................... 21
2.1.4.6 Turbulent Dispersion ............................................................................ 22
2.2 Discussion of previous CFD Studies in Liquid-Liquid Systems ...................... 22
2.2.1 Rotating Disc Contactor (RDC) Columns ................................................ 23
2.2.1.1 Multiphase and Turbulence Modeling Approach ................................. 24
2.2.1.2 Treatment of Drop Diameter................................................................. 25
2.2.1.3 Interphase Forces .................................................................................. 26
2.2.2 Stirred/Mechanically Agitated Tanks ....................................................... 26
2.2.2.1 Multiphase and Turbulence Modeling Approach ................................. 31
2.2.2.2 Treatment of Drop Diameter................................................................. 32

iv
2.2.2.3 Interphase Forces .................................................................................. 32
2.2.3 Static Mixers ............................................................................................. 33
2.2.3.1 Multiphase and Turbulence Modeling Approach ................................. 34
2.2.3.2 Treatment of Drop Diameter................................................................. 35
2.2.3.3 Interphase Forces .................................................................................. 35
2.2.4 Jet Homogenizer ....................................................................................... 35
2.2.4.1 Multiphase/Turbulence Modeling Approach, Treatment of Drop
Diameter and Interphase Forces............................................................................ 36
2.2.5 Pipelines.................................................................................................... 36
2.2.5.1 Multiphase and Turbulence Modeling Approach ................................. 38
2.2.5.2 Treatment of Drop Diameter................................................................. 40
2.2.5.3 Interphase Forces .................................................................................. 41
3 RESEARCH OBJECTIVE AND METHODOLOGY.............................................. 48
4 REVIEW OF EXPERIMENTAL INVESTIGATIONS IN VERTICAL LIQUID-
LIQUID PIPE FLOWS ..................................................................................................... 51
4.1 Analysis of Liquid-Liquid Pipe Flow Experiments .......................................... 51
4.1.1 Discussion of Experimental Investigations............................................... 54
4.1.2 Selection of Experimental Data for CFD Validation................................ 57
4.1.2.1 Selected Experimental Data Points....................................................... 59
5 CFD MODELING OF LIQUID-LIQUID DISPERSIONS ...................................... 62
5.1 Selection of an Appropriate Multi-fluid Model for Simulating Immiscible
Liquid Dispersions ........................................................................................................ 62
5.1.1 A synopsis of the Eulerian-Eulerian Approach ........................................ 64
5.1.1.1 Conservation of Mass (Equation of Continuity)................................... 64
5.1.1.2 Conservation of Momentum ................................................................. 64
5.2 Selection of a Turbulence model for Simulating Immiscible Liquid Dispersions
........................................................................................................................... 66
5.2.1 Two-Equation Standard k-ε Model........................................................... 66
5.2.1.1 Near-Wall Modeling Approaches ......................................................... 67
5.2.1.2 A Synopsis of the Standard k-ε Turbulence Model .............................. 68
5.2.1.2.1 Turbulence in the Continuous Phase............................................... 69
5.2.1.2.2 Turbulence in the Dispersed Phase ................................................. 71
5.3 Interphase Closure ............................................................................................ 71
5.3.1 Drag Forces in Liquid-Liquid Systems..................................................... 72
5.3.1.1 Drag on a Single Drop .......................................................................... 73
5.3.1.2 Drag on a Drop in the Presence of Adjacent Drops.............................. 77
5.3.2 Non-drag Lateral Forces in Dispersed Liquid-Liquid Flows.................... 80
5.3.2.1 Lift Forces............................................................................................. 81
5.3.2.2 Turbulent Dispersion ............................................................................ 85
5.3.2.2.1 A Synopsis of the Simonin and Viollet (1990) Turbulent Dispersion
Model ......................................................................................................... 87
6 RESULTS AND DISCUSSION ............................................................................... 91
6.1 Analysis of Interphase Forces........................................................................... 91
6.1.1 Assessment of Drag Forces....................................................................... 91

v
6.1.1.1 Drag on a Single Drop .......................................................................... 92
6.1.1.1.1 Comparative Evaluation of Drag Coefficient Expressions for Single
Entities (using CFD Simulations) ..................................................................... 94
6.1.1.2 Drag on a Single Drop in the Presence of Adjacent Drops................... 96
6.1.1.2.1 Comparative Evaluation of Drag Coefficient Expressions that
Account for the Presence of Adjacent Entities (using CFD Simulations) ........ 98
6.1.2 Assessment of Lift Forces......................................................................... 99
6.1.2.1 Constant Lift Coefficient ...................................................................... 99
6.1.2.2 Expression for CL proposed by Moraga et al. (1999) ......................... 101
6.1.2.3 Expression for CL proposed by Troshko et al. (2001) ........................ 102
6.1.3 Assessment of Turbulent Dispersion ...................................................... 105
6.1.4 Re-assessment of the Drag Force taking into account other Interphase
Forces ................................................................................................................. 107
6.2 CFD Simulation of Liquid-Liquid Dispersions .............................................. 109
6.2.1 Simulation Results .................................................................................. 113
6.2.1.1 Data Set of Farrar and Bruun (1996) .................................................. 114
6.2.1.2 Data Set of Hamad et al. (2000).......................................................... 116
6.2.1.3 Data Set of Al-Deen and Bruun (1997) .............................................. 118
6.2.1.4 Data Set of Lang (1999)...................................................................... 122
6.2.1.5 Data Set of Vigneaux et al. (1988)...................................................... 127
6.2.2 Conclusions from CFD Simulations ....................................................... 131
6.2.3 Validation of the Proposed Interphase Closure Guidelines .................... 133
6.2.3.1 Validation Results............................................................................... 135
6.2.3.1.1 Vertical Holdup Profile Data Set of Soleimani et al. (1999) ........ 136
6.2.3.1.2 Horizontal Holdup Profile Data Set of Angeli (1996) .................. 139
6.2.3.2 Conclusions and Recommendations ................................................... 140
7 OVERALL CONCLUSIONS AND RECOMMENDATIONS.............................. 142
7.1 Conclusions..................................................................................................... 142
7.2 Recommendations........................................................................................... 144
REFERENCES ............................................................................................................... 146
APPENDICES ................................................................................................................ 158
A1.1 CFD SIMULATION METHODOLOGY.............................................................. 159
A1.1.1 Numerical Solution of Differential Equations ................................................ 159
A1.1.2 Placement of the First Grid Point.................................................................... 162
A1.1.3 Solution of the Governing Equations.............................................................. 163
A1.1.3.1 Pressure-Velocity Coupling ..................................................................... 164
A1.1.3.2 Turbulence Modeling in Fluent 6.1.22..................................................... 165
A1.1.4 Boundary Conditions ...................................................................................... 165
A1.1.5 Simulation Conditions..................................................................................... 167
A1.1.6 Solution Convergence ..................................................................................... 168
A1.1.7 Post-processing ............................................................................................... 170
A1.1.8 Additional Considerations for 3D Simulations............................................... 170
A2.1 GRID INDEPENDENCE STUDIES ..................................................................... 172
A3.1 SOURCE CODES.................................................................................................. 176

vi
A3.1.1 Accounting for the Drag Force in Fluent 6.1.22 ............................................. 176
A3.1.1.1 User-Defined Function (UDF) Implementation Aspects ......................... 176
A3.1.2 User-Defined Functions .................................................................................. 178
A3.1.3 Analysis of Drag Expressions ......................................................................... 183
A3.1.4 Calculation of the Position of the First Grid Point.......................................... 187

vii
LIST OF TABLES
Table 2.1 Summary of Previous Liquid-Liquid CFD studies ........................................... 42
Table 4.1 Experimental investigations on liquid-liquid pipe flows.................................. 52
Table 4.2 Characteristic dispersed phase holdup profiles observed in vertical pipe
upflows of liquid-liquid dispersions ................................................................................. 56
Table 4.3 Summary of liquid-liquid experimental data points selected for CFD validation
........................................................................................................................................... 61
Table 5.1 Standard k-ε turbulence model constants (Launder and Spalding, 1972) ........ 69
Table 5.2 Drag expressions for single drops..................................................................... 75
Table 5.3 Typical dimensionless groups used in drag/lift coefficient expressions........... 76
Table 5.4 Drag expressions for drops which take into account the presence of adjacent
drops.................................................................................................................................. 79
Table 6.1 Physical data and constants used in the analysis of various drag models......... 92
Table 6.2 Average holdup predictions for various single-entity drag models at low phase
ratio (5%) .......................................................................................................................... 94
Table 6.3 Average holdup predictions for various single-entity drag models at high phase
ratio (30%) ........................................................................................................................ 95
Table 6.4 Average holdup predictions for various drag models taking into account the
presence of adjacent entities at high phase ratio (30%).................................................... 98
Table 6.5 Average holdup predictions for various drag models at high phase ratio (30%),
taking into account the combined effect of drag, lift and turbulent dispersion .............. 108
Table 6.6 Summary of experimental flow conditions for selected data points............... 113
Table 6.7 Experimental conditions and CFD simulation results .................................... 129
Table 6.8 Summary of experimental flow conditions for horizontal pipes .................... 135

viii
LIST OF FIGURES
Figure 1.1 Multi-fluid flow patterns for horizontal pipe flow (Wörner, 2003) .................. 2
Figure 2.1 Modeling approaches for multiphase flows .................................................... 11
Figure 2.2 Schematic representation of scales in turbulent flows (adapted from Ferziger
and Peric, 1995) ................................................................................................................ 15
Figure 2.3 Quick comparison of DNS, LES and RANS (Ranade, 2002) ......................... 16
Figure 2.4 Simulation and modeling approaches for turbulent flows (Adapted from
Ranade, 2002) ................................................................................................................... 17
Figure 2.5 Nature of drag forces in multiphase systems................................................... 19
Figure 2.6 Lift force acting on a dispersed entity ............................................................. 20
Figure 2.7 CFD studies on various liquid-liquid contactors ............................................. 22
Figure 2.8 RDC Schematic (Koch Modular Process Systems, 2000)............................... 23
Figure 2.9 Schematic representation of a stirred/mechanically agitated tank .................. 27
Figure 2.10 Concentration distribution contours in a static mixer (LaRoche, 1998) ....... 33
Figure 2.11 Key components of a Jet Homogenizer (Soon et al., 2001) .......................... 35
Figure 2.12 Fully mixed flow pattern for oil-water system in a horizontal pipe (Angeli
and Hewitt, 2000).............................................................................................................. 37
Figure 2.13 Radial dispersion of CCl4 drops in water (Domgin et al., 1997)................... 39
Figure 2.14 Radial distribution of volume fraction for kerosene-water system:
Experiment (+) CFD Simulation ( ) (Hamad et al., 1999) ............................................. 40
Figure 5.1 Flow structure near the wall (Fluent, 2004) .................................................... 67
Figure 5.2 Various approaches to model near-wall Turbulence (Adapted from Fluent,
2004) ................................................................................................................................. 68
Figure 5.3 Internal circulation in a drop without (left) and with (right) surface-active
agents (Adapted from Patro, 2003)................................................................................... 73
Figure 5.4 Inviscid lift forces experienced by an entity in shear flow.............................. 83
Figure 6.1 Comparison of various single-entity drag models........................................... 93
Figure 6.2 Comparison of various drag models that take into account the presence of
adjacent entities at de = 2 mm ........................................................................................... 96
Figure 6.3 Comparison of various drag models that take into account the presence of
adjacent entities at de = 5 mm ........................................................................................... 97
Figure 6.4 Effect of constant lift coefficients on the radial distribution of the dispersed
phase ............................................................................................................................... 100
Figure 6.5 Variation of the lift coefficient with Re Re∇ as proposed by Moraga et al.
(1999) and Troshko et al. (2001) .................................................................................... 102
Figure 6.6 Comparison of the expression proposed by Troshko et al. (2001) with constant
lift coefficients ................................................................................................................ 104
Figure 6.7 Radial distribution of the lift coefficient for the expression proposed by
Troshko et al. (2001) as compared to constant lift coefficients...................................... 104
Figure 6.8 Effect of various dispersion coefficients (CTD) on the dispersed phase holdup
for the Lopez de Bertodano (1992) model...................................................................... 106
Figure 6.9 Effect of various dispersion prandtl numbers (DPN) on the dispersed phase
holdup for the Simonin and Viollet (1990) model.......................................................... 106

ix
Figure 6.10 Effect of varying drop diameter on the holdup for CL = 0.005 ................... 111
Figure 6.11 Effect of varying drop diameter on the holdup for the lift coefficient
expression proposed by Troshko et al. (2001) ................................................................ 111
Figure 6.12 Lift coefficient distribution for the expression proposed by Troshko et al.
(2001).............................................................................................................................. 112
Figure 6.13 Simulated phase distribution profiles for the data set of Farrar and Bruun
(1996), (DPN = 7.5; CL = Troshko et al. (2001); de = 5 mm)......................................... 114
Figure 6.14 Effect of constant (negative) lift coefficients on the dispersed phase holdup,
(DPN = 7.5, de = 5 mm) .................................................................................................. 115
Figure 6.15 Comparison of experimental and simulated continuous phase velocity
profiles for the data set of Farrar and Bruun (1996), (DPN = 7.5; CL = Troshko et al.
(2001); de = 5 mm).......................................................................................................... 115
Figure 6.16 Simulated phase distribution profiles for the data set of Hamad et al. (2000),
H10 (DPN = 0.01; CL = Troshko et al. (2001); de = 3.25 mm), H20 (DPN = 0.075; CL =
Troshko et al. (2001); de = 3.5 mm)................................................................................ 117
Figure 6.17 Simulated phase distribution profiles for the data set of Al-Deen and Bruun
(1997), CL = Troshko et al. (2001), A5 (DPN = 0.024; de = 3 mm), A10 (DPN = 0.075; de
= 3.5 mm), A20 (DPN = 0.412; de = 4 mm), A30 (DPN = 7.5; de = 4 mm)................... 119
Figure 6.18 Effect of constant lift coefficients on phase distribution for data point A20
(DPN = 7.5; de = 4 mm).................................................................................................. 120
Figure 6.19 Radial variation of lift coefficient for data point A5 (DPN = 0.024, de = 3
mm) ................................................................................................................................. 121
Figure 6.20 Predicted and experimental ‘actual continuous phase velocity’ distribution
for data point A30 (DPN = 7.5; CL = Troshko et al. (2001); de = 4 mm) ....................... 121
Figure 6.21 Comparison of experimental and simulated continuous phase velocity
profiles for the data set of Al-Deen and Bruun (1997), A5 (DPN = 0.024; CL = Troshko et
al. (2001); de = 3 mm)..................................................................................................... 122
Figure 6.22 Simulated phase distribution profile from the data set of Lang (1999), L20B
(DPN = 0.06; CL = -0.05; de = 2 mm)............................................................................. 123
Figure 6.23 Simulated phase distribution profiles for the data set of Lang (1999), L20A
(DPN = 0.75; CL = -0.075; de = 1 mm), L40 (DPN = 3.0; CL = -0.05; de = 1 mm)........ 123
Figure 6.24 Comparison of two near-wall modeling approaches for data point L40 (DPN
= 3.0; CL = -0.05; de = 1 mm) ......................................................................................... 124
Figure 6.25 Predictions obtained when the lift expression proposed by Troshko et al.
(2001) is used for data point L20B (de = 2 mm)............................................................. 125
Figure 6.26 Comparison of predicted and experimental velocities for the dispersed phase
from the data set of Lang (1999), (DPN = 0.06; CL = -0.05; de = 2 mm)....................... 126
Figure 6.27 Phase distribution profiles for the data set of Vigneaux et al. (1988), V5
(DPN = 1.59; CL = -0.05; de = 2.5 mm), V10 (DPN = 0.328; CL = -0.05; de = 2.75 mm),
V30 & V50 (DPN = 4.125; CL = Troshko et al. (2001); de = 5 mm).............................. 128
Figure 6.28 Recommendation for turbulent dispersion in terms of DPN ....................... 132
Figure 6.29 Upper and lower limits for DPN beyond which no significant differences in
the holdup distribution are observed............................................................................... 133

x
Figure 6.30 Transient phase distribution profiles for the data set of Soleimani et al.
(1999), S40A (DPN = 7.5; CL = -0.05; de = 0.5 mm) ..................................................... 136
Figure 6.31 Transient phase distribution profiles for the data set of Soleimani et al.
(1999), S54 (DPN = 7.5; CL = -0.05; de = 0.5 mm) ........................................................ 138
Figure 6.32 Transient phase distribution profiles for the data set of Soleimani et al.
(1999), S40B (DPN = 7.5; CL = -0.05; de = 0.5 mm) ..................................................... 138

xi
NOMENCLATURE
CD0 Drag coefficient of a single entity [-]
CDM Drag coefficient in the presence of adjacent entities [-]
CL Lift coefficient [-]
CTD Turbulent dispersion coefficient [-]
CV Coefficient of virtual/added mass [-]
Cµ Constant in standard k-ε turbulence model (= 0.09) [-]
D Diffusivity [m2/s]
de Equivalent drop diameter [m]
DPN Dispersion Prandtl Number [-]
g Acceleration due to gravity (= 9.81) [m/s2]
ID Pipe internal diameter [m]
k Turbulent Kinetic energy [m2/s2]
L Pipe length [m]
p Pressure [N/m2]
Re Drop/Bubble Reynolds Number [-]
t Time [s]
u Velocity [m/s]
U Velocity [m/s]
v Velocity [m/s]
VS Relative/Slip velocity [m/s]

Greek letters
α Volume fraction/Holdup [-]
ε Turbulent Kinetic energy dissipation rate [m2/s3]
µ Viscosity [kg/ms]
ρ Density [kg/m3]
σ Oil-water interfacial tension [N/m]
τ Shear stress [N/m2]

xii
Subscripts
e Effective/Equivalent quantity
m Mixture
p, d Dispersed phase (drop)
q, c Continuous phase (water)
t Turbulent
Superscripts
¯ Mean component
Fluctuating component


Vector quantity
T Turbulent

xiii
ACKNOWLEDGEMENTS
I would like to express deep appreciation to my research supervisors, Dr. A. M. Al
Taweel and Dr. Murat Koksal for their guidance, encouragement, support and criticism
throughout the course of this work. I am also grateful to the members of the supervisory
committee, Dr Y.Gupta, Dr J. Chuang, and Dr T. Dabros for their continuous guidance
and helpful comments.In addition I would like to thank all faculty members and staff at
the chemical engineering department for their help and cooperation.

I gratefully acknowledge the support and help offered by FLUENT engineers, Dr. K.
Dhanasekharan, Dr. A. Troshko and Dr. B. Bell, whose lectures and tutorials were
immensely useful in saving a lot of time and effort. The technical guidance provided by
the academic support team at Fluent cannot be underestimated.

I acknowledge the help rendered by the Sexton Design & Technology library in
collecting the necessary literature to complete the research work. I am also thankful to
my research supervisor, the Department of Chemical Engineering, the Faculty of
Engineering and the Faculty of Graduate studies for providing teaching assistantships and
scholarship funding throughout the course of my study.

Special thanks are expressed to F. Azizi, K. Podila, P. Appasani, C. Li, R. Venkataraman,


S. A. Somasuntharam, S. Pallapothu, A. Rudrake, R. S. Jadeja, Naveen B and S. Rao for
their support, help, and companionship throughout the time we spent sharing the same
office. I am also thankful to my roommate Sameer for his continuous cooperation and
companionship. Last but not the least I am highly obliged to my parents, Mr. Madhavan
and Mrs. Srimathi Madhavan for their constant support and affection, and their boundless
dedication to my education. I am also grateful to by elder brother, Mr. Sriharsha
Madhavan for his encouragement and support throughout.

xiv
ABSTRACT
Immiscible liquid dispersions are extensively encountered in the chemical, process and
petroleum industries. Some common examples are solvent extraction, emulsification,
polymerization, caustic treatment, heterogeneous chemical synthesis etc. In most of these
chemical processes, the rate of interphase heat/mass transfer is known to strongly affect
the overall performance and depends on the interfacial contact area between the two
phases. Experiments can yield significant insight into the factors affecting the interfacial
area. However, problems such as the limited range of applicability of the empirical
correlations so developed, difficulty in extrapolating the inferences to other geometries or
even scaled-up versions of the same equipment, overall cost and time frame limitations
etc. have led to more attention being focused on numerical approaches such as
Computational Fluid Dynamics (CFD). This is particularly true for multi-fluid flow in
complex/intricate geometries. The proven success of CFD to accurately predict the
hydrodynamics in most single-phase flows cannot however, be readily translated to the
case of multiphase dispersions. The presence of the dispersed phase invariably introduces
additional complexities. Nevertheless, if fundamental multiphase characteristics such as
the dispersed phase holdup could be accurately predicted, it would be easier to estimate
the contact area provided the dynamics of the dispersed phase (size distributions,
breakup, coalescence etc.) are properly accounted for. This can eventually be
accomplished, provided CFD analysis is carried out meticulously by slowly increasing
the problem complexity.
With specific reference to dispersed liquid-liquid flows, it is seen that the two-fluid
approach is widely used for multiphase modeling, especially when detailed predictions
are desirable over a wide range of holdups in exchange for a reasonable amount of
computation power. This is made possible in the two-fluid framework by appropriate
averaging of the governing equations. Unfortunately, such manipulations result in more
unknowns than the number of available equations. In order to obtain a closed set of
equations, certain terms require modeling. These are the interphase forces and turbulent
stresses (i.e. the closure problem).
The present study is focused towards identifying and quantifying the various significant
forces encountered in turbulent liquid-liquid dispersions. The vertical pipeline was
chosen as the geometry of interest as it featured simple hydrodynamics in addition to an
extensive database of accurate experimental results. Preliminary simulations confirmed
the experimental observations that drag, lift and turbulent dispersion were the significant
interphase forces affecting the spatial holdup distribution. Several expressions for these
forces were therefore tested by implementing them in a commercial CFD package (Fluent
6.1.22). CFD predictions so obtained were subsequently compared with experiments
featuring a wide range of pipe diameters, superficial velocities, and dispersed phase
holdups. Interphase closure guidelines for liquid-liquid bubbly flows were developed
based on simulation results that yielded the best agreement with the experimental data.
The suitability of the closure guidelines was then tested by applying them to the case of
horizontal dispersed flows.

xv
1

CHAPTER 1

1 INTRODUCTION
1.1 Multiphase Flow
A phase can be generally defined as any identifiable class of matter (e.g. temperature,
composition, density etc.), where matter can be categorized into thermodynamic phases
such as gas, liquid and solid. Any system that features only one of these unique phases is
called a single phase system. Consequently, the flow of any of these phases in their
entirety can be considered as single phase flow. On the other hand, a system that features
a combination of different phases can be referred to as a multiphase system. It would
therefore be appropriate to extend this logical definition to multiphase flow as the flow of
a mixture of any of the above phases. However, the definition of multiphase flow extends
slightly beyond this simple and fairly obvious characterization. Multiphase flow
encompasses not only the combined flow of two or more distinct thermodynamic phases,
but also any flow of heterogeneous nature. For instance, the collective flow of two solid
spheres made of the same material but different sizes (a heterogeneous class), is also
considered as multiphase flow.

Multi-fluid systems are a sub-group of multiphase systems in which the dispersed phase
is made up of bubbles and/or drops. This includes liquid-in-gas, gas-in-liquid as well as
liquid-in-liquid dispersions, which are encountered in a broad range of industries (such as
the production, storage and transport of oil and gas resources, the chemical process
industry, biotechnology, mineral and metal processing, energy conversion, separation
technology, and environmental technology). Operations such as distillation, absorption,
stripping, multiphase reactions, liquid-liquid extraction, direct contact heat transfer,
biological wastewater treatment, ozonation, evaporation, steam condensation, spraying,
and fuel atomization involve multi-fluid systems. The performance of these operations
(for e.g. conversion of chemical reactions, heat and mass transfer rates) is dictated by the
2

detailed structure of the multi-fluid flow and the resulting complex interphase
interactions.

Multi-fluid flows can appear in quite different topological or morphological


configurations (usually referred to as flow regimes or flow patterns) depending on
operating conditions and geometry of process equipment. Various physical interphase
transfer processes are known to strongly depend on the nature of the flow regime. Figure
1.1 illustrates typical flow patters encountered in horizontal multi-fluid pipe flow.

(a) bubbly flow; (b) plug flow; (c) stratified (segregated flow); (d) wavy flow;
(e) slug flow; (f) annular flow; (g) spray/drop flow

Figure 1.1 Multi-fluid flow patterns for horizontal pipe flow (Wörner, 2003)
When two phases move together with one floating over the other (Figure 1.1 c & d),
neither of the phases exists in a discontinuous particle-like form. Such as situation, would
feature a fairly clear and well-defined interface of contact between the two phases. In
multiphase jargon, this type of multi-fluid motion is commonly referred to as 'stratified'
or segregated flow. The density difference between the phases is primarily responsible
for this phenomenon. A typical example is the flow of oil and water in a horizontal pipe
at low phase velocities. On the other hand, if the external surface of one phase is
completely encompassed by the other for most of the time, the former phase is said to
3

have been dispersed into the latter. The latter is then aptly referred to as continuous
phase, as it occupies continuously connected regions of space. The former is usually
referred to as dispersed phase. Such flow patterns are identified as bubbly flows (Figure
1.1 a). A very common example is the dispersion of gas into a liquid which can be easily
observed, for instance in a fish tank, where air is dispersed as bubbles in water for
aeration purposes.

1.2 Immiscible Liquid Dispersions and their Importance


As the present work is focused on two-phase liquid-liquid dispersions, it will be useful at
the very outset to present a quick introduction on their nature and importance. Liquid-
Liquid dispersions ideally comprise of drops of one liquid scattered or dispersed into the
other (bubbly flow). Such immiscible mixtures are extensively encountered in the
chemical, process and petroleum industries.

In chemical process industries, liquid-liquid dispersions are commonly encountered in


solvent extraction, emulsification, polymerization, heterogeneous chemical synthesis,
direct contact heat transfer, homogenization processes etc. For optimal design and
efficient operation of equipments handling such dispersions, it is imperative that a
qualitative and quantitative understanding of the hydrodynamics (i.e. flow fields,
turbulence, and dispersion phenomena) be developed. Also, the transfer of mass and heat
to and from the dispersed phase assumes/ additional importance in the production,
processing and/or transport of intermediate/end products.

Dispersed entities typically feature varying shapes and sizes, a factor which can affect
both the hydrodynamics and the heat/mass transfer rates. In chemical processes involving
multiphase dispersions, it is well known that mass transfer rates can be easily enhanced if
the interfacial contact area of the dispersed phase is increased. This can be inferred from
Equation 1.1, where the rate of mass transfer (dφ dt ) is shown to directly depend on the

( )
concentration gradient φ * − φ and the interfacial area of contact ( A) . The constant of
proportionality (k L ) is commonly referred to as the overall mass transfer coefficient.
4


Mass transfer rate =
dt
(
= kL A φ * − φ ) [1.1]

In comparison to the driving force (φ * − φ ) , the contact area ( A) can be more easily
manipulated; consequently, most efforts concentrate on finding efficient means to
increase the value of A . One of the first steps towards realizing a high contact area is the
choice of a flow pattern. With reference to Figure 1.1, it can be seen that flow patterns
such as bubbly or spray/droplet flows are most desirable.

1.2.1 Dispersed Phase Holdup and its Significance


One of the fundamental multiphase characteristics of particular interest in bubbly flows
(such as those encountered in stirred tanks, mechanically agitated columns, static mixers,
pipelines, bubble/spray columns etc.) is the dispersed phase holdup, which is defined as
the volume fraction of the total domain that is occupied by the dispersed phase. It is often
used as a measure of the interfacial area ( A) available for phase interaction, which in turn
is a function of the drop size. In addition to serving as an indicator of how concentrated
the two-phase mixture is, it also affects several transport processes, determines the global
residence time of the dispersed phase and strongly influences pressure drop in the system.
Information concerning the distribution of the holdup is necessary to accurately predict
local reactor performance and therefore efficiently design the contactor. As heat, mass
and momentum transfer rates depend strongly on these quantities, the dispersed phase
holdup and its distribution constitute important design parameters in majority of the
chemical processes.

1.2.2 The Need for Numerical Studies


Extensive and often expensive experiments need to be conducted under a wide range of
operating conditions in order to clearly understand all parameters that influence the
holdup and its distribution (especially the most sensitive ones). However, empirical
conclusions derived from the often time-consuming experiments, can suffer from severe
shortcomings such as:
5

• Their limited range of application, which rules out the possibility of any
extrapolated inferences,

• The simplifying assumptions used in their development that render them


inaccurate when applied to flows of practical interest,

• Lack of confidence in their predictions due to inaccuracy of measurements (for


instance, when using intrusive measurement techniques that alter the flow field
significantly),

• The inability to successfully apply the proposed correlations to different or even


scaled-up geometries.

The situation becomes more convoluted when complex/intricate geometries are involved.
A very promising avenue is the use of cost-effective Computational Fluid Dynamics
(CFD) technology to simulate the system in question.

1.3 Overview of Computational Modeling and CFD


Computational fluid dynamics or simply CFD is a tool used to model fluid-flow based on
well-established equations that govern the flow (usually referred to as the Navier-Stokes
equations). The roots of CFD are embedded in the rudimentary ‘Computational
Modeling’ approach.

1.3.1 Computational Modeling


‘Computational Modeling’ is about elaborating real-world problems (e.g. non-optimal
design and thus inefficient operations, scale-up issues etc.) and developing solutions to
these problems using numerical calculations.

Several decades ago, computations were done by hand and modeling was restricted to a
small data set. It therefore addressed a narrow range of conditions and produced a small
set of useful inferences. In the recent past, a new modeling approach was devised that
involved dividing the region of interest into several compartments (referred to as sub-
regions or cells), assuming a simple case of average (uniform) hydrodynamics in each
cell and applying the basic principles of mass, energy and momentum conservation to it.
6

Models were typically based on the assumption that each cell represents a completely
mixed zone and heat/mass/momentum transfer occurs through finite exchange with
neighboring cells. The outputs from one cell were used as inputs to calculate the results
for the cell(s) connected to it and so on. This simplistic approach would finally yield the
overall picture. For instance, Mann and Mavros (1982) were able to successfully predict
stirred tank reactor performance using a ‘Network of zones’ model to represent the flow
generated by an impeller, where the flow through different zones was prescribed using
experimental measurements of velocity fields.

One of the benefits of this method was that local distributions of flow and dispersion
parameters could be easily obtained. On the other hand one had to make a compromise,
in that, the error involved in estimating the parameters of interest was high, primarily
because of the coarse number of cells that were chosen. Also, if the geometry were to be
scaled up, the whole process would have to be carried out again as the model parameters
would have changed significantly.

With the advent of powerful computers, the modeling scenario experienced a major
change. Increasing the number of cells by a factor of 1000 or more was possible, giving
rise to a much finer grid composed of several cells. More ambitious numerical
calculations became possible and Computational Fluid Dynamics (or simply CFD),
started to surface in the modeling world.

1.3.2 Computational Fluid Dynamics (CFD)


Accurate simulation of fluid flows by solving the governing equations (conservation of
mass, momentum, energy etc.) is the primary objective of CFD. Typical steps in this
direction involve, meshing, specifying boundary conditions and finally solving the
discretized governing equations iteratively using a suitable numerical method. The
process of meshing involves dividing the domain of interest into a large number of cells
(which typically range from a few thousand to a million or even more, depending on the
complexity of the problem and available computation power), followed by creation of a
grid for computation. The nature of the grid (for instance the cell density near solid
boundaries) largely depends on the flow being simulated.
7

Initially, CFD codes were quite specific and focused on specific cases rather than a
generic problem. As processing capabilities of computers increased, it became easier to
generalize the codes. This resulted in a feasible complementary alternative to
experimental analysis, which soon gained immense popularity due to the fact that the new
technology (CFD) allowed for detailed analysis of fluid flow problems, faster and earlier
in the design cycle than possible with experiments. Moreover, it costs less money and
involves reduced risk. This trend continues to grow as computers are becoming
increasingly cheaper and more powerful. Listed below are some of the several advantages
that CFD features:

• Low cost: Computational prediction costs many orders of magnitude lower than
most experimental investigations. This factor becomes increasingly important as
the physical situation under study becomes larger and more complex,

• Prompt analysis devoid of any scale-up issues: Many different configurations of


the same equipment (setup) can be investigated in a shorter time frame than that
possible with experiments,

• The ability to simultaneously model heat/mass transfer effects in addition to fluid


flow.

1.4 Problem Statement


There are several known approaches to modeling dispersed two-phase flows using CFD.
Some of the commonly used ones are the two-fluid (Eulerian-Eulerian), discrete phase
(Eulerian-Lagrangian), and interface tracking (Volume of fluid) approaches. A summary
of each approach is provided later in Chapter 2. Amongst the aforementioned approaches,
the two-fluid approach is widely used on account of the adequate level of information
(flow detail) it provides in exchange for a reasonable amount of computation power. The
governing equations for two-fluid models are obtained by appropriate averaging
procedures (such as time-averaging, volume-averaging, ensemble averaging etc.). This is
done in order to restrict the total amount of information handled by the model. A direct
consequence of averaging is that individual flow properties such as velocities and
8

volumes are simplified to phasic quantities (i.e. phase volume fractions instead of
individual entity volumes etc.). This is particularly true for interphase forces, which are
vital to the accurate prediction of the distribution of dispersed phase holdup as they
dictate the motion of the dispersed entity. To this end, there have been several attempts to
arrive at ideal interphase closure relations for turbulent bubbly flows. However, a large
number of these studies are focused on gas-liquid systems (Friberg, 1998; Thakre and
Joshi, 1999; Joshi, 2001; Sokolichin et al., 2004; Deen et al., 2004). It is well known that
concepts and results related to gas-liquid systems cannot be readily applied to liquid-
liquid systems owing to small density differences and high viscosity ratios in the latter
(Jin et al., 2000; Brauner, 2002; Shi et al., 2003). Given the fact that immiscible liquid
dispersions are encountered in a diverse range processes and equipments, it is quite
surprising to note that, the closure requirements for such systems have received relatively
lesser attention. The present study therefore attempts to fill this void by means of an
organized study which clearly distinguishes and quantifies the various significant forces
acting on the dispersed phase in immiscible liquid dispersions.
9

CHAPTER 2
2 USING CFD TO SIMULATE LIQUID-LIQUID
SYSTEMS: PREVIOUS WORK
This chapter presents a detailed report on the various numerical investigations pertinent
to immiscible liquid dispersions with the objective to help provide guidance and gain
insight into the liquid-liquid CFD modeling picture. The following sections review the
various techniques that have been used to simulate turbulent dispersed liquid-liquid flows
using CFD. In particular, the discussion entails the various multiphase and turbulence
modeling approaches used, followed by the handling of interphase forces that are largely
responsible for accurate prediction of dispersed phase distribution, and the treatment of
drop diameter which assumes special importance in situations that feature high phase
fractions.

However, before discussing previous work, a brief review of multiphase and turbulence
modeling approaches and a summary of interphase forces typically encountered in two-
phase dispersions will be presented in order to better understand the discussion pertaining
to the literature review.

2.1 CFD Simulation of Multi-fluid Flows


The subject of modeling multi-fluid flows is quite vast and encompasses an extensive
spectrum of sub-topics. Owing to the relative importance of dispersed multi-fluid flows
in chemical reaction engineering and more importantly, its striking relevance to the case
at hand, this section reviews multi-fluid modeling aspects of only dispersed flows. As
most fluid flows encountered in practice are almost exclusively turbulent in nature, the
present section also discusses various aspects of turbulence modeling in multi-fluid
systems. Special attention is given to the treatment of interphase forces for the reasons
elaborated in Section1.4.

It is however, important to understand the need for modeling flow fields and turbulence
as opposed to directly resolving them. In this regard, an exclusive technique which allows
10

for an exact computation of the instantaneous flow fields and their turbulent fluctuations
without resorting to any kind of modeling is first discussed. The limitations of applying
this technique are highlighted and thus, the need for modeling is illustrated.

2.1.1 Direct Numerical Simulations (DNS)


For any kind of fluid flow, it is theoretically possible to compute an entire three-
dimensional turbulent flow field without resorting to any kind of modeling. This type of
direct solution of the governing equations (referred to as Direct Numerical Simulation) is
limited in its accuracy only by the numerical method employed. This method generates a
lot of information containing the time history of all flow variables at every point in the
domain. DNS therefore features a very profitable way to investigate and understand
fundamentals of fluid flow dynamics and turbulence phenomena and thereby help in the
evaluation and validation/development of existing models.

However, the DNS approach is not feasible for practical engineering problems primarily
because of the prohibitive mesh sizes (proportional to Re (9/4)) that are needed to resolve
all scales of motion in the three spatial dimensions (Fluent, 2004). Also, the specification
of initial and boundary conditions is regarded as one of the most important and difficult
steps in applying DNS (Ranade, 2002). Adding to the complexity is the fact that the
simulation would have to be a transient one with very small time steps. As a result, the
computational power required to resolve dense (i.e. high dispersed phase fraction) and
often turbulent, multi-fluid flows is way beyond the capabilities of even modern
computers.

2.1.2 Dispersed Multi-fluid Flow Modeling


There are four known approaches to modeling dispersed multi-fluid flows (Ranade,
2002):

1. Volume of Fluid (Eulerian framework for both phases with reformulation of the
interface forces on a volumetric basis)

2. Eulerian-Lagrangian (Eulerian framework for the continuous phase and


Lagrangian framework for the dispersed phase)
11

3. Eulerian-Eulerian (Eulerian framework for both phases without explicitly


accounting for the interface between phases)

4. Mixture (An approach that couples the continuous and dispersed phases together
as a single mixture phase)

The above approaches are depicted schematically in Figure.2.1.

VOF Eulerian- Eulerian - Mixture


Lagrangian Eulerian

Continuous Dispersed Mixture


phase phase phase

Figure 2.1 Modeling approaches for multiphase flows


A brief explanation of each approach follows:

2.1.2.1 Volume of Fluid (VOF)


The VOF approach tracks the individual phase motion which helps it trace the interface
indirectly. In this approach, the participating fluids share a single set of conservation
equations with mixture properties whenever a control volume is not entirely occupied by
one phase. When one phase completely occupies the control volume, its corresponding
properties are used. This avoids abrupt changes in the physical properties across a very
thin interface. To ensure this, all interfacial forces are replaced by smoothly varying
volumetric forces. Typically, when the shape and flow processes occurring near the
interface are of interest, the VOF approach is beneficially employed. For instance, the
VOF approach can be effectively used to simulate the shape deformation of dispersed
fluid particles (e.g. bubbles) due to the surrounding liquid flow (Delnoij, 1999). In
systems that feature large dispersed phase fractions, enormous computational resources
12

are required to resolve the flow field around each dispersed entity. Thus, although this
approach is conceptually the simplest when compared to the Eulerian-Eulerian or the
Eulerian-Lagrangian approach (discussed in the following sections), it is best suited to
application in situations that feature simple problems such as sloshing of fluid in a tank
(Hyung, 2004), flow around single entities (i.e. particles, drops or bubbles) and dispersed
multi-fluid flows that feature very low phase fractions of the dispersed phase (Stover et
al., 1997; Delnoij, 1997; Krishna and van Baten, 1999).

2.1.2.2 Eulerian-Lagrangian (E-L)


In the Eulerian-Lagrangian approach, the explicit motion of the interface is not modeled.
Thus small-scale fluid motions around individual dispersed entities are neglected. Their
influence is modeled indirectly while considering the motion (trajectories) of the
dispersed entities which are obtained by solving an equation of motion for each dispersed
phase entity. Averaging over a large number of trajectories is done to derive the required
information to model the continuous phase using an Eulerian approach. This approach is
capable of accurately capturing dispersed phase dynamics. Thus particle-level processes
(e.g. reactions, heat/mass transfer effects etc.) can be simulated in adequate detail. Also,
size distributions and inter-particle interactions can be easily accounted for in this
framework. However, in turbulent flows it becomes necessary to simulate a very large
number of particle trajectories in order to obtain meaningful averages. Consequently, as
the number of dispersed entities increases, the computational cost increases at an
alarming rate, again restricting this approach to the simulation of flows featuring low
(less than 10%) volume fractions of the dispersed phase (Domgin et al., 1997; Jaworski
and Pianko-Oprych, 2002).

2.1.2.3 Eulerian-Eulerian (E-E)


The Eulerian-Eulerian approach models the flow of all phases using an Eulerian
framework, and is therefore based on the assumption of interpenetrating continua. This
means that all phases share the domain and may interpenetrate as they move through it.
Each phase is characterized by distinct fields of velocity and volume fraction. In this
approach, appropriate interphase transport models for drag, lift, virtual mass etc. need to
13

be specified in order to effectively handle the coupling between the phases (closure). The
Eulerian-Eulerian approach is more efficient in terms of CPU time owing to the
continuum approach for the dispersed phase. In contrast to the VOF and the Eulerian-
Lagrangian approach, the Eulerian-Eulerian approach is well suited to modeling systems
which feature large volume fractions of the dispersed phase, such as bubble-column
reactors, mechanically agitated tanks etc. Although handling complex phenomena at the
particle level, such as dynamic size distributions of the dispersed phase is relatively
difficult when compared to the more accommodating Eulerian-Lagrangian method, the
aforementioned advantages make it a potential choice for modeling practically
encountered dense multi-fluid flows whose treatment using an Eulerian-Lagrangian
approach is difficult, due to the prohibitive computational resource requirements.

2.1.2.4 Mixture
The mixture approach is a simplified multiphase model that can be used to simulate
multi-fluid flows where the phases may move at different velocities, but assume local
equilibrium over short spatial length scales (Fluent, 2004). This implies that the
accelerating dispersed phase entities attain the terminal velocity after traveling a distance
much smaller than the length scale of the system (Chen et al., 2005). The assumption of
local equilibrium is necessary in order to prescribe an algebraic equation for the relative
velocity between the phases (Hossain et al., 2003, Chen et al., 2005). Unlike the VOF
approach, the multi-fluid mixture model allows the two phases to be interpenetrating.
Each dispersed phase is represented by a species concentration (volume fraction)
equation. The transport equation for each dispersed phase allows for relative movement
(slip) between the dispersed and continuous phase. Conceptually, the mixture approach
treats the combination of continuous and dispersed phases as a single entity. This is
referred to as a mixture phase, and correspondingly its physical properties (i.e. the
mixture density and mixture viscosity), which are calculated on the basis of individual
phase fractions are used in the governing equation. A prerequisite is that the coupling
between the two phases should be strong. The mixture model can also be used to simulate
homogeneous multi-fluid flows with very strong coupling where the phases move
14

approximately at the same velocity (i.e. negligible slip). This model is computationally
least expensive when compared to the other models discussed so far, as it solves the least
number of equations. However, there is a strong compromise in the level of flow detail it
can provide when compared to the comprehensive Eulerian-Eulerian approach.
Moreover, the mixture framework does not conceptually allow accounting for non-drag
interphase forces such as the lift and virtual mass forces. Typical applications of the
mixture model include sedimentation, particle-laden flows with low loading, and bubbly
flows where the dispersed phase volume fraction remains low.

2.1.3 Treatment of Turbulence in Multi-fluid Flows


Multiphase dispersions encountered in CPI are inherently turbulent in nature and are
therefore characterized by fluctuating velocity fields. These fluctuations mix transported
quantities such as momentum, energy, and species concentration, and cause them to
fluctuate as well. Since these fluctuations can be of small scale and high frequency, they
are computationally very expensive to simulate directly using DNS as discussed in
Section 2.1.1. Instead, the instantaneous governing equations can be time-averaged,
ensemble-averaged, or otherwise manipulated to remove the small scales (LES), resulting
in a modified set of equations that are computationally less expensive to solve. However,
the modified equations contain additional unknown variables, and turbulence models are
needed to determine these variables in terms of known quantities (closure).

2.1.3.1 Large Eddy Simulations (LES)


Large Eddy Simulations (LES) are based on the hypothesis that the relevant scales in
turbulent flows can be split into large-scale and small-scale (also referred to as sub-grid)
components as shown in Figure 2.2. In this approach, it is implicitly assumed that such
separation does not significantly affect the evolution of large-scale turbulent motions.
Large-scale motions are typically much more energetic than the small-scale motions and
thus contribute more to the transport of conserved quantities. Therefore, LES attempts to
simulate these large-scale motions much more precisely than small-scale motions, which
are believed to be more universal in character and hence more easily modeled.
15

Large scales

Small scales

Figure 2.2 Schematic representation of scales in turbulent flows (adapted from Ferziger
and Peric, 1995)

Consequently, mesh resolution requirements and time-step sizes are less restrictive when
compared to the DNS approach. Despite these advantages, LES still suffers from some of
the shortcomings of DNS such as difficulties in specifying boundary conditions and
generating a huge amount of information which may not be useful for practical purposes
(Bakker and Oshinowo, 2004).

The limited ability (due to computational constraints) of applying the DNS or LES
approaches to flows of practical interest in CPI coerces the use of computationally more
tractable turbulence models based on the Reynolds Averaged Navier Stokes (RANS)
approach (Prasad et al., 1998; Bakker and Oshinowo, 2004), which is discussed in the
following section. It should be mentioned that recent advances in modeling have led to a
hybrid approach that combines RANS modeling with LES, known as Detached Eddy
Simulation (DES). Essentially, DES reduces to RANS in regions close to the walls and
switches over to LES in regions away from the walls (Constantinescu and Squires, 2003).
This reduces the computational effort significantly while providing more accurate flow
detail when compared to using RANS throughout.

2.1.3.2 Turbulence Models based on RANS


In the RANS-based approach, the instantaneous value of any flow variable ( ϕ ) is
decomposed into a mean ( ϕ ) and a fluctuating component ( ϕ ′ ):
ϕ = ϕ + ϕ′ [2.1]
Instantane ous value Time Averaged mean Fluctuatin g component
16

The mean value can be determined by averaging over an appropriate time interval.
Reynolds averaging obeys the following properties:
ϕ = ϕ and ϕ ′ = 0 [2.2]
where the overbar denotes time averaging. Equation 2.1 is substituted in the basic
governing equations for the flow variable ( ϕ ) (e.g. velocity of a phase) followed by time
averaging, subject to the conditions listed in Equation 2.2 in order to yield governing
equations for the mean quantities. When simplified, the new averaged equation features
an extra term which accounts for the turbulent transport of ‘ ϕ ’. As the variation of time-
averaged quantities occurs at much larger scales, resolving the smaller spatial and
temporal scales is unnecessary in the RANS-based approach.
Figure 2.3 describes the variation of fluid velocity with time and shows how the RANS
based approach compares with the DNS and LES approaches.

Figure 2.3 Quick comparison of DNS, LES and RANS (Ranade, 2002)

As a result of averaging, the RANS approach requires far less computational power
compared to the LES or DNS methods. Unfortunately however, time-averaging the basic
governing equations gives rise to new terms which leads to a closure problem similar to
the one discussed earlier in Section 1.4. These new terms may be interpreted as apparent
stress gradients and heat/mass fluxes associated with turbulent motion (Ranade, 2002). It
is possible theoretically, to derive governing equations for these new terms. However, the
resulting equations would also contain more unknown terms. It therefore becomes
17

imperative to introduce a turbulence model, which relates these unknowns to known ones
in order to close the set of governing equations.

Figure 2.4 Simulation and modeling approaches for turbulent flows (Adapted from
Ranade, 2002)

Over the last three decades, several turbulence models have been proposed and used in
simulations with varying degrees of success. A detailed description highlighting the
advantages, limitations and the range of applicability of each is beyond the scope of this
thesis. Instead, for the sake of completeness a very brief summary of the most widely-
used ones is provided in graphical form (Figure 2.4).
18

2.1.4 Treatment of Interphase Forces in Dispersed Multi-fluid flow


It is well known that in dispersed multiphase flows, interphase forces play a significant
role in determining the local phase distribution (Jakobsen et al., 1997; Friberg, 1998;
Deen, 2001; Joshi, 2001; Ranade, 2002; Worner, 2003; Chen et al., 2005). Listed below,
are some of the typical interphase forces encountered in multi-fluid dispersions:

1. Gravitational and Buoyancy forces (arises from a difference in the density of the
two phases),

2. Drag force (Prime force that comes into play whenever there is a relative motion
between the phases),

3. Lift force (Force which acts in a direction normal to the relative velocity between
the phases),

4. Virtual/Added mass force (Significant force when dispersed entities


accelerate/decelerate),

5. Basset force (Force associated with past movements of a dispersed entity),

6. Wall lubrication force (A near-wall force that drives dispersed entities normally
away from the wall),

7. Turbulent dispersion (Accounts for the spread of the dispersed phase entities
under the action of turbulent fluctuations in the continuous phase).

A brief explanation of each force follows:

2.1.4.1 Drag Force


In dispersed multiphase systems, the force that opposes the relative velocity between the
phases is called the drag force. The nature of this force is clearly illustrated in Figure 2.5.
The Total drag force on the entity is made up of two components, namely the Form drag
and the Skin drag. While Skin drag results due to the skin shear component in the
direction opposing the relative velocity, Form drag is the force due to the fluid pressure
acting on the dispersed entity. The Total drag is the sum of integrals of both these
quantities, each evaluated over the entire surface of the body in contact with the fluid.
19

Variations in shape and volume of an entity are characterized by a dimensionless


parameter referred to as the Drag coefficient. The importance of this force and its
treatment in bubbly flows is discussed later in Section 5.3.1.

Drag force
C
Dispersed C − A Direction of relative velocity
entity

A
A C

Fluid velocity vectors CASE I: When the entity moves slower than the fluid

C
Drag force

C − A Direction of relative velocity

Dispersed
entity
A
A C

Fluid velocity vectors CASE II: When the entity moves faster than the fluid

Figure 2.5 Nature of drag forces in multiphase systems

2.1.4.2 Lift Force


When a dispersed phase entity moves through a non-uniform flow field, it will experience
a lift force due to the vorticity or shear in the continuous phase field. This is illustrated in
Figure 2.6.
20

Dispersed phase velocity

Lift Force Lift Force

Continuous phase velocity profile

Figure 2.6 Lift force acting on a dispersed entity

The lift force acts on the dispersed entities in a direction perpendicular to the relative
motion between the two phases. Phase distribution in bubbly flows is severely affected by
this force. Several studies have shown that lift forces are not negligible and tend to
increase with increasing entity diameter (Dandy, 1993; Tomiyama, 1998; Moraga et al.,
1999). Drew and Lahey (1987) proposed an expression for the lift force which depends
on the relative velocity between the phases and the continuous phase velocity gradients
(Equation 2.3).

Flift = − C L ρ q α p (v p − v q )× (∇ × v q ) [2.3]

Analogous to the Drag coefficient, the dimensionless constant, CL in Equation 2.3 is


referred to as the Lift coefficient.

Peirano (1998) summarized several studies on the lift force and found that most
investigations were restricted to small Reynolds numbers. Recently however, there have
been some investigations that explore the nature of lift forces at high Reynolds numbers
(Moraga et al., 1999; Kulkarni and Joshi, 2004). The nature and treatment of this force in
turbulent bubbly flows is discussed elaborately in Section 5.3.2.1.

2.1.4.3 Virtual/Added Mass Force


When a dispersed entity accelerates relative to the continuous phase, a part of the
continuous phase which surrounds the entity is also accelerated. The acceleration of this
extra continuous phase has the effect of added inertia. In essence, the virtual mass force is
the force required to accelerate the apparent mass of the continuous phase in the
21

immediate vicinity of the dispersed phase. It therefore has the effect of dampening the
natural tendency of the particle to accelerate in any direction. This force can be important
near dispersed phase inlets and outlets (where large acceleration de-acceleration
conditions are encountered) and plays a particularly significant role when the density of
the dispersed phase is much smaller than that of the continuous phase (e.g. in the case of
bubbles).

2.1.4.4 Basset Force


Also known as the History term, this force is associated with past movements of the
particle. As it accounts for the deviation in flow pattern from steady state, it assumes
relevance only for unsteady flows, and in most cases its magnitude is much less than the
drag force. Including this term increases the instantaneous flow resistance when a particle
is being accelerated at a high rate by an external force. This effect may therefore be
important in creeping or laminar flows or in cases where the fluid density is almost equal
to the dispersed phase density. It involves a history integral which is time-consuming to
evaluate. This force decays as t-n with n > 2 (Mei, 1993) for intermediate time and is
therefore neglected in most turbulent flows (Ranade, 2002).

2.1.4.5 Wall Lubrication Force


When a rising dispersed phase entity comes in close proximity to a wall, the normal
uniform continuous fluid drainage around it changes dramatically. The presence of the
wall slows the fluid drainage rate between the entity and the wall. This leads to an
increased drainage rate on the opposite side. The net effect is a force which tries to push
the bubble away from the wall (Friberg, 1998). A physical argument for the wall
lubrication force is that the centroid of a spherical dispersed entity of a given diameter
cannot come any closer to the wall than half the entity's diameter (Wörner, 2003). This
force is non-zero only very close to the wall and purportedly counterbalances opposing
forces that act in a direction normal to the wall (for instance, the lift force in bubbly
upflows in vertical pipes), so that a physically reasonable decrease of the void fraction in
close vicinity of the wall is ensured.
22

2.1.4.6 Turbulent Dispersion


This pseudo-force induces a diffusive flux that accounts for dispersion (or spread) of
dispersed phase entities due to the random influence of the turbulent eddies present in the
continuous phase. It is strongly influenced by the fluctuating continuous phase turbulence
field. This force also counteracts the lateral lift force (e.g. in turbulent bubbly pipe flows)
in an attempt to homogenize the dispersed phase distribution. A comprehensive account
of this pseudo-force is presented later in Section 5.3.2.2.

2.2 Discussion of previous CFD Studies in Liquid-Liquid Systems

CFD INVESTIGATIONS
PERTINENT TO
IMMISCIBLE LIQUID
DISPERSIONS

I. Rotating Disc Contactor II. Stirred / Mechanically


(RDC): Agitated Tank:

1. Rieger (1994) 1. Koh and Xantidis (1999)


2. Rieger et al. (1996) 2. Alopaeus et al. (1999)
3. Modes and Bart (2001) 3. Alopaeus et al. (2002)
4. Vikhansky and Kraft 4. Baldyga and Podgorska
(2004) (2002)
5. Alexopoulos et al. (2002)
6. Agterof et al. (2003)

III. Static Mixer: IV. Jet Homogenizer: V. Vertical Pipe:

1. Jaworski and Pianko- 1. Soon et al. (2001) 1. Domgin et al. (1997)


Oprych (2002) 2. Hamad et al. (1999)

Figure 2.7 CFD studies on various liquid-liquid contactors


23

A critical review of previous work reveals that CFD investigations relevant to immiscible
liquid dispersions in general, are quite limited in number (Figure 2.7). As a result, this
section will review all previous investigations irrespective of contactor geometry. In the
ensuing paragraphs the CFD modeling approaches used for each contactor will be
summarized followed by a discussion of the relevant findings. The terms drop and
dispersed phase are interchangeably used throughout the following discussions.

2.2.1 Rotating Disc Contactor (RDC) Columns


The Rotating Disc Contactor (Figure 2.8) has been extensively used in several industrial
liquid-liquid extraction operations since the early fifties. It finds application in areas such
as purification of heat sensitive materials, furfural extraction of lubricating oils,
desulphurization of gasoline, recovery of phenol from wastewaters etc. The system as
such is complex both in construction and in hydrodynamics.

Figure 2.8 RDC Schematic (Koch Modular Process Systems, 2000)


A typical setup features horizontal doughnut-shaped or annular baffles (commonly
referred to as stators) that split the tower into virtual compartments. A centrally located
24

smooth and flat horizontal disc (rotor) within each compartment provides the agitation. A
variable speed motor drives this disc. The light liquid enters the column at the bottom and
the heavy liquid is fed from the top in order to achieve counter-current operation (Figure
2.8).

In the past, the design of such extraction processes required a highly tedious empirical
approach and the scale-up of laboratory results often failed due to lack of geometric and
hydrodynamic similarity (Rieger, 1994). As a result, it was unfeasible to design industrial
size units effectively in the absence of reliable pilot-scale experimental data.

The widespread progress of multiphase models over the last decade has resulted in the
use of CFD to simulate the complex hydrodynamics and mass transfer characteristics of
typical industrial contacting units like the RDC. Ironically, recent CFD studies (Fei et al.,
2000; Wang et al., 2002) appear to focus on single-phase velocity field predictions in the
RDC as opposed to two-phase investigations carried out in the past decade. Four different
CFD investigations associated with the RDC were found in the open literature. These
have been reviewed below in detail.

Early CFD studies appear to have focused on predicting velocity fields, local volumetric
distribution of the dispersed phase fraction and the analysis of turbulence-related
dispersion processes (Rieger et al., 1994; Rieger et al., 1996). More recent studies
however, concerted their efforts towards predicting the residence time distributions of the
dispersed phase (Modes and Bart, 2001) and the use of drop population balance modeling
within the CFD framework (Vikhansky and Kraft, 2004). A discussion of the multiphase
and turbulence CFD models typically used and other important details (such as the
treatment of inter-phase forces and dispersed phase diameter etc.) follows.

2.2.1.1 Multiphase and Turbulence Modeling Approach


Two multiphase CFD approaches were commonly used to model RDC hydrodynamics,
namely the Eulerian-Lagrangian and the Eulerian-Eulerian approach. In spite of the
complications associated with modeling the turbulent interactions between the phases,
most studies preferred the latter approach owing to the fact that it manifests as a logical
25

extension of single-phase hydrodynamics and its ease of applicability to situations which


feature high-dispersed phase loading (Rieger et al., 1996). It should however be
mentioned that Vikhansky and Kraft (2004) proposed and used an extension of the
Eulerian-Lagrangian approach in order to apply it to cases where the volume fraction of
the dispersed phase is high. Also, the Eulerian-Lagrangian approach showed encouraging
results when comparing experimental and simulated residence time distributions (RTD)
(Modes and Bart, 2001).

All investigators used the standard k-ε turbulence model in their simulations, but it is
only some of them that accounted for the turbulence interaction between the phases. In
particular, Rieger (1994) modified the standard k-ε turbulence model to account for the
effect of the dispersed phase on the continuous phase turbulence (Simonin and Viollet,
1990b). Vikhansky and Kraft (2004) considered turbulence generation due to the
presence of the dispersed phase in addition to accounting for the effect of the drops on the
flow field, as described by Laín et al. (2002). On the other hand, Modes and Bart (2001),
who studied single drop motion in the RDC did not mention whether interphase
turbulence interactions were taken into account.

2.2.1.2 Treatment of Drop Diameter


All studies, except the one by Vikhansky and Kraft (2004) used a constant uniform drop
diameter throughout their CFD simulations. This simplification was considered justified
as being adequate for the purpose of a preliminary study by Rieger (1994). As the work
of Modes and Bart was based on single-droplet studies to begin with, they also used a
constant drop diameter throughout their simulations. On the other hand, Vikhansky and
Kraft (2004) considered size distributions of the dispersed phase in conjunction with
treatment of drop coalescence and breakage using a single-particle Monte-Carlo
stochastic simulation method as described in Sommerfeld (2001) and Ho and
Sommerfeld (2003).
26

2.2.1.3 Interphase Forces


With the exception of Modes and Bart (2001), who considered drag on single drops using
expressions proposed by Hu and Kintner (1955), Grace et al. (1976) etc., all investigators
used the standard Drag coefficient expression for a single rigid sphere as specified in
Clift et al. (1978).

Only Modes and Bart (2001) considered the effect of lift (coriolis) forces in their CFD
simulations, although there was no mention of the value of the lift coefficient or the
formulation/expression used.

In two of the studies (Rieger, 1994; Rieger et al., 1996), pressure gradient, virtual mass
and basset forces were ignored in view of the fact that relative velocities are low in
liquid-liquid systems as opposed to gas-liquid systems. In contrast, Modes and Bart
(2001) and Vikhansky and Kraft (2004) considered the effect of virtual mass on the bulk
flow in their CFD simulations. However, no justification for the need to include this
additional effect was given.

It should be noted that although the buoyancy force in RDCs is not expected to be very
significant when the agitation provided by the rotor can give rise to comparatively
stronger centrifugal forces (Vikhansky and Kraft, 2004), it is still the force responsible
for the overall counter-current motion between the phases.

2.2.2 Stirred/Mechanically Agitated Tanks


The stirred tank is the most commonly used mixing equipment for chemical processes. It
finds widespread use in crystallization, suspension/emulsion polymerization, extraction
and emulsification processes etc. Typically, it consists of a baffled (to reduce tangential
motion (i.e.) prevent the formation of a vortex) vertical cylindrical vessel, agitated using
an impeller (e.g. open straight blade, vaned disk, vertical curved blade etc.) mounted on a
shaft that is driven by a variable speed motor such as the one shown in Figure 2.9. The
tank bottom is usually rounded to eliminate sharp corners or regions into which fluid
currents cannot penetrate. This also reduces power consumption when compared to a flat
bottom. It features rather complex hydrodynamics, especially in the highly turbulent
27

region near the impeller which is under constant shear and where maximum turbulent
energy dissipation rates are observed.

Figure 2.9 Schematic representation of a stirred/mechanically agitated tank

The function of the impeller is to enhance mixing by creating a flow pattern in the vessel
(e.g. axial, radial) causing the fluid(s) to circulate through the vessel and return to the
impeller eventually. Stirred tanks form the core of chemicals synthesis in process
industries and their optimal design is therefore considered to be very important. It is
therefore not very surprising that the number of CFD studies is the highest in stirred tank
reactors.

Conventional stirred tank design and optimization involves extensive pilot-scale tests.
Such empirical tests are useful mostly for obtaining approximations to general bulk
behaviour such as impeller power consumption/discharge flow, mixing/circulation time
etc. However, these methods provide little or no information about flow patterns, mixing
mechanisms, or homogeneity of mixing, all of which may be crucial to the success of the
mixing process at hand. Although experimental approaches to discern the flow behaviour
in mixing vessels are quite useful, it should be noted that the combination of certain
inevitable factors such as the unsteady turbulent flow, intricate impeller geometry and
complex hydrodynamics etc. render quantitative measurements and flow visualization
both expensive and time-consuming (Alexopoulos et al., 2002). It is here that valuable
28

CFD predictions, which provide a wealth of information (local velocity fields, turbulent
kinetic energy/dissipation rates, dispersed phase distribution etc.), can serve to reveal
actual flow behaviour, circulation patterns, vortex structures etc. which in turn would
yield an improved insight into the complex nature of turbulent mixing.

Turbulent single-phase flow in agitated vessels has been the subject of several numerical
investigations to date. In particular, the application of single-phase CFD simulations to
stirred tanks has already been well examined by several authors (Placek et al., 1986;
Kresta and Wood, 1991; Harris et al., 1996; Vivaldo-Lima et al., 1997; Derksen and Van
der Akker, 1999; Deen et al., 2002; Alexopoulos et al., 2002). As already mentioned, a
key issue in many industrial processes involving the use of agitated tanks is the
extent/quality of mixing. Traditionally, most mixing applications reported in the literature
tended to treat the entire stirred vessel as a field of homogeneous isotropic turbulence
which is an oversimplification, especially when one compares the region close to the
impeller with the bulk. Several researchers have clearly identified (and in most cases,
later verified using single-phase CFD simulations) that turbulence characteristics (e.g.
energy dissipation rate) are highly anisotropic and inhomogeneous near the impeller
stream and can typically feature 2-3 orders of magnitude difference when compared to
the bulk (Okamoto et al., 1981; Placek et al., 1986; Calabrese and Stoots, 1989; Baldyga
et al., 1995; Alopaeus et al., 1999). As a result, single-phase simulations have not only
been found to yield helpful information concerning qualitative flow behaviour, but also
serve as useful starting points for the more complicated two-phase simulations.

Although comprehensive CFD simulations of two-phase flows in stirred tanks are in the
early stages, it is encouraging to note that there have been several recent studies (Koh and
Xantidis, 1999; Alopaeus et al., 1999; Alopaeus et al., 2002; Podgórska and Bałdyga,
2002; Alexopoulos et al., 2002; Agterof et al., 2003) that have addressed turbulent liquid-
liquid systems.

Koh and Xantidis (1999) simulated a critical stage in resin manufacture by suspension
polymerization that involved a liquid-liquid dispersion of a heavy organic phase within a
light aqueous phase. Their objective was to obtain the velocity fields, dispersed organic
29

phase holdup and the turbulent energy dissipation rates in three different reactors (300
ml, 200 litre, and 4 cubic meters) and to use the information so obtained, to ultimately
design the 4 cubic meter reactor. This was done by developing a method for correlating
the experimental particle sizes against maximum turbulent dissipation rates predicted
using CFD.

Other studies have focused more on compartmental approaches that typically transform a
continuously varying quantity into a quantity that is constant within a series of ideal
compartments or zones, thus simplifying the problem significantly (Alexopoulos et al.,
2002). In other words, the compartment-type approaches split the stirred tank into a
network of well mixed zones in order to better describe the turbulence inhomogenity of
the liquid-liquid dispersion.

For instance, Alopaeus et al. (1999) developed a multi-block model to simulate drop
populations in a stirred tank. Their overall objective was to determine the drop size
distribution and from it, the effective mass transfer area. As a complete CFD model with
population balances calculated in each cell was considered too complicated for a full
simulation, the results of the complicated models were averaged, and a simplified flow
model comprising a limited number of control volumes was used. Consequently, the tank
was divided into 11 different sub-regions thus giving rise to the multi-block model.
Differences observed in their simulation results of drop populations between single and
multi-block models led to the conclusion that drop rate parameters should be fitted with
the more realistic multi-block model. The multi-block model also showed that the energy
dissipation rate is non-homogeneous and varies spatially in the stirred tank and cannot be
simply averaged when used in optimizing the contactor design for efficient operation.
Hence, this model could predict inhomogeneity of dispersions in addition to some scale-
up phenomena.

In a later study, Alopaeus et al. (2002) tested the previously developed multi-block model
in a parameter fitting procedure where the drop breakage and coalescence parameters
were fitted against drop size measurements from dense and fully turbulent liquid-liquid
dispersions. Local flow field and turbulence data obtained using CFD were used to
30

provide inputs to the multi-block model. The fitting procedure revealed that the multi-
block model is a useful tool when used along with population balances. Further division
of the multi-block model was suggested as a method of obtaining more accurate
simulation results. An interesting observation from their results was that, measuring and
fitting the parameters with local drop size distributions described the dispersion in the
vessel more accurately when compared to using average drop sizes.

In yet another study, Podgórska and Bałdyga (2002) discussed the effects of fine and
large-scale inhomogeneity of turbulence on the drop size distribution in stirred tanks.
Large-scale distributions of turbulence properties were considered using a network of
well mixed zones. CFD was used to compute turbulence quantities in these zones. Once
again, simulation results revealed that the turbulent energy dissipation rate indeed has a
non-uniform character in the stirred tank. A comparison of the 10-cell model predictions
with experimental data of Konno et al. (1988) for both short and long agitation times
revealed good agreement between the simulation and measured values. Their model
which took into account both types (fine and large scale) of inhomogeneities was also
able to explain the effect of scale of the system on the drop size.

On the basis of a thorough review of previous compartment-based models (in most of


which, compartment volumes were arbitrarily defined and compartment parameters
assumed to remain constant with changes in the agitation rate and/or mixing
vessel/impeller design), Alexopoulos et al. (2002) suggested that the simplest possible
representation of a non-homogeneous turbulent stirred vessel was to consider just two
homogeneous compartments - a small region around the impeller characterized by large
energy dissipation rates and a larger circulation region far away from the impeller where
the turbulent flow field is nearly homogeneous. In their work, CFD was used to generate
spatial distributions of energy dissipation rates within an un-baffled mixing vessel
agitated by a flat two-blade impeller. A general methodology was developed for
extracting from the CFD simulations, three model parameters, namely the volume ratio of
the impeller over the circulation zone, the ratio of average turbulent energy dissipation
rates in the two zones and the exchange flow rate between the two compartments. To
31

assess the effectiveness of the two-compartment model in predicting the effect of non-
homogeneities, the proposed model was used to predict drop size distribution in dense
liquid-liquid dispersions (Magioris et al., 2000). A comparison of the measured and
predicted drop size distribution revealed that the agreement was quite satisfactory.

A very recent study by Ageterof et al. (2003), who developed an approach to predict the
drop size evolution of an oil-in-water emulsion, used single-phase CFD simulations to
resolve the velocity field for later use in the convective terms of a transport equation (for
the moments of drop size distribution). Their method allowed for drop size prediction in
any equipment. The evolution of drop size distribution as a function of time for two
rotational speeds of the stirrer was then studied. A fairly good agreement between the
simulations and the experimental data was reported. Even though their method was
simple, it must be noted that it cannot be used when the volume fraction of the dispersed
phase is not negligible. This is because in their method, it is assumed that there is no
effect whatsoever on the continuous phase flow quantities due to the presence of the
dispersed phase.

2.2.2.1 Multiphase and Turbulence Modeling Approach


Across the board, it is seen that most of the recent work on stirred tanks (Alopeaus et al.,
1999; Podgórska and Bałdyga, 2002; Alexopoulos et al., 2002; Agterof et al., 2003) have
used single-phase CFD predictions to obtain velocity fields and turbulence characteristics
such as the energy dissipation rate. The ultimate objective in these studies was to use the
energy dissipation rates from CFD simulations to account for drop breakage and
coalescence effects and hence predict the drop size distribution and/or its evolution with
time. However, in the work of Koh and Xantidis (1999), the two-fluid Eulerian-Eulerian
approach was used to determine the distribution of the organic phase in the reactor at
varying stirring speeds. In contrast, Alopeaus et al. (2002) used the Algebraic Slip
Mixture Model (ASMM) throughout their simulations.

Once again, from a review of previous liquid-liquid CFD studies in stirred tanks it is seen
that the standard k-ε turbulence model has been the universal choice throughout, with the
exception of Agterof et al. (2003) who do not mention which turbulence model was used.
32

The choice of the k-ε model was justified by Alopeaus et al. (2002) on the basis of
previous work reported on baffled vessels (Majander and Manninen, 1996 and 1998). On
the other hand, Podgórska and Bałdyga (2002) tested the Reynolds Stress Model (RSM)
in their simulations and reported that it severely under-predicted the energy dissipation
rate in the impeller swept zone when compared to the standard k-ε model. As a result,
they used the k-ε model in subsequent calculations. Although the standard k-ε model
suffers from inherent shortcomings, notably in describing curvilinear and highly swirling
flows, Alexopoulos et al. (2002) argued that the numerical robustness and
computationally less-expensive nature of the model make it a reasonable choice given
their observation that even the more advanced RSM model produced only marginally
better results. It should be noted that Koh and Xantidis (1999) who also used the standard
k-ε model took into account the effect of turbulent eddies, in order to facilitate the spread
of the dispersed phase by setting the turbulent prandtl number to unity.

2.2.2.2 Treatment of Drop Diameter


As previously mentioned in Section 3.2.1, most of the studies were focused on
accounting for dispersed phase breakage and coalescence using compartment-type
models outside the CFD framework, and hence used single-phase CFD simulations to
solve for the flow and turbulence fields in the stirred tank. On the other hand, the
investigations of Koh and Xantidis (1999) and Alopeaus et al. (2002) can be considered a
complete application of CFD to two-phase systems. While the former used a constant
uniform size for the dispersed phase all through their simulations, the latter used twenty
drop size groups in a population balance model which was implemented within the
framework of CFD. The slip velocity however, was neglected in the simulations of
Alopeaus et al. (2002), as the drops were very small and the densities of the two liquids
were quite close to each other.

2.2.2.3 Interphase Forces


While Koh and Xantidis (1999) considered the effect of buoyancy forces in their
simulations, Alopeaus et al. (2002), again neglected this effect due to the small density
difference between the phases, and the small drop sizes. Koh and Xantidis (1999) used
33

the standard rigid sphere expression to account for interphase drag in their simulations.
Alopeaus et al. (2002) used the standard rigid sphere drag expression in some of their
preliminary simulations and several expressions for the drag on dispersed fluid particles
in the rest of their simulations (the exact correlations used or the differences observed
when compared to rigid spheres were unfortunately not discussed). There was also no
mention of any other non-drag interphase forces (such as the lift force) being considered
in any of the CFD studies pertinent to stirred tanks.

2.2.3 Static Mixers


Static mixers are devices which consist of a series of metal inserts placed inside a pipe.
Typically, as the name suggests, they do not have any moving parts. One of the main
types is the helical-element mixer also referred to as the kenics static mixer.

Figure 2.10 Concentration distribution contours in a static mixer (LaRoche, 1998)

The kenics static mixer is widely used in food and chemical industries for in-line
blending of liquids, usually under laminar conditions. Although these mixers have been
in use for over 30 years, experiments to probe the interior are still very difficult to
perform (Bakker, 2001). In order to improve the efficiency of mixing, CFD can be used
as a tool to effectively design this device.

The kenics mixer consists of a number of left and right-hand helical elements. Each
element, 1 to 1.5 pipe diameters in length, divides the incoming stream into two halves,
following which it gives the stream a 180° twist and delivers it to the next element, which
is set at a 90° angle to the trailing edge of the preceding element. This element again
34

divides this stream and twists it by 180° in the opposite direction. In order to visualize
this procedure, consider Figure 2.10, where the fluid enters from the bottom left hand
side. The helical mixing elements direct the flow of material radially toward the pipe
walls and back to the center. With reference to Figure 2.10 which shows the
concentration distribution on the surface of a kenics static mixer, it is seen that a blue
tracer injected in the center is almost fully mixed with the fluid after six elements.

In addition to the absence of any moving parts, the static mixer features several other
advantages such as lower capital cost, high interfacial area density (which largely
depends on type of element used), short residence times, minimal space requirements etc.
It is therefore quite surprising to note that over the last decade only one CFD
investigation (Jaworski and Pianko-Oprych, 2002) pertinent to liquid-liquid systems has
been reported in the open literature.

Jaworski and Pianko-Oprych (2002) used CFD to model two-phase liquid-liquid flows
through a kenics static mixer. The Algebraic Slip Mixture Model (ASMM) and the
Eulerian-Lagrangian approaches were individually studied to determine their advantages
and drawbacks. CFD predictions were validated by comparing the variation of pressure
drop across one mixer insert with the Reynolds number, using the experimental data of
Aylianawati et al. (2002). A good agreement between the predictions and experimental
data was reported.

2.2.3.1 Multiphase and Turbulence Modeling Approach


As already mentioned, Jaworski and Pianko-Oprych (2002) used the ASMM and the
Eulerian-Lagrangian approaches in their numerical investigation. The ASMM was used
to predict the pressure drop across the inserts, local velocities and the volume fraction of
both phases. On the other hand, the Lagrangian approach was used to track the trajectory
of dispersed fluid elements (drops) in the static mixer. The particle history was then
analyzed in terms of the residence time in the mixer. As the flow was laminar in all the
simulations, turbulence modeling was not necessary.
35

2.2.3.2 Treatment of Drop Diameter


A constant uniform drop diameter as computed from an expression proposed by Chen
and Libby (1978) for kenics static mixers was used throughout the simulations.

2.2.3.3 Interphase Forces


Unfortunately, there was no mention of the drag law used in their simulations. Also
missing, was any mention of whether any non-drag forces were considered in their work.

2.2.4 Jet Homogenizer


The jet homogenizer (Figure 2.11) is a device used to make ultra fine emulsions which
find extensive use in the dairy industry. The sequence of operation consists of a
downward stroke during which a crude dispersion is drawn into the piston chamber.
Next, during an upward stroke, the dispersion is forced through a small orifice producing
a very high velocity jet that travels through a pipe of varying cross sectional area before
impinging on a target. The disrupted material is then cooled by contact with the walls of
the disruption chamber which are held at a low temperature by a recirculating coolant.

Figure 2.11 Key components of a Jet Homogenizer (Soon et al., 2001)


36

It is well known that the stability of immiscible liquid dispersions and emulsions is
affected by the drop size distribution (Chatzi and Lee, 1987; Boye et al., 1996; Lo et al.,
1998). Prediction of drop size distribution requires quantitative information about the
fluid energy dissipation rate in the vicinity of the drops (Batchelor, 1953; Hinze, 1955;
Coulaloglou and Tavlarides, 1977; Lo et al., 1998; Soon et al., 2001). As in the case of
static mixers, experiments to determine transient continuous phase jet velocity and the
associated energy dissipation rate are complicated to perform. However, CFD can be
used to predict the above flow quantities and thereby overcome these limitations.

Soon et al. (2001) assessed the capacity of a high velocity jet homogenizer (originally
designed for cell disruption) to work as an emulsification device. Their objective was to
predict the evolution of drop size distribution as a function of the number of passes
through the orifice (residence time) and the operating pressure. Experimental data on
drop size distributions was combined with the numerical results of fluid flow (obtained
from CFD predictions) to establish the effects of operating conditions on drop properties.
For a non-coalescing, lean, oil-in-water dispersion, they found that when the maximum
energy dissipation rate at the orifice (obtained from CFD) was used in a population
balance equation, the resulting drop size distributions compared very well with
experimental data.

2.2.4.1 Multiphase/Turbulence Modeling Approach, Treatment of Drop Diameter


and Interphase Forces
For the purposes of analyzing of the flow, the liquid-liquid mixture in this investigation
was treated as a continuum having the physical properties of the continuous aqueous
phase. This assumption was considered reasonable given the relatively low dispersed
phase concentration (5 % v/v) and the fact that the density ratio between the phases was
very close to unity. Since the flow of the jet was highly turbulent, the high-Reynolds
number form of the k-ε turbulence model (RNG k-ε) was used for turbulence predictions.

2.2.5 Pipelines
Liquid-Liquid pipe flows (Figure 2.12) are encountered in a broad range of operations
such as the production, processing and transport of petroleum resources, direct contact
37

heat transfer and various reaction engineering operations encountered in the CPI. Domgin
et al. (1997) studied the turbulent dispersion of drops in a vertical pipe. They used the
Eulerian-Lagrangian approach (after optimization of the lagrangian relationship to
turbulent diffusion) to simulate the experimental work of Calabrese and Middleman
(1979) on the turbulent dispersion of drops in a vertical pipe (50.8 mm ID and 9.1 m
long). More recently, Hamad et al. (1999) simulated kerosene-water upflows in a vertical
pipe (78 mm ID and 1.5 m long) and compared their results with the experimental data of
Farrar (1988).

As the present study is focused on the CFD simulation of liquid-liquid flows in pipes,
reasons for the choice of pipes are discussed first. As already seen, most of the previous
CFD related studies have focused on mechanically agitated tanks (Alopaeus et al., 2002;
Podgórska and Bałdyga, 2002; Alexopoulos et al., 2002; Agterof et al., 2003) or
extraction columns (Rieger, 1994; Rieger et al., 1996; Modes and Bart, 1999; Vikhansky
and Kraft, 2004). These configurations suffer from the presence of complex
hydrodynamics with very large spatial variation in energy dissipation rates and feature a
wide range of circulation times.

Figure 2.12 Fully mixed flow pattern for oil-water system in a horizontal pipe (Angeli
and Hewitt, 2000)
38

The results obtained from such configurations cannot therefore be expected to yield
accurate predictions of the fundamental two-phase flow characteristics (such as the local
dispersed phase holdup, relative velocity between the phases etc.) without recourse to a
large degree of empiricism and know-how. In pipes, the hydrodynamics are much simpler
and the turbulence characteristics of the continuous phase (wall-generated turbulence in
particular) are very well investigated (Stock and Michaelides, 1989; Kostoglou and
Karabelas, 1998; Nigmatulin et al., 2000). In addition, an extensive database of accurate
and detailed experimental results is available for dispersed liquid-liquid pipeline flow
(Foussat and Hulin, 1984; Farrar, 1988; Farrar and Bruun, 1988; Vigneaux et al., 1988;
Simonian, 1993; Farrar and Bruun, 1996; Lang and Auracher, 1996; Al-Deen and Bruun,
1997; Ali et al., 1999; Lang, 1999; Soleimani et al., 1999; Fordham et al., 1999; Hamad
et al., 2000).

2.2.5.1 Multiphase and Turbulence Modeling Approach


Domgin et al. (1997) used the Eulerian-Lagrangian approach in their work, as one of their
objectives was to couple a lagrangian model to a commercial CFD code and improve the
basic k-ε formalism using algebraic stress relations for non-isotropic turbulent flows.
Their lagrangian model did not take into effect particle rotation, particle-particle
collisions or turbulence modulation of the carrier (continuous) phase due to the presence
of the dispersed phase (i.e. their model can be classified as being a “one-way coupling”
approach). Turbulence was modeled using the k-ε turbulence model which was
supplemented with algebraic stress relations deduced from a second order closure scheme
(as described in Rodi, 1979). An alternate approach (called the Reynolds Stress Model or
RSM) was also tested where the turbulence in the carrier flow was predicted using a
second order turbulence model, which would directly compute the fluctuating velocity
correlations. However, since the improvements brought by the RSM turbulence model
did not justify the additional CPU time requirements, it was not used in their final
simulations. Also, in their work the value of σk, a constant in the transport equation for
the continuous phase turbulent kinetic energy was modified from its default value of 1 to
2.3, in order to obtain a numerical profile for the turbulent kinetic energy which was in
39

good agreement with the experimental data of Sabot (1976) for turbulent pipe flow.
When comparing the time-variation of the simulated and measured radial dispersion
(from the experiments of Calabrese and Middleman, (1979)), it was seen that the
simulations were in reasonably good agreement with the experimental data, except for
short times, where the radial drop dispersion was clearly underestimated (Figure 2.13).

Figure 2.13 Radial dispersion of CCl4 drops in water (Domgin et al., 1997)

It was speculated that this difference could result from the injection of the dispersed
phase at the center of the pipe which in turn might disturb the flow of the continuous
phase. To overcome this, it was suggested that, the turbulence intensity be increased
locally in the simulation, thus giving rise to a higher degree of dispersion.

In another investigation, Hamad et al. (1999) used the two-fluid Eulerian-Eulerian model
to predict the dispersed phase volume fraction and phase velocity distribution for
kerosene-water up-flows in a vertical pipe. Their preliminary single phase predictions
(velocity and turbulence intensity profiles) reported very good agreement with
experimental data. In their liquid-liquid CFD simulations, the velocity profiles for both
phases were predicted quite accurately. However, they were unable to precisely predict
the distribution of the dispersed phase volume fraction, particularly the characteristic
near-wall peak (Figure 2.14). They attributed this discrepancy to uncertainty in the
40

experimental data as well as the insufficient treatment of interphase forces. While


transport equations for the turbulent kinetic energy and the turbulent energy dissipation
rate were solved for the continuous phase, there was no mention of how the turbulence in
the dispersed phase was accomodated. It was however mentioned that their k-ε model did
not take into account drop-induced turbulence (e.g. in a fashion similar to the models
developed by Lance and Bertodano (1992) and Bertodano et al. (1994) for bubbles).

Figure 2.14 Radial distribution of volume fraction for kerosene-water system:


Experiment (+) CFD Simulation ( ) (Hamad et al., 1999)
2.2.5.2 Treatment of Drop Diameter
Domgin et al. (1997) used a constant uniform drop diameter in all their simulations which
were based on the experimental conditions of Calabrese and Middleman (1979).
Calabrese and Middleman (1979) had observed in their experiments that the drop
diameter did not have a significant influence on the radial dispersion. As a result, they
proposed a single curve for the radial drop dispersion irrespective of the drop diameter.
Interestingly, it was seen from the simulations of Domgin et al. (1997) that unlike the
experimental observations of Calabrese and Middleman (1979), the degrees of dispersion
for different diameter drops were clearly distinct (Figure 2.13), particularly for drops
41

more dense than water (e.g. CCl4). In the other study by Hamad et al. (1999) there was no
mention of how the drop diameter was treated in their CFD simulations.

2.2.5.3 Interphase Forces


Both Domgin et al. (1997) and Hamad et al. (1999) used the standard drag expression for
a single rigid sphere throughout their simulations. Hamad et al. (1999) considered non-
drag interphase forces (such as the lift force, virtual mass force etc.) to be minor
additional forces and consequently did not account for them in their simulations. On the
other hand, Domgin et al. (1997) studied the effect of virtual/added mass, pressure
effects, and the basset (or history) terms in their simulations. They justified the inclusion
of the virtual mass force in their study on the basis of a wide range of density ratios
(ranging from 0.695 to 1.595) involved in their investigations (due to the three different
systems being simulated: Butyl Benzoate-Water (neutrally dense), CCl4-Water (heavy),
n-Heptane-Water (buoyant)). Nevertheless, it was seen from their simulations that the
basset and pressure terms had a weak influence on the results. This was not very
surprising considering the fact that relative fluid-particle acceleration was negligible
throughout their simulations. Moreover, since the dispersion area was mainly restricted to
the center region of the pipe (r/R < 0.5), velocity gradients there were almost negligible,
rendering the pressure term insignificant. As a result, these two forces were neglected in
their future calculations. It should be noted that Domgin et al. (1997) also did not
consider the effect of lift forces in their simulations.

Table 2.1 summarizes all previous work discussed in this Chapter.


Table 2.1 Summary of Previous Liquid-Liquid CFD studies

Turbulence Treatment of Inter- Liquid-Liquid


phase forces System(s) investigated Computer
model (s)
CFD Multiphase hardware
S. Contacting tested for Boundary
package approach(es) Reference used/time
No unit the conditions (BC)
used used taken for a
continuous Continuous Dispersed
Drag force Lift force simulation
phase phase(s) phase(s)

Density =
Standard k- No-slip BC at the 800
as extended Rigid column walls, kg/m3,
1 sphere – Density = Constant
RDC Not by Rodi Eulerian- Not Velocity inlet BC Not
standard
considered
Rieger (1994) 1000 uniform
Column mentioned (1988) for Eulerian at column inlets mentioned
drag kg/m3 drop
the turbulent correlation and outflow BC at
stress tensor. column exits. diameter,
d = 2 mm

Organic
solution
(lighter),
Eulerian- Rigid Three
sphere – Plug flow across Aqueous constant
RDC Eulerian and Not Rieger et al. Not
2 FIRE Standard k- Standard
considered
the inlets for both solution uniform
Column Eulerian- drag (1996) mentioned
phases (heavier) diameters,
Lagrangian correlation
d = 0.5,
1.0 and
2.0 mm

Single
drops (Hu Constant
and Corolis uniform
Kintner forces were No-Slip BC at Not drop
RDC Eulerian- Modes and Not
3 FIDAP Standard k- (1955), considered in column walls and mentione diameter,
Column Lagrangian Vignes their Bart (2001) mentioned
stator. d
(1965) and simulations d = 2.6
Grace et al. mm
(1976))

42
Turbulence Treatment of Inter- Liquid-Liquid
phase forces System(s) investigated Computer
model (s)
CFD Multiphase hardware
S. Contacting tested for Boundary
package approach(es) Reference used/time
No unit the conditions (BC)
used used taken for a
continuous Continuous Dispersed
Drag force Lift force simulation
phase phase(s) phase(s)

Density =
Standard k- 881 kg/m3
Eulerian-
modified to
Lagrangian Poly-
account for Rigid
extended to dispersed
the effect of sphere – Vikhansky Density =
RDC In-house accommodate Not with 5 to Not
4 droplets on standard
considered
and Kraft Not mentioned 998
Column code high volume 15 mentioned
the flow drag (2004) kg/m3
fractions of correlation droplets
field and
the dispersed in each
turbulence
phase. spatial
generation.
cell.

System 1:
Density =
Not applicable 805 kg/m3
as only single System System 2:
phase 1:
Not
Density =
simulations Density =
applicable
Not 923 kg/m3
were applicable as 1064
performed to as only
Stirred Not single
only single
Alopaeus et kg/m3 Poly- Not
5 CFX4 extract the phase
phase Not mentioned dispersed
Tank mentioned simulations al. (1999) System mentioned
local flow simulations with 15
fields and were 2:
were drop
performed Density =
turbulent performed classes
energy 1193 used in
dissipation kg/m3 the
rates. Populatio
n Balance
Equation.

43
Turbulence Treatment of Inter- Liquid-Liquid
phase forces System(s) investigated Computer
model (s)
CFD Multiphase hardware
S. Contacting tested for Boundary
package approach(es) Reference used/time
No unit the conditions (BC)
used used taken for a
continuous Continuous Dispersed
Drag force Lift force simulation
phase phase(s) phase(s)

Dense
organic
phase
(Non-
Standard k- , Newtonia
a turbulent Rigid
Koh and Aqueous n)
sphere –
Stirred prandtl Eulerian- Not Not
6 CFX 4.1 standard
considered
Xantidis Not mentioned solution Initial
Tank number of Eulerian drag mentioned
(1999) (lighter) condition:
1.0 was correlation
used. Constant
uniform
drop
diameter
of 100 µm

Exxsol,
Algebraic Slip (Organic
Mixture solvent)
Model
Rigid Density =
(ASMM) with spheres – 800 kg/m3
and without standard
slip velocities drag Poly-
Standard k-
Stirred (Alopaeus, V., correlation Not Alopaeus et dispersed Not
7 Fluent 5.4 with wall and drag considered
Not mentioned Water
Tank Department of al. (2002) with 20 mentioned
functions. coefficient
Chemical drop size
Technology, expression
s for fluid groups
Kemistinte, spheres used in
Personal the
communicatio Populatio
n, 2004). n Balance
Equation

44
Turbulence Treatment of Inter- Liquid-Liquid
phase forces System(s) investigated Computer
model (s)
CFD Multiphase hardware
S. Contacting tested for Boundary
package approach(es) Reference used/time
No unit the conditions (BC)
used used taken for a
continuous Continuous Dispersed
Drag force Lift force simulation
phase phase(s) phase(s)

Not applicable Sunflower


as only single oil SiG O2,
phase
Not
Water Poly- R10K
simulations Not
applicable containin dispersed processor,
were applicable as
as only g few with 10- 256MB
performed to only single
Stirred Not single Agterof et al. percent 20 drop RAM, IRIX
8 STAR-CD extract the phase
phase Not mentioned
Tank mentioned simulations (2003) CMC to classes 6.3,
local flow simulations
were increase used in simulation
fields and were
performed its the took half a
turbulent performed
viscosity. Populatio day to
energy
n Balance complete.
dissipation
rates. Equation

Not applicable
as only single
phase
Reynolds Not
simulations Not
Stress Model applicable
were applicable as
(RSM) and as only
performed to only single Podgorska & Density = Density =
Stirred Standard k- single No-slip BC at all Not
9 Fluent 6.0 extract the phase
phase Baldyga 1000 1000
Tank with solid boundaries. mentioned
local flow simulations
simulations (2002) Kg/m3 Kg/m3
standard were
fields and were
wall performed
turbulent performed
functions.
energy
dissipation
rates.

45
Turbulence Treatment of Inter- Liquid-Liquid
phase forces System(s) investigated Computer
model (s)
CFD Multiphase hardware
S. Contacting tested for Boundary
package approach(es) Reference used/time
No unit the conditions (BC)
used used taken for a
continuous Continuous Dispersed
Drag force Lift force simulation
phase phase(s) phase(s)

Not applicable
as only single
phase
simulations Not SiG R10K,
Not
were applicable Significant
Reynolds applicable as Non-equilibrium
performed to as only 50 vol % CPU time
only single
Stirred Stress Model single Alexopoulos BC in ε was used Aqueous
10 Fluent 4.32 extract the phase
phase n- Butyl was
Tank (RSM) and simulations et al. (2002) at the solid system
local flow simulations Chloride required for
standard k- were boundaries.
fields and were convergenc
performed
turbulent performed e.
energy
dissipation
rates.

Silicon oil
50:
Density =
Algebraic Slip 949 kg/m3
Not Velocity inlet BC
Mixture Aqueous
applicable as Jaworski and at the inlet, Silicon oil
Kenics Model CMC:
the flow Not Not Pianko- pressure outlet BC 500: Not
11 Static Fluent 5.4.8 (ASMM) and mentioned considered
Density =
being Oprych at the exit, No-slip Density = mentioned
Mixer Eulerian- 998
investigated (2002) BC for all solid- 957 kg/m3
Lagrangian kg/m3
was laminar. liquid boundaries.
approach. Constant
uniform
drop
diameter

46
Turbulence Treatment of Inter- Liquid-Liquid
phase forces System(s) investigated Computer
model (s)
CFD Multiphase hardware
S. Contacting tested for Boundary
package approach(es) Reference used/time
No unit the conditions (BC)
used used taken for a
continuous Continuous Dispersed
Drag force Lift force simulation
phase phase(s) phase(s)

For the
High- purposes of Distilled
Reynolds analysis of the water
number form flow, the with
of the k-ε Liquid-Liquid some
Jet Laborator
turbulence mixture was Not Not Soon et al. concentra Not
12 Homogeniz CFX 4.1 applicable applicable
Not mentioned. y grade
model (RNG treated as a (2001) tion of mentioned
er Castor Oil
k-ε) was continuum surfactant
used for all (having (SDS)
the properties of dissolved
predictions the aqueous in it
phase).

Reynolds
Stress Model 1. Butyl
(RSM) and Rigid Benzoate,
standard k- , sphere –
Vertical Eulerian- Not Domgin et al. Not
13 PHOENICS standard Not mentioned. Water 2. CCl4,
Pipe value of Lagrangian drag
considered (1997) mentioned
constant (σk) correlation 3. n-
was Heptane
modified.

Standard k- ,
different Rigid
single-phase sphere –
Vertical Eulerian- Not Hamad et al. No-slip BC at the Not
14 PHOENICS logarithmic standard
considered
Water Kerosene
Pipe Eulerian drag (1999) wall. mentioned
law
constants correlation
were used.

47
48

CHAPTER 3

3 RESEARCH OBJECTIVE AND METHODOLOGY


The overall objective of this research project is to develop CFD models that can
accurately predict the behavior of multiphase systems. These models will then be used to
improve the efficiency by which energy resources are developed and utilized.

The objective of the present study is to identify and quantify the various significant
interphase forces encountered in turbulent bubbly flows of immiscible liquid dispersions.
The knowledge so gained can be beneficially employed to develop generally applicable
CFD guidelines for interphase closure in dispersed liquid-liquid systems.

To accomplish the aforementioned objective, CFD modeling and validation was


undertaken. The overall approach adopted is outlined below:

1. An in-depth review of previous work focusing on CFD studies of liquid-liquid


dispersions (irrespective of geometry) was first conducted. Several technical
databases such as Compendex, Science Direct, CISTI etc. were queried for this
purpose. Particular importance was given to the treatment of various inter-phase
forces in each of the previous investigations. An online MySQL-based, in-house
database (http://f106-6.chemeng.dal.ca) with a user-friendly front-end was then
created to organize all the literature.

2. The vertical pipeline was chosen as the geometry of interest in the present study.
Elaborate reasons for the selection of a pipeline are discussed in Section 2.2.5. A
‘vertical’ configuration was preferred in order to avoid stratification and backflow
effects (typically observed in horizontal and inclined flows respectively) which
could easily alter the flow regime, thereby complicating the hydrodynamics.

3. Experimental data (e.g. dispersed phase holdup distribution, relative velocity


distribution, individual phase velocity distributions etc.) for vertical liquid-liquid
pipeline up-flows was collected for validating the CFD predictions. The
49

experimental phase distribution profiles were analyzed to classify the observed


trends. Data sets for CFD validation were then selected on an appropriate basis.

4. One of the popular and user-friendly commercial CFD packages (Fluent 6.1.22)
was used to solve the governing equations in this study. The CFD solver/post-
processor (Fluent 6.1.22) and pre-processor (GAMBIT 2.1.2) were installed on a
GNU/Linux workstation. A Pentium IV 2.53 GHz machine (1 GB DDR SDRAM
and 1 GB swap space) running Red Hat Linux 9 was used to carry out the
simulations. In addition, a 4-node parallel cluster was also successfully created
and configured for use with Fluent.

5. A review of previous work related to interphase forces encountered in immiscible


liquid dispersions was then carried out. It was seen that drag, lift and turbulent
dispersion forces were predominant and crucial to phase distribution in dispersed
liquid-liquid systems. Investigations pertinent to the drag on single drops, and the
effect of adjacent drops were then collected and analyzed. A similar study for the
lift forces and turbulent dispersion was also performed.

6. In order to effectively utilize the multifarious capabilities of Fluent, particularly


the user-defined functions (UDF) which allow the end-user to incorporate custom
expressions in a modular fashion, training was undertaken at Fluent Inc.
headquarters. Several drag/lift coefficient correlations were then incorporated as
subroutines written in the 'C' programming language. In addition, the effect of
turbulent dispersion was also studied using the in-built model of Simonin and
Viollet (1990).

7. Preliminary simulations of liquid-liquid vertical up-flows were then conducted.


The two-fluid (Eulerian-Eulerian) approach was used to model the multi-fluid
flow. The standard k-ε model was used to account for turbulence in the
continuous phase. Turbulence in the dispersed phase was modeled in accordance
with Tchen’s theory of dispersion of particles. The assumptions underlying the
50

multiphase and turbulence modeling approaches, and the basis for their selection
are discussed elaborately in Chapter 5.

8. The predicted dispersed phase volume fraction distributions were compared with
several sets of experimental data spanning a wide range of pipe diameters and
phase holdups (Table 4.3). For each experimental data set validated, drag/lift
coefficient formulations/values and turbulent dispersion coefficients which result
in the best matches with experimental data were identified.

9. On the basis of the results obtained in (8), generally applicable guidelines for
interphase closure were proposed. The validity/suitability of the proposed
guidelines to effectively predict useful flow quantities (such as the local dispersed
phase holdup, phasic velocities etc.) in liquid-liquid dispersions was then tested
by application to the case of dispersed liquid-liquid flows in horizontal pipes.
51

CHAPTER 4

4 REVIEW OF EXPERIMENTAL INVESTIGATIONS IN


VERTICAL LIQUID-LIQUID PIPE FLOWS
As mentioned in Chapter 3, the vertical pipe was chosen as the geometry of interest in the
present study. This chapter therefore discusses experimental investigations of liquid-
liquid flows with specific reference to vertical pipelines. An analysis of the phase
distribution profiles obtained from the experiments was done to correlate the nature of
observed profiles with relevant physical dimensions of the domain and flow parameters.
Attention was also focused on the measurement techniques used, pipe diameters and
length to diameter (L/D) ratios and the range of dispersed phase holdup investigated etc.
Experimental data sets for CFD validation were then chosen on a suitable basis.

4.1 Analysis of Liquid-Liquid Pipe Flow Experiments


In general, experimental liquid-liquid investigations of pipe flows are relatively scarce
when compared to investigations on gas-liquid systems, which are overwhelmed with
such studies (e.g. as summarized by Podila, 2005). It is therefore not surprising to note
that a comprehensive review of the literature revealed only nine individual experiments
on vertical bubbly liquid-liquid pipe up-flows that included phase distribution
measurements and/or velocity and turbulence intensity profiles (Table 4.1). Although,
experimental data on vertical pipe down-flows of immiscible liquid dispersions would
have proved equally useful in the present study, no such investigation could be found in
the open literature.

Unlike gas-liquid vertical pipe up-flows, typical liquid-liquid flows do not feature the
slug flow regime (Hewitt, 1997; Brauner 2002). This can be attributed to the fact that
typically, the oil-water interface is not stable enough and any slugs formed would get
broken up very quickly into small drops (Vigneaux et al., 1988).
Table 4.1 Experimental investigations on liquid-liquid pipe flows

Pipe Length Measurement Range of


Observed Total
and Diameter technique used* dispersed
S. Drop Size volumetric
Flow Type Reference Sparger range
phase
No. flow rate (QT ratios (%)
(mm) = QC +QD)
ID (QD/QT)
L (m) Holdup Velocity
(mm)

Vertical upward Separate


1. flow
Foussat and Hulin, 1984 12.0 150 inlets
3–5 LHFP ADVP 15 m3/hr NS

Vertical upward Farrar, 1988, Pipe Y- (5, 10, 15,


2. flow Farrar and Bruun, 1996
1.50 78 junction
5 – 10 HFA HFA 11.1 m3/hr
20, 25, 30)
Vertical upward Pipe (5, 10, 30,
3. flow
Vigneaux et al., 1988 14.0 200 junction
2–6 LHFP NM 27 m3/hr
50, 70, 90)

Injectors
Vertical upward (4.3, 2.4) (18.5,
4. Simonian, 1993 1.50 78 (port 2–4 DP DP
flow
holes) m3/hr 29.5)

(9.7, 10.4,
Vertical upward Pipe Y- (5, 10, 20,
5. flow
Al-Deen and Bruun, 1997 1.50 77.8 junction
NS HFA HFA 11.5, 13.3,
30, 40)
15.5) m3/hr

Vertical upward Lang and Auracher, 1996, Pipe T - 0.3619 & (20, 40, 60,
6. 1.05 16 NS LHFP LHFP
flow Lang, 1999 junction 1.0857 m3/hr 80)

(9.9, 10.4, 11.7, (5, 10, 20,


Same as
Vertical upward 13.4, 15.6,
7. flow
Ali et al., 1999 1.50 77.8 Bruun, NS HFA HFA 18.7, 23.4,
30, 40, 50,
(1995) 60, 70)
31.2) m3/hr

Same as
Vertical upward (11.1, 11.7)
8. Hamad et al., 2000 4.20 77.8 Bruun, 1–7 DOP DOP (10, 20)
flow
(1995) m3/hr

52
Pipe Length Measurement Range of
Observed Total
and Diameter technique used* dispersed
S. Drop Size volumetric
Flow Type Reference Sparger range
phase
No. flow rate (QT ratios (%)
(mm) = QC +QD)
ID (QD/QT)
L (m) Holdup Velocity
(mm)

Vertical upward (2.1 – 28.4)


9. Fordham et al., 1999 10.0 78.0 NS NS LFOP LFOP (5 – 95)
flow m3/hr

Inclined upward
Pipe
10. flow (15˚ from Vigneaux et al., 1988 14.0 200 junction
2–6 LHFP NM 40 m3/hr (5, 50, 90)
vertical)

15˚: (4, 4.3) 15˚: (15,


Inclined upward 3
Injectors m /hr 22.2)
flow (15˚, 30˚ &
11. 45˚ from
Simonian, 1993 1.50 78 (port 4–7 DP DP 3
30˚: (6.7) m /hr 30˚: (26.3)
holes)
vertical) 3
45˚:(2.9) m /hr 45˚:(38.2)

Pipe T - (37.5,
12. Horizontal flow Angeli, 1996 9.50 24 junction
0.5 – 1 LHFP NM 2.7 m3/hr
62.5)

Pipe T - LHFP,
13. Horizontal flow Soleimani et al., 1999 9.70 25.4 junction
NS NM 5.5 m3/hr (40, 54)
GDS
*
HFA: Hot Film Anemometer; LHFP: Local High Frequency Impedance probe; DP: Dual probe (A combination of a hot-film and two optical probes); PL: Pulsed Laser; ADVP: Acoustic
Doppler Velocimeter probe; DOP: Dual Optical probe; LFOP: Local Fibre Optical probe; GDS: Gamma Densitometer system; NS: Not specified; NM: Not measured

53
54

As a result, the bubbly flow regime persists even at relatively higher dispersed phase
holdups (up to 50% or even more in the presence of contaminants/surface active agents).
However, it should be noted that such behavior strongly depends on the physical
properties (density, viscosity, surface tension) of the oil involved, as any changes in these
properties can directly influence the stability of the oil-water interface, resulting in a
strongly modified flow structure (Vigneaux et al., 1988). Another important factor that
should be borne in mind is the phenomenon of phase inversion, wherein a small change
in operating conditions can cause the continuous and dispersed phases to spontaneously
invert. Typically, this condition is realized only beyond high dispersed phase ratios
(greater than 50 %). As extensions to the two-fluid multiphase model in CFD, that can
take into account flow regime transitions are still under active research (for instance, the
work of Navarro-Valenti et al., 1993), the present study was confined to dispersed flow
conditions below the phase inversion point. Each of the nine investigations on vertical
pipe up-flows is discussed in the following section.

4.1.1 Discussion of Experimental Investigations


One of the foremost investigations of liquid-liquid vertical up-flows that featured both
phase distribution and velocity measurements was that of Foussat and Hulin (1984). They
studied the flow of a kerosene-water mixture over a wide range of dispersed phase hold-
ups (4 – 93%). Details concerning their experimental investigation can be found in Table
4.1. Their observations showed that the relative velocity between the phases is high at
low volume fractions of the dispersed phase and vice-versa. Similar trends were reported
by other investigators as well (Vigneaux et al., 1988; Simonian, 1993). They also
observed that the dispersed phase velocity and volume fraction profiles were highly
curved (i.e. low concentration/velocity near the walls and high concentration/velocity
near the pipe axis, a trend commonly referred to as coring) at low dispersed phase
holdups (less than 5%). This was attributed to the effect of the wall on drop dynamics. At
higher dispersed phase fractions (greater than 15%), the volume fraction profiles for the
dispersed phase were observed to become extremely flat except at distances less than 10
mm from the walls, where a small peak was seen (a trend commonly referred to as wall
55

peaking). On the other hand, the dispersed phase velocity profiles assumed a saddle-like
shape featuring a minimum at the pipe axis. As the dispersed phase concentration was
further increased, the velocity profiles of the dispersed phase were observed to flatten
out.

These observations were ascribed to the fact that at low dispersed phase concentrations,
the drop velocity is determined by its distance from the pipe wall, while at high
concentrations, the interactions with neighboring drops assumed greater importance. An
analogy with the hindered settling effect (which is typically observed above a critical
concentration in the case of solid suspensions) was drawn to explain the findings at high
dispersed phase holdups. In a later investigation, Vigneaux et al. (1988) reported very
similar observations for the dispersed phase concentration profiles. The coring trend
observed at low dispersed phase concentrations was explained on the basis of large
velocity and volume fraction gradients being present throughout the pipe cross-section.
Thus the interaction with the wall played a major role at low holdups.

On the other hand, for pipes of somewhat smaller internal diameter (ID ≈ 78 mm),
several investigators (Farrar, 1988; Farrar and Bruun, 1996; Al-Deen and Bruun, 1997;
Ali et al., 1999; Hamad et al., 2000) consistently reported a slight increase in the
dispersed phase concentration near the wall at low dispersed phase holdups which
subsequently manifested into a near-wall peak at higher volume fractions (20 - 30%).
However, in contrast to the above observations, Simonian (1993) observed coring even at
dispersed phase concentrations as high as 10%. This irregularity clearly indicates that
there are quite a few lateral forces at play, depending on the flow conditions. These
forces are elaborately discussed in Chapter 5. At still higher dispersed phase holdups, (20
- 40 %), most investigators (Farrar, 1988; Farrar and Bruun, 1996; Al-Deen and Bruun,
1997; Ali et al., 1999; Hamad et al., 2000) report significant near-wall peaking of the
dispersed phase volume fractions. This is in stark contrast to the observations in gas-
liquid flows, where dispersed phase void fraction profiles are near-wall peaked at low
dispersed phase holdups and exhibit coring at higher concentrations. Furthermore, Farrar
(1988) and Farrar and Bruun (1996) observed that at high dispersed phase concentrations,
56

the continuous phase velocity profiles are more non-uniform (inverted ‘V’ shaped) and
feature large gradients and therefore could be expected to be responsible for the observed
near-wall peaking of the dispersed phase holdup. It should also be noted that all
investigators who observed the drop sizes in their experiments (Foussat and Hulin, 1984;
Farrar, 1988; Vigneaux et al., 1988; Simonian et al., 1993; Farrar and Bruun, 1996;
Hamad et al., 2000), reported an increase in the average drop size with increasing
dispersed phase holdup.

In other investigations, Lang and Auracher (1996) and Lang (1999) studied the phase
distribution and heat transfer characteristics of n-heptane-water mixtures in a very small
diameter tube and concluded on the basis of their observations that only the continuous
phase was in direct contact with the wall, at all times. Their observations also showed
that the dispersed phase concentration profiles exhibited consistent coring type behavior
for high dispersed phase holdups (20 – 60%). Unfortunately no experimental data was
obtained at low dispersed phase concentrations.

Table 4.2 Characteristic dispersed phase holdup profiles observed in vertical pipe
upflows of liquid-liquid dispersions

No of Type of profile
Pipe Diameter – (ID mm) Dispersed phase holdup
investigations observed *

Low holdup (< 10%) ND

Small Diameter - (16 mm) 1 Intermediate holdup (20 – 60%) VC

High holdup ( 80%) WP

Low holdup (≤ 10%) NWP and VC


Intermediate Diameter - (≈
5 Intermediate holdup (20 – 60%) VC and NWP
78 mm)

High holdup (> 80%) ND

Low holdup (≤ 10%) VC


Large Diameter - (150 – 200
2 Intermediate holdup (20 – 60%) WP
mm)

High holdup (> 80%) VC

* ND: No data; VC: Coring; NWP: Near-wall peaking; WP: Wall peaking
57

From the above discussion, it is clear that vertical liquid-liquid pipe upflows typically
feature non-homogeneous phase distributions. Table 4.2 categorizes experimentally
observed phase distribution profiles on the basis of dispersed phase concentration and
pipe diameters.

It is seen from Table 4.2 that it is not possible to arrive at a conclusive general behavior
for the dispersed phase distribution profiles. Given the number of investigations on
intermediate diameter pipes, it is reasonable to accept that the observed profiles at low
and intermediate holdups are indeed general trends for 3 pipes. However, for the other
pipe diameters, it is difficult to reach a similar conclusion on the basis of just one or two
experiments.

Having discussed the experimentally observed dispersed phase distributions in detail, it is


important to select appropriate data sets from Table 4.1 for CFD validation studies.

4.1.2 Selection of Experimental Data for CFD Validation


In the following discussion, each phase distribution measurement (obtained from a
particular set of experimental conditions) across the pipe cross-section will be referred to
as a data point. A data point is designated using the corresponding phase ratio for the
dispersed phase. For instance, the data set of Hamad et al. (2000) is comprised of two
data points (10% and 20%). For quick and easy reference, these data points will be
referred to as H10 and H20 respectively. The following paragraph discusses the selection
criteria used for each data point in the present study.

The choice of data points for validation studies depends on several factors starting with
the most basic requirement: availability of adequate information required to carry out a
two-phase CFD simulation. Also, consistent and reliable experimental data are crucial for
successful CFD validation studies (e.g. as discussed by Benay et al., 2003). Other
important factors are the measurement technique used (invasive versus non-invasive),
number of measured parameters, their accuracy and suitability. These factors are
discussed elaborately in the present section.
58

Non-invasive measurement techniques are usually more accurate than invasive ones
owing to their ability to measure quantities without probing into the domain and thus
altering any flow fields. For instance, in phase distribution measurements, LDA (Laser
Doppler Anemometry) is usually preferred over hot-film anemometry or high frequency
impedance measurements, which are traditional invasive techniques. However, the
application of LDA to liquid-liquid flows is limited to very low volume fractions of the
dispersed phase, due to the requirement of a free optical path (Al-Deen and Bruun, 1997;
Ali et al., 1999). As a result, it is difficult to find liquid-liquid investigations with phase
distribution data that have been acquired using such non-invasive measurement
techniques. Hand-in-hand, experiments which perform simultaneous measurements of
flow fields, turbulence quantities, drop size distributions, interfacial area density etc. are
given more preference in the current validation study, as such data sets can be considered
relatively ‘complete’ in nature. This is because, successful validation with such data will
increase confidence in the computational models used provided the data is accurate
within acceptable limits. Accuracy of experimental data is essential simply because CFD
validations are typically carried out by comparing quantitative predictions with
experimental data. No experiment is 100% accurate, and therefore it is imperative to
gauge the correctness of experimental data beforehand. An aspect which is directly
related to the measurement accuracy is the suitability of measured data. For instance, a
very fundamental test for suitability of phase distribution data is to check whether the
area-weighted average of the dispersed phase volume fraction across the measurement
section (i.e. the in-situ dispersed phase holdup) is less than the input phase ratio of the
dispersed phase. This condition is necessary to ensure a non-zero positive slip between
the dispersed and continuous phases when the dispersed phase moves faster than the
continuous phase. Therefore for the dispersed phase, a phase ratio lower than the average
holdup represents a physically inconsistent situation (negative slip), thus rendering the
data point largely inaccurate.
59

4.1.2.1 Selected Experimental Data Points


Based on the guidelines stipulated in Section 4.1.2, appropriate data points for CFD
validation were selected. A brief summary of the data sets analyzed and the data points
selected is provided in the present section.

Although extensive, the data set of Foussat and Hulin (1984) could not be used for
validation in the current study due to the unavailability of inlet conditions (phase
superficial velocities). A personal communication with the corresponding author
confirmed the unavailability of requested data as old as 1984. It could be argued that one
might match experimental phase distribution data by employing trial and error methods
to guess the phase ratios and subsequently use the total volumetric flow rate (QT) to
calculate the individual phase velocities. But that would mean using inlet velocities and
consequently, mass flow rates that could well differ from those actually used in their
experiments. As a result, there would be less confidence in the resulting CFD prediction
even if the match with experimental data was excellent.

Data points from Farrar and Bruun (1996) were the found to be the same as those of
Farrar (1988). The first three data points (5%, 10% and 15%) were rejected as they did
not satisfy the suitability criteria for phase distribution. As a result, only the last three
data points (20%, 25% and 30%) which were found to be appropriate (based on the
guidelines proposed in Section 4.1.2) were used for CFD validation.

Although the data set of Simonian (1993) comprised of fifteen data points, only two data
points (18.5, 29.5) were available with phase velocity data for both phases. Input phase
ratios for the other data points were also not specified. Simonian (1993) reported both
average dispersed phase holdup and average slip velocities in his experiments. When the
average slip velocities obtained from direct measurements were used to calculate the
average dispersed phase holdup for comparison with the measured average dispersed
phase holdup, a percentage error greater than 20% was obtained for both data points.
Moreover, Simonian (1993) reported relative velocities in excess of 20 cm/s for most of
his data points even at low dispersed phase holdups, which is quite unreasonable as can
60

be inferred from the discussion in Section 6.1.1.1. Consequently, the data set of Simonian
(1993) was not used for validation.

The first four data points (5%, 10%, 20%, 30%) of Al-Deen and Bruun (1997) were
found to satisfy all requirements stated in the previous section, and were therefore used in
the present validation study. In their investigation, Al-Deen and Bruun (1997) expressed
reproducibility concerns for the last data point (40%) which appeared to exhibit both
near-wall peaking and coring trends. They concluded that further studies would be
required to determine the cause for such behavior. As a result, this ambiguous data point
(40%) was not used for validation purposes.

Unfortunately, all the data points (below the phase inversion point) in the data set of Ali
et al. (1999) did not satisfy the suitability criteria and consequently could not be used for
validation.

The data set of Lang (1999) was found to include all data points in the Lang and
Auracher (1996) data set. Out of a total of six data points (mostly below the phase
inversion point), only four (L20A, L20B, L40, L60) were found to satisfy the suitability
criteria and were thus used for validation.

All data points (below the phase inversion point) from the data set of Vigneaux et al.
(1988) were used for validation as they satisfied both accuracy and suitability
requirements. Although the data set of Hamad et al. (2000) did not feature phase
distribution data close to the wall, it was found to meet the suitability criteria if the
volume fraction of the dispersed phase was assumed to be zero at the wall. On the basis
of this assumption, all data points in their data set were also selected for validation.

While Fordham et al. (1999) successfully demonstrated the use of a novel fiber-optical
sensor for measuring phase distribution in kerosene-water vertical pipe upflows, they
only presented a single data point which was unfortunately very close to the phase
inversion point (αd ≈ 70%). Moreover, the individual phase velocities used to obtain the
data point were also not explicitly stated. Consequently, their data set could not be used
in the present validation study. A personal communication with the corresponding author
61

for other useful phase distribution data could not be established. The data sets and
corresponding data points selected for validation in the present CFD study are organized
in Table 4.3.
Table 4.3 Summary of liquid-liquid experimental data points selected for CFD validation

Data set Selected data points

Farrar and Bruun (1996) F20, F25, F30

Vigneaux et al. (1988) V5, V10, V30, V50

Al-Deen and Bruun (1997) A5, A10, A20, A30

Lang (1999) L20A, L20B, L40, L60

Hamad et al. (2000) H10, H20


62

CHAPTER 5

5 CFD MODELING OF LIQUID-LIQUID DISPERSIONS


Having discussed the experimental investigations pertinent to liquid-liquid upflows in
vertical pipes and selected appropriate experimental data for CFD validation, this chapter
focuses on the selection of appropriate multiphase and turbulence modeling approaches
that can efficiently handle immiscible liquid dispersions.

5.1 Selection of an Appropriate Multi-fluid Model for Simulating


Immiscible Liquid Dispersions
The different modeling approaches for dispersed multi-fluid systems were discussed in
Section 2.1.2. At first glance, all four modeling approaches (VOF, E-L, E-E and Mixture)
may seem suitable to simulate immiscible liquid dispersions, at least at low volumetric
loading. However, the choice needs to be made on a more practical and rational basis
keeping in mind the objective and other computational constraints. However, before
deciding on an approach, it is important to identify the extent of coupling between the
dispersed and continuous phase. The different types of coupling are therefore discussed
first.

Dispersed entities (solid or fluid particles) are not passive contaminants and their
presence can very well influence the continuous phase hydrodynamics. Typical liquid-
liquid flows of practical interest are invariably turbulent in nature with phase fractions
varying in the wide range (0-100 %). The presence of small dispersed phase entities at
very low volumetric concentrations has virtually no effect on the continuous phase
turbulence. This condition is often referred to as One-way coupling. However, as the
volume fraction of the dispersed phases increases, this effect starts to become more
pronounced and Two-way coupling is realized. Here, the influence of either phase on the
other's turbulence is considerable. When the number density of the dispersed phase is
sufficiently large to allow for strong particle-particle interactions, the system qualifies as
a Four-way coupling type (two-way coupling in conjunction with inter-particle collisions
63

and viscous stresses due to particles). The extent of coupling can be analyzed by
examining relevant length and time scales. Elghobashi (1991) proposed a regime map for
this purpose. It is inferred from this map that dense multiphase flows typically feature
Four-way coupling. The suitability of each of the four multi-fluid approaches can now be
discussed.

The VOF approach is not suited to modeling high dispersed phase fractions (greater than
10%) for the reasons cited in section 2.1.2.1. Also, as the present work is not focused on
accurately modeling interface-related forces such as surface tension or adhesion forces,
the VOF approach is not particularly useful. Typically, the Eulerian-Lagrangian approach
would then be a natural choice owing to the straightforward treatment of the coupling
between the turbulence fields of the continuous and dispersed phases (Picart et al., 1986).
However, computation can become very time consuming when a large number of
particles are to be tracked. While recent research (Vikhansky and Kraft, 2004) suggests
modifications/extensions to the Eulerian-Lagrangian formulation to make it
computationally manageable at high-phase fractions, the discrepancy between simulated
predictions and experimental data they identified when using this modified E-L approach
suggests that this technique needs more refinement and validation before being
considered as a viable option. On the other hand, although the multi-fluid mixture
approach would solve a single momentum equation for the liquid-liquid mixture thereby
significantly reducing computation time and power requirements, it is not possible to
include the effect of other non-drag forces such as virtual mass, lift, and turbulent
dispersion effects in this framework (as discussed earlier in Section 2.1.2.4). As a result,
the mixture model is also not suitable for the present study.

In contrast, the Eulerian-Eulerian approach not only provides an effective structure to


incorporate and examine several inter-phase forces, but is also well suited to modeling
dispersed flows at high phase fractions that typically feature Four-way coupling. It is also
well known that the predictions of the Eulerian-Eulerian approach are more accurate than
those of the mixture approach in flows where the dispersed phases are concentrated in
certain portions of the domain as opposed to uniformly distributed throughout (Ranade,
64

2002). It is therefore not surprising to note that most CFD studies on immiscible liquid
dispersions have almost exclusively adopted the Eulerian-Eulerian approach (Table 2.1).

On the basis of the above facts, the Eulerian-Eulerian approach was chosen for
multiphase modeling in the present study.

5.1.1 A synopsis of the Eulerian-Eulerian Approach


While there are several formulations of the Eulerian-Eulerian approach, the form of
governing equations as proposed by Ishii (1975) is most commonly used in fluid-fluid
flows. The basis for the governing equations has been discussed in detail by van Wachem
and Almstedt (2003). The Two-fluid Eulerian-Eulerian approach used in the current study
solves the following conservation equations:

5.1.1.1 Conservation of Mass (Equation of Continuity)


Conservation of mass for any phase (q) is given by the following transport equation:

∂ dq ρq
(α q ρ q ) + ∇ ⋅ (α q ρ q v q ) =
n
m pq − α q [5.1]
∂t p =1 dt
Convection term
Accumulati on term
Diffusion term

For steady-state incompressible flow in the absence of mass transfer, this simplifies to:

∇ ⋅ (α q v q ) = 0 [5.2]

The solution of this equation for each phase, taking into consideration the condition that
the phase volume fractions sum to unity, allows for the calculation of individual phase
volume fractions.

5.1.1.2 Conservation of Momentum


The conservation of momentum for a continuous fluid phase ‘q’ in a non-accelerating
frame of reference is given by:


(α q ρ q vq ) + ∇ ⋅ (α q ρ q vq vq ) = −α q ∇p + ∇ ⋅τ q + α q ρ q g
∂t
[5.3]
+ (Fq + Flift ,q + Fvm,q ) + (K (v − v q ) + m pq v pq )
n

pq p
p =1
65

Where the qth phase stress tensor, τq, for laminar flow is given by:

( )
τ q = α q µ q ∇v q + ∇v q T + α q λ q − µ q ∇ ⋅ v q I
2
3
[5.4]

Here ‘µq’ and ‘λq’ are the shear and bulk viscosity of phase ‘q’. The relative velocity

between the continuous and dispersed phase is given by (v p − v q ) , where ‘ I ’ is a unit

tensor and ‘p’ is the pressure shared by all phases. The bulk viscosity is typically ignored
in fluid-fluid flows as it assumes importance only in gas-solid flows (Lun et al., 1984)
and flows that feature shock waves, absorption and attenuation of acoustic waves etc.
(Ranade, 2002).

For steady-state incompressible flow in the absence of mass transfer, external body forces
(Fq) such as the centrifugal force, and virtual/added mass effects (which assume
significance only when high-frequency fluctuations of the relative velocity are
predominant; Drew, 1983; Chen et al., 2005), the momentum conservation equation
simplifies to:

(
∇ ⋅ α q ρ q vq vq ) = − α q ∇p + ∇ ⋅τ q +
Change of momentum due to convection Pr essure force per unit volume Viscous force per unit volume

[5.5]
(Flift ,q ) (K pq (v p − v q ))
n
αq ρq g + +
Gravitatio nal force per unit volume Lift force per unit volume p =1
Drag & Turbulent dispersion force per unit volume

Appropriate expressions for the viscous stresses (∇ ⋅ τ q ) and the interphase terms ( Flift ,q

and K pq (v p − v q ) ) need to be supplied in order to obtain a closed set of equations. These

are discussed in the following sections.


66

5.2 Selection of a Turbulence model for Simulating Immiscible Liquid


Dispersions
It is an unfortunate fact that no single turbulence model is universally accepted as being
superior for all classes of problems. The choice of a turbulence model depends on
considerations such as the physics encompassed in the flow, the level of accuracy desired
and the available computational resources. In most cases, there is a need to strike a
balance between these three factors. In the context of turbulent flows, the two-equation
standard k-ε turbulence model is the most extensively studied/validated and is therefore
used as a baseline approach in RANS-based models. Also, the proven success of the
standard k-ε model in predicting turbulence quantities for single-phase pipe flows cannot
be underestimated (Martinuzzi and Pollard, 1989; Pollard and Martinuzzi, 1989).
Moreover, with specific reference to immiscible liquid dispersions, irrespective of the
domain geometry, it is seen that the two-equation standard k-ε turbulence model is the
most successful (reasonable accuracy, numerically robust and computationally
economical) and therefore extensively used (Table 2.1). The mechanics of this model,
including the modifications needed to account for the presence of the dispersed phase, are
discussed in the present section.

5.2.1 Two-Equation Standard k-εε Model


In the standard k-ε model, the solution of two separate transport equations (turbulent
kinetic energy, k, and its dissipation rate, ε) allows the turbulent velocity and length
scales to be independently determined. It is a semi-empirical model, and the derivation of
the model equations relies on phenomenological considerations and empiricism. Put
simply, it represents the effect of turbulence by calculating a quantity (using k and ε)
referred to as the turbulent viscosity, which augments the laminar viscosity in the
governing equations (i.e. based on the eddy viscosity hypothesis). In this model, it is
assumed that the flow Reynolds number is high, turbulence is nearly homogeneous
throughout, and the effects of molecular viscosity are negligible. Therefore, the k-ε
model is valid only for fully turbulent flows (e.g. the flow in regions somewhat far away
67

from walls). Consideration therefore needs to be given as to how to make it suitable for
wall-bounded flows.

5.2.1.1 Near-Wall Modeling Approaches


Numerous experiments in the past have shown that the near-wall region can be divided
into three layers. In the innermost layer, called the viscous sub-layer, the flow is almost
laminar, and the molecular viscosity plays a dominant role in momentum, heat and mass
transfer. In the outermost layer (usually referred to as the fully-turbulent core), turbulence
plays a major role. In the interim region between the viscous sub-layer and the fully
turbulent core, the effects of molecular viscosity and turbulence are equally important.
These divisions of the near-wall region are shown in Figure 5.1.

Figure 5.1 Flow structure near the wall (Fluent, 2004)

Traditionally, there are two approaches to modeling the near-wall region:

1. In one approach, the viscosity-affected inner region (viscous sub-layer and buffer
layer) is not resolved. Instead, semi-empirical relations called ‘Standard wall
68

functions’ are used to bridge the viscosity-affected region between the wall and
the fully-turbulent region (core). The use of wall functions obviates the need to
modify the turbulence models to account for the presence of the wall.

2. In another approach, the turbulence models are modified to enable the viscosity-
affected region to be resolved with a mesh that extends all the way to the wall,
including the viscous sub-layer (Y+ < 5). This is referred to as the enhanced wall
treatment approach.

These two approaches are depicted schematically in Figure 5.2.

Figure 5.2 Various approaches to model near-wall Turbulence (Adapted from Fluent,
2004)

It should be noted that for most high-Reynolds-number flows (Re > 50,000) in pipes, the
wall function approach substantially saves computational resources, because the
viscosity-affected near-wall region, in which the solution variables change most rapidly,
does not need to be resolved.

5.2.1.2 A Synopsis of the Standard k-εε Turbulence Model


Multiphase turbulence modeling typically relies on some of the base practices adopted in
single-phase flows. To account for the obvious difference in single and multiphase flows,
some ad hoc modifications are introduced to account for the presence of dispersed phase
69

entities. The treatment of continuous and dispersed phase turbulence in the context of the
standard k-ε turbulence model is discussed in the following sections.

5.2.1.2.1 Turbulence in the Continuous Phase


Continuous phase turbulence is usually modeled using transport equations for ‘k’ and ‘ε’
which are written in a form similar to those found in single phase flows (Equation 5.6):


(α q ρ qϕ q ) + ∇ ⋅ α q ρ qU qϕ q = ∇ ⋅ α q µ t ,q ∇ϕ q + Sϕq
( ) [5.6]
∂t σϕ

Where, q can be the turbulent kinetic energy or the turbulent energy dissipation rate of
the continuous phase ‘q’. The symbols µt,q and σϕ refer to the turbulent viscosity and the
turbulent prandtl number for ‘k’ and ‘ε’ of the continuous phase.
Table 5.1 Standard k-ε turbulence model constants (Launder and Spalding, 1972)

Model parameter Default (Suggested) Value

C1ε 1.44

C2ε 1.92

Cµ 0.09

σk 1.0

σε 1.3

The standard k-ε turbulence model contains certain empirical constants which are listed
in Table 5.1. These are the same as those found in the single phase k-ε turbulence model,
which were determined from experiments with air and water for fundamental turbulent
shear flows which include homogeneous shear flows and decaying isotropic grid
turbulence (Fluent, 2004). In principle, the values of the constants will depend on particle
loading and the ratio of particle relaxation time (describes the inertia of a particle and
measures its ability to respond to different scales of the local turbulence) to eddy lifetime
70

(Lahey, 1987; Squire and Eaton, 1994). Although, there have been some investigations
discussing the variation of certain constants in the standard k-ε turbulence model (for
instance the study by Cao and Ahmadi (1995) which discusses the variation of parameter
Cµ with the dispersed phase volume fraction), it is generally agreed upon that more
research is needed to accurately predict turbulence in the continuous phase. In view of
this uncertainty, the use of single phase constants without any modifications can be
considered justified.

The continuous phase turbulent viscosity is calculated using an expression similar to that
used in single phase flows:

k q2
µ t ,q = ρ q C µ [5.7]
εq

The term ‘S q’ in Equation 5.6 is the source term for the quantity, . This can be further
expanded as follows:

S kq = α q Gk ,q − α q ρ q ε q + α q ρ q ∏ k q
εq
S εq = α q
kq
[C 1ε ]
G k , q − C 2ε ρ q ε q + α q ρ q ∏ ε q
[5.8]

Here, Gk,q represents the turbulent kinetic energy generation due to mean velocity
gradients in phase ‘q’, and is again calculated in a fashion similar to single-phase flows,
where:

∂u j
G k = − ρ u i′u ′j [5.9]
∂xi

Extra generation or dissipation of turbulence due to the presence of the dispersed phase
(i.e. turbulence modulation) is accounted for in the last terms of the expressions for Skq
and Sεq respectively (i.e. α q ρ q ∏ k q and α q ρ q ∏ ε q ). There have been several attempts to

develop models in order to represent such extra terms and these are reviewed by Lahey
(1987) and more recently by Peirano and Leckner (1998). However, in the absence of
adequate information, the extra generation/dissipation terms are usually set to zero. Once,
71

‘k’ and ‘ε’ have been solved using the transport equations (Equation 5.6), the turbulent
viscosity (µt,q) can be obtained from Equation 5.7. This can then be used to determine the
stress tensor for turbulent flows using the expression given below:

τq′′ = −
2
3
( ) (
ρ q k q + ρ q µ t ,q ∇.U q I + ρ q µ t ,q ∇U q + ∇U q T ) [5.10]

The above expression for τq′′ is added to the laminar stress tensor contribution ( τ q ) in the

momentum conservation equation (Equation 5.5) to account for turbulence in the


continuous phase.

5.2.1.2.2 Turbulence in the Dispersed Phase


Turbulence in the dispersed phase may be physically understood as the particle velocity
fluctuations caused by inter-particle collisions and interactions of the particles with the
turbulent continuous phase (Balzer et al., 1995; Simonin, 1995; Enwald et al., 1996).
However, if the concentration of the dispersed phase is assumed to be dilute (i.e. one-way
coupling), inter-particle collisions are negligible and the random motion of the dispersed
phase is dominated by the turbulence in the continuous phase. Fluctuating quantities of
the dispersed phase can therefore be given in terms of the mean characteristics of the
continuous phase and the ratio of the particle relaxation time and eddy-particle interaction
time. This approach is applicable when there is clearly one continuous phase and the rest
are dispersed dilute secondary phases. Predictions for turbulence quantities for the
dispersed phases are then obtained using the Tchen theory of dispersion of discrete
particles by homogeneous turbulence (Tchen, 1947).

The procedure used to calculate the dispersed phase turbulent viscosity and kinetic
energy and from it, the drift velocity which accounts for the turbulent dispersion effect is
explained in Section 5.3.2.2.1.

5.3 Interphase Closure


As mentioned in section 5.1.1.2, apart from the turbulent stress tensor, several interphase
forces need to be specified in order to close the momentum conservation equation. A
72

review of interphase forces encountered in liquid-liquid systems was therefore carried out
to determine the most significant ones. It was seen that interphase drag, lift forces and
turbulent dispersion effects were very commonly encountered in liquid-liquid bubbly
flows. This section presents a detailed account of the nature of these interphase forces.
Also discussed are the various expressions/formulations available in the open literature to
account for them.

5.3.1 Drag Forces in Liquid-Liquid Systems


Traditionally, it is seen that the drag on rigid spheres has been investigated thoroughly.
Several expressions of the drag coefficient for a single rigid sphere have been proposed in
the past. These expressions (given in terms of the drag coefficient) are reviewed by Clift
et al. (1978). One of the most popular expressions is the one proposed by Schiller and
Naumann (1935):

C D0 =
( )
24 1 + 0.15 Re 0.687 Re Re ≤ 1000
[5.11]
0.44 Re > 1000

It was seen from a review of the literature on liquid-liquid flows (Table 2.1) that some
investigators have directly used this expression to account for the drag on drops. The drag
on fluid particles (i.e. bubbles and drops) is however more complex in nature when
compared to the drag on rigid spheres. From fragmentary experimental results of
previous work, it has been found that a liquid drop in another liquid medium does not
always behave like a rigid sphere. According to Hu and Kintner (1955), this discrepancy
can be attributed to the fact that the nature of a liquid-liquid interface differs markedly
from that of a solid-liquid interface. This is because unlike rigid spheres, fluid particles
typically feature certain unique characteristics such as internal circulation, deformation
(changes in shape), oscillation (changes in volume) etc. Changes in the shape or volume
directly affect the equivalent drop diameter (de) and hence the drag force on the drop.
Internal circulation on the other hand, works to reduce the effect of drag on the drop. The
flow field around the drop induces a circulation in the fluid that makes up the drop
(Figure 5.3). Energy required for this circulation is directly obtained from the drag force.
73

As a result, a portion of energy from the drag force is used up to maintain the internal
circulation, thus reducing the effective drag on the drop. The situation becomes even
obscure when surface-active agents (SAA) are involved. SAA act as contaminants and
render the surface of the drop immobile, thus reducing the internal circulation (Figure
5.3). A direct consequence of this behavior is a reduction in the relative (rise) velocity of
the drop when compared to the clean (uncontaminated) situation.

Internal circulatory
currents
Drop

Contaminants
Fluid streamlines

Figure 5.3 Internal circulation in a drop without (left) and with (right) surface-active
agents (Adapted from Patro, 2003)

Further complications are introduced when a liquid drop moves in a liquid medium that is
contains a mixture of several similar drops. Although ‘fluid particles’ seldom occur in
isolation in processes of practical interest, it is essential to understand the behavior of
single ‘fluid particles’ before a complete understanding of their movement in a medium
featuring several adjacent ‘fluid particles’ can be achieved (Grace et al., 1976).

5.3.1.1 Drag on a Single Drop


The drag force on a single drop can be calculated in a fashion very similar to that of a
single rigid sphere. Differences between the two are easily accommodated by suitable
modifications to the expression for single entity drag coefficient (CD0). The conventional
drag force (FD) on a single entity moving through a fluid is given by:

1
FD = π d e2 C D 0 ρ CVS2 [5.12]
8
74

Here, ‘de’ is the equivalent diameter, which for fluid particles is defined as the diameter
of a sphere having the same volume as that of the fluid particle. ‘ VS ’ is the relative (slip)

velocity between the drop and the liquid through which it moves.

Suitable modifications to the above expression need to be made in order to accommodate


it within the two-fluid model framework. If ‘αd’ be the volume fraction of the dispersed
phase, then by definition:

Total volume of the dispersed phase


αd = [5.13]
Total volume of the dispersed phase + Total volume of the continuous phase

For a unit volume of the domain,

Total volume of the dispersed phase + Total volume of the continuous phase = 1 [5.14]

Also,

Total volume of the dispersed phase = (No of dispersed phase particles ) ∗


[5.15]
(Volume of one dispersed phase particle )

In other words:

π d e3
Total volume of the dispersed phase = n p [5.16]
6

Where ‘np’ represents the number of dispersed phase particles in the domain. Substituting
Equation 5.14 into Equation 5.13 and using Equation 5.16, we can obtain an expression
for ‘np’:

6α d
np = [5.17]
π d e3
The total drag force on all dispersed phase entities in the domain is then given by:

1
FD = π d e2 C D 0 ρ C VS2 n p [5.18]
8

Using Equation 5.17, this expands to:


75

3 C D 0α d ρ cVS2
FD = [5.19]
4 de

For single drops, various expressions for the Drag coefficient (CD0) have been proposed
on the basis of experiments (Table 5.2).
Table 5.2 Drag expressions for single drops

Investigator (s) Proposed expression for the Drag coefficient (CD0)

(4 3)X 1.275 2 < Y ≤ 70


Y=
(0.045)X 2.37 Y > 70
Hu and Kintner (1955)

(
Y = C D 0 We P 0.15 , X = Re P 0.15 + 0.75 )
22.2 C D−50.18 We −0.169 d ≤ dc
Re =
0.00418 C D2.091 We −1.81 d > dc
Klee and Treybal (1956)

d c = 0.33 ρ c− 0.14 ∆ρ −0.43 µ c0.30σ 0.24

H = 4 3 Eo M −0.149 (µc 0.0009)−0.14

0.94 H 0.757 2 < H ≤ 59.3


J=
3.42 H 0.441 H > 59.3

µc
U= M − 0.149 (J − 0.857 )
ρc d e

Grace et al. (1976) 4 gd e ∆ρ


CD0 =
3ρ cU 2

Provided ,
Eo < 40, M < 10 −3 , Re > 0.2
and
(de D ) < (0.08 + 0.02 log Re ) 0.2 < Re ≤ 100
or
(de D ) < 0.12 Re > 100
76

Investigator (s) Proposed expression for the Drag coefficient (CD0)

( )
24 1 + 0.1 Re 0.75 Re Re < 1000
Viscous regime
0.45 Re ≥ 1000
2
Ishii and Zuber (1979) C D0 = Eo Distorted regime
3
8
Spherical cap regime
3

As can be seen from Table 5.2, the Drag coefficient (CD0) is usually given in terms of
several dimensionless groups. Typical dimensionless groups are organized in Table 5.3.

Table 5.3 Typical dimensionless groups used in drag/lift coefficient expressions

Dimensionless Group Expression

d eVs 

Bubble Reynolds Number (Re) Re = c

gµ c4 ∆ρ
Modified Morton Number (M) M =
ρ c2σ 3

−1
gµ c4 ∆ρ
Physical Property Group (P) P=
ρ c2σ 3

d eVS2 ρ c
Weber Number (We) We =
σ

gd e2 ∆ρ
Eötvös Number (Eo) Eo =
σ

∇ × U C d e2
Shear Reynolds Number (Re∇) Re∇ = 

c
77

5.3.1.2 Drag on a Drop in the Presence of Adjacent Drops


Analogous to the Hindered Settling phenomena observed in solid-liquid systems, the drag
on a single drop is markedly different from that of a drop moving in the presence of
adjacent drops. The analysis of the motion of single drops can be extended to the case of
multi-drop systems if the differences between the two are properly recognized and
accounted for. There are two important factors that need to be accounted for when
evaluating the drag force on a drop in multi-drop systems:

1. The drop is no longer moving in a purely continuous medium of density ‘ρc’, but
in a two-phase medium of effective density ‘ρe’ which can be given by:

ρ e = ρ d (1 − α c ) + ρ cα c [5.20]

Logically, the buoyancy force on the drop should then be estimated on the basis
of an effective mixture density (ρe), which is lower than the continuous phase
density (ρc) for lighter drops (e.g. kerosene, ρd ≈ 780) that typically rise when
introduced in a dense continuous phase (e.g. water, ρc ≈ 1000). Accounting for a
decreased buoyancy force has the effect of reduced relative velocity of the drop as
buoyancy is the prime driving force for drop movement. However, since all the
drops move together relative to the continuous phase and not relative to the two-
phase mixture, it would be inappropriate to calculate the buoyancy force on the
basis of the effective mixture density (Barnea and Mizrahi, 1973). A possible
solution to this problem is to decrease the drag force experienced by the drop by
multiplying either the drag coefficient expression or the drag force for a single
drop with the volume fraction of the continuous phase (αc), which serves as an
index of how concentrated the two-phase mixture is (Ishii and Zuber, 1979;
Kumar and Hartland, 1985; Chen et al., 2005). This indirectly accounts for the
reduced buoyancy effect in an effective and acceptable manner. Interestingly, an
auxiliary benefit of using this approach is that it ensures that the drag force tends
to zero whenever the continuous phase is not present in the domain (e.g. in the
disengagement section of air-lift reactors).
78

2. The drag force on the drop is directly affected by the presence of adjacent drops.
In particular, the drop experiences more drag while moving in the presence of
adjacent drops because its motion is now impeded by the other drops. Similar to
the effect of mixture density (ρe) discussed above, the increase in drag can be
explained on the basis of an effective mixture viscosity (µe), which could be
calculated in a fashion similar to ‘ρe’. For simplicity and purposes of discussion,
the following rudimentary expression for mixture viscosity will be used:

µ e = µ d (1 − α c ) + µ c (α c ) [5.21]

The fact that the drop is now moving in a medium with viscosity (µe) and not the
pure continuous phase viscosity (µc) causes a reduction in the rise velocity of the
drop. The true drag force corresponds to that which would exist in free movement
of the drop through a fluid medium of higher viscosity as µe > µc (subject to the
constraint µd > µc). A practical example can be used to easily illustrate this
behaviour. Considering the case of a kerosene-water system where kerosene drops
rise relative to water, we have:

If µ d ≈ 0.0024 and µ c ≈ 0.001 µ e > 0.001 ∀ (0 ≤ α c ≤ 1) as long as µ d > µ c [5.22]

A similar approach was employed by Barnea and Mizrahi (1973) and Ishii and
Zuber (1979), where different formulations for the mixture viscosity was used to
define a Reynolds number which was in turn used in a suitable drag coefficient
correlation.

The drag force on a drop moving in the presence of adjacent drops can be written in a
fashion similar to Equation 5.19:

3 C DM α d ρ cV S2
FD = [5.23]
4 de

Where, the new drag coefficient (CDM) takes into account the effect of both, reduced
buoyancy and increased drag due to the presence of adjacent drops, as discussed above.
While most investigators (Ishii and Zuber, 1979; Kumar and Hartland, 1985; Rusche and
79

Issa, 2000; Behzadi et al., 2004) have accounted for the reduced buoyancy effect using
the continuous phase volume fraction (αc) in their definition for the drag coefficient
(CDM), the increased drag due to the presence of adjacent drops has been accommodated
in different ways. As already mentioned, Ishii and Zuber (1979) used a mixture viscosity
approach to achieve this effect. On the other hand, Kumar and Hartland (1985) proposed
an empirical expression connecting the relative velocity between the immiscible liquid
phases to the dispersed phase holdup, on the basis of 998 published experimental results
for 29 liquid-liquid systems from 14 different data sources. They reported good
agreement between experimental and predicted relative velocities over a wide range of
holdup (0.01≤ αd ≤ 0.76) and bubble Reynolds numbers (0.16 ≤ Re ≤ 3169).

Table 5.4 Drag expressions for drops which take into account the presence of adjacent
drops

Investigator (s) Proposed expression for the Drag coefficient (CDM)

d eV s ρ c
µ m = µ c (1 − α d )− 2.5(µ d + 0.4 µc ) (µ d + µc ) and Re m =
µm
24
C DM = Stokes regime
Re m

C DM =
24
Re m
(
1 + 0.1 Re 0m.75 ) Undistorted particle regime
Ishii and Zuber (1979) 2
1 + 17.67[ f (α d )]
6
µc
f (α d ) = 1 − α d
7
(Dense fluid particles) and E (α ) =
µm 18.67 f (α d )
C DM = 0.45 E (α ) } Newton ' s regime

Eo (E (α ))
2
C DM = Distorted particle regime
3

C DM =
8
(1 − α d )2 Churn turbulent flow regime
3

Kumar and Hartland


C DM = 0.53 +
24
Re
(1 + 4.56 α d
0.73
)
(1985)

Rusche and Issa (2000) C DM = C D 0 f (α ) where f (α ) = exp (2.10 α d ) + α d0.249


80

More recently, Rusche and Issa (2000) proposed a correction to the ‘single drop’ drag
coefficient (CD0) which increased with an increase in the dispersed phase holdup (αd).
They developed this correction factor on the basis of a comprehensive assessment of the
various drag models available in the literature, which involved comparing several model
predictions with an extensive set of experimental data. Similar to Ishii and Zuber (1979),
their correction ensures that the correction factor for the drag coefficient tends to unity as
the dispersed phase volume fraction approaches zero, thus reverting to the case of a
single drop. The formulations discussed above are presented in Table 5.4.

5.3.2 Non-drag Lateral Forces in Dispersed Liquid-Liquid Flows


Non-drag lateral forces can be broadly classified into wall lubrication forces, lift forces
and turbulent dispersion. Unlike gas-liquid systems where the effects of these forces have
been studied extensively (Antal et al., 1991; Jakobsen et al., 1997; Hjertager, 1998;
Cokljat et al., 2000; Lucas et al., 2001; Troshko et al., 2001; Troshko and Hassan, 2001;
Tomiyama et al., 2002; Behzadi et al., 2004), very few studies focusing on non-drag
lateral forces encountered in liquid-liquid dispersions were found in the open literature.
Nevertheless, some liquid-liquid experimental investigations have speculated the
presence of turbulent dispersion (Domgin et al., 1997; Soleimani et al., 1999) and lift
forces (Farrar and Bruun, 1996; Soleimani et al., 1999). Unfortunately, there were no
reports discussing the nature of wall lubrication forces in the context of liquid-liquid
systems. While Farrar and Bruun (1996) attributed the appearance of near-wall peaks in
their dispersed phase volume fraction profiles to the action of radial lift forces, Soleimani
et al. (1999) offered an explanation for the observed accumulation of drops in the center
of a horizontal pipe, based on the combined action of turbulent dispersion and lift forces.
Furthermore, Hamad et al. (1999) attributed discrepancies between experimental holdup
distributions and their CFD predictions to the inadequate treatment of turbulent
dispersion and lift forces in their work. It is therefore important to review the modeling of
turbulent dispersion and lift forces. These are discussed in the following sections.
81

5.3.2.1 Lift Forces


A review of the literature revealed very few investigations pertaining to the lift force on a
liquid drop. In particular, Dandy (1993) conducted numerical studies of steady uniform
and shear flows past stationary drops and particles over a wide range of bubble Reynolds
numbers (0.1 ≤ Re ≤ 100), Weber numbers (0.001 ≤ We ≤ 10) and viscosity ratios (0.001
≤ (µd/µc) ≤ 1000). The numerical simulations of Dandy (1993) were carried out for low
and intermediate Reynolds numbers only. In the low Reynolds number limit (Re < 1), the
lift coefficient was found to vary as Re-0.5 for a constant shear rate. For intermediate
Reynolds numbers (40 ≤ Re ≤ 100), it was observed that the lift coefficient was
approximately constant for non-zero shear rates. In both cases, the lift coefficient was
found to be positive. Unfortunately, no correlation relating the lift coefficient to the
bubble Reynolds number was presented in the work of Dandy (1993). However,
analogous to the drag force, the lift force on a rigid sphere has been the subject of
numerous investigations (Eichhorn and Small, 1964; Saffman, 1965; Auton, 1987; Drew
and Lahey, 1987; Auton et al., 1988; Drew and Lahey, 1990; Mei et al., 1992; Moraga,
1998; Moraga et al., 1999). According to the inviscid theory, the lift force on a sphere
depends on the diameter of the sphere, the relative velocity between the sphere and the
fluid, and the average vorticity at the sphere’s centroid (Moraga et al., 1999). When
incorporated in a CFD model, the lift force assumes the following form (Equation 5.24),
where the volume of the sphere is replaced by the volume fraction of the dispersed
phase (α p ) :

F = −C ρ α v − v × ∇ × v [5.24]
lift L q p q p q

This expression was independently derived by both Drew and Lahey (1987, 1990) and
Auton (1987). They determined that the value of lift coefficient (CL) in the above
expression is 0.5 for inviscid flow around a rigid sphere. The presence of a shear field
around a rigid particle can exert a lift force on the entity in either direction (i.e. towards
the low velocity region or the high velocity region) depending on the relative velocity
distribution (as clearly discussed by Matas et al., 2004). This can be explained
82

graphically using Figure 5.4, where two different cases of simple shear flow in a pipe are
considered:

1. CASE I: Where the dispersed entity is moving slower than the continuous liquid
phase (i.e. C < A and C < B ). This is equivalent to the case of a shear flow

around a tethered rigid sphere.


2. CASE II: Where the dispersed entity is moving faster than the continuous liquid
phase (i.e. C > A and C > B ). This is equivalent to the case of a lighter dispersed

phase (e.g. air bubbles or kerosene drops) rising in a slow shear flow of a dense
continuous phase (e.g. water).

If the difference in magnitude of the liquid and dispersed phase velocity vectors is
analyzed, it is seen that for CASE I, the relative velocity distribution is qualitatively
similar to the liquid velocity distribution, whereas for CASE II, an exactly opposite trend
is predicted. According to Bernoulli’s principle, the total energy of the fluid along a
streamline is constant in the absence of any losses. Hence, a velocity gradient in the free
stream approaching the entity will result in a pressure difference about the entity. Thus,
for instance in CASE II, the pressure above the entity is higher than the pressure below
the entity. This results in an unbalanced force that acts to push the entity towards the
wall. This phenomenon was first observed by Saffman (1956, 1965) who obtained the lift
force on a sphere moving through a simple shear flow.

In a later experimental study, Hall (1988) reported that Saffman’s observation was true
for higher Reynolds numbers as well. Recent studies however, indicate that the inviscid
lift force may not be the only lift force experienced by a dispersed entity in shear flow
(Taeibi-Rahni and Loth, 1996; Loth et al., 1997; Moraga et al., 1999).

Many researchers (Wang et al., 1987; Kariyasaki, 1987; Bel Fdhila, 1991; Lahey et al.,
1993; Grossetete, 1995) observed that in order to match their experimental data, the value
of the lift coefficient needed to be significantly less than the default inviscid value of 0.5,
and in some cases even negative. In a recent study, Moraga et al. (1999) explored the
causes for this apparent sign reversal of the lift force.
83

CASE I CASE II

Axis Axis

A A
C C
B Dispersed entity
B
Dispersed entity

Wall Wall

A A
C C

C−A C−A

B B
C C

C−B C−B

Relative velocity distribution Relative velocity distribution

High velocity Low pressure Low velocity High Pressure

C−A C−A
Inviscid Lift
force
Inviscid Lift

C−B C−B force

Low velocity High Pressure High velocity Low pressure

Figure 5.4 Inviscid lift forces experienced by an entity in shear flow


84

They attributed this phenomenon to the mechanism of vortex shedding (wake effects).
Their explanation for the same is discussed below. When a vortex is shed, the space it
occupied behind the dispersed entity is replenished by the surrounding liquid moving in.
Typically, this liquid moves slower than the rotational velocity of the vortex. As the
incoming liquid has to make a sharp turn to occupy the volume immediately behind the
dispersed entity, its velocity is significantly reduced. Again, according to Bernoulli’s
principle, this velocity reduction generates an increase in the pressure, which gives rise to
a transient lateral force on the dispersed entity.

If after averaging over a sufficiently long period of time, the wake is asymmetric (with
respect to a plane perpendicular to the direction of the lateral force, parallel to the main
direction of the flow and passing through the center of the dispersed entity), then an
unbalanced lateral force will act on the entity. Based on the observations of Sakamoto
and Haniu (1995) who investigated lateral forces on a fixed sphere in uniform shear flow,
and results from their own experiments, Moraga et al. (1999) proposed that vortex
shedding induced lift forces act in a direction opposite to that predicted by the inviscid
lift theory. They also determined that the lift force due to vortex shedding has the same
functional dependence as inviscid lift (Equation 5.24). This is not surprising considering
the fact that most published data on dispersed two-phase flows (Wang et al., 1987; Lahey
et al., 1993; Alajbegovi et al., 1994) suggest values of the lift coefficient that range from
0.01 to 0.1, which are considerably lower than the value predicted by the inviscid theory
(0.5), thus implying that wake effects induce lift forces opposing inviscid lift, but with
the same functional dependence. It is therefore possible to accommodate wake effects by
choosing an appropriate (effective) lift coefficient that takes into account both the
inviscid and vortex shedding contributions.

For a long time, the lift coefficient was treated as a constant, independent of the local
relative velocity, equivalent diameter etc., due to the lack of experimental or theoretical
work that could quantify it as a function of such variables. In practice however, the lateral
force on a dispersed entity depends on local flow parameters. On the basis of their
experiments on a tethered rigid sphere submerged in a turbulent uniform shear flow of
85

water, Moraga et al. (1999) proposed an expression for the effective lift coefficient that
correlated with the product of bubble and shear Reynolds numbers (ReRe∇). Both
dimensionless numbers (Re and Re∇) are calculated based on local flow variables (Table
5.3). Their expression (Equation 5.25) is valid in the range 6000 ≤ Re Re ∇ ≤ 50000000

only for flows with characteristic turbulence length scales smaller than the size of the
dispersed entity.

C L = [0.12 − 0.2 exp(− Re Re ∇ 360000 )] exp(Re Re ∇ 30000000 ) [5.25]

On the other hand, some studies have produced evidence that the non-dimensional
product (Re Re∇) is by itself, insufficient to describe the behaviour of the lift coefficient
(CL) since the bubble Reynolds number (Re) which largely determines the wake
structure, plays an equally important role. In particular, Eichhorn and Small (1964) and
later, Dandy and Dwyer (1990) found that the lift coefficient does not correlate with the
product (ReRe∇) for Re < 250. This was attributed to the bubble Reynolds number being
too small for the hypothesis of the inviscid theory to be valid. In light of these findings
and based on their own observations, Moraga et al. (1999) concluded that only the wake
and inertia effects on the dispersed entity were correlated by the product (ReRe∇). As a
result, viscous contributions were not taken into account in Equation 5.25.

Nonetheless, Moraga et al. (1999) advocate the use of their correlation in two-fluid
models as it would eliminate the need for fitting global lateral lift coefficients to
experimental data, which limits a priori prediction capabilities. In a recent study, Troshko
et al. (2001) successfully validated a modified version of the correlation proposed by
Moraga et al. (1999) with experimental data sets of bubbly flows in three different
geometries. Their modifications are discussed later in Section 6.1.2.1.

5.3.2.2 Turbulent Dispersion


In turbulent two-phase flows, dispersed entities are transported from regions of high
concentration to regions of low concentration by the turbulent eddies in the continuous
phase. This phenomenon is referred to as turbulent dispersion.
86

In the context of dispersed liquid-liquid systems, Soleimani et al. (1999), who performed
experiments to quantify phase distribution in horizontal pipe flows, observed that at high
mixture velocities, oil drops tend to concentrate in the center of the pipe. They attributed
this coring behavior to the combined effect of inviscid lift forces and turbulent
dispersion. In particular, they reported that turbulent dispersion was strongest near the
wall. As a result, drops in the vicinity of the wall tend to be moved out into the core
where the turbulence was relatively lower. They also reported that drops in the core
region had the same diffusivity as the continuous phase.

In another investigation, Domgin et al. (1997) predicted the radial dispersion of drops in a
turbulent flow field by accounting for turbulent dispersion within the Lagrangian
framework. Later, Hamad et al. (1999) speculated that the observed near-wall peaking
tendency in 3 ID vertical pipes at intermediate dispersed phase holdups was due to the
combined action of reduced turbulent dispersion and stronger lift forces. It can therefore
be inferred from the above discussion that turbulent dispersion effects can be quite
significant in dispersed liquid-liquid flows.

Several models are available in the open literature that account for turbulent dispersion.
Some of the commonly used ones are those of Simonin and Viollet (1990), Lopez de
Bertodano (1992), Gossman et al. (1992), Carrica et al. (1999), Behzadi et al. (2001) and
the favre averaged drag model proposed by Burns et al. (2004). Most of these models
have a user-defined non-dimensional empirical constant, usually referred to as the
dispersion coefficient, which can be used as a means to control the degree of turbulent
dispersion.

Equation 5.26 shows the model proposed by Lopez de Bertodano (1992) which is
typically used as a baseline model to aid the development of other sophisticated models.

M TD = CTD ρ q k q ∇α p [5.26]

where, ‘ CTD ’ is the turbulent dispersion coefficient which varies in the range 0.1 ≤ CTD ≤
0.5. In order to derive the above expression for the turbulent dispersion force, an analogy
87

was made with the thermal diffusion of air molecules in the atmosphere. The fact that air
molecules are not held against the ground by gravity can be attributed to the thermal
kinetic energy of the air molecules. Thermal motion that keeps the molecules away from
the ground can in turn be considered to be equivalent to a thermal diffusion force that acts
against gravity. The dispersed phase (drops) in a turbulent liquid-liquid mixture can
therefore be expected to behave like the air molecules in constant motion, except that
their motion is now caused by the turbulence in the carrier (continuous) phase instead of
the thermal energy of the air molecules.

Burns et al. (2004) present a comprehensive comparative evaluation of various turbulent


dispersion models found in the literature. In the present study, the model proposed by
Simonin and Viollet (1990), which is inbuilt in Fluent 6.1.22, is used to account for
turbulent dispersion. Justification for using this model is provided in Section 6.1.3, where
it is compared with the baseline model proposed by Lopez de Bertodano (1992) and
shown to be equivalent in predictions. A synopsis of the Simonin and Viollet (1990)
model is provided below.

5.3.2.2.1 A Synopsis of the Simonin and Viollet (1990) Turbulent Dispersion Model
The present study accommodates the turbulent dispersion effect using the model
proposed by Simonin and Viollet (1990). This is achieved by introducing an additional
contribution in the momentum equation. A detailed description of the same follows.

Time and length scales that characterize the fluid motion are used to evaluate correlation
functions and the turbulent kinetic energy of the dispersed phase:

The characteristic time ( τ t ,q ) of the energetic eddies in turbulent flows is defined as:

3 kq
τ t ,q = C µ [5.27]
2 εq

The length scale of the turbulent eddies ( Lt ,q ) is calculated using the expression:
88

3
3 kq 2
Lt ,q = Cµ [5.28]
2 εq

The Lagrangian integral time scale ( τ t , pq ), defined as the time a dispersed entity spends

within a turbulent eddy, is calculated along particle trajectories and is mainly affected by
the crossing trajectory effect (Csanady, 1963). This time scale is given as follows:

τ t ,q
τ t , pq = [5.29]
(1 + C βξ )
2

Where,

v pq τ t ,q
ξ= [5.30]
Lt ,q

C β = 1.8 − 1.35 cos 2 θ [5.31]

Here, ‘ ’ is the angle between the mean dispersed phase velocity and the mean relative
velocity between the phases. The characteristic particle relaxation time ( τ F , pq ) connected

with inertial effects acting on a dispersed phase ‘p’ is given by:

ρp
τ F , pq = α p ρ q K pq
−1
+ CV [5.32]
ρq

The ratio between the lagrangian integral time scale ( τ t , pq ) and the characteristic particle

relaxation time ( τ F , pq ) is then written as:

τ t , pq
η pq = [5.33]
τ F , pq

Following Simonin and Viollet (1990), the turbulence quantities for the dispersed phase
are calculated as shown below:

The dispersed phase turbulent kinetic energy (kp) is estimated using the parameter ‘b’,
added mass coefficient, CV (= 0.5) and 

pq:
89

−1
ρp
b = (1 + CV ) + CV [5.34]
ρq

b 2 + η pq
k p = kq [5.35]
1 + η pq

The continuous-dispersed phase covariance ( k pq ) is then calculated as:

b + η pq
k pq = 2k q [5.36]
1 + η pq

Using ‘kpq’, the diffusivity (Dp) can be determined from dispersed phase turbulent
viscosity (Dt,pq) and turbulent kinetic energy (kp) in the following fashion:

1
Dt , pq = k pqτ t , pq [5.37]
3

2 1
D p = Dt , pq + k p − b k pq τ F , pq [5.38]
3 3

Turbulent dispersion effects can thus be easily accounted for by adding to the mean drag
in the momentum conservation equation, an additional contribution in the form of a drift
velocity ( vdr ) which depends on the diffusivities (Dp and Dq):

Dp Dq
v dr = − ∇α p − ∇α q [5.39]
σ pqα p σ pqα q

where ‘σpq’ is a user-adjustable dispersion prandtl number and is by default, usually


given a value of 0.75. Setting the dispersion prandtl number to a higher value signifies
lesser turbulent dispersion and vice-versa. When using Tchen’s theory of dispersion in
multiphase flows, it is assumed that Dp = Dq = Dt,pq. When the drift velocity (v dr ) is

multiplied by the exchange coefficient (Kpq), it serves as a correction to the momentum


exchange term for turbulent flows:
90

( )
K pq (v p − v q ) = K pq U p − U q − K pq v dr [5.40]

Equation 5.40 can now be substituted in the momentum conservation equation (Equation
5.5) to account for turbulent dispersion in addition to the drag force.
91

CHAPTER 6

6 RESULTS AND DISCUSSION


This chapter is divided into two major sections. The first section analyzes several
expressions for the interphase forces discussed in chapter 5. In the second section,
selected expressions for the drag coefficient, lift coefficient and turbulent dispersion are
used as interphase closure models to predict phase distribution in vertical liquid-liquid
up-flows corresponding to flow conditions of data sets listed in Table 4.3. The CFD
simulation results obtained are compared with experimental data and discussed as well.

6.1 Analysis of Interphase Forces


A review of the literature revealed three predominant interphase forces that are
commonly encountered in liquid-liquid dispersions. These are drag, lift and turbulent
dispersion. Various correlations proposed to account for each of these forces were
analyzed prior to carrying out the CFD simulations. Preliminary two-phase CFD
simulations were then carried out to determine whether significant differences existed
between the predictions of the various drag expressions. This was done in order to assist
in the selection of a single expression for the drag coefficient that could possibly act as an
appropriate representative for all the expressions. The effect of constant lift coefficients
and the expression proposed by Troshko et al. (2001) on the phase distribution were also
analyzed. Finally, the model of Simonin and Viollet (1990) was compared with the model
proposed by Lopez de Bertodano (1992) in order to identify a suitable approach to
account for turbulent dispersion.

6.1.1 Assessment of Drag Forces


As already discussed in Section 5.3.1.1, several correlations are available to account for
the drag on single drops. It is therefore important to make a comparative evaluation of the
various drag coefficient expressions. Also, as explained in Section 5.3.1.2, the drag on a
92

drop in the presence of adjacent drops differs from a case of the drag on a single drop. As
a result, this section analyzes expressions suited to both situations in detail.

6.1.1.1 Drag on a Single Drop


In order to compare the drag expressions proposed for single drops, the relative (rise)
velocities (VS ) were calculated for various equivalent diameters (d e ) using the following

expression (Equation 6.1) which can be obtained from a rudimentary force balance on the
drop. It should be noted that (VS ) represents the terminal velocity of the drop.

4 gd e ∆ρ
VS = [6.1]
3ρ c C D 0

Various expressions for the single drop drag coefficient (C D 0 ) are listed in Table 5.2. In

order to compare these expressions, programs were written in the ‘C’ language
(Appendix A3.1.3) to calculate the relative velocities of kerosene drops rising in water as
a function of the equivalent diameter, which was varied from 1 to 16 mm. Table 6.1 lists
the physical properties of water (continuous phase) and kerosene (dispersed phase) used
in this analysis.
Table 6.1 Physical data and constants used in the analysis of various drag models

Physical property /
Value
Constants

ρc 998.2 kg/m3

ρd 780.0 kg/m3

σ 0.0404 N/m

µc 0.001003 kg/ms

µd 0.0024 kg/ms

g 9.81 m/s2
93

An iterative solution algorithm based on the Newton-Raphson scheme was adopted for
non-standard algebraic expressions such as the expression for rigid spheres proposed by
Schiller and Naumann (1935). The data generated is graphically presented in Figure 6.1,
where the variation of relative (rise) velocity with equivalent diameter is shown.

Figure 6.1 clearly reveals that all expressions predict similar relative velocities up to an
equivalent diameter of 3 mm. After this point, all the expressions for a single drop
consistently predict lower relative velocities when compared to the expression for a rigid
sphere.

0.35
Klee and Treybal
Hu and Kintner
0.30 Grace et al.
Ishii and Zuber
Schiller and Naumann
0.25
Relative velocity (m/s)

Single rigid spheres


0.20

0.15

0.10

0.05
Single drops

0.00
0 2 4 6 8 10 12 14 16
Equivalent diameter (mm)

Figure 6.1 Comparison of various single-entity drag models

Another observation is that the relative velocities for single drops tend to become
constant after a certain equivalent diameter. On the other hand, the expression for rigid
spheres predicts relative velocities that constantly increase with equivalent diameter.
94

6.1.1.1.1 Comparative Evaluation of Drag Coefficient Expressions for Single


Entities (using CFD Simulations)
In two-phase CFD simulations, an accurate estimation of the drag force is imperative as
this influences the relative velocity between the phases, which in turn directly affects the
average dispersed phase holdup. The predicted dispersed phase holdup was therefore
used to compare the different single entity drag coefficient expressions using CFD
simulations. The experimental conditions of Al-Deen and Bruun (1997) were used as an
example.

As differences in the calculated relative (rise) velocities for all drag models are observed
in the range 2 mm ≤ de ≤ 8 mm (Figure 6.1), three equivalent diameters (2 mm, 5 mm and
8 mm) were selected for the CFD simulations. This range also covers most of the drop
diameters observed in Table 4.1. A dispersed phase ratio of 5% was chosen, primarily to
emulate single-drop conditions as much as possible. Table 6.2 presents the area-weighted
average of dispersed phase holdup obtained from CFD simulations, at the outlet of the 1.5
m pipe (3 ID), for the three different equivalent diameters. In all simulations, the
volumetric flow rate of the continuous phase (water) was the same as that in the actual
experiment, 0.0026 m3/s.
Table 6.2 Average holdup predictions for various single-entity drag models at low phase
ratio (5%)

Expression for CD0 de = 2 mm de = 5 mm de = 8 mm

Schiller and Naumann (1935) 0.04429 0.03894 0.03651

Ishii and Zuber (1979) 0.04405 0.04095 0.04094

Grace et al. (1976) 0.04405 0.04023 0.04055

Hu and Kintner (1955) 0.04432 0.04014 0.04032

Klee and Treybal (1956) 0.04428 0.04208 0.04204

It can be seen from Table 6.2, that at de = 2 mm, all expressions yield similar average
dispersed phase holdups (≈ 0.044). As the equivalent diameter is increased to 5 mm, it is
95

seen that the expression for rigid spheres under-predicts the average dispersed phase
holdup when compared to the predictions of all single drop expressions by approximately
5%. This effect is magnified at de = 8 mm, where percentage differences of in the order of
11 % are observed. These observations indicate that, as the equivalent diameter increases
beyond 2 mm, using the rigid sphere expression results in increasing errors in the
estimation of relative velocity and consequently the average dispersed phase holdup.

To test the predictions of all single-entity drag expressions at higher dispersed phase
holdups, experimental conditions corresponding to data point A30 from the data set of
Al-Deen and Bruun (1997) were chosen. The volumetric flow rate of the continuous
phase was the same as before (i.e. 0.0026 m3/s). Table 6.3 lists the holdup predictions for
different single-entity drag models at high dispersed phase holdups.
Table 6.3 Average holdup predictions for various single-entity drag models at high phase
ratio (30%)

Expression for CD0 de = 2 mm de = 5 mm de = 8 mm

Schiller and Naumann (1935) 0.2832 0.2640 0.2530

Ishii and Zuber (1979) 0.2824 0.2707 0.2707

Grace et al. (1976) 0.2832 0.2680 0.2692

Hu and Kintner (1955) 0.2848 0.2702 0.2678

Klee and Treybal (1956) 0.2828 0.2733 0.2732

A very similar trend is observed when compared to the low holdup simulations, although
the percentage error between the rigid sphere prediction and the single drop predictions is
now slightly lower (≈ 3% at de = 5mm and ≈ 7% at de = 8 mm).

Amongst the single drop expressions, it is seen that the percentage difference in relative
velocity prediction between the expression proposed by Ishii and Zuber (1979) and the
other single drop expressions was in the range of ± 2.5%, at both low and high dispersed
phase ratios. As a result, the expression proposed by Ishii and Zuber (1979) was used as a
96

representative expression for the drag on single drops. It is however important to reach a
similar conclusion when all three interphase forces are in place and comparison with
experiments has been made.

6.1.1.2 Drag on a Single Drop in the Presence of Adjacent Drops


As previously discussed in section 5.3.1.2, it is well known that the drag on a drop is
affected by the presence of adjacent drops. To analyze this behavior, correlations listed in
Table 5.4 were used to study the effect of adjacent drops. This was done by observing the
variation of the relative velocity with increasing dispersed phase holdup. The following
expression for the relative velocity was used:

4 gd e ∆ρ (1 − α d )
VS = [6.2]
3ρ c C DM

This expression for the relative velocity is based on a drag coefficient (C DM ) which
accounts for the reduced buoyancy effect as well as the effect of adjacent entities.

0.09
0.08
Relative velocity (m/s)

0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Dispersed phase holdup (-)

Ishii and Zuber (Dense fluid particles)


Ishii and Zuber (corrected w ith Rusche and Issa (2000)
Kumar and Hartland
Ishii and Zuber (Single drop)

Figure 6.2 Comparison of various drag models that take into account the presence of
adjacent entities at de = 2 mm
97

In order to test the correction factor proposed by Rusche and Issa (2000), the single drop
expression developed by Ishii and Zuber (1979) was used (on the basis of discussions in
Section 6.1.1.1). Also the churn turbulent flow regime in the dense fluid particles model
of Ishii and Zuber (1979) was not considered in this analysis, as all experimental data
points were obtained in the bubbly flow regime.

For each of the correlations listed in Table 5.4, programs were written in the ‘C’ language
(Appendix A3.1.3) to calculate the relative velocity for varying dispersed phase holdups
at different equivalent diameters. Physical properties and constants were maintained the
same as before (Table 6.1). Figures 6.2 and 6.3 present the calculated data for two
different equivalent diameters (2 mm, 5 mm).

0.16
0.14
Relative velocity (m/s)

0.12
0.1
0.08
0.06
0.04
0.02
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Dispersed phase holdup (-)

Ishii and Zuber (Dense fluid particles)


Ishii and Zuber (corrected w ith Rusche and Issa (2000)
Kumar and Hartland
Ishii and Zuber (Single drop)

Figure 6.3 Comparison of various drag models that take into account the presence of
adjacent entities at de = 5 mm
It can be seen that all correlations shown in Figures 6.2 and 6.3, revert to the drag on a
single drop when the dispersed phase volume fraction tends to zero, except for the drag
98

coefficient expression proposed by Kumar and Hartland (1985) which is applicable only
for dispersed phase holdups greater than or equal to 1%.

6.1.1.2.1 Comparative Evaluation of Drag Coefficient Expressions that Account for


the Presence of Adjacent Entities (using CFD Simulations)
In order to verify the trends observed in Figures 6.2 and 6.3, CFD simulations were
carried out using the experimental conditions of Al-Deen and Bruun (1997) at a dispersed
phase ratio of 30%. The simulation results are presented in Table 6.4. The ‘single drop’
drag predictions of Ishii and Zuber (1979) are also shown for comparison.

Table 6.4 Average holdup predictions for various drag models taking into account the
presence of adjacent entities at high phase ratio (30%)

Expression for CDM de = 2 mm de = 5 mm

Ishii and Zuber - Dense fluid particles (1979) 0.2855 0.2751

Kumar and Hartland (1985) 0.2890 0.2797

Ishii and Zuber (1979) drag expression for single drops,


modified to account for the presence of adjacent drops 0.2902 0.2812
using the correction factor of Rusche and Issa (2000)

Ishii and Zuber (1979) drag expression for single drops 0.2824 0.2707

Across the board, it is observed that all three expressions predict similar holdups. It is
also seen that the percentage difference in relative velocity prediction between the
expression proposed by Kumar and Hartland (1985) and the other drag expressions which
account for the presence of adjacent entities, is in the range of ± 1%. In addition, the area-
weighted average of the dispersed phase holdup observed from experiments is
approximately 29%, for the chosen experimental conditions. Comparing this with CFD
predictions in Table 6.4 suggests that considering the drag on a drop by taking into
account the presence of adjacent drops yields a slightly better prediction than the drag
expression for single drops. At this stage, it would appear that the expression proposed by
Kumar and Hartland (1985) would suitably account for the presence of adjacent entities
since its predictions of relative velocity lie between the other two approaches. However,
99

no definite conclusions can be reached at this point without the inclusion of other
interphase forces and subsequent comparison with experimental data.

6.1.2 Assessment of Lift Forces


In the context of liquid-liquid systems, the lift force has rarely been included in CFD
simulations. Its importance however, has been stressed by several investigators as
discussed in section 5.3.2. Table 4.2 clearly shows that non-uniform radial volume
fraction profiles (wall peaking, near-wall peaking and/or coring) are experimentally
observed over a wide range of holdups. As lift forces are primarily responsible for non-
homogeneous radial distribution of the dispersed phase holdup (particularly in turbulent
bubbly flows), it is important to include their effects in CFD simulations of such
experiments. The present section will therefore analyze different values/formulations of
the lift coefficient. In particular, attention will be focused on three major schemes for
prescribing the lift coefficient:

• Constant lift coefficient,

• The expression for lift coefficient proposed by Moraga et al. (1999),


• An approach similar to that of Moraga et al. (1999) in which validity limits for the
lift coefficient expression have been modified in accordance with the
recommendations made by Troshko et al. (2001).

6.1.2.1 Constant Lift Coefficient


The constant lift coefficient is by definition, a single value which can be either positive or
negative. It therefore, does not vary with local hydrodynamic conditions or other flow
properties. Several gas-liquid studies in the past have successfully fitted both positive and
negative constant lift coefficients to experimental data (Wang et al., 1986; Bel Fdhila,
1991; Lahey et al., 1993; Grossetete, 1995). For instance, for fully developed gas-liquid
flows in a vertical pipe, the values of constant lift coefficient that resulted in a good fit to
the experimental data were found to be in the range 0.01 ≤ CL ≤ 0.15 (Wang et al., 1986;
Lahey et al., 1993). Unfortunately, there are no reports of the usage of lift forces in
dispersed liquid-liquid pipe flows. It is however useful to examine the effect of different
100

constant lift coefficients on the radial dispersed phase holdup distribution. To study this,
experimental conditions corresponding to data point (H20), which depicts a near-wall
peak experimentally, were chosen in view of the fact that a near-wall peak is a clear
indication of the presence of lateral forces.
For this data point, the phase ratio of the dispersed phase was 20% and the volumetric
flow rate of the continuous phase (water) was 0.00278 m3/s. The expression for drag
coefficient proposed by Kumar and Hartland (1985), and a constant equivalent diameter
of 6 mm were used throughout the simulations. Lift coefficients in the range -0.01 ≤ CL ≤
0.01 were tested to show the effect of constant (positive and negative) lift coefficients.
Figure 6.4 shows the effect of various constant lift coefficients on the radial distribution
of dispersed phase holdup.

0.5

CL = 0.0 CL = -0.01
0.45

CL = -0.005 CL = -0.001
0.4

CL = +0.001 CL = +0.005
0.35
Dispersed phase holdup (-)

CL = +0.01 H20 - EXP


0.3

0.25

0.2

0.15

0.1

0.05

0
0 0.2 0.4 0.6 0.8 1
Normalized radius (-)

Figure 6.4 Effect of constant lift coefficients on the radial distribution of the dispersed
phase
101

While positive constants (+0.001, +0.005, +0.01) for the lift coefficients tend to predict
increasing wall peaks, negative constants (-0.001, -0.005, -0.01) shift the dispersed phase
holdup profiles towards an increased coring tendency. Referring to Figure 6.4, it is seen
that apart from the asymmetric nature of the holdup profiles when positive and negative
constants of the same magnitude are used, the use of extreme lift coefficients (i.e. +0.01
and -0.01) results in very low or very high dispersed phase holdups at the wall. The flat
holdup distribution for CL = 0.0 corresponds to the case when only drag forces are
accounted for.

6.1.2.2 Expression for CL proposed by Moraga et al. (1999)


As discussed in Section 5.3.2.1, Moraga et al. (1999) developed an expression relating
the lift coefficient (CL) to the product of bubble and shear Reynolds numbers (ReRe∇) as
opposed to proposing a constant lift coefficient (Equation 6.3).
C L = [0.12 − 0.2 exp(− Re Re ∇ 360000 )] exp(Re Re ∇ 30000000 ) [6.3]

The above relationship (Equation 6.3) is presented graphically in Figure 6.5, which
shows the variation of (CL) with the product (ReRe∇).

As the expression was proposed for fluid flow around tethered rigid spheres, CL assumes
a negative value (implying inviscid lift dominance) of approximately -0.07 at low Re Re∇
(≈ 6000). This gradually increases to zero at ReRe∇ ≈ 183897. After this point, CL begins
to assume a positive value implying vortex-shedding dominance. If the expression were
to be applied to drops and bubbles which rise faster relative to the surrounding fluid, it
would be necessary to reverse the direction of inviscid and vortex shedding lift forces on
the basis of the discussion in Section 5.3.2.1. The modified expression which is valid for
drops and bubbles is shown in Equation 6.4.
C L = −[0.12 − 0.2 exp(− Re Re ∇ 360000 )] exp(Re Re ∇ 30000000 ) [6.4]

This would result in an inverted profile for CL as shown by the broken curve in Figure
6.5. As a result, when CL > 0 inviscid lift is predominant and pushes drops/bubbles
towards the wall (i.e. towards the low velocity region) as discussed in Section 5.3.2.1.
102

0.7

0.5

0.3

0.1
CL (-)

-0.1
Moraga et al. (1999)

CL = 0.0
-0.3
Troshko et al. (2001)

ReReÑ = 183897
-0.5
Moraga et al. (1999) applied to drops
and bubbles
-0.7
1 100 10000 1000000 100000000

Re Re∇ (−)

Figure 6.5 Variation of the lift coefficient with Re Re∇ as proposed by Moraga et al.
(1999) and Troshko et al. (2001)

On the contrary, when CL < 0 vortex shedding effects are dominant and drops/bubbles
experience a force pushing them away from the wall.

6.1.2.3 Expression for CL proposed by Troshko et al. (2001)


When the expression proposed by Moraga et al. (1999) as applied to drops and bubbles,
was implemented and tested in a two-fluid model for bubbly flows, Troshko et al. (2001)
reported unphysical predictions and/or convergence difficulties. This was found to be
true in the present liquid-liquid study as well. To overcome this problem, Troshko et al.
(2001) made some modifications to the correlation proposed by Moraga et al. (1999). In
particular, the upper limit of validity for the product Re Re∇ was reduced from 50000000
to 190000. The value of CL at this new upper limit was obtained by substituting Re Re∇ =
103

190000 in Equation 6.4. The resulting upper limit value of CL (≈ -0.02) is maintained ∀
Re Re∇ 190000 in the modified expression. The lower validity limit was also modified
by setting CL = 0.07671 ∀ Re Re∇ ≤ 6000. In the intermediate range (6000 < Re Re∇ <
190000), CL was calculated using Equation 6.4. The modified expression proposed by
Troshko et al. (2001) is summarized in Equation 6.5.

+ 0.0767 Re Re∇ ≤ 6000


[ ( )] (
CL = − 0.12 − 0.2 exp − Re Re∇ 36 ∗104 exp Re Re∇ 3 ∗107 ) for 6000 < Re Re∇ < 190000
[6.5]
− 0.002 Re Re∇ ≥ 190000

The modified expression was also successfully validated by Troshko et al. (2001) with
bubbly flow experiments (Wang et al., 1987; Lopez de Bertodano, 1992; Bel Fdhila et al.,
1990) in three different geometries (i.e. vertical pipe, vertical triangular duct and sudden
pipe expansion).

To demonstrate the effect of the modified lift coefficient correlation proposed by Troshko
et al. (2001) on the dispersed phase holdup, the data point F30, from the experimental
data set of Farrar (1988) was chosen as it predicts a near-wall peak experimentally. For
this data point, the phase ratio of the dispersed phase was 30% and the total mixture flow
rate (QT = QC + QD) was 0.003083 m3/s. The drag expression proposed by Kumar and
Hartland (1985) and an equivalent diameter of 5 mm were used throughout the
simulations.

Figure 6.6 compares the expression proposed by Troshko et al. (2001) with different
constant lift coefficients. Positive lift coefficients were chosen for comparison as they
typically predict wall peaks as opposed to negative lift coefficients (e.g. as shown in
Figure 6.4). Although a high dispersed phase concentration is predicted by both the
constant lift coefficients, and the expression proposed by Troshko et al. (2001) in the
near-wall region, there are significant differences that can be observed. In particular, the
expression proposed by Troshko et al. (2001) ensures that a physically reasonable near-
wall peak is predicted as opposed to constant lift coefficients, which predict a higher non-
zero dispersed phase concentration at the wall.
104

0.5
0.45 Wall peak

Dispersed phase holdup (-)


0.4
0.35 Near-wall peak
0.3
0.25
Troshko et al. (2001)
0.2
CL = +0.001
0.15
CL = +0.005
0.1 CL = +0.01
0.05 exp F30
0
0 0.2 0.4 0.6 0.8 1
Normalized radius (-)

Figure 6.6 Comparison of the expression proposed by Troshko et al. (2001) with
constant lift coefficients
CL = Troshko et al. (2001) CL = 0.0
CL = -0.005 CL = -0.001
CL = -0.01 CL = 0.01
CL = 0.005 CL = 0.001

0.09
0.08
0.07
0.06
Lift coefficent (-)

0.05
0.04
0.03
0.02
0.01
0
-0.01
-0.02
0 0.2 0.4 0.6 0.8 1
Normalized radius (-)

Figure 6.7 Radial distribution of the lift coefficient for the expression proposed by
Troshko et al. (2001) as compared to constant lift coefficients
105

To achieve this effect using constant lift coefficients some investigators have introduced
a pseudo wall lubrication force that drives the dispersed phase away from the wall (Kuo
et al., 1997; Lucas et al., 2001). However, as pointed out by Troshko et al. (2001) this
additional pseudo force is unnecessary if the modified expression for the lift coefficient
(Equation 6.5) is used, as it takes into account the pressure differences caused by the
mean liquid velocity gradient (i.e. the lubrication effect).

Figure 6.7 shows the lift coefficient distribution for data point F30, as predicted by the
expression proposed by Troshko et al. (2001). Also shown are the other constant positive
and negative lift coefficients. It can be seen that, for the expression proposed by Troshko
et al. (2001), the lift coefficient is positive for a large portion in the core region where
inviscid lift is predominant. However, in the near-wall region (r/R > 0.9) it attains a
negative value signifying vortex-shedding dominance. This lift coefficient distribution is
responsible for the distinct ‘near-wall’ peak observed in Figure 6.6.

6.1.3 Assessment of Turbulent Dispersion


In the context of dispersed liquid-liquid systems, the importance of turbulent dispersion
has been recognized by quite a few investigators (Domgin et al., 1997; Soleimani et al.,
1999; Hamad et al., 1999). As discussed in Section 5.3.2.2, there are several models
proposed to account for turbulent dispersion effects. However, a detailed comparison of
all existing models is beyond the scope of this thesis. As all models work towards
homogenizing the dispersed phase distribution depending on the intensity of turbulence,
the effect of turbulent dispersion will be demonstrated using just two models (Simonin
and Viollet, 1990 and Lopez de Bertodano, 1991) as an example. For purposes of
studying the effect of turbulent dispersion, the F30 data point was used for the same
reason as before (i.e. the presence of a near-wall peak). Once more, an equivalent
diameter of 5 mm and the drag expression proposed by Kumar and Hartland (1985) were
employed. The expression for the lift coefficient proposed by Troshko et al. (2001) was
used this time, to demonstrate changes in the wall peak as turbulent dispersion is
increased. Figures 6.8 and 6.9 show the effect of increased turbulent dispersion on the
106

dispersed phase distribution, using the models proposed by Lopez de Bertodano (1992)
and Simonin and Viollet (1990) respectively.

0.4

0.35

Dispersed phase holdup (-)


0.3

0.25

0.2
CTD 0.1 CTD 0.4
0.15

0.1
CTD 10
0.05

0
0 0.2 0.4 0.6 0.8 1
Normalized radius (-)

Figure 6.8 Effect of various dispersion coefficients (CTD) on the dispersed phase holdup
for the Lopez de Bertodano (1992) model

0.4

0.35
Dispersed phase holdup (-)

0.3

0.25

0.2

0.15 DPN 0.075 DPN 0.75

0.1
DPN 7.5
0.05

0
0 0.2 0.4 0.6 0.8 1
Normalized radius (-)

Figure 6.9 Effect of various dispersion prandtl numbers (DPN) on the dispersed phase
holdup for the Simonin and Viollet (1990) model
107

It is clear from the figures that both models exhibit similar behavior. As turbulent
dispersion is increased, the wall peak starts to diminish resulting in a more homogeneous
profile for the dispersed phase holdup. The only discernable difference between the two
models is that, both have different ways of specifying the degree of turbulent dispersion.
While the model proposed by Lopez de Bertodano (1992) uses a higher Dispersion
coefficient (CTD) to account for increased turbulent dispersion, the Dispersion prandtl
number (DPN) can be lowered to account for the same effect using the model proposed
by Simonin and Viollet (1990).

In other words, the ‘Dispersion Prandtl number’ can be thought of as an inverse


‘Dispersion coefficient’ or vice-versa. It should however be noted that it is possible for
both models to yield the same phase distribution for a particular set of different constants
for (CTD and DPN), which can be obtained through optimization studies employing a
finer search grid. Burns et al. (2004) have performed a similar study to examine the
differences between several turbulent dispersion models. As the model of Simonin and
Viollet (1990) was inbuilt in Fluent 6.1.22, it was conveniently used to account for
turbulent dispersion in the present study.

6.1.4 Re-assessment of the Drag Force taking into account other Interphase
Forces
Having assessed several models for lift and turbulent dispersion, it is now possible to
select an appropriate drag model for use in CFD simulations of liquid-liquid dispersions.
Clearly this drag model should be capable of accounting for the presence of adjacent
drops.

In Section 6.1.1.2, it was speculated that the expression for drag coefficient proposed by
Kumar and Harland (1985) could be a suitable candidate. However, it was difficult at that
point to decisively make such a choice. In the present section, CFD simulations were
again carried out using data point A30 for comparing the various drag models that take
into account the presence of adjacent drops. Experimental conditions corresponding to
data point A30 were chosen because of two reasons. Firstly, data point A30 represents a
high phase ratio (30%) of the dispersed phase and therefore, the effect of adjacent entities
108

is expected to be prominent. Secondly, it also allows for easier comparison with the
predictions of other drag coefficient expressions (e.g. in Tables 6.3 and 6.4). Both the lift
forces and turbulent dispersion were accounted for this time, using the models proposed
by Troshko et al. (2001) and Simonin and Viollet (1990). A constant Dispersion Prandtl
number of 0.75 was used throughout.

Table 6.5 Average holdup predictions for various drag models at high phase ratio (30%),
taking into account the combined effect of drag, lift and turbulent dispersion

Expression for CDM de = 2 mm de = 5 mm

Ishii and Zuber - Dense fluid particles (1979) 0.2916 0.2746

Kumar and Hartland (1985) 0.2952 0.2796

Ishii and Zuber (1979) single-entity drag modified


for the presence of adjacent drops using the
0.2965 0.2812
correction factor proposed by Rusche and Issa
(2000)

The results of the CFD simulations are presented in Table 6.5. It is seen that at both
equivalent diameters (2 mm and 5 mm), all models predict fairly close values. Once
again, it is seen that the expression for drag coefficient as proposed by Kumar and
Hartland (1985) predicts an average dispersed phase holdup that lies between the
predictions of the other two models. It should also be noted that the expression proposed
by Kumar and Harland (1985) was developed after a comprehensive analysis of an
extensive set of experimental data, and is valid over a wide range of holdups (1 – 76 %)
and bubble Reynolds numbers (0.16 - 3169). In view of the above facts, the expression
for drag coefficient proposed by Kumar and Hartland (1985) was chosen as a
representative of the three drag models in Table 5.4, and was therefore used to account
for the drag force between the dispersed and continuous phase throughout the present
study.
109

6.2 CFD Simulation of Liquid-Liquid Dispersions


This section presents CFD simulation results for turbulent liquid-liquid up-flows in a
vertical pipe. Experimental conditions corresponding to the selected data points listed in
Table 4.3 were used in this study. In all simulations, the combined effect of drag, lift and
turbulent dispersion was considered using the interphase models assessed in Section 6.1.
On the basis of the discussion in Section 6.1.4, the drag coefficient expression proposed
by Kumar and Hartland (1985) was used throughout the simulations. Attention was
therefore focused on lift forces and turbulent dispersion. The effect of turbulent
dispersion on the radial phase distribution was discussed in Section 6.1.3, where it was
shown that the prime effect of turbulent dispersion was to homogenize the phase
distribution across the pipe cross-section. Lift forces on the other hand, significantly
contribute towards non-uniform phase distributions (Figure 6.4). On the basis of these
observations, it can be concluded that lift forces and turbulent dispersion oppose each
other. Therefore, an attempt was made to find a suitable combination of lift and turbulent
dispersion coefficients (for the experimental conditions of each data point) that would
accurately predict flow quantities such as the dispersed phase holdup distribution across
the pipe cross-section, individual phase velocities etc. This was done by comparing each
simulation with experimental data. Wherever possible, the residual sum of squares was
used to compare different simulations with the experiment in order to identify the lift and
turbulent dispersion coefficients that yield the best match statistically.

Table 6.6 lists the selected data points and relevant flow conditions used for conducting
the simulations. It can be seen that the 17 individual data points listed in Table 6.6 span a
wide range of equivalent drop diameters (1 – 5 mm). As exact mean diameters for the
dispersed phase were not prescribed by any of the investigators in their experiments, the
equivalent drop diameters had to be guessed. A rough estimate of the equivalent diameter
was obtained by calculating the average slip velocity (VS ) from the average dispersed

phase holdup as reported in the experiments (α d ,exp ) using Equation 6.6:


110

x d − α d ,exp
VS = VM
α d ,exp (1 − α d ,exp )
[6.6]

where ‘ xd ’ is the phase ratio of the dispersed phase and ‘ VM ’ is the average velocity of

the mixture, defined as the sum of superficial velocities of both phases. The equivalent
diameter (d e ) was then estimated using Equation 6.2 and the expression for drag

coefficient as proposed by Kumar and Hartland (1985). It should be noted that Equation
6.6 can be derived for two-phase pipe flows using the fundamental definition of slip
velocity as shown by Vigneaux et al. (1988). However, care was taken to ensure that the
equivalent drop diameter used in the simulations was well within the experimentally
observed range as stipulated by each investigator. The reason for placing additional
importance on the choice of drop diameter is explained below.

It is apparent that an increase in the drop diameter decreases the average dispersed phase
holdup when only the drag force is present (Table 6.5). However, when lateral forces also
exist, the effect of varying drop diameters on the phase distribution has to be investigated.
This was done by again carrying out CFD simulations as discussed below.

In order to observe the effect of varying drop diameter when lateral forces exist, it is
desirable to choose a data point that features an inhomogeneous phase distribution. For
this purpose, data point F25 which features a near-wall peak was selected as an example.
Simulations were conducted using the drag coefficient expression proposed by Kumar
and Hartland (1985) and a constant dispersion Prandtl number (0.75) throughout. The
simulation results are presented in Figures 6.10 and 6.11 for two cases (constant lift
coefficient and the expression proposed by Troshko et al., 2001), where the predicted
phase distributions for varying equivalent diameters (1 to 5 mm) are compared. It can be
inferred from Figure 6.10 that for a constant lift coefficient (0.005 in this case) an
increase in the equivalent drop diameter results in an increase in the position of the wall
peak for the dispersed phase holdup.
111

0.35

0.3

Dispersed phase holdup (-)


0.25

0.2

0.15 de = 5 mm de = 4 mm

0.1 de = 3 mm de = 2 mm

0.05
de = 1 mm
0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.10 Effect of varying drop diameter on the holdup for CL = 0.005

0.6
de = 5 mm de = 4 mm
0.5
Dispersed phase holdup (-)

de = 3 mm de = 2 mm
0.4

de = 1 mm
0.3

0.2

0.1

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.11 Effect of varying drop diameter on the holdup for the lift coefficient
expression proposed by Troshko et al. (2001)
112

0.09
0.08
0.07
0.06
Lift coefficient (-)

0.05
de = 5 mm de = 4 mm
0.04
0.03 de = 3 mm de = 2 mm

0.02
de = 1mm CL = 0.0
0.01
0
-0.01
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.12 Lift coefficient distribution for the expression proposed by Troshko et al.
(2001)
On the other hand, when the expression proposed by Troshko et al. (2001) is used, a
significantly different trend is observed. In particular, it can be observed from Figure 6.11
that the holdup at the wall first increases for a while (up to a drop diameter of 3 mm) and
then starts to decrease as the drop diameter approaches 5 mm. This can be explained by
analyzing the lift coefficient distribution for each equivalent drop diameter (Figure 6.12),
where it is seen that the lift coefficient at the wall is positive up to de = 3 mm after which
it changes sign. Physically, around this point (i.e. for 3 ≤ de ≤ 4) vortex-shedding induced
lift forces start opposing the inviscid lift forces in the near-wall region..

The equivalent drop diameter can therefore easily influence the phase distribution
especially when using expressions for the lift coefficient that account for changes in the
direction of lift forces.
113

6.2.1 Simulation Results


The present section discusses the simulation results obtained for experimental conditions
corresponding to data points listed in Table 6.6. As discussed in Section 4.1.1, it is
difficult to propose a general conclusive behavior for most phase distribution trends
observed in the experiments. As a result, the following discussion of simulation results
are organized according to the each experimental data set for both low and high dispersed
phase holdups.
Table 6.6 Summary of experimental flow conditions for selected data points

Data Set Data Continuous phase Dispersed phase Average Equivalent drop
Point superficial superficial velocity experimental diameter (mm)
velocity (m/s) (m/s) dispersed phase
holdup (-)

F20 0.4935 0.1363 0.1912 5.00


Farrar and
F25 0.4634 0.1637 0.2275 5.00
Bruun (1996)
F30 0.4263 0.1972 0.2783 5.00

Hamad et al. H10 0.5855 0.0651 0.0873 3.25

(2000)
H20 0.5855 0.1464 0.1764 3.50

A5 0.5441 0.0286 0.0493 3.00

Al-Deen and A10 0.5441 0.0605 0.0917 3.50

Bruun (1997)
A20 0.5441 0.1360 0.1872 4.00

A30 0.5441 0.2332 0.2992 4.00

L20A 0.4000 0.1000 0.1851 1.00

L20B 1.2000 0.3000 0.1809 2.00


Lang (1999)
L40 0.3000 0.2000 0.3692 1.00

L60 0.2000 0.3000 0.5474 1.00

V5 0.2268 0.0119 0.0323 2.50

Vigneaux et V10 0.2149 0.0239 0.0661 2.75

al. (1988)
V30 0.1671 0.0716 0.2350 5.00

V50 0.1194 0.1194 0.4308 5.00


114

6.2.1.1 Data Set of Farrar and Bruun (1996)


Figure 6.13 shows the simulated volume fraction profiles for data points (F20, F25 and
F30). Related experimental data are shown as well.

0.4

0.35

0.3
Dispersed phase holdup (-)

0.25

0.2

0.15
F20 - SIM F20 - EXP

0.1
F25 - SIM F25 - EXP

0.05 F30 - SIM F30 - EXP

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.13 Simulated phase distribution profiles for the data set of Farrar and Bruun
(1996), (DPN = 7.5; CL = Troshko et al. (2001); de = 5 mm)
Preliminary simulations showed that constant (positive) lift coefficients which typically
predict high dispersed phase concentrations in the near-wall region, could not capture the
experimentally observed near-wall peak in the dispersed phase holdup profiles (Figure
6.6). Although certain values of negative lift coefficients did predict a near-wall peak
(Figure 6.14), they were unable to accurately capture its position and trend even at high
dispersion Prandtl numbers (≈ 7.5), which signify reduced turbulent dispersion effects.

On the contrary, the expression for the lift coefficient proposed by Troshko et al. (2001)
was found to give the best predictions, as it appropriately reverses the sign of the lift
coefficient close to the wall (Figure 6.7). The physical significance of this sign reversal
115

can be explained on the basis of vortex-shedding (wake) effects as discussed in Section


5.3.2.1.

0.35

0.3

Dispersed phase holdup (-)


0.25

0.2
F25 - SIM (CL = -0.001)
0.15
F25 - SIM (CL = -0.005)
0.1
F25 - SIM (CL = -0.01)

0.05
F25 - EXP

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.14 Effect of constant (negative) lift coefficients on the dispersed phase holdup,
(DPN = 7.5, de = 5 mm)

1.2

1
Normalized velocity (m/s)

0.8

0.6
F20 - SIM

0.4
F20 - EXP

0.2

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.15 Comparison of experimental and simulated continuous phase velocity


profiles for the data set of Farrar and Bruun (1996), (DPN = 7.5; CL = Troshko et al.
(2001); de = 5 mm)
116

Setting the dispersion Prandtl number to a high value (≈ 7.5) was found to accurately
predict the position of the wall peak in all three cases. The effect of varying turbulent
dispersion has already been discussed in Section 6.1.3.

Among the three data points, the prediction for the F30 data point is very good. However,
there are minor discrepancies observed for the other data points (F20 and F25), especially
the under-prediction in the central region between the pipe centerline (core) and the wall,
particularly before the near-wall peak. It is difficult to determine the exact cause for such
behavior at this point. Nevertheless, the fact that very similar under-predictions are
observed in the predictions for other data points (A20 and A30) (discussed later), and
deficient when using constant lift coefficients (Figures 6.6 and 6.14), does suggest that
the chances of experimental error being the cause are relatively low, although the
possibility cannot be totally ruled out. It is however likely that the under-predictions can
be attributed to using a constant dispersion Prandtl number as opposed to one that varies
depending on the local turbulence intensity.

Another interesting observation is the consistent non-zero phase volume fraction,


predicted at the wall (Radial position = 0.039 m) by most constant lift coefficients and the
expression proposed by Troshko et al. (2001). Unfortunately, this issue is still under
debate. While some experimental investigators (Foussat and Hulin, 1984; Vigneaux et al.
1988) report similar non-zero phase fractions at the wall, others (Lang and Auracher,
1996; Lang, 1999) maintain that only the continuous phase wets the wall. As a result, it
cannot be concluded that this behavior is an artifact of interphase models and/or
boundary conditions.

Figure 6.15 shows a comparison of normalized velocity predictions for the F20 data
point. It can be seen that the overall agreement with experimental data is very good.

6.2.1.2 Data Set of Hamad et al. (2000)


For the two data points (H10 and H20) from the data set of Hamad et al. (2000), the CFD
predictions and experimental holdup data are presented in Figure 6.12.
117

Once again, constant values for the lift coefficient (both positive and negative) were
found to be incapable of accurately predicting the near-wall peak regardless of the value
of DPN. In particular, it can be inferred from Figure 6.4 that all constant lift coefficients
start to predict high or low phase concentrations after approximately 0.035 m from the
pipe axis (which corresponds to an r/R ≈ 0.85), while the experimental data show a
change in phase distribution much earlier, at approximately 0.028 m from the pipe axis.

0.4

0.35
H10 - SIM H10 - EXP
Dispersed phase holdup (-)

0.3

H20 - SIM H20 - EXP


0.25

0.2

0.15

0.1

0.05

0
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
Radial position (m)

Figure 6.16 Simulated phase distribution profiles for the data set of Hamad et al. (2000),
H10 (DPN = 0.01; CL = Troshko et al. (2001); de = 3.25 mm), H20 (DPN = 0.075; CL =
Troshko et al. (2001); de = 3.5 mm)

On the other hand, the expression proposed by Troshko et al. (2001) when combined with
suitable dispersion Prandtl numbers, yielded a reasonably good match with experimental
data for both the data points, as shown in Figure 6.16. In particular, it was necessary to
increase turbulent dispersion at low holdups (H10) when compared to high holdups
(H20). Although, this cannot be justified with direct reference to data points (H10 and
118

H20) due to the lack of experimental turbulence-related data, it can be explained using
the experimental observations of Ali et al. (1999), who reported continuous phase
turbulence intensities in a vertical pipe featuring the same physical dimensions, liquid-
liquid system and similar flow velocities. Their observations showed a reduction in
turbulence intensity as the dispersed phase holdup was increased beyond 10% thereby
suggesting that turbulent dispersion effects are stronger at low dispersed phase holdups.
Interestingly, this fact is complimentary evidence that significant lateral forces can exist
even at low dispersed phase fractions as shown in Table 4.2. A possible explanation for
the weak influence of turbulent dispersion at high dispersed phase holdups could be the
presence of closely packed drops which strongly resist any movement in the lateral
direction (for instance due to the turbulent eddies in the carrier phase). A comparison of
the predicted phase distribution with experimental data follows.
Unfortunately, due to the lack of experimental data in the near-wall region (0.035 ≤ r ≤
0.039), it was not possible to compare predicted phase distributions in that area. It should
be noted that this also limits the overall comparisons that can be made, as discussed
below. Although, it is seen that for both the data points, the simulation slightly under-
predicts the experimental dispersed phase holdup especially near the pipe centerline, it is
difficult to identify the cause for the same without information on the relative error for
the experimental measurements. If however, the phase distribution close to the wall was
known from the experiments, the average dispersed phase holdup as observed in the
experiments could be calculated and compared with the phase ratio of the dispersed phase
to determine whether the discrepancy was due to the closure models used or the
experimental data itself. This is more significant in the case of data point H10, where it
can be seen that all experimental holdup measurements are lined up very close to the
input phase ratio (i.e. 10 %).

6.2.1.3 Data Set of Al-Deen and Bruun (1997)


Al-Deen and Bruun (1997) presented a comprehensive data set for phase distribution in
liquid-liquid vertical upflows. For the present study, only four data points (A5, A10, A20
and A30) from their data set were found suitable for CFD validation, as discussed in
119

Section 4.1.2.1. The results obtained from the CFD simulations for experimental
conditions corresponding to these data points are presented in this section. Predicted
phase distributions are compared with experiments in Figure 6.17. An excellent
agreement with experimental data is observed at low dispersed phase holdups (A5, A10).
At high holdups (A20, A30), while the overall agreement is good, a slight under-
prediction can be observed in the region adjoining the ‘near wall’ peak. This behavior
was also observed while simulating the data set of Farrar and Bruun (1996). Again, using
constant lift coefficients resulted in much lesser under-prediction in this region. However,
near-wall predictions were found to be highly inaccurate if constant lift coefficients were
used even at high values of DPN. This is illustrated for data point A20 in Figure 6.18.

0.6
A5 - SIM A5 - EXP
A10 - SIM A10 - EXP
0.5
A20 - SIM A20 - EXP
Dispersed phase holdup (-)

A30 - SIM A30 - EXP


0.4

0.3

0.2

0.1

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.17 Simulated phase distribution profiles for the data set of Al-Deen and Bruun
(1997), CL = Troshko et al. (2001), A5 (DPN = 0.024; de = 3 mm), A10 (DPN = 0.075; de
= 3.5 mm), A20 (DPN = 0.412; de = 4 mm), A30 (DPN = 7.5; de = 4 mm)
120

0.3

0.25

Dispersed phase holdup (-)


0.2

0.15

A20 - SIM, CL = +0.005


0.1
A20 - SIM, CL = -0.005

0.05
A20 - EXP

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.18 Effect of constant lift coefficients on phase distribution for data point A20
(DPN = 7.5; de = 4 mm)
On the other hand, the expression for lift coefficient proposed by Troshko et al. (2001)
was again found to yield very good predictions when compared to the experimental data
at all holdups. For the predictions shown in Figure 6.17, it was again necessary to
decrease turbulent dispersion at higher dispersed phase holdups (A20, A30). The
apparent reason for higher turbulent dispersion at low dispersed phase holdups (A5, A10)
was explained in Section 6.2.1.2 for similar experimental conditions. Also, the absence of
a significant ‘wall’ or ‘near-wall’ peak at low dispersed phase holdups (A5, A10) can be
attributed to the combined effect of inviscid lift forces and higher turbulent dispersion.
For instance, Figure 6.19 shows the lift coefficient distribution across the pipe cross-
section for experimental conditions corresponding to data point A5. It is seen that the lift
coefficient is positive throughout, thus implying inviscid lift dominance. The low DPN
value (0.024) on the other hand signifies increased turbulent dispersion. On the other
hand, for higher dispersed phase holdups, as explained in Section 6.2.1.1, vortex
shedding dominance in the near-wall region is primarily responsible for the distinct ‘near-
wall’ peak observed in the predictions.
121

0.09
0.08

0.07

Lift coefficient (-)


0.06

0.05
0.04

0.03 Troshko et al. (2001)

0.02
0.01

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.19 Radial variation of lift coefficient for data point A5 (DPN = 0.024, de = 3
mm)

0.9
Continuous phase absolute velocity (m/s)

0.8

0.7

0.6

0.5 A30 - SIM

0.4

0.3 A30 - EXP

0.2

0.1

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.20 Predicted and experimental ‘actual continuous phase velocity’ distribution
for data point A30 (DPN = 7.5; CL = Troshko et al. (2001); de = 4 mm)
122

1.2

N o rm aliz ed velo city (-)


0.8

0.6 SIM - A5

0.4
EXP - A5
0.2

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure 6.21 Comparison of experimental and simulated continuous phase velocity


profiles for the data set of Al-Deen and Bruun (1997), A5 (DPN = 0.024; CL = Troshko et
al. (2001); de = 3 mm)
Figure 6.20 shows a comparison of predicted actual velocity distribution with
experimental data at high dispersed phase holdup (A30). Figure 6.21, on the other hand
shows a comparison of predicted and experimental normalized continuous phase
velocities at low dispersed phase holdup (A5). It can be seen that the overall agreement
between the simulated and experimental phase velocities is quite good.

6.2.1.4 Data Set of Lang (1999)


All the discussions so far (Sections 6.2.1.1-6.2.1.3) were focused on radial phase
distributions and velocities for kerosene-water systems at varying holdups in a 78 mm ID
vertical pipe. Lang (1999) performed similar measurements for the n-heptane–water
system in a much smaller (16 mm ID) vertical pipe. Unfortunately however, all data
points from the data set of Lang (1999) were focused only on relatively higher dispersed
phase holdups (L20A, L20B, L40 and L60). Phase distribution predictions for data points
(L20A, L20B and L40) are presented in Figures 6.22 and 6.23.
123

0.3

0.25

Dispersed phase holdup (-)


0.2

0.15
L20B - SIM

0.1

0.05 L20B - EXP

0
0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008
Radial position (m)

Figure 6.22 Simulated phase distribution profile from the data set of Lang (1999), L20B
(DPN = 0.06; CL = -0.05; de = 2 mm)

0.9

0.8 L20A - SIM L20A - EXP


Dispersed phase holdup (-)

0.7
L40 - SIM L40 - EXP
0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.002 0.004 0.006 0.008
Radial position (m)

Figure 6.23 Simulated phase distribution profiles for the data set of Lang (1999), L20A
(DPN = 0.75; CL = -0.075; de = 1 mm), L40 (DPN = 3.0; CL = -0.05; de = 1 mm)
124

With the exception of minor inconsistencies in the near-wall region, it can be seen from
these figures that the overall agreement between the predictions and experimental data is
excellent.

0.6

0.5
Dispersed phase holdup (-)

0.4

0.3 L40 - SIM (Standard wall functions)

0.2 L40 - SIM (Enhanced wall


treatment)

0.1 L40 - EXP

0
0 0.002 0.004 0.006 0.008
Radial position (m)

Figure 6.24 Comparison of two near-wall modeling approaches for data point L40 (DPN
= 3.0; CL = -0.05; de = 1 mm)
Contrary to earlier findings (discussed in Sections 6.2.1.1-6.2.1.3), constant (negative) lift
coefficients (-0.05, -0.075) were found to yield the best predictions. Also for data points
L20A and L40, the Enhanced-Wall Treatment approach was used for near-wall
turbulence modeling as opposed to Standard Wall Functions, in order to better resolve
flow quantities especially near the wall. This was necessary as very low flow Reynolds
numbers (≈ 3000 - 7000) were encountered while simulating these data points. As the
hypotheses underlying the Standard Wall Functions cease to be valid in situations where
low-Reynolds number effects are pervasive, (Fluent, 2004) the use of the Enhanced-Wall
Treatment approach, which is valid even in the viscosity affected near-wall region, is
considered justified. Figure 6.24 shows a comparison of the two approaches for data
point L40. Clearly the ‘Enhanced wall treatment’ approach gives superior predictions for
the same simulation conditions.
125

0.5
L20B - SIM (Troshko et al. (2001), DPN 7.5)
0.45
L20B - EXP
0.4

Dispersed phase holdup (-)


L20B - SIM (Troshko et al. (2001), DPN 0.75)
0.35

0.3

0.25

0.2

0.15

0.1

0.05

0
0 0.002 0.004 0.006 0.008
Radial position (m)

Figure 6.25 Predictions obtained when the lift expression proposed by Troshko et al.
(2001) is used for data point L20B (de = 2 mm)

When the expression for lift coefficient as proposed by Troshko et al. (2001) was used,
highly inaccurate phase distribution profiles were obtained. Increasing turbulent
dispersion only resulted in a more homogeneous profile. Figure 6.25 shows an example
of the same for data point L20B. The apparent failure of the lift coefficient expression
proposed by Troshko et al. (2001) to predict the correct trend for phase distribution can
be explained on the basis of bubble Reynolds numbers as follows.

The expression proposed by Troshko et al. (2001) was largely based on the original
model developed by Moraga et al. (1999) as discussed in Section 5.3.2.1. One limitation
of this original model is that it does not take into account the effect of viscous
contributions. This may be acceptable in cases where inertial effects are predominant (i.e.
at high bubble Reynolds numbers). However, at low bubble Reynolds numbers, viscous
effects tend to predominate. This raises concerns over the applicability of the lift
coefficient expression proposed by Moraga et al. (1999) under such conditions.
Fortunately, this issue was discussed by Moraga et al. (1999) in their paper where they
pointed out that other researchers (Eichhorn and Small, 1964; Dandy and Dwyer, 1990)
126

were also unable to correlate the lift coefficient with the product of bubble and shear
Reynolds numbers when Re ≤ 250. This was attributed to the bubble Reynolds number
being too small for the hypothesis of the inviscid theory to be valid. For the present
investigation, when the bubble Reynolds numbers for data points L20A, L20B and L40
were obtained from the simulations, they were found to be significantly lower than 250.
This explains the inability of the expression proposed by Troshko et al. (2001) to
properly predict the phase distribution.
In addition to negative lift coefficients, including turbulent dispersion effects was also
found to be equally important. It was again observed that it was necessary to decrease
turbulent dispersion at higher dispersed phase holdups (data point L40) so that better
predictions could be obtained.
For the predicted phase distributions in Figure 6.23, the minor under-predictions
observed in the near-wall region can be attributed to the use of constant lift coefficients
and dispersion Prandtl numbers as opposed to locally varying values. The assumption of
a mono-dispersion of drops could also be responsible for this behaviour.

2.5
Actual velocity of dispersed phase (m/s)

1.5

1 SIM - L20B

0.5 EXP - L20B

0
0 0.002 0.004 0.006 0.008
Radial position (m)

Figure 6.26 Comparison of predicted and experimental velocities for the dispersed phase
from the data set of Lang (1999), (DPN = 0.06; CL = -0.05; de = 2 mm)
127

Figure 6.26 presents a comparison of actual velocity predictions for the dispersed phase
with corresponding experimental data for data point L20B. It can be seen that there is a
good agreement between the simulation and the experiment.

Several attempts to predict phase distribution corresponding to the experimental


conditions of data point L60 at steady state resulted in severe numerical instabilities. In
order to determine the cause for this behavior, the same experimental conditions were
simulated using an unsteady (time-dependant) solution approach. Under these conditions,
although the solution remained stable, a steady state solution could not be attained. This
can be explained as follows. Based on their experiments, Lang and Auracher (1996)
determined that phase inversion took place somewhere in the region of data point L60. It
is therefore possible that data point L60 is situated in the transition region, which cannot
be simulated assuming steady state conditions, thus causing the observed instability. This
would also explain why a steady-state solution could not be attained even when using a
time-dependant solution approach.

6.2.1.5 Data Set of Vigneaux et al. (1988)

Vigneaux et al. (1988) studied the up-flow of kerosene-water mixtures in a vertical pipe
of ID 200 mm. All four data points (V5, V10, V30 and V50) from their data set were
found suitable for CFD validation as discussed in Section 4.1.2.1. The predicted phase
distributions for these data points are shown in Figure 6.27, where it can be seen that the
overall predictions are very good, especially at low dispersed phase holdups.Constant lift
coefficients were found to give excellent predictions at these low holdups, while the
expression proposed by Troshko et al. (2001) yielded better predictions at high holdups.
At low dispersed phase holdups (V5 and V10), bubble Reynolds numbers less than 250
were encountered. The use of constant lift coefficients at these holdups can therefore be
justified based on a similar reasoning to that discussed in Section 6.2.1.4. Again, when
the expression proposed by Troshko et al. (2001) was tested at low holdups, significant
discrepancies similar to those described in Section 6.2.1.4 were observed. At higher
holdups, although constant (positive) lift coefficients yielded similar predictions to those
shown in Figure 6.27, the expression proposed by Troshko et al. (2001) was preferred as
128

it estimated the lift coefficients directly based on the local flow properties. In order to
obtain the predictions shown in Figure 6.27, it was necessary to account for higher
turbulent dispersion at low holdups (V5, V10) when compared to high holdups (V30 and
V50).

0.8
V5 - SIM V5 - EXP
0.7
V10 - SIM V10 - EXP

V30 - SIM V30 - EXP


Dispersed phase holdup (-)

0.6
V50 - SIM V50 - EXP
0.5

0.4

0.3

0.2

0.1

0
0 0.02 0.04 0.06 0.08 0.1
Radial position (m)

Figure 6.27 Phase distribution profiles for the data set of Vigneaux et al. (1988), V5
(DPN = 1.59; CL = -0.05; de = 2.5 mm), V10 (DPN = 0.328; CL = -0.05; de = 2.75 mm),
V30 & V50 (DPN = 4.125; CL = Troshko et al. (2001); de = 5 mm)
It is difficult to exactly pin down the cause for the slight over-prediction at high dispersed
phase holdups as seen in Figure 6.27. It is likely however, that this is due to the
assumption of an equivalent drop diameter instead of a drop size distribution. This is
considered plausible as Vigneaux et al. (1988) observed a fairly wide drop size
distribution (2 – 6 mm) throughout their experiments.

A summary of the experimental conditions and CFD Simulation results discussed so far
is presented in Table 6.7.
Table 6.7 Experimental conditions and CFD simulation results
EXPERIMENTAL CFD SIMULATION

Phase
Pipe Total Observed Area- Area-
ratio of
Expression Expression Dispersion Equivalent
dimensions volumetric dispersed averaged averaged
dispersed
Data used for or constant prandtl drop
Investigator phase flow rate phase dispersed bubble
point drag used for lift number diameter
ID QD (Q D + QC ) holdup
coefficient* coefficient* (DPN) (mm)
phase Reynolds
Length
(m QD + QC (m3/s) trend* holdup number
(m)
m)

F20 1.500 78.0 0.20 0.00308 NWP KH1985 T2001 7.500 5.00 0.1929 452.53
Farrar and
Bruun F25 1.500 78.0 0.25 0.00308 NWP KH1985 T2001 7.500 5.00 0.2357 420.59
(1996)
F30 1.500 78.0 0.30 0.00308 NWP KH1985 T2001 7.500 5.00 0.2908 384.45

Hamad et al. H10 4.201 77.8 0.10 0.00310 NWP KH1985 T2001 0.010 3.25 0.0895 282.32

(2000)
H20 4.201 77.8 0.20 0.00348 NWP KH1985 T2001 0.075 3.50 0.1875 259.70

A5 1.500 77.8 0.05 0.00272 NWP KH1985 T2001 0.024 3.00 0.0441 283.24

Al-Deen and
A10 1.500 77.8 0.10 0.00287 NWP KH1985 T2001 0.075 3.50 0.0904 318.71
Bruun
(1997) A20 1.500 77.8 0.20 0.00323 NWP KH1985 T2001 0.412 4.00 0.1854 323.29

A30 1.500 77.8 0.30 0.00369 NWP KH1985 T2001 7.500 4.00 0.2846 272.35

L20B 1.050 16.0 0.20 0.00030 C KH1985 -0.050 0.060 2.00 0.1881 159.97

Lang (1999) L20A 1.050 16.0 0.20 0.00010 C KH1985 -0.075 0.750 1.00 0.1733 50.543

L40 1.050 16.0 0.40 0.00010 C KH1985 -0.050 3.000 1.00 0.3420 44.588

129
EXPERIMENTAL CFD SIMULATION

Phase
Pipe Total Observed Area- Area-
ratio of
dimensions Expression Expression Dispersion Equivalent
dispersed volumetric dispersed averaged averaged
Data used for or constant prandtl drop
Investigator phase flow rate phase dispersed bubble
point drag used for lift number diameter
ID QD (Q D + QC ) holdup
coefficient* coefficient* (DPN) (mm)
phase Reynolds
Length
(m QD + QC (m3/s) trend* holdup number
(m)
m)

V5 14.00 200 0.05 0.00750 C KH1985 -0.050 1.593 2.50 0.0340 221.51

Vigneaux et V10 14.00 200 0.10 0.00750 C KH1985 -0.050 0.328 2.75 0.0691 229.93

al. (1988)
V30 14.00 200 0.30 0.00750 WP KH1985 T2001 4.125 5.00 0.2354 421.71

V50 14.00 200 0.50 0.00750 WP KH1985 T2001 4.125 5.00 0.4364 306.15
*
NWP: Near-wall peaking; C: Coring profile; WP: Wall peaking; KH1985: Kumar and Hartland (1985); T2001: Troshko et al. (2001)

130
131

6.2.2 Conclusions from CFD Simulations


Following conclusions can be drawn based on the numerical study of liquid-liquid up-
flows in vertical pipes covering a wide range of experimental conditions (Table 6.6):
1. Liquid-Liquid bubbly up-flows in vertical pipes typically feature Wall peaking,
Near-wall peaking and Coring trends for the dispersed phase holdup distribution.
In order to successfully predict such inhomogeneous phase distributions,
accounting for three significant interphase forces is imperative. These are the
drag and lift forces and turbulent dispersion.
2. An analysis of several drag coefficient expressions clearly reveals that the drag on
drops differs significantly from the drag on rigid spheres, particularly at larger
equivalent drop diameters. Also, accounting for the presence of adjacent drops
when estimating the drag yields slightly better predictions. Although properly
accounting for the drag force yields a good prediction for the average dispersed
phase holdup, it cannot predict the local holdup accurately.
3. Non-drag lateral forces such as the lift force and turbulent dispersion dictate the
overall phase distribution in turbulent bubbly flows. With reference to the lift
forces, the expression for lift coefficient proposed by Troshko et al. (2001) yields
very good predictions when the bubble Reynolds number is greater than 250.
However, when bubble Reynolds numbers lower than 250 are encountered,
constant (negative) values for the lift coefficient in the range (-0.05 to -0.075) are
recommended.
4. Turbulent dispersion is found to be more significant at lower dispersed phase
holdups than at higher dispersed phase holdups. For the model proposed by
Simonin and Viollet (1990), Dispersion Prandtl Numbers in the range 0.01 ≤
DPN ≤ 0.05 are recommended for use at low input dispersed phase ratios (i.e. xd

≤ 0.1) and DPN values in the range 0.075 ≤ DPN ≤ 7.5 are recommended for use
at intermediate dispersed phase ratios (i.e. 0.1 < xd ≤ 0.2). At high dispersed
132

phase ratios (i.e. xd > 0.25), a DPN value of atleast 7.5 is recommended. The

above guidelines for turbulent dispersion are graphically presented in Figure 6.28.

8
Dispersion Prandtl Number (-)
7

4
≤ ≤
3

1 ≤ ≤
0
0 10 20 30 40 50 60
Phase ratio of the dispersed phase (%)

Figure 6.28 Recommendation for turbulent dispersion in terms of DPN


It is important at this juncture to point out the lower and upper limits of significance for
the value of DPN. All values of DPN 7.5 essentially yield similar predictions, thus
making 7.5 the significant upper limit. Likewise, similar predictions are observed for all
values of DPN ≤ 0.0075, thus making 0.0075 the significant lower limit. Consequently,
considerable changes in the holdup distributions are observed only in the intermediate
range (i.e. 0.0075 ≤ DPN ≤ 7.5). This can also be inferred from Figure 6.9. The dispersed
phase holdup predictions for the aforementioned upper and lower limits are presented
graphically in Figure 6.29 using the experimental conditions corresponding to datapoint
F30 as an example.
133

0.4

0.35

Dispersed phase holdup (-)


0.3

0.25

0.2 DPN 7.5 DPN 75

0.15 DPN 750 DPN 10000000000

0.1 No turbulent dispersion DPN 0.004

0.05 DPN 0.005 DPN 0.0075

0
0 0.2 0.4 0.6 0.8 1
Ra dial position (m)

Figure 6.29 Upper and lower limits for DPN beyond which no significant differences in
the holdup distribution are observed
6.2.3 Validation of the Proposed Interphase Closure Guidelines
Having proposed closure guidelines for interphase forces encountered in turbulent liquid-
liquid vertical upflows it becomes important to test the validity of the recommendations
by application to other situations where immiscible liquid dispersions are commonly
encountered. In particular, applying the closure guidelines to a system that features non-
axisymmetric and non-homogeneous dispersed phase distribution would serve as a
constructive test to assess the suitability of recommendations. The horizontal pipe was
therefore chosen as a test bed for this validation exercise. Accordingly, the experimental
conditions of Angeli (1996) and Soleimani et al. (1999) were chosen for CFD simulation
(Table 4.1).

As simulation of liquid-liquid horizontal flows in two dimensions would result in an


oversimplification of the actual flow dynamics, it was necessary to carry out 3D
simulations. This resulted in a 12 to 16 fold increase in the grid size when compared to
134

2D grids. However, as the vertical plane divides the pipe into two identical sections, it
was possible to exploit symmetry and reduce the grid size to half its original value.
Details concerning the grid creation technique, boundary conditions etc. are elaborated in
Section A1.1.8.

Despite exploiting symmetry to reduce the number of grid cells and thus the turnaround
time, it was necessary to carry out the computations either using a symmetric multi-
processor (SMP) machine (64 bit dual Intel Xeon 3.4 GHz featuring 2.5 GB RAM) or at
least a network of parallel workstations (two Pentium IV 2.4 GHz workstations featuring
1 GB RAM each, connected over 10/100 ethernet) to ensure manageable computation
times (≈ 8 - 16 hours to process 1000 timesteps of 0.001 s each). The maximum number
of iterations per time step was set to a suitable value (20-60) in order to ensure
convergence within each time step. This was necessary in order to capture any transient
behavior. It should however be noted that this value (i.e. the maximum iterations per time
step) strongly depends on the choice of under-relaxation factors (URF). For a fixed time
step size, lower URFs demand more iterations per time step when compared to higher
ones.

Preliminary attempts to run the simulations assuming steady-state resulted in severe


numerical instabilities. As a result, transient (time-dependant) simulations were carried
out. Ideally, one would expect to compute a time-dependant simulation until a steady
state was reached and then compare the final predictions with experimental data.
However, before a steady-state solution could be attained, sporadic instabilities in the
CFD solution were observed. These instabilities magnified with time and ultimately
resulted in solution divergence. A possible reason for the same could be the increasing
numerical errors in the constituents of the momemtum equation that are caused by the
almost negligible slip velocities encountered throughout the solution domain as time
progresses. The robustness of the numerical solution could be another source for the
problems observed.
135

6.2.3.1 Validation Results


The present section discusses transient simulation results for four data points (Table 6.8).
As before, each data point is labelled using the corresponding input phase ratio for the
dispersed phase.

As Soleimani et al. (1999) did not present experimental data concerning the drop
diameter an estimate of the same was obtained at similar flow conditions from the data
set of Angeli (1996). This is justified since phase distribution data for both data sets were
acquired using the same experimental facility. However, care was taken to ensure that the
drop diameter was obtained for the exact pipe material as was used in both data sets. This
estimate for the drop diamter was also successfully verified using the procedure described
in Section 6.2.
Table 6.8 Summary of experimental flow conditions for horizontal pipes

Average
Continuous Dispersed Equivalent
experimental
Data phase phase drop
Data Set dispersed
Point superficial superficial diameter
phase holdup
velocity (m/s) velocity (m/s) (mm)
(-)

Angeli (1996) –
AN37.5 1.0625 0.6375 NS* 1.0
Transplite pipe

S40A 1.8000 1.2000 0.39 0.5


Soleimani et al. (1999)
S54 1.3800 1.6200 0.53 0.5
– Stainless steel pipe

S40B 1.2720 0.8480 0.39 0.5


*
NS: Not Specified.

As the average in-situ dispersed phase holdup was very close to the input phase ratio of
the dispersed phase for most of the data points, this resulted in very low estimates for the
slip velocity magnitude (≤ 0.005 m/s). This was also observed by Soleimani et al. (1999)
in their experiments. Such behaviour is quite reasonable considering the fact that
buoyancy is no longer a driving force for relative motion in the direction of flow. When
136

such low slip velocities were combined with drop sizes in the range (0.5 to 1 mm), bubble
Reynolds numbers significantly lower than 250 were obtained. As a result, based on the
closure recommendations for lift forces, a constant (negative) lift coefficient (CL = -0.05)
was used throughout. Also, as all data points featured input phase ratios greater than 25%
(for the dispersed phase), a high dispersion Prandtl number (DPN = 7.5) was used. A
discussion of the CFD simulation results follows.

6.2.3.1.1 Vertical Holdup Profile Data Set of Soleimani et al. (1999)


Soleimani et al. (1999) performed a series of experiments to quantify the spatial
distribution of oil and water in horizontal pipe flow. Two different measurement
techniques (high frequency impendence probe, gamma densitometer system) were used
to obtain the dispersed phase distribution along the vertical 90° diameter as shown in
Figure 6.30. The simulations results obtained are discussed below.

0.025
Upper wall

0.02
Flow
Pipe diameter (m)

0.015 Gravity

EXP-HFP
0.01
EXP-GDS

SIM-0.05s
0.005 SIM-0.25s

SIM-0.5s

0
0 0.2 0.4 0.6 0.8 1
Dispersed phase holdup (-)

Figure 6.30 Transient phase distribution profiles for the data set of Soleimani et al.
(1999), S40A (DPN = 7.5; CL = -0.05; de = 0.5 mm)
137

For flow conditions corresponding to data point S40A, the time-varying CFD predictions
of the dispersed phase holdup are shown in Figure 6.30. The experimental results
obtained using two different measurement techniques show some discrepancy with the
GDS results being considered more reliable because of the non-invasive nature of this
technique (Soleimani et al., 1999). Although the dispersed phase holdup is almost
uniform throughout the pipe, definite reduction in phase fraction can be observed in the
lower sections.

CFD predictions are in reasonable agreement with the experimental trends and the
tendency to form lower concentration regions at the bottom of the pipe was predicted.
However, the time-varying predictions appear to indicate a certain trend. In particular, it
is seen that as the flow time increases (i.e. from 0.05 s to 0.25 s to 0.5 s solution time) the
segregation effect becomes more prominent. The predicted increase in dispersed phase
holdup in the upper regions of the pipe was not experimentally observed.

The discrepancy between experiments and CFD predictions could have been caused by
the severely diminished value of relative velocities between the phases under the
experimental conditions investigated (as low as 0.001 m/s), which can give rise to
numerical instabilities. In addition, it is important not to underestimate the effect of a
dynamic drop size distribution and/or spatial variation in turbulent dispersion. Further
CFD investigations are therefore necessary to determine the exact cause for such
behaviour.

Figure 6.31 compares the CFD predictions with the experimental holdup results for flow
conditions corresponding to data point S54. The CFD predictions show trends similar to
those discussed earlier in conjunction with data point S40A. Reasonable agreement was
obtained between the CFD simulation and the experimental results (obtained using the
high-frequency impedence probe). However, CFD predictions overestimated the
segregation tendencies in the vicinity of the upper and lower walls and could not predict
the coring tendencies measured using the GDS technique. These factors combined with
the difficulty encountered in achieving steady state solution suggest the need for
additional investigation taking drop breakage and coalescence into account.
138

0.025
Upper wall

0.02

Pipe diameter (m) 0.015 Flow

Gravity
0.01
EXP-HFP
EXP-GDS
SIM-0.05s
0.005
SIM-0.25s
SIM-0.5s
0
0 0.2 0.4 0.6 0.8 1
Dispersed phase holdup (-)

Figure 6.31 Transient phase distribution profiles for the data set of Soleimani et al.
(1999), S54 (DPN = 7.5; CL = -0.05; de = 0.5 mm)

0.025
Upper wall

0.02
Pipe diameter (m)

0.015 Flow
EXP-HFP

Gravity EXP-GDS
0.01
SIM-0.05s

0.005 SIM-0.25s

SIM-0.5s

0
0 0.2 0.4 0.6 0.8 1
Dispersed phase holdup (-)

Figure 6.32 Transient phase distribution profiles for the data set of Soleimani et al.
(1999), S40B (DPN = 7.5; CL = -0.05; de = 0.5 mm)
139

Time-varying CFD predictions of dispersed phase distribution for flow conditions


corresponding to the third data point (S40B) are compared with experimental data in
Figure 6.32. It is seen that the general agreement between the two is not satisfactory.
While the reduced oil concentration near the lower wall is well predicted, the
experimental results do not confirm the predicted increase in oil concentration in the
region near the upper wall. In this regard, it should be noted that Soleimani et al. (1999)
expressed concerns regarding the discrepancy between the experimental measurements
near the upper wall. The observed trends can again be attributed to several factors similar
to those discussed for data point S40A.

6.2.3.1.2 Horizontal Holdup Profile Data Set of Angeli (1996)


Angeli (1996) carried out local phase fraction measurements using a high frequency
impedence probe for a wide range of mixture velocities (0.9 to 1.7 m/s). These conditions
resulted in the stratified-mixed, three-layer and mixed flow patterns. Only those data
points featuring mixed flow conditions (AN37.5, AN62.5) were chosen for validation in
the present study. The AN62.5 data point however, could not considered for simulation as
it was obtained using flow conditions around the phase inversion point. As a result,
experimental conditions corresponding to data point AN37.5 were only simulated. Phase
distribution measurements corresponding to data point AN37.5 were made along the 0°
diameter which is normal to the gravity vector. This is indicated in Figure 6.33 where
CFD predictions are compared with experimental data.

It is seen that the transient simulation results under-predict the experimental data. This is
similar to the trends observed while simulating the results of Soleimani et al. (Section
6.2.3.1.1) and the dispersed phase holdups are expected to approach the experimental
values as the flow time increases. Unfortunately numerical instabilities limited the total
solution time that could be reached. The problems encountered can again be attributed to
several factors similar to those discussed for the case of the vertical holdup distribution.
140

0.6

0.5

Dispersed phase holdup (-)


0.4

0.3

0.2
EXP-AN37.5 SIM-0.3s

0.1
SIM-0.05s
Gravity
0
0 0.002 0.004 0.006 0.008 0.01 0.012

Distance from pipe wall (m)

Figure 6.33 Transient phase distribution profiles for the data set of Angeli (1996),
AN37.5 (DPN = 7.5; CL = -0.05; de = 1.0 mm)
6.2.3.2 Conclusions and Recommendations
Based on the findings discussed in Section 6.2.3.1, the following conclusions can be
drawn.

1. The horizontal pipe configuration severely tests the ability to use the Eulerian
model to simulate multiphase flow.

2. It was imperative to switch from a steady-state 2-D solution approach to that of


transient 3D simulations in order to achieve a stable non-divergent solution.
Furthermore, it was necessary to use a hexahedral grid with at least 150,000 cells
to adhere to the Y+ and cell aspect ratio requirements. This resulted in a drastic
increase in computational requirements forcing the use of more powerful
computational resources for faster grid generation and CFD solution.
141

3. The very low relative velocity between the phases is possibly responsible for the
intermittent numerical instabilities observed, that ultimately caused the CFD
solution to diverge. As a result, it is recommended to use a more robust solution
algorithm.

4. As a result of the numerical problems encountered, it is not possible to draw


definite conclusions on the suitability of the proposed interphase closure
guidelines. However, the limited CFD predictions obtained provided reasonable
estimates of the holdup values and trends.

It is therefore recommended that the cause of the numerical instability be assessed. Also,
the nature of the lift forces acting on small drops moving in dense dispersions should be
studied. In addition, it is important to determine whether other lateral forces such as the
wall lubrication force could have a significant effect on the phase distribution. Since
these factors are strongly affected by the drop size, it is imperative that such
investigations be conducted while accounting for the possibility of changing drop size
distributions (i.e. drop breakage and coalescence).
142

CHAPTER 7

7 OVERALL CONCLUSIONS AND


RECOMMENDATIONS

A comprehensive review of the literature revealed that despite the apparent importance of
liquid-liquid dispersions in several industrial processes, very few attempts focused on
using CFD to model the hydrodynamics and accurately predict fundamental multiphase
characteristics such as the dispersed phase distribution. Within the few CFD studies
identified the importance of interphase closure was underestimated and many terms were
arbitrarily overlooked. In particular, general closure guidelines for turbulent liquid-liquid
dispersions were not developed. This contrasts with gas-liquid simulation studies in
which the forces acting of the dispersed entity have been thoroughly investigated.

The present study attempts to fill this void by implementing and examining several
formulations for the interphase forces in a commercial CFD code. The case of immiscible
liquid dispersions flowing upwards in a vertical pipe was chosen as the system of interest
for reasons discussed in Section 2.2.5. Closure guidelines were developed by comparing
CFD simulations with a database of experimental results published in the open literature
(16 data sets of experimental results). Attempts were then made to test the ability of the
proposed guidelines to predict the dispersed phase distribution in horizontal pipes.

Based on the findings of the present investigation, the following conclusions can be
inferred.

7.1 Conclusions
• For small equivalent drop diameters (≤ 3 mm), all available drag expressions for
single entities predict essentially similar holdups. On the other hand, significantly
lower dispersed phase holdups were predicted when the drag expression for a
rigid sphere was used in conjunction with equivalent diameters larger than 5 mm.
143

These are encountered at higher dispersed phase holdups where the increased
coalescence tendency shifts the drop size distribution towards larger drops.

• The various drag expressions for single drops predict essentially similar holdups.
The drag expression proposed by Ishii and Zuber (1979) was selected as a
representative of the all single drop drag expressions.

• The drag expressions that account for the presence of adjacent drops (Kumar and
Hartland, 1985; Ishii and Zuber, 1979; Ruche and Issa, 2000) were found to yield
somewhat similar average holdups but the drag expression proposed by Kumar
and Hartland (1985) was found to most accurately predict the average dispersed
phase holdup under a wide range of experimental conditions (0.2 < Vm < 1.5 m/s;
0.04< αd < 0.44; 0.016 < ID < 0.2 m).

• Lift forces were found to play a significant role at both low and high dispersed
phase holdups and their use is necessary to predict the experimentally observed
wall peaking and coring profiles. The lift forces can best be predicted using the
lift coefficient expression proposed by Troshko et al. (2001) except at low drop
Reynolds numbers (Re ≤ 250) in which case a negative coefficient (CL = - 0.05)
should be used.

• Turbulent dispersion effects can be simulated using the approach proposed by


Simonin and Viollet (1990) which accounts for the response of drops to turbulent
eddies in the continuous phase. The effect of virtual mass on the turbulent
dispersion was found to be insignificant at the relatively high density ratios
encountered in the liquid-liquid dispersions investigated.

• Turbulent dispersion was found to be significant at low dispersed phase holdups


but is practically negligible at dispersed phase ratios larger than 25%. Similar
conclusions apply for the dispersion model proposed by Lopez de Bertodano
(1992).
144

• The aforementioned guidelines for interphase closure could not be effectively


tested for the case of horizontal liquid-liquid flows because of issues related to
numerical instability. However, for the most part, a comparison of transient CFD
predictions revealed a reasonably good agreement with experimental data.

7.2 Recommendations
Based on the findings of the present study, the following recommendations are made in
order to improve the ability of CFD to accurately simulate the hydrodynamics of liquid-
liquid dispersions:
• Modeling recommendations
o Interphase forces
In the case of low drop Reynolds numbers (Re ≤ 250), the dependence of
the lift coefficient on the local flow properties may be better expressed
using the approach of Mei (1992) which focuses on the lift force
experienced by a sphere in a slow shear flow.
In a fashion similar to the treatment of drag force in the presence of
adjacent drops, the attempts should be made to account for the effect of
adjacent drops on the lift force. The approaches of Beyerlein et al. (1985),
Petersen (1992), Behzadi et al. (2004) and Kulkarni and Joshi (2004) may
provide some guidance.
o Turbulence
The ability to accurately predict turbulence in the continuous phase should
be tested.
The effect of turbulence modulation (both enhancement and suppression)
should be included in future simulations.. This would assist in avoiding
the need to modify the Dispersion Prandtl Number in order to account for
the effect of the dispersed phase on turbulent dispersion.
The use of other approaches to turbulence modeling of multi-fluid
dispersions (e.g. RSM, DES, LES etc.) should be assessed.
o Drop dynamics
145

The effect of accounting for a dynamic drop size distribution of on CFD


predictions should be investigated particularly where wide drop size
distributions were encountered.
o Closure validation
Further validation of the proposed closure guidelines should be undertaken
preferably in stirred tanks, mechanically agitated columns and other
industrially relevant contacting units. Depending on the flow dynamics, it
might be necessary to include the effect of other interphase forces in
addition to drag, lift and turbulent dispersion (e.g. the centrifugal force).

• Experimental recommendations
o Detailed experimental data on vertical liquid-liquid pipe flows needs to be
supplied by investigators. In particular, experimentally obtained mean drop
sizes for each radial holdup measurement, data pertaining to actual phase
velocity distributions, and measured radial profiles for turbulence quantities
(e.g. turbulence intensity) of both phases would help in developing better
closure models for CFD.
o Similar to certain gas-liquid studies (Wang et al., 1987; Hibiki et al., 1999;
Hibiki et al., 2001), experimental data on developing vertical up-flows and
down-flows of immiscible liquid dispersions would prove immensely useful
to test the recommended interphase closure models, particularly the lift
models. It should be noted that in developing flows, it might be necessary to
exactly replicate inlet conditions used in the experiments.
146

REFERENCES
Abduvayt, P., R. Manabe, T. Watanabe and N. Arihara, “Analysis of Oil-Water Flow
Tests in Horizontal, Hilly-Terrain, and Vertical Pipes”, presented at the SPE
Annual Technical Conference and Exhibition, Houston, Texas, U.S.A., 26-29
September 2004.
Agterof, W. G. M., G. E. J. Vaessen, G. A. A. V. Haagh, J. K. Klahn, J. J. M. Janssen,
“Prediction of emulsion particle sizes using a computational fluid dynamics
approach”, Colloids Surf., B 31, 141-148 (2003).
AIAA Guide G-077-1998, “Guide for the Verification and Validation of Computational
Fluid Dynamics Simulations”, American Institute of Aeronautics and
Astronautics (1998).
Alajbegovic, A., A. Assad, F. Bonetto and R.T. Lahey Jr, "Phase distribution and
turbulence structure for solid/fluid up flow in a pipe", Int. J. Multiphase Flow 20,
453-479 (1994).
Al-Deen, M. F. N. and H. H. Bruun, “A comparative study of single normal, X type and
split-film anemometer probe measurements in kerosene/water two-phase flow”,
Meas. Sci. and Technol. 8, 885-893 (1997).
Alexopoulos, A. H., D. Maggioris and C. Kiparissides, “CFD analysis of turbulence non-
homogeneity in mixing vessels A two-compartment model”, Chem. Eng. Sci. 57,
1735-1752 (2002).
Ali, J., Z. Hassan and M. Ngah, “Calibrations and measurements of hot-film anemometry
in two-phase flow”, Jurnal Teknologi 30, 33-46 (1999).
Alopaeus, V., J. Koskinen and K. I. Keskinen, “Simulation of the population balances for
liquid-liquid systems in a nonideal stirred tank. Part 1 Description and qualitative
validation of the model”, Chem. Eng. Sci. 54, 5887-5899 (1999).
Alopaeus, V., J. Koskinen, K. I. Keskinen and J. Majander, “Simulation of the population
balances for liquid-liquid systems in a nonideal stirred tank. Part 2 -parameter
fitting and the use of the multiblock model for dense dispersions”, Chem. Eng.
Sci. 57, 1815-1825 (2002).
Angeli, P., “Liquid-liquid dispersed flows in horizontal pipes”, Ph.D. Thesis, Imperial
College, London (1996).
Angeli, P. and G. F. Hewitt, “Flow structure in horizontal oil-water flow”, Int. J.
Multiphase Flow 26, 1117-1140 (2000)
Antal, S.P., Lahey, R. T. Jr and J.E. Flaherty, "Analysis of Phase Distribution in Fully
Developed Laminar Bubbly Two-Phase Flow", Int. J. Multiphase Flow 17, 635-
652 (1991).
147

Auton, T. R., J.C.R. Hunt and M. Prud’Homme, “The force exerted on a body in inviscid
unsteady non-uniform rotational flow”, J. Fluid Mech. 197, 241-257 (1988).
Auton, T.R., "The Lift Force on a Spherical Body in a Rotational Flow", J. of fluid Mech.
183, 199-218 (1987).
Bakker, A. “Kenics Mixer – Then and Now” [online].
Available:
www.fluent.com/about/news/newsletters/01v10i1/a14.htm [2001].
Bakker, A. and L. M. Oshinowo, “Modeling of Turbulence in stirred vessels using Large
Eddy Simulation”, Chem. Eng. Res. Des. 82, 1169-1178 (2004).
Baldyga, J. and W. Podgorska, “Drop break-up in intermittent turbulence: Maximum
stable and transient sizes of drops”, Can. J. Chem. Eng. 76, 456-470 (1998).
Baldyga, J., W. Podogorska and R. Pohorecki, “Mixing-precipitation model with
application to double feed semibatch precipitation” Chem. Eng. Sci. 50, 1281-
1300 (1995).
Balzer, G., A. Boelle and O. Simonin, “Eulerian gas-solid flow modeling of dense
fluidized bed”, Fludization VIII, Int’ Symp’ of the Engineering Foundation, Tours
1125 (1995).
Barnea, E. and J. Mizrahi, “A generalized approach to the fluid dynamics of particulate
systems: Part I – General correlation for fluidization and sedimentation in solid
multiparticle systems”, Chemical Engineering Journal 5, 171-189 (1973).
Batchelor, G. K., Theory of Homogeneous Turbulence Cambridge University Press,
London, UK (1953).
Behzadi, A., R.I. Issa and H. Rusche, "Effects of turbulence on inter-phase forces in
dispersed flow", ICMF 2001, 4th Int. Conf. Multiphase Flow, New Orleans, LA,
U.S.A. (2001).
Behzadi, A., R.I. Issa and H. Rusche, "Modelling of Dispersed Bubble and Droplet Flow
at High Phase Fractions", Chem. Eng. Sci. 59, 759-770 (2004).
Bel Fdhila, R'Bei, "Analyse experimentale et modeelisation d'un ecoulement vertical a
bulles un elargissement brusque", PhD thesis, L'Institute National Polytechnique
de Tolouse, Tolouse, France (1991).
Benay, R., B. Chanetz and J. Délery, “Code verification/validation with respect to
experimental data banks”, Aerospace Science and Technology 7, 239-262 (2003).
Bertodano, M. L., R. T. Lahey and O. C. Jones, “Phase distribution in bubbly two-phase
flow in vertical ducts”, International Journal of Multiphase Flow 20 805-818
(1994).
148

Beyerlein, S.W., R.K. Cossmann and H.J. Richter, "Prediction of Bubble Concentration
Profiles in Vertical Turbulent Two-Phase Flow", Int. J. Multiphase. Flow 11, 629-
641 (1985).
Boye, A. M., M. Y. A. Lo and P. A. Shamlou, “Effect of Two-Liquid Phase Rheology on
Breakage in Mechanically Stirred Vessels”, Chem. Eng. Comm. 143, 149-167
(1996).
Brauner, N., "Modelling and Control of Two-Phase Phenomena - Liquid-liquid two-
phase flow systems" Technical Report , 1-59 (2002).
Burns, A. D., T. Frank, I. Hamill and J-M. Shi, "The Favre Averaged Drag Model for
Turbulent Dispersion in Eulerian Multi-phase Flows", ICMF 2004, 5th Int. Conf.
Multiphase Flow, Yokohama, Japan (2004).
Calabrese, R. and C. M. Stoots, “Flow in the impeller region of a stirred tank”, Chemical
Engineering Progress 85 43-50 (1989).
Calabrese, R. V. and S. Middleman, “The Dispersion of Discrete Particles in a Turbulent
Fluid Field”, AIChE Journal 25, 1025-1035 (1979).
Cao, J. and G. Ahmadi, “Gas-particle two phase turbulent flow in a vertical duct”,,
Internaional Journal of Multiphase Flow 21, 1203-1228 (1995).
Carrica, P.M., D.A. Drew and R.T. Lahey, "A polydisperse model for bubbly two-phase
flow around a surface ship", Int. J. Multiphase Flow 25, 257-305 (1999).
Chatzi, E. and J. M. Lee, “Analysis of Interactions for Liquid-Liquid Dispersions in
Agitated Vessels”, Ind. Eng. Chem. Res. 26, 2263-2267 (1987).
Chen, P., M. P. Dudukovic and J. Sanyal, “Three-Dimensional Simulation of Bubble
Column Flows with Bubble Coalescence and Breakup”, AIChE Journal 51, 696-
712 (2005).
Clift, R., J. R. Grace and M. E. Weber, Bubbles, Drops and Particles Academic Press,
New York (1978).
Cokljat, D., V. A. Ivanov, F. J. Sarasola and S. A. Vasquez, “Multiphase k-epsilon
models for unstructured meshes”, ASME 2000 Fluids Engineering Division
Summer Meeting, June 11-15, Boston, Massachusetts, U.S.A. (2000).
Constantinescu, G.S. and K.D. Squires, “LES and DNS Investigations of Turbulent Flow
over a Sphere at Re = 10,000”, Flow, Turbulence and Combustion 70, 267-298
(2003).
Coulaloglou, C. A. and L. L. Tavlarides, “Description of interaction processes in agitated
liquid-liquid dispersions”, Chem. Eng. Sci. 32, 1289-1297 (1977).
Csanady, G. T., "Turbulent Diffusion of Heavy Particles in the Atmosphere", J. Atmos.
Science 20, 201-208 (1963).
149

Dandy, D. S., “Theoretical Studies of droplet deformation in finite Reynolds numbers


flows”, SPIE – The International Society for Optical Engineering 1862, 237-248
(1993).
Dandy, D.S. and H.A. Dwyer, "A sphere in shear flow at finite Reynolds number: effect
of shear on particle lift, drag, and heat transfer", J. Fluid Mech. 216, 381-410
(1990).
Deen, N. G., M. V. S. Annaland and J. A. M. Kuipers, "Multiscale modeling of dispersed
gas-liquid two-phase flow" Chemical Engineering Science 59, 1853-1861 (2004).
Deen, N., "An Experimental and Computational Study of Fluid Dynamics in Gas-Liquid
Chemical Reactors" Ph.D. Thesis , 1-133 (2001).
Delnoij, E., J. A. M. Kuipers and W. P. M. van Swaaij, "A three-dimensional CFD model
for gas-liquid bubble columns" Chemical Engineering Science 54, 2217-2226
(1999).
Delnoij, E., J. A. M. Kuipers and W. P. M. van Swaaij, "Computational fluid dynamics
applied to gas-liquid contactors" Chemical Engineering Science 52, 3623-3638
(1997).
Domgin, J., D. G. F. Huilier, H. Burnage and P. Gardin, “Coupling of a Lagrangian
model with a CFD code: Application to the numerical modelling of the turbulent
dispersion of droplets in a turbulent pipe flow”, Journal of Hydraulic Research 35,
473-488 (1997).
Drew, D. A. and R. T. Lahey Jr., “The Virtual Mass and Lift Force on a sphere in
Rotating and Straining Inviscid Flow” Int. J. Multiphase Flow 13, 113-121
(1987).
Drew, D.A. and R. T. Lahey Jr., “Some supplemental analysis concerning the virtual
mass and lift force on a sphere in a rotating and straining inviscid flow”, Int. J.
Multiphase Flow 16, 1127-1130 (1990).
Eichhorn R. and Small S., “Experiments on the Lift and Drag of Spheres Suspended in a
Poiseuille Flow” J. Fluid Mech. 20, 513-527 (1964).
Elgobashi, S. E., “Particle-laden turbulent flows: Direct simulation and closure models”,
Appl. Sci. Res. 48, 301-314 (1991).
Enwald, H., E. Peirano and A. E. Almstedt, “Eulerian two phase flow theory applied to
fluidization”, International Journal of Multiphase Flow 22, 21-66 (1996).
Farrar, B. and H. H. Bruun, “A computer based hot-film technique used for flow
measurements in a vertical kerosene-water pipe flow”, Internaional Journal of
Multiphase Flow 22, 733-751 (1996).
Farrar, B. and H. H. Bruun, “Hot-film probe measurements in kerosene-water flows” in
Second UK National Conference on Heat Transfer, University of Strathclyde,
UK, 14-16 September, 1269-1279 (1998).
150

Farrar, B., “Hot-film anemometry in dispersed oil-water flows”, Ph.D. Thesis,


Department of Mechanical and Manufacturing Engineering, Bradford University
(1988).
Fei, W. Y., Y. D. Wang and Y. K. Wan, “Physical modeling and numerical simulation of
velocity fields in rotating disc contactor via CFD simulation and LDV
measurement”, Chemical Engineering Journal 78, 131-139 (2000).
Ferziger, J.H. and M. Peric, Computational Methods for Fluid Dynamics Springer
Verlag, Berlin
Fluent ”Fluent 6.1.22 user’s guide” [online].
Available:
www.fluentusers.com/fluent61/doc/ori/html/ug/main_pre.htm [2004].
Fordham, E. J., A. Holmes, R.T. Ramos, S. Simonian, S-M Huang and C. P. Lenn,
“Multi-phase-fluid discrimination with local fibre-optical probes: I Liquid/liquid
flows”, Meas. Sci. Technol. 10, 1329-1337 (1999).
Foussat, A. J. M. and J. P. Hulin, “Vertical liquid-liquid and liquid-gas Two-phase Flow
Measuerments with a Vortex Flowmeter”, Proc. IUTAM Symp. On Measuring
Techniques in Gas-Liquid Two-Phase flow, Nancy, France, July 5-8 (1983).
Frantz, J., ‘g3data’ Data digitizer [online].
Available:
www.acclab.helsinki.fi/~frantz/software/g3data.php [2003].
Friberg, P. C., "Three dimensional modelling and simulation of gas-liquid flows" Ph.D.
Thesis, 1-107 (1998).
Gosman, A.D., C. Lekakou, S. Politis, R.I. Issa, and M.L. Looney, "Multidimensional
Modeling of Turbulent Two-Phase Flows in Stirred Vessels", AIChE Journal 38,
1946 (1992).
Grace, J. R., T. Wairegi and T. H. Nguyen, “Shapes and velocities of single drops and
bubbles moving freely through immiscible liquids”, Trans. Inst. Chem. Engrs. 54,
167-173 (1976).
Grossetete, C., “Caraterisation experimentale et simulations de l’evolution d’un
ecoulemnt diphasique a bulles ascendant dans un conduite verticale”, Collection
de notes internes de la Direction des Études et Recherches. Electricité de France,
December (1995).
Hall, D., “Measurements of the mean force on a particle near a boundary in turbulent
flow”, J. Fluid Mech. 187, 451 (1988).
Hamad, F. A., B. K. Pierscionek and H. H. Bruun, “A dual optical probe for volume
fraction, drop velocity and drop-size measurements in liquid-liquid two-phase
flow”, Meas. Sci. and Technol. 11, 1307-1318 (2000).
151

Hamad, F. A., M. K. Khan and H. H. Bruun, "A Comparison of Predicted and


Experimental Phase Distribution and Turbulence in Two-Phase Flow" The
Phoenics Journal of Computational Fluid Dynamics and its Applications 12, 19-
41 (1999).
Hewitt, G. F., “From Gas-Liquid to Liquid-Liquid two-phase flow: A difficult journey”,
International Symposium on Liquid-Liquid Two-Phase Flow and Transport
Phenomena, Anatalya, Turkey, November 3-7 (1997).
Hinze, O, “Fundamentals of the Hydrodynamics Mechanism of Splitting in Dispersion
Processes”, AIChE Journal 1, 289-295 (1955).
Hjertager, B. H., “Computational Fluid Dynamics (CFD) analysis of Multiphase
Chemical Reactors”, Trends in Chem. Eng. 4, 45-92 (1998).
Ho, C. A., and M. Sommerfeld, “Modeling of micro-particle agglomeration in turbulent
flows”, Chem. Eng. Sci. 57, 3073-3084 (2003).
Hossein, A., J. Naser, K. McManus and G. Ryan, “CFD Investigation of Particle
Deposition and Distribution in a Horizontal Pipe”, Presented at the third
international conference on CFD in the Minerals and Process Industries,
Melbourne, Australia, 10-12 December 2003.
Hu, S. and R. C. Kintner, “The fall of single liquid drops through water”, AIChE Journal
1, 42-48 (1955).
Hyung, R. S., “A validation study for tank sloshing using a Navier-Stokes solver”,
Proceedings of the International Conference on Offshore Mechanics and Arctic
Engineering 3, 461-471 (2004).
Ishii, M. and N. Zuber, "Drag Coefficient and Relative Velocity in Bubbly, Droplet or
Particulate Flows" AIChE Journal 25, 843-855 (1979).
Ishii, M., “Thermo-fluid dynamic theory of two-phase flow”, Direction des Etudes et
Recherches d’Electricité de France, Eyrolles, Paris, France (1975).
Jakobsen, H. A., B. H. Sannæs, S. Grevskott and H. F. Svendsen, "Modeling of Vertical
Bubble-Driven Flows" Industrial and Engineering Chemistry Research 36, 4052-
4074 (1997).
Jaworski, Z. and P. Pianko-oprych, “Two-phase laminar flow simulations in a kenics
static mixer”, Trans IChemE 80 Part A, 910-916 (2002).
Jin, N., W. Wang, X. Liu and S. Tian, “Characterization of Oil-Water Two-phase Flow
Patterns in Vertical Upward Flow Pipes Based on Fractal and Chaotic Time Series
Analysis”, presented at the SPE International Oil and Gas Conference and
Exhibition, Beijing, China, 7-10 November 2000.
Joshi, J. B., “Computational flow modeling and design of bubble column reactors”,
Chem. Eng. Sci. 56, 5893-5933 (2001).
152

Kariyasaki, A., “Behavior of single gas bubble in a liquid flow with a linear velocity
profile” in Proceedings of the 1987 ASME-JSME Thermal Engineering Joint
conference, 261-267 (1987).
Klee, A. J. and R. E. Treybal, “Rate of rise and fall of liquid drops”, AIChE Journal 2,
444-447 (1956).
Koch Modular Process Systems, “Rotating Disc Contactor (RDC) Column” [online].
Available:
http://www.liquid-extraction.com/images/rdc-column.jpg [2000]
Koh, P. T. L. and F. Xantidis, “CFD modelling in the scale-up of a stirred reactor for the
production of resin beads”, in “Second International Conference on CFD in the
Minerals and Process Industries”, CSIRO, Melbourne, Australia, 369-374, Dec. 6-
8, (1999).
Kostoglou, M. and A. J. Karabelas, “Comprehensive modeling of precipitation and
fouling in turbulent pipe flow”, Ind. Eng. Chem. Res. 37, 1536-1550 (1998).
Krishna, R. and J. M. van Baten, "Rise characteristics of gas bubbles in a 2D rectangular
column - VOF simulations vs Experiments" International Communications in
Heat and Mass Transfer 26, 965-974 (1999).
Kulkarni, A.A. and J.B. Joshi, The Lift Forces on Bubbles in a Swarm: Experimental
Analysis Using LDA, "Third International Symposium on Two Phase Modelling
and Experimentation", Pisa, Italy, 2004.
Kumar, A. and S. Hartland, “Gravity settling in Liquid/Liquid Dispersions”, Can. J.
Chem. Eng. 63, 368-376 (1985).
Kuo T. C., C. Pan and C.C. Chieng, "Eulerian-Lagrangian Computations on Phase
Distribution of Two-Phase Bubbly Flows", Int. J. Num. Methods in Fluids 24,,
579–593 (1997).
Lahey Jr., R.T., M. Lopez de Bertodano and O.C. Jones Jr., "Phase distribution
phenomena in complex geometry conduits", Nuclear Eng. and Design 141, 177-
201 (1993).
Lahey, R. T., “Turbulence and phase distribution phenomena in two-phase flow”,
ICHMT International Seminar on Transient Phenomena in Multiphase flow,
Dubrovnik, Yugoslavia, May 24-30 (1987).
Laín, S., D. Broder, M. Sommerfeld and M. F. Goz, “Modeling hydrodynamics and
turbulence in a bubble column using Euler-Lagrange procedure”, International
Journal of Multiphase Flow 28, 1381-1407 (2002).
Lance, M. and M. L. Bertodano, “Phase distribution phenomena and wall effects in
bubbly two-phase flows, Proceedings of third international workshop on two-
phase flow fundamentals”, London, UK, June (1992).
153

Lang, P. and H. Auracher, “Heat Transfer to Nonmiscible Liquid-Liquid Mixtures


Flowing in a Vertical Tube”, Experimental Thermal and Fluid Science 12, 364-
372 (1996).
Lang, P., “Experimentelle Untersuchung von Phasenverteilung und Wärmeübergang in
einer Wasser/n-Heptan Flüssig/Flüssig-Zweiphasenströmung”, Ph.D. Thesis,
Technischen Universität Berlin (1993).
LaRoche, R. D. and A. Bakker, “The colorful fluid mixing gallery - Kenics Static Mixer”
[online].
Available:
http://www.bakker.org/cfm/km2.gif [1998].
Launder, B. E. and D. B. Spalding, Lectures in Mathematical Models of Turbulence
Academic Press, London, England (1972).
Lo, M. Y. A., A. T. Gierczycki, N. J. Titchner-Hooker and P. A. Shamlou, “Power Input
and Drop Size Distribution in Two Immiscible Liquid Phase Operations in
Vessels Agitated by Up-and-Down Moving impellers”, Can. J. Chem. Eng. 76,
471-478 (1998).
Lopez de Bertodano, M., "Two fluid model for two-phase turbulent jet", Nucl. Eng. Des.
179, 65 (1998).
Lopez de Bertodano, “Turbulent Bubbly Two Phase Flow in a Triangular Duct”, Ph. D.
Thesis, Rensselaer Polytechnic Institute, New York, 1992.
Loth, E., M. Taeibi-Rahni and G. Tryggvason, “Deformable bubbles in a free shear
layer”, Int. J. Multiphase Flow 23, 977-1001 (1997).
Lucas, D., E. Krepper and H.-M. Prasser, “Development of bubble size distributions in
vertical pipe flow by consideration of radial gas fraction profiles”, 4th
International conference on Multiphase flow, New Orleans, May 27 – June 1,
Paper 378 (2001).
Lun, C. K. K., S. B. Savage, D. J. Jeffrey and N. Chepurniy, “Kinetic Theories for
Granular Flow: Inelastic particles in Couette Flow and Slightly Inelastic Particles
in a General Flow Field”, Journal of Fluid Mechanics 140, 223-256 (1984).
Majander, J. and M. Manninen, “Numerical simulation of flow induced by a pitched
blade turbine: The multiple reference frame technique” Technical Report LVT
3/98, VTT Energy, Finland (1998).
Majander, J. and M. Manninen, “Numerical simulation of flow induced by a pitched
blade turbine: Comparison of the sliding mesh technique and an averaged source
term method”, in Proceedings of the Third Colloquium on Process Simulation,
Helsinki University of Technology, Espoo, Finland, 12-14 June (1996).
Mann, R. and P. Mavros, Proceedings of 4th European Conference in Mixing, p 35
(1982).
154

Martinuzzi, R. and A. Pollard, “Comparitive study of turbulence models in predicting


turbulent pipe flow. I – Algebraic stress and k-epsilon models”, AIAA Journal 27,
29-36 (1989).
Matas, J.P., J.F. Morris and E. Guazzelli, “Lateral forces on a Sphere”, Oil & Gas
Science and Technology – Rev. IFP 59, 59-70 (2004).
Mei, R., “An approximate expression for the shear lift force on a spherical particle at
finite Reynolds number”, Int. J. Multiphase Flow 18, 145-147 (1992).
Mei, R., “History force on a sphere due to a step change in the free stream velocity”,
International Journal of Multiphase Flow 19, 509-525 (1993).
Modes, G. and H.-J. Bart, “CFD Simulation of Nonideal Dispersed Phase Flow in Stirred
Extraction Columns”, Chem. Eng. Technol. 24, 1242-1245 (2001).
Moraga, F. J., “Lateral forces on rigid spheres in a turbulent uniform shear flow”, Ph.D.
Thesis, Renesselaer Polytechnic Institute, Troy, New York (1998).
Moraga, F. J., F. J. Bonetto and R. T. Lahey, "Lateral forces on spheres in turbulent
uniform shear flow" International Journal of Multiphase Flow 25, 1321-1372
(1999).
Navarro-Valenti, S., R. T. Lahey Jr., D. A. Drew and M. Mills, “Two-fluid modeling of
the bubbly/slug flow regime transition”, Transactions of the American Nuclear
Society Winter Meeting, 69, 495-496 (1993).
Nigmatulin, T. R., F. J. Bonetto, A. E. Larrereguy, R. T. Lahey Jr and J. B. Mcquillen,
“An Experimental study of dispersed liquid-liquid two-phase upflow in a pipe”,
Chem. Eng. Comm. 182, 121-162 (2000).
Okamoto, Y., M. Nishikawa and K. Hashimoto, “Energy dissipation rate distribution in
mixing vessels and its effects on liquid-liquid dispersion and solid-liquid mass
transfer”, International Chemical Engineering 21, 88-94 (1981).
Patro, R., Bubbleology Research, “Bubble Hydrodynamics ” [online].
Available:
http://www.bubbleology.com/Hydrodynamics.html [2000]
Peirano, E., “Modeling and simulation turbulent gas-solid flows applied to fluidization”,
Ph.D. Thesis, Chalmers University of Technology, Goteborg, Sweden (1998).
Peirano, E. and B. Leckner, “Fundamentals of turbulent gas-solid flow applied to
circulating fluidized bed combustion”, Proc. Energy Combustion Sci. 24, 259-296
(1998).
Petersen K O “Etude Experimentale Numerique Des Ecoulements Disphasiques Dans Les
Reacteurs Chimiques“, Ph.D Thesis, L’ Universite Claude Bernald, Lyon (1992).
Placek, J., L. L. Tavlarides, G. W. Aminth and I. Font, “Turbulent flow in Stirred tanks –
Part II. A two-scale model of turbulence”, AIChE Journal 32, 1771-1786 (1986).
155

Podila, K., “CFD modeling of Turbulent Bubbly Flows in Vertical Pipes”, M. A. Sc.
Thesis, Department of Chemical Engineering, Dalhousie University, Halifax, NS
(2005).
Pollard, A. and R. Martinuzzi, “Comparitive study of turbulence models in predicting
turbulent pipe flow. II – Reynolds stress and k-epsilon models”, AIAA Journal
27, 1714-1721 (1989).
Prasad R.O., Bakker A., Marshall E.M., “Numerical Modeling of Mixing Processes -
What Can LES Offer?” Presented at the 1998 Annual Meeting of the AIChE,
Session 238, Nov. 15-20, 1998, Miami Beach.
Ranade, V., Computational Flow Modeling for Chemical Reactor Engineering,
Academic press, 2002.
Rieger, R., “Application of a 3D two-phase CFD code to simulate an RDC column”,
Comput. Chem. Eng. 18, S229-S233 (1994).
Rieger, R., C. Weiss, G. Wigley, H.-J. Bart and R. Marr, “Investigating the process of
liquid-liquid extraction by means of Computational Fluid Dynamics”, Comput.
Chem. Eng. 20, 1467-1475 (1996).
Rodi, W., “Turbulence models for Environmental Problems”, Lecture series 2979-2, Von
Karman Institute (1979).
Rusche, H. and R. Isaa, I., The Effect on Voidage on the Drag Force on Particles in
Dispersed Two-Phase Flow, "Japanese-European two phase flow meeting",
Tsukuba, Japan, 2000.
Sabot, J., “Etude de la Cohérence Spatiale et Temporelle de la Turbulence Etablie en
Conduite Circulaire’”, Ph.D. Thesis, Lyon University (1976).
Saffman, P.G., "The Lift on a Small Sphere in a Slow Shear Flow", J. Fluid Mech. 22,
385-400 (1965).
Saffman, P.G., “On the motion of small spheroidal particles in a viscous liquid”, J. Fluid
Mech. 1,540-553 (1956).
Sakamoto, H., H. Haniu, "The formation mechanism and shedding frequency of vortices
from a sphere in uniform shear flow", J. Fluid Mech. 287, 151-171 (1995).
Schiller, L. and Z Naumann., "Über die grundlegenden Berechungen bei der
Schwerkraftbereitung ", Z. Ver. Deutsch. Ing. 77, 318-320 (1935).
Shi, H., W. P. Jepson and L. D. Rhyne“Segregated Modeling of Oil-Water Flows”,
presented at the SPE Annual Technical Conference and Exhibition, Denver,
Colarado, U.S.A., 5-8 October 2003.
Simonian, S., “Measurement of Oil-Water flows in deviated pipes using thermal
anemometry and optical probes”, Ph.D. Thesis, City University (1993).
156

Simonin, C. and P.L. Viollet, "Predictions of an Oxygen Droplet Pulverization in a


Compressible Subsonic Coflowing Hydrogen Flow", Numerical Methods for
Multiphase Flows, FED91, 65-82 (1990).
Simonin, O. and P. Viollet, “Modeling of turbulent two-phase jets loaded with discrete
particles” in Proceedings of International Seminar on Phase Interface Phenomena
in Two-Phase Flow, Dubrovnik, ICHMT Belgrad (1990b).
Simonin, O., “Summerschool on ‘Numerical modeling and prediction of dispersed phase
flows’”, IMVU, Meserburg, Germany (1995).
Sokolichin, A., G. Eigenberger and A. Lapin, “Simulation of buoyancy driven bubbly
flow: Established simplifications and open questions”, AIChE Journal 50, 24-45
(2004).
Soleimani, A., C. J. Lawrence and G. F. Hewitt, “SpatialDistribution of Oil and Water in
Horizontal Pipe Flow”, presented at SPE Annual TechnicalConference and
Exhibition, Houston, Texas, 3-6, October 1999.
Sommerfeld, M., “Validation of a stochastic Lagrangian modeling approach for inter-
particle collisions in homogeneous isotropic turbulence”, International Journal of
Multiphase Flow 27, 1829-1859 (2001).
Soon, S. Y., J. Harbidge, N. J. Titchener-Hooker and P. A. Shamlou, “Prediction of Drop
Breakage in an Ultra High Velocity Jet Homogenizer”, J. Chem. Eng. Jpn. 34,
640-646 (2001).
Squire, K. D. and J. K. Eaton, “Effect of selective modification of turbulence on two
equation models for particle laden turbulent flows”, ASME J. FED 116, 778
(1994).
Stock, D. E. and E. E. Michaelides, “Turbulence Modification in Dispersed Multiphase
flows” in ASME Fluids Eng. Div. Publ., San Diego, CA, USA, July 9-12 (1989).
Stover, R. L., C. W. Tobías and M. M. Denn, “Bubble coalescence dynamics”, AIChE
Journal 43, 2385-2392 (1997).
Taeibi-Rahni, M. and E. Loth, “Forces on a large cylindrical bubble in an unsteady
rotational flow”, AIChE Journal 42, 638 (1996).
TChen, C. M., “Mean value and correlation problems connected with the motion of small
particles suspended in a turbulent fluid”, Ph.D. Thesis, Delft University of
Technology, The Netherlands (1947).
Thakre, S. S. and J. B. Joshi, “CFD simulation of bubble column reactors: importance of
drag force formulation”, Chem. Eng. Sci. 54, 5055-5060 (1999).
Tomiyama, A., "Struggle with Computational Bubble Dynamics", Multi. Sci. Tech. 10,
369-405 (1998).
Tomiyama, A., H. Tamai, I. Zun and S. Hosokawa, "Transverse Migration of Single
Bubbles in Simple Shear Flows", Chem. Eng. Sci. 57, 1849-1858 (2002).
157

Troshko, A., V. Ivanov and S. Vasques, “Implementation of a general lift coefficient in


the CFD model of turbulent bubbly flows”, Proceedings of the ninth Annual
Conference of the CFD Society of Canada, May 27-29, Waterloo, Ontario (2001).
Van Wachem, B. G. M. and A. E. Almstedt, “Methods for multiphase computational
fluid dynamics”, Chemical Engineering Journal 96, 81-98 (2003).
Versteeg, H. K. and W. Malalasekera, ”An Introduction to Computational Fluid
Dynamics: The Finite Volume Method”, Addison-Wesley, 1995.
Vigneaux, P., P. Chenais and J. P. Hulin, “Liquid-Liquid Flows in an Inclined Pipe”,
AIChE Journal 34 781-789 (1988).
Vikhansky, A. and M. Kraft, "Modelling of a RDC using a combined CFD-population
balance approach" Chemical Engineering Science 59, 2597-2606 (2004).
Wang, S.K.,S.J. Lee, J. Jones, O. C. and J. Lahey, R. T., "3-D Turbulence Structure and
Phase Distribution Measurements in Bubbly Two-Phase Flows", Int. J.
Multiphase Flow 13, 327-343 (1987).
Wang, Y. D., W. Y. Fei, J. H. Sun and Y. K. Wan, “Hydrodynamics and mass transfer
performance of a modified rotating disc contactor (MRDC)”, Chem. Eng. Res.
Des. 80, 391-399 (2002).
Wörner, M., "A Compact Introduction to the Numerical Modeling of Multiphase Flows"
Technical Report, 1-47 (2003).
158

APPENDICES
159

APPENDIX 1

A1.1 CFD SIMULATION METHODOLOGY


A1.1.1 Numerical Solution of Differential Equations
Mathematical models of flow processes are typically non-linear partial differential
equations. Analytical solutions to such equations are possible only for some simple cases.
As a result, in the majority of flow processes, numerical solution of the governing
equations is generally considered as the norm. Typically, numerical solution replaces
continuous information contained in the exact solution of partial differential equations by
discrete information that is available at a finite number of locations (usually referred to as
grid points). The values of dependent variables at these grid points are unknown. A
suitable numerical method then provides a set of algebraic equations (called discretized
equations) for these unknowns and prescribes an algorithm to solve the equations as well.
For a given differential equation, there can be several ways to arrive at a set of discretized
equations (finite element, finite difference and finite volume). Of the three, the finite
volume method, which calculates the values of the conserved variables averaged across
the volume, ensures integral (as in inbuilt) conservation of mass, momentum and energy
over any group of control volumes even if these control volumes are associated with a
small number of grid points. For the limiting case of one computational cell the finite
volume equations become equivalent to those used for a completely mixed reactor
(Ranade, 2002). In this method, the differential equation is integrated over the volume of
each computational cell to obtain the discretized equations. One advantage of the finite
volume method over the finite difference method is that it does not necessarily require a
structured mesh (i.e. the finite volume method can be applied to a grid where there is no
restriction on the shape of the control volume and the number of neighboring nodes)
although a structured mesh can also be used. Moreover, it is preferable to other methods
(finite-element, finite-difference) due to the fact that boundary conditions can be applied
non-invasively.
160

Objectives of Flow modeling – Developing guidelines for interphase closure

Development of modeling approach


1. Define objectives of the flow model – (Multiphase and Turbulence modeling)
2. Space and time scale analysis of problem under consideration (3D and steady state)
3. Evaluate possible simplifications (Axisymmetry in 2D and symmetry in 3D)
4. Devise an approach to achieve model objectives within the resource constraints (Standard wall functions
wherever possible, finer grid resolution near the wall, network of parallel workstations)
5. Identify key fluid dynamic issues (turbulence and interphase forces – drag, lift and turbulent dispersion)
6. Develop model equations (Multiphase: Eulerian-Eulerian, Turbulence: Standard k-ε)

Geometry modeling and grid generation


1. Finalize solution domain and boundary conditions (Velocity inlet, Pressure outlet, Wall, Axis/Symmetry)
2. Geometry modeling (Build from scratch using GAMBIT)
3. Examine resolution requirements/grid spacing & distribution (Structured quad grid; Y+: Refer Section A1.1.2)
4. Formulate grid sequencing/refining strategies (Modify cell aspect ratio in order to bring it closer to 1:1)

Numerical solution of model equations


1. Specification of required system data and boundary conditions (Obtain from experimental conditions)
2. Selection of appropriate numerical method:
a. Discretization scheme (Refer Table A1.1)
b. Algorithm (Finite volume: Fluent 6.1.22; Pressure-Velocity coupling: Refer Section A1.1.3.1)
c. Solution of Algebraic equations (Momentum, Continuity, Turbulent kinetic energy and dissipation rate)
3. Specification of suitable numerical parameters:
a. Under-relaxation factors (Refer Table A1.4)
b. Time steps (N/A for steady state, 0.001 seconds for unsteady state)
c. Internal iterations (≈ 20 per time step)
d. Convergence criterion (Refer Table A1.3)
4. Generate data files containing solution of model equations

Analysis & interpretation of solution


1. Accessing influence of numerical parameters on simulated results
a. Grid spacing (Refer Appendix 2)
b. Time step (N/A for steady state)
2. Qualitative evaluation – Are key flow features captured? (velocity/phase distribution profiles)
3. Quantitative evaluation/Validation – Understanding limitations of simulations (physical as well as numerical)
4. Abstract useful information from simulated results (phase and slip velocity distributions, actual velocities)

Achieve objectives of flow modeling

Figure A1.1 CFD solution procedure for immiscible liquid dispersions (Adapted from
Ranade, 2002)
This is true because the values of the conserved variables are located within the volume
element, and not at nodes or surfaces (Versteeg and Malalasekera, 1995). Moreover, the
finite-difference methods are restricted to application in simple geometries and special
care needs to be taken to ensure global conservation. Although the finite element methods
161

can be successfully applied to complex geometries, it is difficult to develop


computationally efficient solution methods in their framework when the governing
equations are strongly coupled and non-linear (Ranade, 2002). The overall procedure for
CFD simulations of liquid-liquid dispersions is presented in Figure A1.1.

The necessary computational tools required to carry out these steps can be generally
classified into three categories:

1. Pre-processing (includes geometry modeling and grid generation tools)

2. Solving (implementation of numerical methods to solve the governing equations)

3. Post-processing (analysis and interpretation of simulation results)

In this work, grid generation was done using GAMBIT 2.1.2, a commercial geometry
modeling suite. As multi-fluid up-flow in vertical pipes is aligned with the grid and
features simple hydrodynamics, a typical structured ‘quad grid’ (quadrilateral cells) was
used. A typical 2D structured ‘quad grid’ is shown in Figure A1.2. The pipe inlet is
located far to the left (not shown in figure).

First grid point

+ve

+ve

Figure A1.2 A typical 2D structured grid


162

Grid resolution is usually maintained finer near the wall, as walls are the main source of
turbulence and hence, large gradients in solution variables are expected in their vicinity.
A finer resolution would therefore increase the accuracy of near-wall predictions. In
GAMBIT, it is necessary to create the grid such that the pipe centerline lies on the
positive X-axis, even if the pipe is physically vertical. Setting the gravity to - 9.81 m/s2 in
the X-direction ensures that a vertical pipe up-flow is simulated.

A1.1.2 Placement of the First Grid Point


For Standard wall functions, Y+ values in the range 30 to 60 (preferably closer to the
lower bound), and for Enhanced wall treatment, Y+ values in the range 1 to 5 are
recommended (Fluent, 2004). In view of the above recommendations, a Y+ value of 35
when using Standard wall functions and a value of 5 when using Enhanced wall
treatment were used to estimate the distance of the first grid point from the wall. At least
ten computational cells were placed within the viscosity affected region near the wall
(Rey ≤ 200) when the Enhanced wall treatment approach was used. This was necessary in
order to effectively resolve the mean velocity and turbulence quantities in that region
(Fluent, 2004). ‘Rey’ is defined as the turbulent Reynolds number. It is calculated based
on the normal distance from the wall (y) and the turbulent kinetic energy of the
continuous phase (kc) as follows:

ρc y kc
Re y = [A1.1]
µc

After a converged solution, the radial distribution of Rey was checked to ensure that, at
least the first ten grid points from the wall were placed within Rey ≤ 200.

An estimate of the distance of the first grid point from the wall is obtained using the
Blasius resistance formula for pipelines (Schlichting, 1979):

µ cY +
Distance of first node point from the wall = [A1.2]
τ w ρc
163

It should be noted that the distance of the first grid point from the wall is twice the
distance of the first node point from the wall. The distance of the first grid point from the
wall is also referred to as the first length in GAMBIT. The wall shear stress (τ w ) in

Equation A1.2 is calculated based on a dimensionless coefficient of resistance (λ ) and the


continuous phase superficial velocity (u c ) (Equation A1.3).

τ w = λρ c u c2 8 [A1.3]

The dimensionless coefficient of resistance is obtained directly from the flow Reynolds
number (Reflow, which is calculated based on the pipe diameter and superficial velocity of
the continuous phase) as shown below (Equation A1.4).

λ = 0.3164 Re 0flow
.25
[A1.4]

It should be noted that the above method of estimating the first grid point is only valid if
Re flow ≤ 100000 . A rudimentary program (written in the ‘C’ language) to calculate the

distance of the first grid point from the wall, given the Y+ and superficial velocity (u c ) is

also available in Appendix A3.1.4.

A1.1.3 Solution of the Governing Equations


All governing equations (Equations 5.2, 5.5 and 5.6) were solved in a sequential fashion
(i.e. segregated from one another) using the commercial finite volume-based CFD solver,
Fluent 6.1.22. Simulations were carried out in 2D in order to simplify the problem and
also reduce the computation overhead. As liquid-liquid vertical pipe flows feature steady-
state behavior, except possibly near the phase inversion point, steady state conditions
were assumed. A time-dependant (unsteady state) solution procedure was adopted if the
inlet flow conditions gave rise to numerical instabilities during the simulation. For all
such cases, a fixed time step size of 0.001 s and a maximum of 20 iterations per time step
were employed. For discretization, a second-order upwind scheme which is usually
recommended (Fluent, 2004; AIAA, 1998), was employed as it is more attractive from an
accuracy viewpoint when compared to the more numerically diffusive and only first-
164

order accurate, first-order upwind scheme. In addition, the inherent ability of second-
order schemes to suppress any non-physical ‘wiggles’, in the CFD solution warrants their
preference (Ranade, 2002). Discretization schemes employed in the present study are
summarized in Table A1.1. A double-precision solver was used in place of a single-
precision solver as disparate length scales (e.g. a long thin pipe) were encountered in
some of the simulations. Under these circumstances, a single-precision solver would fail
to adequately represent any node coordinates (Fluent, 2004). Moreover, several
preliminary simulations were carried out with high-aspect-ratio grids (where the width of
a cell was four to five times its height). Such grids are well known to impair the
convergence and/or accuracy of the final solution if a single-precision solver is used
(Fluent, 2004).
Table A1.1 Summary of discretization schemes used in the present study

Flow variable Discretization scheme

Momentum 2nd Order Upwind

Volume fraction 2nd Order Upwind

Turbulent kinetic energy 2nd Order Upwind

Turbulent dissipation rate 2nd Order Upwind

A1.1.3.1 Pressure-Velocity Coupling


A unique feature of the momentum equation which distinguishes it from other governing
equations is the role played by pressure. There is no obvious method to determine the
pressure gradients (∇p) which appear in the source terms of the momentum equation. The
pressure field is indirectly specified through the continuity equation. It therefore becomes
necessary to estimate the pressure field in such a fashion that the resulting velocity field
satisfies the continuity equation. Typically, special treatments are used to convert the
indirect information in the continuity equation into a direct algorithm (Pressure-Velocity
coupling) which would determine the pressure. Implicit/Semi-Implicit pressure correction
methods are invariably used for this purpose in most incompressible flows (Ranade,
165

2002). One of the widely used popular methods proposed by Patankar and Spalding
(1972) is called SIMPLE (semi-implicit method for pressure linked equations). In this
method, discretized momentum equations are solved using a guessed pressure field. An
extension of the SIMPLE algorithm to multiphase flows called PC-SIMPLE (phase-
coupled SIMPLE) where velocities are solved coupled by phases in a segregated fashion,
is used in the present liquid-liquid study. For steady incompressible multiphase flow in
the absence of mass transfer, the pressure-correction equation takes the form:

{∇ ⋅ α }
n

k v k′ + ∇ ⋅ α k v k* = 0 [A1.5]
k =1

Where ‘ vk′ ’ is the velocity correction and ‘αk’ is the volume fraction for phase ‘k’. The

value of v k at the current iteration is given by vk* . The velocity corrections are themselves

expressed as functions of the pressure corrections.

A1.1.3.2 Turbulence Modeling in Fluent 6.1.22

The present study employs the dispersed approach to multiphase turbulence in Fluent
6.1.22 (Cokljat et al., 2000). Interphase momentum source is enabled through the text
user interface (TUI) in Fluent in order to account for turbulent dispersion using the
Simonin and Viollet (1990) approach. The degree of turbulent dispersion was varied by
modifying the Dispersion Prandtl Number in the Viscous model panel of Fluent 6.1.22.
As turbulence modulation was not considered in the present study, Interphase k-ε source
was left disabled in the TUI.

A1.1.4 Boundary Conditions


The following boundary conditions were used to limit/define the solution domain:

1. Pipe inlet: Velocity inlet boundary condition was used to define the flow velocity,
along with all relevant scalar properties of the flow, at the flow inlet. A constant,
166

uniform velocity (across the cross section), normal to the inlet boundary was
specified in all the simulations.

2. Pipe outlet: A Pressure outlet boundary condition which requires the specification
of a static (gauge) pressure at the outlet boundary was used at the pipe exit. In the
present work, a zero gauge pressure which corresponds to an absolute pressure
equal to the atmospheric pressure was specified at the outlet. An alternative to the
pressure outlet boundary condition is the Outflow boundary condition which
imposes a zero diffusive flux in a direction normal to the exit plane on all flow
variables. However, the use of a pressure outlet boundary condition over the
outflow boundary condition is justified in view of the fact that the outflow
condition is strictly valid only for fully developed flows, which may not entirely
true for some of the experimental conditions (Farrar, 1988; Al-Deen and Bruun,
1997) simulated in the present study. Also, convergence may be affected if there
is recirculation through an outflow boundary. This is numerically plausible even if
the final solution is not expected to feature any kind of physical backflow,
particularly for turbulent flow simulations (Fluent, 2004). The pressure outlet
boundary condition does not suffer from these limitations and is therefore a
proper choice of a boundary condition for the pipe exit.

3. Wall: Wall boundary conditions were used to bind fluid and solid regions. In
viscous flows, the 'no-slip' boundary condition is enforced at walls. The shear
stress between the fluid and wall is computed based on the flow details in the
local flow field.

4. Axis: Vertical liquid-liquid pipe flows are typically symmetric about the pipe
axis. An axisymmetric grid was therefore used to reduce the number of
computational cells and therefore decrease the time and computational power
required for the simulations. To enforce this condition, an Axis boundary
condition was applied at the pipe centerline. The appropriate physical value for a
particular variable at a point on the axis is determined using the cell value in the
adjacent cell (Fluent, 2004).
167

A1.1.5 Simulation Conditions


In all the simulations it was assumed that phases are incompressible (constant density)
and viscosities were assumed to be constant as well. Also, no heat/mass transfer effects
were considered in any of the simulations. For all kerosene-water data sets, standard
physical properties of both liquids were used as they were not specified by the
investigators. However, for the n-heptane-water system, physical properties as specified
by the investigators (Lang and Auracher, 1996; Lang, 1999) were used. The physical
properties used in the present study are listed in Table A1.2.

An initial guess for the dispersed phase volume fraction was provided at the inlet and exit
boundary. Typically, for a non-zero positive slip velocity (i.e. when the dispersed phase
moves faster than the continuous phase) between the phases, any non-zero value less than
the phase ratio of the dispersed phase can be given as the initial guess. However, if a
reasonable guess is provided, the solution process would be faster. As a result, for each
data point, the area-averaged dispersed phase holdup as measured by the experiments or
calculated from the experimental data was provided as the initial guess.
Table A1.2 Physical properties of the liquids used in the present study

Continuous phase Dispersed phase

Liquid-Liquid system properties properties

(Dispersed-Continuous)
Density Viscosity Density Viscosity
3 3
(kg/m ) (kg/ms) (kg/m ) (kg/ms)

Kerosene-water 998.2 0.001003 780 0.002400

n-heptane-water 994.0 0.000828 677 0.000365

Physical (actual) velocities which are calculated based on the superficial phase velocities
were specified at the velocity inlet (Equation A1.6).

Superficial velocity of the phase


Physical velocity of a phase = [A1.6]
Volume fraction of the phase
168

For the phase volume fraction in Equation A1.6, the initial volume fraction guess was
used.

Turbulence boundary conditions at inlet and exit were specified using the hydraulic
diameter (equal to the pipe diameter) and turbulence intensity. The initial guess for
turbulence intensity (I) was calculated based on the hydraulic diameter (DH) as follows:

I = 0.16 Re −D1H8 [A1.7]

‘ Re DH ’ is the flow Reynolds number calculated based on the hydraulic diameter.

Inlet guesses for the turbulent kinetic energy (k) and dissipation rate (ε) were calculated
based on the physical velocity (u c α c ) and the turbulence length scale ( l ) using

Equation A1.8.

2 3
2
3 uc 3 k
k= I and ε = Cµ 4
[A1.8]
2 αc l

The turbulence length scale ( l ) in fully-developed duct flows is given by l = 0.07 D ,


where ‘D’ is the pipe diameter.

Providing volume fraction and turbulence intensity guess values at the exit is useful
because, if at any point during the iterative process backflow of either phase is
encountered (due to numerical solution instabilities), these non-zero quantities will be
used to help reduce convergence difficulties.

A1.1.6 Solution Convergence


Solution convergence was judged by monitoring residuals (defined as the imbalance in a
discretized governing equation summed over all the cells in the computational domain)
and more importantly, relevant flow quantities (e.g. velocity magnitude of the fluid at a
fixed point in the solution domain, area-weighted average of phase volume fraction at the
exit etc.). Convergence criteria for the flow variables solved in the present study are
given in Table A1.3. When solving the governing equations at steady state, the
169

convergence criteria for continuity was lowered from 10-3 to 10-13 in order to judge
convergence exclusively based on the iteration history of other flow quantities as
discussed above. This is useful in cases where the initial guesses may not be so good. In
such situations, the initial residuals can be large enough that even a three-order drop in
residuals does not necessarily guarantee convergence (Fluent 2004).

Table A1.3 Convergence criteria used in the present study

Convergence
Flow variable
criterion
Continuity 10-13

Volume fraction 10-03

Axial/radial velocity 10-03

Turbulent kinetic energy 10-03

Turbulent energy dissipation rate 10-03

Table A1.4 Under-relaxation factors used in the present study

Flow variable Value

Pressure 0.3

Density 1.0

Body forces 1.0

Momentum 0.7

Volume fraction 0.7

Turbulent kinetic energy 0.8

Turbulent dissipation rate 0.8

Turbulent viscosity 1.0


170

When the residuals and monitored flow quantities stabilized, or exhibited uniform
fluctuations about a mean value, the mass balance was checked. The simulation was
stopped when the difference in mass flow rate between the inlet and exit was less than at
least 0.001. If numerical issues such as solution divergence were encountered, under-
relaxation factors (typically used to control the magnitude of change of a variable within
each iteration), were lowered until the solution stabilized. It should be noted that
converged solutions are independent of the under-relaxation factors chosen (Ranade,
2002). Under-relaxation factors that were used in the present study are listed in Table
A1.4.

A1.1.7 Post-processing
The several post-processing facilities included with Fluent solver (vector/contour plots,
XY plots etc.) were used to visualize and interpret the simulation results. ‘g3data’
(Frantz, 2003), a versatile data digitizer was used to extract data points from experimental
graphs for comparison with CFD predictions.

A1.1.8 Additional Considerations for 3D Simulations


Certain modifications to the CFD methodology described so far are necessary when 3D
simulations are carried out. As most of the solution procedure is similar to that described
earlier in this chapter, only the differences encountered while pre-processing are
explicitly discussed.

One of the foremost pre-processing aspects relates to the choice of a suitable 3D grid. As
the geometry being modeled (cylindrical pipe) was quite simple, a structured hexahedral
grid was used. Figure A1.3 shows a closer view near the inlet section of a typical 3D grid
created using GAMBIT. The boundary layer option in GAMBIT was used to specify the
location of the first grid point from the wall and also ensure a finer mesh resolution in the
near-wall regions. The cooper meshing scheme was employed to mesh the volume using
pre-meshed inlet and outlet faces as sources. The number of grid points on the semi-
circumference of the inlet/outlet faces was controlled by manually meshing the curved
edges before applying the boundary layer.
171

Wall
Symmetry plane

Gravity
Inlet

Figure A1.3 3D structured grid composed of hexahedral cells


Again, care was taken to ensure that the entire geometry was created in the positive
quadrant for all three dimensions. For the simulation of liquid-liquid flows in horizontal
pipes, a symmetry boundary condition in the vertical plane was applied to reduce the
number of computational cells. The procedure for estimating the location of the first grid
point was the same as before (i.e. as described in Section A1.1.2). The size of a typical
(coarse) grid was at least 150,000 cells.
172

APPENDIX 2

A2.1 GRID INDEPENDENCE STUDIES


Grid-independence tests were carried out to determine whether an increase in the grid
size (finer discretization) would have a significant impact on the results. Predicted
velocity and dispersed phase holdup profiles at different experimental conditions (pipe
diameters, phase velocities, dispersed phase holdups etc.) were compared at different grid
sizes for this purpose. The study was performed by increasing radial and axial nodes until
a grid-independent solution was achieved. Care was taken to ensure that the Y+ was in the
range of 30 - 35 when using ‘Standard wall functions’. Also, cell aspect ratios never
exceeded 1:5 in any grid that was tested. This is usually recommended in order to avoid
numerical divergence issues (Fluent, 2004). All simulation parameters except the grid
size were kept constant throughout the study. Results from two experimental data sets
featuring different dispersed phase holdups as well as different pipe diameters are shown
here for illustration purposes. The corresponding experimental conditions and grid sizes
tested are listed in Table A2.1.
Table A2.1 Summary of the grid independence study

No. of radial cells in


Investigator Grid size the axisymmetric Pipe dimensions
domain

1750 cells 8
Al-Deen and Bruun (1997) Length = 1.5 m
3000 cells 11
Data point A5 ID = 77.8 mm
9760 cells 17

4900 cells 8
Vigneaux et al. (1988) Length = 14 m
8000 cells 11
Data point V30 ID = 200 mm
23400 cells 14
173

Results of the grid independence study are presented in Figures A2.1 through A2.4.

0.7

Continuous phase velocity (m/s)


0.6

0.5

0.4 9760 cells

0.3 3000 cells

0.2 1750 cells

0.1

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure A2.1 Comparison of simulated continuous phase velocity profiles for the data set
of Al-Deen and Bruun (1997), A5 (DPN = 0.024; CL = Troshko et al. (2001); de = 3 mm)

0.07

0.06
Dispersed phase holdup (-)

0.05

0.04

0.03

0.02
9760 cells 3000 cells

0.01 1750 cells

0
0 0.01 0.02 0.03 0.04
Radial position (m)

Figure A2.2 Comparison of simulated phase distribution profiles for the data set of Al-
Deen and Bruun (1997), A5 (DPN = 0.024; CL = Troshko et al. (2001); de = 3 mm)
174

0.25

Continuous phase velocity (m/s)


0.2

0.15
23400 cells

0.1 8000 cells

4900 cells
0.05

0
0 0.02 0.04 0.06 0.08 0.1
Radial position (m)

Figure A2.3 Comparison of simulated continuous phase velocity profiles for the data set
of Vigneaux et al. (1988), V30 (DPN = 4.125; CL = Troshko et al. (2001); de = 5 mm)
0.3

0.25
Dispersed phase holdup (-)

0.2

23400 cells
0.15

8000 cells
0.1

4900 cells
0.05

0
0 0.02 0.04 0.06 0.08 0.1
Radial position (m)

Figure A2.4 Comparison of simulated phase distribution profiles for the data set of
Vigneaux et al. (1988), V30 (DPN = 4.125; CL = Troshko et al. (2001); de = 5 mm)
175

For data point A5 from the Al-Deen and Bruun (1997) data set, it is seen from Figures
A2.1 and A2.2 that, grid independence is achieved around 3000 cells. However, the grid
composed of 9760 cells was used for final simulations as it featured cells with a highly
desirable aspect ratio (1:1).

For data point V30, simulation results of continuous phase velocity and dispersed phase
holdup for three different grid sizes are presented in Figures A2.3 and A2.4 respectively.
It is seen that grid independence is realized with about 8000 cells. However, the grid
composed of 23400 cells was used for final simulations as it featured cells with an aspect
ratio 1:1.
176

APPENDIX 3

A3.1 SOURCE CODES


Source code samples written in the ‘C’ programming language that were used to test
various drag/lift and turbulent dispersion expressions are presented in this section.
However, a concise introduction to the implementation aspects is first provided.

A3.1.1 Accounting for the Drag Force in Fluent 6.1.22


Interphase drag force is implemented using an interphase exchange term (Kpq) in Fluent.
The interphase exchange coefficient (Kpq) used in the present work is given by:

3 C D α d ρ cV S
K pq = [A3.1]
4 de

Where:

VS = (v p − v q ) [A3.2]

Equation A3.1 is then substituted for Kpq in the momentum conservation equation
(Equation 5.5). The product (K pq (v p − vq )) in the momentum equation represents the

overall drag force per unit volume. With regard to interphase exchange in the two-fluid
model, it should be noted that the force exerted by the continuous phase on the dispersed
phase is equal and opposite to the force exerted by the dispersed phase on the continuous
phase (i.e. K pq = − K qp ). Also, the interaction force between fluid elements of the same

phase is set to zero (i.e. K pp = K qq = 0 ).

A3.1.1.1 User-Defined Function (UDF) Implementation Aspects


Most drag coefficient expressions listed in Tables 5.2 and 5.4, including the drag on a
rigid sphere (Equation 5.11) feature a fairly steady continuous relationship (i.e. featuring
no local maxima/minima) between the drag coefficient and another quantity (usually the
bubble Reynolds number). This characteristic can be usefully exploited to simplify the
177

regime selection method when incorporating the drag expressions in a UDF. For instance,
the drag coefficient expression proposed by Schiller and Naumann (1935) can be re-
written as Equation A3.3 without any loss of accuracy.

( ( )
C D 0 = max 24 1 + 0.15 Re 0.687 Re , 0.44 ) [A3.3]

This is clearly illustrated in Figure A3.1 where Equations 5.11 and A3.3 are plotted
graphically for comparison.

4.5
4
3.5 Equation 5.11
3
2.5 Equation A3.3
CD0

2
1.5
1
0.5
0
0 500 1000 1500 2000
Bubble Reynolds number (Re)

Figure A3.1 Two different approaches for implementing the drag coefficient

The drag coefficient expression proposed by Grace et al. (1976) shown in Table 5.2 is
valid only for the distorted fluid particle regime. However, when implemented in CFD, it
is imperative to make accommodations to account for the viscous regime as well. It is
well known that at sufficiently small bubble Reynolds numbers, fluid particles behave
like rigid spheres (Clift et al., 1978; Ishii and Zuber, 1979). Hence, the drag coefficient
for a single drop in the viscous flow regime is easily approximated by the Schiller and
Naumann (1935) correlation.
178

A3.1.2 User-Defined Functions


/* Version: 0.0.0 Author: pUl| (smadhava@dal.ca, msrinath80@yahoo.com)
#include "udf.h"
#include "prop.h"
#include "sg_mphase.h"

/* ------------------------------------------------------------ */
/* some predefined quantities, usually constants (in SI units) */
/* ------------------------------------------------------------ */

/* Oil-Water interfacial tension */


#define sigma 0.0404

/* gravity */
#define g 9.81

/* turbulent diffusion coefficient for the Lopez de bertodano model */


#define Ctd 0.1

/* ------------------------------------------------------------ */

/* -------------------------- */
/* Kumar-Hartland (1985) drag */
/* -------------------------- */

DEFINE_EXCHANGE_PROPERTY(kumar_hartland, c, mixture_thread,
second_column_phase_index, first_column_phase_index)
{
Thread **pt = THREAD_SUB_THREADS(mixture_thread);

/* some declarations for variables used in calculation */


real d, urelx, urely, urel, relre, vof, drag_coeff, f, tau_p, k_pq;

/* grab dispersed phase diameter from the panel - SI units */


d = C_PHASE_DIAMETER(c,pt[first_column_phase_index]);

/* calculation of the relative (slip) velocity - SI units */

urelx = C_U(c,pt[first_column_phase_index]) -
C_U(c,pt[second_column_phase_index]);

urely = C_V(c,pt[first_column_phase_index]) -
C_V(c,pt[second_column_phase_index]);

urel = sqrt(urelx*urelx + urely*urely);

/* calculation of the bubble reynolds number - dimensionless - we do in


SI */
179

relre = d * urel * C_R(c,pt[second_column_phase_index]) /


C_MU_L(c,pt[second_column_phase_index]);

/* grab the local volume fraction of the dispersed phase - no units


obviously */
vof = C_VOF(c,pt[first_column_phase_index]);

/* ><><><><><><><><><><><><><><><><><><><><><><><><><> */
/* Kumar and Hartland (1985) drag formulation function */
/* ><><><><><><><><><><><><><><><><><><><><><><><><><> */
real kumar_hartland(void)
{
/* Refer paper by Augier, 2003 */
drag_coeff = ((0.53 + (24. / relre)) * (1. + (4.56 *
pow(vof,0.73))));

/* store the drag coefficient in UDMI 0 for reference */


C_UDMI(c,mixture_thread,0) = drag_coeff;

/* store the slip velocity in UDMI 4 for reference */


C_UDMI(c,mixture_thread,4) = urel;

/* store the Relative reynolds number in UDMI 5 for reference */


C_UDMI(c,mixture_thread,5) = relre;

return drag_coeff;
}
/* ><><><><><><><><><><><><><><><><><><><><><><><><><> */

/* calculation of drag function f - dimensionless */


f = kumar_hartland() * relre / 24.;

/* ------------------------------------------------------------------
*/
/* Calculation of the interphase exchange coefficient (k_pq) proceeds
*/
/* ------------------------------------------------------------------
*/
/* calculation of tau_p - units of time (seconds) */
tau_p = ((C_R(c,pt[first_column_phase_index]) * pow(d,2.)) / (18.
* C_MU_L(c,pt[second_column_phase_index])));

/* calculation of k_pq - 'kg / m3 s' - SI units */

if (vof > 0.999999 && vof < 1.000001) { k_pq = 0.0; } else {
k_pq = ((vof * C_R(c,pt[first_column_phase_index]) * f) / tau_p); }

/* store the Inter-phase exchange coefficient in UDMI 6 for reference


*/
C_UDMI(c,mixture_thread,6) = k_pq;

/* return inter-phase exchange coefficient k_pq */

return k_pq;
180

/* common code ends for drag model :) */


/* ------------------------------------------------------ */
}

/*--------------*/
/* Troshko lift */
/*--------------*/

DEFINE_EXCHANGE_PROPERTY(troshko_lift, c, mixture_thread,
second_column_phase_index, first_column_phase_index)
{
Thread **pt = THREAD_SUB_THREADS(mixture_thread);
real d, dudy, dvdx, wz, wabs, urelx, urely, urel, reb, res, prod,
lift_coeff;

/* grab dispersed phase diameter from the panel - SI units */


d = C_PHASE_DIAMETER(c,pt[first_column_phase_index]);

/* calculation of the vorticity components - SI units again */

dudy = C_U_G(c,pt[second_column_phase_index])[1];
dvdx = C_V_G(c,pt[second_column_phase_index])[0];
wz = dvdx - dudy;
wabs = fabs(wz);

/* calculation of the relative (slip) velocity - SI units */

urelx = C_U(c,pt[first_column_phase_index]) -
C_U(c,pt[second_column_phase_index]);

urely = C_V(c,pt[first_column_phase_index]) -
C_V(c,pt[second_column_phase_index]);

urel = sqrt(urelx*urelx + urely*urely);

/* calculation of the bubble Reynolds number -


dimensionless */

reb = (d * urel * C_R(c,pt[second_column_phase_index]) /


C_MU_L(c,pt[second_column_phase_index]));

/* calculation of the shear Reynolds number */

res = (wabs * pow(d,2.) * C_R(c,pt[second_column_phase_index]) /


C_MU_L(c,pt[second_column_phase_index]));

/* calculation of the product */

prod = res * reb;


181

/* store product in UDMI for easy reference */

C_UDMI(c,mixture_thread,12) = prod;

/* calculation of the lift coefficient according to Moraga et al.


(1999) */

lift_coeff = -(0.12 - 0.2 * exp(-prod / 360000.)) * exp(prod /


30000000.);

/*inviscid lift dominance - substitute prod = 6000 in the above


correlation to get 0.076709631 */
if (prod < 6000.)
{
lift_coeff = 0.07671;
}

/* vortex shedding dominance - substitute prod = 190000 in the


above correlation to get -0.002029928 */
if (prod > 1.9e5) lift_coeff = -0.00203;

/* store lift coefficient in UDMI 1 for easy reference */

C_UDMI(c,mixture_thread,1) = lift_coeff;

return lift_coeff;
}

/*--------------------------- */
/* Turbulent dispersion force */
/*--------------------------- */

DEFINE_ADJUST(tdf_adjust, domain)
{

Thread **pt;
Thread *t;
cell_t c0;

Domain *pDomain = DOMAIN_SUB_DOMAIN(domain,P_PHASE);

Alloc_Storage_Vars(pDomain,SV_VOF_RG,SV_VOF_G,SV_NULL);
Scalar_Reconstruction(pDomain, SV_VOF,-1,SV_VOF_RG,NULL);
Scalar_Derivatives(pDomain,SV_VOF,-
1,SV_VOF_G,SV_VOF_RG,Vof_Deriv_Accumulate);

mp_thread_loop_c (t,domain,pt)
if (FLUID_THREAD_P(t))
{
Thread *tp = pt[P_PHASE];

begin_c_loop (c0,t)
{
182

C_UDMI(c0,t,8) = -
C_UDMI(c0,t,7)*C_R(c0,tp)*C_K(c0,tp)*C_VOF_G(c0,tp)[0];
C_UDMI(c0,t,9) = -
C_UDMI(c0,t,7)*C_R(c0,tp)*C_K(c0,tp)*C_VOF_G(c0,tp)[1];

/* store the volume fraction gradients in UDMI 10 and 11


for quick reference */

C_UDMI(c0,t,10) = C_VOF_G(c0,tp)[0];
C_UDMI(c0,t,11) = C_VOF_G(c0,tp)[1];
}
end_c_loop (c0,t)
}
Free_Storage_Vars(pDomain,SV_VOF_RG,SV_VOF_G,SV_NULL);
}

/*---------------- */
/* source terms ;) */
/*---------------- */

DEFINE_SOURCE(x_td_water, cell, thread, dS, eqn)


{
real source;
Thread *tm = THREAD_SUPER_THREAD(thread);

source = C_UDMI(cell, tm, 8) ;

dS[eqn] = 0.;

return source;
}

DEFINE_SOURCE(x_td_oil, cell, thread, dS, eqn)


{
real source;
Thread *tm = THREAD_SUPER_THREAD(thread);

source = -C_UDMI(cell, tm, 8) ;

dS[eqn] = 0.;

return source;
}

DEFINE_SOURCE(y_td_water, cell, thread, dS, eqn)


{
real source;
Thread *tm = THREAD_SUPER_THREAD(thread);

source = C_UDMI(cell, tm, 9);

dS[eqn] = 0.;

return source;
183

DEFINE_SOURCE(y_td_oil, cell, thread, dS, eqn)


{
real source;
Thread *tm = THREAD_SUPER_THREAD(thread);

source = -C_UDMI(cell, tm, 9);

dS[eqn] = 0.;

return source;
}

A3.1.3 Analysis of Drag Expressions


/* Program to generate drop diameter-rise velocity data using the
Schiller-Naumann drag correlation */
/* Version 0.0.0 Author: pUl| (smadhava@dal.ca, msrinath80@yahoo.com)*/
#include <stdio.h>
#include <math.h>

/* Some constants in SI units */


#define rho_c 998.2
#define rho_d 780.
#define mu_c 0.001003
#define g 9.81

/* Some logical defines */


#define PVE 1
#define NVE 0

/* increment x guess for checking positive to negative transition (or


vice versa) in y - decreasing this value increases the accuracy of the
final output */
#define increment 0.0001

int main(void)
{
double u, usphere1, usphere2;

double d;

double min(double f1, double f2)


{
double f;

f = f1;

if (f2 < f1)


{
f = f2;
}
184

return f;
}

double max(double f1, double f2)


{
double f;
f = f1;
if (f2 > f1)
{
f = f2;
}
return f;
}

/* Newton-Raphson based routine to calculate usphere1 automagically


*/
double usph(double d)
{
double x, x_guess, x_new, y, y_new;
int y_flag = 0;
int y_new_flag = 0;

/* initialize the first guess - choose a guess that will not cause
the solution to blow up :P */
x = 0.001;

/* check for d = 0 */

if (d > -0.000001 && d < +0.000001)

{
return 0.0;
}

else

/* define f(x), f'(x) and f"(x) */

double f_of_x(double x)
{
double fx;
fx = (((4. * g * (rho_c - rho_d) * d) / (3. * rho_c * x)) - ((24.
* mu_c / (d * rho_c)) * (1 + (0.15 * pow((d * x * rho_c /
mu_c),0.687)))));
return fx;
}

double fdash_of_x(double x)
{
double fdashx;
185

fdashx = (((4. * g * (rho_d - rho_c) * d) / (3. * rho_c *


pow(x,2.))) - ((24. * mu_c / (d * rho_c)) * ((0.15 * pow((d * rho_c /
mu_c),0.687) * 0.687 * pow(x,-0.313)))));
return fdashx;
}

double fdoubledash_of_x(double x)
{
double fdoubledashx;
fdoubledashx = (((8. * g * (rho_d - rho_c) * d) / (3. * rho_c *
pow(x,3.))) - ((24. * mu_c / (d * rho_c)) * ((0.15 * pow((d * rho_c /
mu_c),0.687) * 0.687 * (-0.313)* pow(x,-1.313)))));
return fdoubledashx;
}

/* Check for change of sign in y value */


while(y_flag == y_new_flag)
{

y = f_of_x(x);

/* printf("y is %f \n",y); */

if (y < 0.0)

{
y_flag = NVE;
}

else

{
y_flag = PVE;
}

x = x + increment;

y_new = f_of_x(x);

if (y_new < 0.0)

{
y_new_flag = NVE;
}

else

{
y_new_flag = PVE;
}
}

/* who is closer to root? */


186

if (fabs(y) < fabs(y_new))

{
x_guess = x - increment;
}

else

{
x_guess = x;
}

/* Check for convergence */


while((f_of_x(x_guess) * fdoubledash_of_x(x_guess)) >
pow(fabs(fdash_of_x(x_guess)),2.))
{
/* prevent solution from blowing up O_o */
if (fdash_of_x(x_guess) > -0.000001 && fdash_of_x(x_guess) <
+0.000001)
{
printf("Error: f'(x) is equal to zero, cannot proceed!");
break;
}
x_new = (x_guess - (f_of_x(x_guess) / fdash_of_x(x_guess)));
x_guess = x_new;
}

/* show the answer :) */


/* printf("Y Answer is %f\n",f_of_x(x_guess)); */
return x_guess;

printf("d (m) \t\t usphere1 (m/s) \t usphere2 (m/s) \t u (m/s) \n");

for (d = 0.0; d < 0.017; d = (d + 0.001))

usphere1 = usph(d);

usphere2 = pow(((4 * g * (rho_c -rho_d) * d) / (1.32 *


rho_c)),0.5);

u = min(usphere1,usphere2);

printf ("%.3f \t\t %.4f \t\t %.4f \t\t %.4f \n", d, usphere1,
usphere2, u);

return 0;
187

A3.1.4 Calculation of the Position of the First Grid Point


/* Program to calculate the first grid point location in Turbulent
flows */
/* Version: 0.0 Author: pUl| (smadhava@dal.ca, msrinath80@yahoo.com)*/
#include <stdio.h>
#include <math.h>
#define DENSITY_WATER 998.2
#define VISCOSITY_WATER 0.001003
/* change this for your case!!! */
#define PIPE_DIAMETER 0.2
int main(void)
{
float U_AVG, N_RE_WATER, LAMBDA, WALL_SHEAR_STRESS;
float YPLUS, Y_P, FIRST_GRID_POINT;
printf ("Please enter a value for U_AVG in SI units\n");
scanf ("%f", &U_AVG);
N_RE_WATER = (PIPE_DIAMETER * U_AVG *
DENSITY_WATER)/(VISCOSITY_WATER);
if (N_RE_WATER > 100000)
{
printf ("Calculation cannot be performed, Reynolds number exceeds
100000 limit\n");
}
else
{
printf ("Reynolds number is %.0f\n", N_RE_WATER);
LAMBDA = 0.3164/pow(N_RE_WATER,0.25);
WALL_SHEAR_STRESS = (LAMBDA*DENSITY_WATER*pow(U_AVG,2)/8);
printf ("Please enter a value for YPLUS\n");
scanf ("%f", &YPLUS);
Y_P =
YPLUS*VISCOSITY_WATER*sqrt(DENSITY_WATER/WALL_SHEAR_STRESS)/DENSITY_WAT
ER;
FIRST_GRID_POINT = 2*Y_P;
printf ("The first grid point should be placed %f metres away
from the wall.\n", FIRST_GRID_POINT);
}
return 0;
}

Вам также может понравиться