Phase Inversion in Dispersed Liquid-Liquid Pipe Flow: Kwun Ho Ngan

Phase Inversion in dispersed liquid-liquid pipe flow
Kwun Ho Ngan
A thesis submitted to University College London in fulfillment of the requirements
for the degree of Doctor of Philosophy
Department of Chemical Engineering
University College London
London WC1E 7JE
October 2010
To my family
and Ting Ting
Abstract

This thesis presents the experimental and theoretical investigations on the development of
phase inversion in horizontal pipeline flow of two immiscible liquids. It aims to provide
an understanding on the flow development across the phase inversion transition as well as
the effect on pressure drop.
Experimental investigation on phase inversion and associated phenomena were

conducted in a 38mm I.D. liquid pipeline flow facility available in the Department of
Chemical Engineering at University College London (UCL). Two sets of test pipelines
are constructed using stainless steel and acrylic. The inlet section of the pipeline has also
been designed in two different configurations – (1) Y-junction inlet to allow dispersed
flow to be developed along the pipeline (2) Dispersed inlet to allow formation of
dispersion immediately after the two phases are joined. Pressure drop along the pipeline
is measured using a differential pressure transducer and is studied for changes due to
redistribution of the phases during inversion. Various conductivity probes (ring probes,
wire probes, electrical resistance tomography and dual impedance probe) are installed
along the pipeline to detect the change in phase continuity and distribution as well as
drop size distribution based on the difference in conductivity of the oil and water phases.
During the investigation, the occurrence of phase inversion is firstly investigated and the
gradual transition during the process is identified. The range of phase fraction at which
the transition occurs is determined. The range of phase fraction becomes significantly
narrower when the dispersed inlet is used. The outcome of the investigation also becomes
the basis for subsequent investigation with the addition of glycerol to the water phase to
reduce the interfacial tension. Based on the experimental outcome, the addition of
glycerol does not affect the inversion of the oil phase while enhancing the continuity of
the water phase.
1
As observed experimentally, significant changes in pressure gradient can be observed
particularly during phase inversion. Previous literatures have also reviewed that phase
inversion occurs at the maximum pressure gradient. In a horizontal pipeline, pressure
gradient is primarily caused by the frictional shear on the fluid flow in the pipe and, in
turn, is significantly affected by the fluid viscosities. A study is conducted to investigate
on the phase inversion point by identifying the maximum mixture viscosity (i.e.
maximum pressure gradient) that an oil-in-water (O/W) and water-in-oil (W/O)
dispersion can sustain. It is proposed that the mixture viscosity will not increase further
with an increase in the initial dispersed phase if the inverted dispersion has a lower
mixture viscosity. This hypothesis has been applied across a wide range of liquid-liquid
dispersion with good results. This study however cannot determine the hysteresis effect
which is possibly caused by inhomogeneous inversion in the fluid system.
A mechanistic model is developed to predict the flow characteristics as well as the

pressure gradient during a phase inversion transition. It aims to predict the observed
change in flow pattern from a fully dispersed flow to a dual continuous flow during phase
inversion transition. The existence of the interfacial height provides a selection criterion
to determine whether a momentum balance model for homogeneous flow or a two-fluid
layered flow should be applied to calculate the pressure gradient. A friction factor is also
applied to account for the drag reduction in a dispersed flow. This developed model
shows reasonable results in predicting the switch between flow patterns (i.e. the
boundaries for the phase inversion transition) and the corresponding pressure gradient.
Lastly, computational fluid dynamic (CFD) simulation is applied to identify the key
interphase forces in a dispersed flow. The study has also attempted to test the limitation
of existing interphase force models to densely dispersed flow. From the study, it is found
that the lift force and the turbulent dispersion forces are significant to the phase
distribution in a dispersed flow. It also provides a possible explanation to the observed
flow distribution in the experiments conducted. However, the models available in CFX
are still unable to predict well in a dense dispersion (60% dispersed). This study is also
2
suggested to form the basis for more detailed work in future to optimize the simulation
models to improve the prediction of phase inversion in a CFD simulation.
3
Acknowledgement
I wish to firstly express my sincere gratitude to my supervisor at University College
London, Dr Panagiota Angeli, for offering me this valuable opportunity to learn about an
interesting topic of multiphase flow under her great guidance and experience. Her personal
devotion and passion on the investigation is always my greatest drive to push beyond the
boundaries for breakthrough. More importantly, I have to express my appreciation on her
personal care in every aspect of my life allowing me to pursue my research work in a worry free
environment. Her support for my participation in various conferences provides further
opportunities to interact with other expertise in the field. It is no doubt the most memorable
academic life I have pursued and a life changing experience I will never regret.
I would also like to thank Chevron Technology Company for the kind sponsorship on my
research project. In particular, I would like to thank Dr Lee Rhyne and Dr Karolina Ioannou for
their regular discussions during my course of studies. Their industrial experience has been a
valuable source of information for me to understand more about applying the outcome of the
investigation in the oil industry. I am also greatly appreciated about spending their precious time
to proof-read this work and making many constructive suggestions to it.
Thanks also to all the colleagues that have spent the past years with me. You are always
the best guides for my memorable student life in London. Without all of you, I would not have
appreciated the vibrant life in this metropolitan city. Thanks also to the staff in the mechanical
and electrical workshop. Your kind help to resolve every small issue that I encounter throughout
the course is what shapes the successful outcome in this work. I wish you all the best in your
future.
On a personal note, I have to express my special thanks to Ting Ting without whom I
would not have the strength and perseverance to pursue this challenging course in a top notched
university. Your continual care and support will always be remembered and appreciated. I wish
you all the success in life and continue to be my guiding light in my life.
Lastly, to my supportive family for always bringing me the feeling of home despite we
are miles apart. Thanks for all the advice and love to me.
4
Table of Contents
Abstract 1
Acknowledgement 4
Table of Contents 5
List of Figures 9
List of Tables 21
Nomenclature 23
Chapter 1: Introduction
1.1 Multiphase Flow in Petroleum Industry 28
1.2 Fundamentals of Multiphase Flow 29
1.2.1 The Occurrence of Phase Inversion 30
1.3 Objectives of Study 31
1.4 Structure of the Thesis 32
Chapter 2: Literature Review

2.1 Overview 34
2.2 Flow Development and Flow Pattern Maps 34
2.3 Phase Inversion 39
2.3.1 Phase Inversion in Stirred Vessels 41
2.3.2 Phase Inversion in Pipe Flow 44
2.3.3 Parametric Study on Phase Inversion 48
2.4 Phase Inversion Mechanisms and Prediction Models 53
2.4.1 Models based on Rate of Drop Break-Up and
53
Coalescence
2.4.1.1 Secondary Dispersions 56
2.4.2 Models based on Minimisation/Equal System and
57
Interfacial Energy
2.4.3 Models based on Zero Shear Stress at Interface 58
2.5 Drop Size Distribution and Characteristic Diameter 59
2.5.1 Types of Drop Size Distribution 60
2.5.2 Characteristic Diameter of Drop Size Distribution 62
5
2.6 Numerical Simulation on Dispersed Flow 63
2.6.1 Conservation Equations used in CFD 66
2.6.2 Models for Simulation of Liquid-Liquid
69
Dispersed Pipe Flow
2.7 Interphase Forces Models 72
2.7.1 Interphase Drag 72
2.7.1.1 Drag Force in Dense Dispersion 73
2.7.2 Lift Force 75
2.7.3 Turbulent Dispersion Force 75
2.7.3.1 Favre Averaged Drag Model 76
2.7.3.2 Lopez de Bertodano Model 76
2.8 Summary of Literature Review 77
Chapter 3: Facility and Instrumentation

3.1 Overview 79
3.2 Test Fluids 79
3.3 Experimental Flow Facility 80
3.4 Summary 100
Chapter 4: The Occurrence of Phase Inversion

4.1 Overview 101
4.2 Experimental Procedure 102
4.3 Visual Observations of Phase Inversion 103
4.4 Detection of Phase Inversion Occurrence 107
4.4.1 In-Situ Phase Distribution Measurements 107
4.4.2 Changes in Phase Continuity (Partial Phase
109
Inversion)
4.5 Effect of Dispersion Initialisation on Phase Inversion 112
4.6 Effect of Mixture Velocity on Phase Inversion 114
4.7 Pressure Gradient during Phase Inversion 116
4.7.1 Effect of Inversion Route on Pressure Gradient 118
4.7.2 Effect of Mixture Velocity on Pressure Gradient 118
4.8 Drop Size Distribution in Oil & Water Dispersion 121
4.8.1 Drop Size Distribution Measurement 122
6
4.9 Effect of Dispersion Inlet on Phase Inversion 126
4.10 Conclusion 130
Chapter 5: Effect of Interfacial Tension on Phase Inversion

5.1 Overview 134
5.2 Experimental Setup 136
5.3 Phase Inversion Results 138
5.4 Effect of Glycerol Addition on Pressure Gradient during
144
Phase Inversion
5.5 Effect of Glycerol on Drop Size Distribution in Oil and
146
Water Continuous Dispersions
5.6 Conclusion 149
Chapter 6: Prediction of Phase Inversion through Fluid Viscosities

6.1 Overview 150
6.2 Suggested Method for Determining the Phase Inversion
151
Point
6.3 Literature Mixture Viscosity Models 154
6.4 Comparison between Predicted Phase Inversion Points
156
and Experimental Results from Literature
6.5 Comparison of the Proposed Methodology with Literature
161
Correlations
6.6 Review on Prediction Model with an Experimental
164
Homogeneous Dispersed Flow
6.7 Conclusion 166
Chapter 7: Prediction of Pressure Gradient during Phase Inversion

7.1 Overview 168
7.2 Experimental Investigation 169
7.3 Development of Model for the Prediction of Pressure
171
Drop
7.3.1 Prediction of Flow Configuration 171
7.3.2 Entrainment Model 173
7
7.3.3 Two Fluid Model 174
7.3.4 Homogeneous Model 176
7.4 Comparison of Model Predictions with Experimental Data 176
7.5 Conclusion 179
Chapter 8: CFD Simulation of Horizontal Two-Phase Pipe Flow

8.1 Overview 180
8.2 Description of Ansys CFX Software Package 180
8.3 Model Geometry and Simulation Conditions 183
8.4 Results and Discussion 187
8.5 Conclusion 195
Chapter 9: Thesis Conclusion and Future Work

9.1 Overview 197
9.2 Conclusions 197
9.2.1 Conclusions from Experimental Studies 197
9.2.2 Conclusions from Theoretical Studies 199
9.3 Recommendations 200
Reference 203
Appendix A – Chord Length Measurement during Phase Inversion 214

Appendix B – Matlab Code for Pressure Gradient Prediction 217
Appendix C – Grid Sensitivity Test for CFD simulation 230
8
List of Figures
Chapter 1
Offshore production pipeline schematic from well centres to
Figure 1.1 28
production platforms and FPSO.
Chapter 2
(a) Flow regime map that is based on the fluid superficial
velocities.
Figure 2.1 35
(b) Flow regime map that is based on input water fraction and
mixture velocity.
(a) Stratified smooth flow (b) stratified wavy flow with the oil
Figure 2.2 phase on the top layer while the water phase on the bottom 37
layer. (Ngan et al., 2007)
Dual continuous flow with several oil drops entrained in the

Figure 2.3 38
water continuous layer. (Ngan et al., 2007)
Annular flow with an oil core and a water annulus. Small

Figure 2.4 amount of drops are also observed within the oil core in the 38
figure. (Ngan et al., 2007)
(a) Dispersed flow of oil drops in water. Drops observed are

Figure 2.5 polydispersed and highly deformable. 39
(b) Oil slug flow in water. (Ngan et al., 2007)
A schematic on the proposed phase inversion mechanism by

Figure 2.6 40
Arirachakaran et al. (1989).
The existence of ambivalent region between the aqueous and

Figure 2.7 organic continuous dispersion across the organic phase fraction 41
(Φo) (Noui-Mehidi et al., 2004).
Droplet in drop for a glycerol/water and chlorobenzene system

Figure 2.8 42
captured by Sony video printer (Pacek et al., 1994).
(a) Pressure gradient measurement of a Exxsol D80/water

system in a 60mm I.D. pipe.
Figure 2.9 45
(b) Pressure gradient measurement of a Marcol 52/water system
in a 60mm I.D. pipe (Ioannou, 2006).
9
Conductivity measurement conducted at the centre of a vertical
Figure 2.10 47
pipe across the phase inversion process (Hu, 2005).
Multiple dispersion of an oil/water system at a mixture velocity

of 1.5m/s and input oil fraction of 31%. The dark region
Figure 2.11 47
represents the oil phase and the light region represents the water
phase. (Liu et al., 2006)
Visual observations of (a) O/W/O and (b) W/O/W dispersions

Figure 2.12 using LIF technique. The oil phase is shown in black while the 57
water phase is shown in luminous green. (Liu et al., 2005)
Comparison between calculated drop size distribution and

experimental drop size data. Dotted line represents the
Figure 2.13 histogram at the circulation region and the solid line represents 60
the histogram at the impeller region. (Coulaloglou and
Tavlarides, 1977).
Comparison between experimental drop size data and selected
Figure 2.14 drop size distribution types in an horizontal pipe flow (Vso = 62
0.75m/s, Vsw=0.75m/s, oil drops). (Vielma et al., 2008).
Schematic representation on the modelling approaches for two-
phase flow: (a) Eulerian-Lagrangian approach and (b) Eulerian-
Figure 2.15 Eulerian approach. The number next to the blue arrow in (b) 65
represents the volume fraction of the dispersed phase within the
control volume. (Rusche, 2002)
Chapter 3
Figure 3.1 Schematic of the pilot scale facility. 81
Figure 3.2 Photograph of the pilot scale facility in stainless steel. 83
Stainless steel Y-inlet where the oil and water phases are joined
Figure 3.3 83
at the end.
Acrylic Y-inlet with a split plate to enhance stratification of the

Figure 3.4 84
oil and water.
Dispersion inlets with staggered nozzles. (a) The dispersed

phase flows within the stainless steel pipe while the continuous
Figure 3.5 phase flows in the acrylic pipe. (b) Dispersed phase is ejected 85
from the nozzles.
10
Junction effect of the coalescer that enhances separation.
Figure 3.6 86
(Information from manufacturer - KnitMesh Technologies).
Setup of the high speed video imaging unit, Kodak HS 4540

Figure 3.7 87
MX.
Section of pipe with quick closing valves in the 38mm stainless

Figure 3.8 steel pipeline. A viewbox is included to aid visual observation 88
and video capturing.
Conductivity ring probe for phase continuity measurement at

Figure 3.9 89
the pipe periphery.
Local conductive wire probe for phase continuity measurement

Figure 3.10 at the localized position. (a) Photo of wire probe (b) Schematic 90
of the interior (side view) of the wire probe.
(a) Stainless steel electrodes embedded across the pipe

periphery.
Figure 3.11 92
(b) Experimental setup of the ERT system with a PC for data
processing.
Sample ERT reconstructed image generated from ITS toolsuite.

Figure 3.12 The colour bar denotes the range of raw conductivity data with 93
respective to the reference frame.
In-situ liquid hold-up comparison between electrical resistance

Figure 3.13 tomography (ERT) and quick closing valves across a range of 94
input phase fractions.
(a) Photograph of the dual impedance probe setup. (b) Probe tip
Figure 3.14 95
configuration (Lovick, 2004).
A sample set of normalized signal from the two impedance

Figure 3.15 wires at a mixture velocity of 3m/s. Dispersed phase will be in 96
contact with probe 1 before it contacts with probe 2.
A sample frequency plot of the time delay between the two

Figure 3.16 impedance wires. The highest peak denotes the time delay 97
where most dispersed drop take to cross the two wires.
Pressure measuring port at stainless steel pipeline. Similar ports

Figure 3.17 99
are also found in the acrylic pipeline.
11
Chapter 4
Visual images are taken in a stainless steel pipe using an high
speed camera (Model: Vision Research Phantom V5.1) at 7m
Figure 4.1 away from the inlet. Mixture velocity is kept constant at 3m/s. 104
The percentage in the bracket represents the input water
fraction which the image is taken.
Visual images are taken in an acrylic pipe using an high speed

camera (Model: Kodak HS 4540) at 7m away from the inlet.
Mixture velocity is kept constant at 3m/s. The percentage in the
Figure 4.2 106
bracket represents the input water fraction at which the image is
taken. The arrows present samples of complex structures
observed in images.
Phase distribution in an acrylic pipe cross section during the

transition from a water continuous to an oil continuous mixture
Figure 4.3 108
at a mixture velocity of 3m/s. The percentages in brackets
represent the input water fraction.
Normalized conductivity data of the oil/water system in an

acrylic pipe at a mixture velocity of 3m/s from the ring and wire
probes as well as the ERT system. The vertical lines limit the
Figure 4.4 phase inversion transition region and are drawn at the first and 110
last near zero conductivity values recorded using the various
probes. The percentages represent the water fraction at which
the different probes show near-zero conductivity values.
Normalized conductivity data of the oil/water system in a

stainless steel pipe at a mixture velocity of 3m/s from the ring
and wire probes as well as the ERT system. The vertical lines
Figure 4.5 limit the phase inversion transition region and are drawn at the 112
first and last near zero conductivity values recorded using the
various probes. The percentages represent the water fraction at
which the different probes show near-zero conductivity values.
12
Normalized conductivity data for the oil/water system in an
acrylic pipe at a mixture velocity of 3m/s from the ring and the
wire probes for the two inversion routes (a) starting from oil
continuous (OC) and (b) starting from water continuous (WC)
Figure 4.6 mixture. The vertical lines limit the phase inversion transition 113
region and are drawn at the first and last near zero conductivity
values recorded from the two probes. The percentages represent
the water fraction at which the different probes show near-zero
conductivity values.
Phase distribution in a pipe cross section during the transition

from a water continuous to an oil continuous dispersion at a
Figure 4.7 115
mixture velocity of 4m/s. The percentages in brackets represent
the input water fraction.
Comparison of normalized conductivity data for the oil/water

mixture at a mixture velocity of 3m/s and 4m/s using the ring
probe and wire probe. The vertical lines determine the phase
Figure 4.8 116
inversion transition at which the first and last near zero
conductivity value are recorded using the various probes –
dotted lines for 3m/s and solid lines for 4m/s.
Experimental pressure gradients in an acrylic pipe at a mixture

velocity of 3m/s. The error bars represent the standard deviation
of the fluctuations at each specific phase fraction. The arrow
Figure 4.9 117
denotes the direction of experiment from water continuous to
oil continuous dispersion and the vertical lines represent the
boundaries of the phase inversion transition region.
Comparison of pressure gradient for the oil/water system in an

acrylic pipe at a mixture velocity of 3m/s between the two
inversion routes (a) starting from oil continuous dispersion
Figure 4.10 119
(OC) (b) starting from water continuous dispersion (WC). The
vertical lines represent the boundaries of the phase inversion
transition region.
Comparison of pressure gradient measurement between mixture

velocity of 3m/s and 4m/s in an acrylic pipe. The error bars
represent the standard deviation of the fluctuations at each
Figure 4.11 phase fraction. The arrow denotes the direction of experiment 120
from water continuous dispersion to oil continuous dispersion.
The vertical lines limit the boundaries for phase inversion
transition - dotted lines for 3m/s and solid lines for 4m/s.
13
Comparison of pressure gradient measurement between mixture
velocity of 3m/s and 4m/s in a stainless pipe. The arrow denotes
Figure 4.12 120
the direction of experiment from water continuous dispersion to
oil continuous dispersion.
Chord length distribution of oil drops in water (40% water

fraction) at a mixture velocity of 3m/s at (a) 6mm (b) 14mm (c)
20mm (d) 26mm (e) 32mm from the top wall of the pipe. The
Figure 4.13 distributions of the two inversion routes are given – starting 123
from a O/W dispersion (□) and from a W/O dispersion (■).
Sample drop size distribution is recorded at 14mm from the top

wall with a mixture velocity of 3m/s (a) oil-in-water dispersion
(60% water fraction). The experimental data is compared with a
Figure 4.14 log-normal distribution with a mean of 1.51mm and a variance 125
of 0.238. (b) water-in-oil dispersion (40% water fraction). The
experimental data is compared with a log-normal distribution
with a mean of 2.07mm and a variance of 0.451.
Sauter mean diameter at different input water fraction across

the phase inversion boundary (1). The phase in brackets
Figure 4.15 represents the initial continuous phase and the number in each 126
column represents the Sauter mean diameter at the specific
sample.
Normalized conductivity data for an oil/pure water mixture at a

mixture velocity of 3m/s using the ring probe and wire probe.
The letter in bracket represents initial dispersed phase – (W)
Figure 4.16 127
water (O) oil. The arrow denotes the direction of experiment
and the vertical line denotes the first input water fraction where
phase inversion occurs.
Normalized conductivity data for an oil/pure water mixture at a

mixture velocity of 3m/s using the ring probe and wire probe
through two initialization conditions – starting from water
continuous dispersion (empty symbols) and starting from oil
Figure 4.17 128
continuous dispersion (solid symbols). The dotted vertical line
represents the phase inversion point starting from water
continuous dispersion and the solid line represents the phase
inversion point starting from oil continuous dispersion.
14
Experimental pressure gradients for a dispersed inlet
configuration at a mixture velocity of 3m/s. The arrow denotes
Figure 4.18 the direction of experiment from water continuous to oil 129
continuous dispersion and the vertical line represents the phase
inversion point.
Comparison of chord length distribution at different input water

Figure 4.19 fraction (fraction in bracket) across the phase inversion process 130
at 3m/s mixture velocity.
Chapter 5
(a) Normalized conductivity data of the oil/water system at a
mixture velocity of 3m/s from the ring and wire probes and the
ERT system. The direction of the experiment from water to oil
continuous is shown by the arrow. The vertical lines denote the
boundaries of the phase inversion transition region and are
drawn at the first and last near zero conductivity values 139
recorded using the various probes. The percentages represent
the water fraction where conductivity approaches zero at the
various locations in the pipe cross section.
Figure 5.1
(b) Phase distribution in a pipe cross section during the
transition from a water continuous to an oil continuous
dispersion at a mixture velocity of 3m/s. The percentages in
brackets represent the input water fraction. Blue denotes the 140
region of the water phase and green denotes the region of the oil
phase. Last tomogram (i.e. water-in-oil dispersion) represents
the completion of phase inversion upon which ERT
measurement fails.
(a) Normalized conductivity data of the oil/1% glycerol solution

system at a mixture velocity of 3m/s from the ring and wire
probes and the ERT system. The direction of the experiment
from water to oil continuous is shown by the arrow. The
vertical lines denote the boundaries of the phase inversion
Figure 5.2 transition region and are drawn at the first and last near zero 141
conductivity values recorded using the various probes. The
percentages represent the water fraction where conductivity
approaches zero at the various locations in the pipe cross
section.
15
(b) Phase distribution in a pipe cross section during the
transition from a 1% aqueous continuous to an oil continuous
dispersion at a mixture velocity of 3m/s. The percentages in
brackets represent the input water fraction. Blue denotes the
Figure 5.2 142
region of the water phase and green denotes the region of the oil
phase. Last tomogram (i.e. water-in-oil dispersion) represents
the completion of phase inversion upon which ERT
measurement fails.
A comparison of the normalised conductivities in the middle of

the pipe and the pipe periphery between the 1% glycerol
Figure 5.3 143
solution and the pure water dispersions at a mixture velocity of
3m/s.
Normalized conductivity data of the oil/0.5% glycerol solution

system at a mixture velocity of 3m/s from the ring and wire
probes and the ERT system. The direction of the experiment
from water to oil continuous is shown by the arrow. The
vertical lines denote the boundaries of the phase inversion
Figure 5.4 144
transition region and are drawn at the first and last near zero
conductivity values recorded using the various probes. The
percentages represent the water fraction where conductivity
approaches zero at the various locations in the pipe cross
section.
Experimental pressure gradient for pure water, 0.5% 1%

glycerol solutions at different input water fractions at a mixture
velocity of 3m/s. The vertical lines indicate the boundaries of
the phase inversion transition region for pure water (solid lines)
Figure 5.5 145
and both glycerol solutions (dotted lines). The percentages
indicate the water fractions where pressure gradient shows a
minimum. The arrow denotes the direction of experiment from
aqueous to oil continuous dispersion.
Sauter mean diameter for pure water, 0.5% and 1% glycerol

solutions starting from (a) a water continuous dispersion and (b)
an oil continuous dispersion. The phase in the bracket
Figure 5.6 148
represents the initial continuous phase. The numbers on the
columns represent the d32 diameter of the respective sample in
mm.
16
Chapter 6
Prediction of phase inversion point for the Exxsol D140-water
dispersion. Phase inversion occurs at the phase fraction where
the difference in viscosity between the oil continuous and the
Figure 6.1 152
water continuous dispersions becomes 0. The Brinkman/Roscoe
(1952) model is used for calculating the viscosity of the
dispersions.
Predicted inversion points using different dispersion viscosity

models against the experimental phase inversion range for a
dispersion of water and (a) Exxsol D80 (1.7mPas), (b) Exxsol
D140 (5.5mPas), (c) Marcol 52 (11mPas) (data in stainless steel
Figure 6.2 pipe by Ioannou, 2006). Points in graph refer to the model 159
predictions while the lines represent the experimental phase
inversion region.

models against the experimental phase inversion ranges in
stainless steel pipe obtained starting from oil continuous and
Figure 6.3 160
from water continuous dispersions for a water-Marcol 52
mixture. Points in graph refer to the model predictions while the
lines represent the experimental phase inversion region.

models against the experimental phase inversion ranges in
stainless steel pipe obtained at different mixture velocities for a
Figure 6.4 160
water-Marcol 52 mixture. Points in graph refer to the model
predictions while the lines represent the experimental phase
inversion range.

models against the experimental phase inversion ranges
Figure 6.5 obtained with different test sections for a water-Marcol 52 161
mixture. Points in graph refer to the model predictions while the
lines represent the experimental phase inversion range.
Comparison between proposed methodology (suggested best

Figure 6.6 163
models) and models for critical phase fraction from literature.
Application of proposed model on various systems from Yeh et

al. (1964) and Arirachakaran et al. (1989). The solid line
Figure 6.7 represents the prediction model by Yeh et al. (1964) (see 164
Equation 6.17) and the dotted line represents the prediction
model by Arirachakaran et al. (1989) (see Equation 6.15).
17
Comparison of pressure gradient at different phase fraction
between experimental data (□) and modeling outcome via
Figure 6.8 166
Brinkman/Roscoe (1952)’s mixture viscosity correlation. The
solid arrow (←) represents the direction for the experiment.
Chapter 7
Pressure gradient at mixture velocity (a) 3m/s and (b) 4m/s
measured experimentally. The solid lines represent the
Figure 7.1 boundaries of the transitional region between water continuous 170
and oil continuous fully dispersed flow. The arrow indicates the
direction of the experiment.
Experimental and predicted pressure gradient at (a) 3m/s and

(b) 4m/s predicted from two-fluid model. The lines represent
the boundaries of the transitional region between water
Figure 7.2 178
continuous and oil continuous fully dispersed flow, solid for
experimental and dotted for predicted fractions. The arrow
indicates the direction of the experiment.
Chapter 8
Schematic diagram on the iterative method for solving the
Figure 8.1 182
conservation equations. (Ansys CFX 11.0 Manual)
(a) Schematic diagram of the grid structure for the 2-D pipe
section for the CFD simulations.
Figure 8.2 184
(b) Part of the mesh of the 2-D pipe section with the thick lines
representing the smooth walls in the simulation.
Experimental oil phase distribution along a vertical pipe

diameter at a mixture velocity of 3m/s. Data is extracted from
Figure 8.3 187
ERT measurements at 7m from the inlet where flow is fully
developed.
Comparison between simulated (base) and experimental (ERT)

dispersed phase fractions at an input water fraction of 80% and
Figure 8.4 mixture velocity of 3m/s. The simulated data is extracted at 3m 188
from the inlet and the experimental data at 7m where the flow is
fully developed.
18
Effect of lift force on phase distribution at mixture velocity of
Figure 8.5 3m/s and input water fraction of 80%. LF in the legend 189
represents the lift coefficient applied to Equation 2.37.
Effect of lift force coefficient on the lift force term at a mixture

velocity of 3m/s and an input water fraction of 80%. ds
Figure 8.6 190
represents the drop size in mm and LF represents the lift force
coefficient applied to Equation 2.37.
Effect of drop size on phase distribution at a constant lift

coefficient (LF=0.5) at a mixture velocity of 3m/s and input
Figure 8.7 water fraction of 80%. ds represents the drop size in mm. 191
Effect of drop size on the lift force of a constant lift coefficient

Figure 8.8 (LF=0.5) at a mixture velocity of 3m/s and input water fraction 192
of 80%. ds represents the drop size in mm.
Effect of turbulent dispersion force coefficient on phase

distribution at a mixture velocity of 3m/s and input water
Figure 8.9 193
fraction of 80%. LB represents the coefficient for the Lopez de
Bertodano model in Equation 2.39.
Comparison between experimental and simulated phase fraction

distributions for 40% and 60% input water fractions at a
Figure 8.10 mixture velocity of 3m/s and drop size of 1mm. Oil is set to be 195
the dispersed phase in the simulation. The lift force coefficient
is 0.5 and the turbulent dispersion force coefficient is 2.
Appendix A
fraction) at various locations from the top wall of the pipe with
Figure A-1 a mixture velocity of 3m/s. The distributions of the two 214
inversion routes are given – starting from a O/W dispersion (□)
and from a W/O dispersion (■).

19
Appendix C
Sample schematics of simulation domain with a vertical height
Figure C-1 230
consisting of 60 cells.
Comparison of results on the dispersed phase fraction along a

Figure C-2 vertical diameter at the end of the pipe with 3m in length. An 231
input dispersed phase fraction of 0.2 is chosen for all 3 cases.
20
List of Tables
Chapter 2
Table 2.1 Flow pattern classification suggested in various literature. 36
Parameters affecting phase inversion in liquid-liquid pipeline

Table 2.2 48
flow.
Coefficients for k-ε turbulence model (Launder and Spalding,

Table 2.3 70
1972).
Table 2.4 Coefficients for k-ε turbulence model (Ansys CFX Manual). 72
Drag coefficient correlations for dense homogeneously

Table 2.5 74
distributed dispersions.
Chapter 3
Table 3.1 Physical properties of working fluids. 79
Comparison of phase distribution contour plot using single

impedance probe (red for oil and blue for water) and ERT
Table 3.2 98
tomogram (green for oil and blue for water) at different input
phase fractions in a 3m/s oil/water mixture flow.
Chapter 5
Properties of the glycerol solutions. The number in bracket
Table 5.1 represents the percentage change in the value of the respective 138
physical property compared to that of pure water.
Chapter 6
Table 6.1 Literatures for dispersion viscosity models. 153
Physical properties of fluids and pipes used in the phase

Table 6.2 157
inversion experiments.
21
Critical phase fraction models for phase inversion from
Table 6.3 163
literature.
Chapter 7
Flow patterns resulting from the application of a critical water
Table 7.1 fraction for phase inversion (WFcritical) in each of the upper and 172
lower layers of a dual continuous flow configuration.
Appendix C
Number of grid points and cell elements in structures used in
Table C-1 231
the grid sensitivity test.
22
Nomenclature
Roman symbols
A cross sectional area
C coefficient in Equation 2.11
CD drag force coefficient
CL lift force coefficient
CLD chord length distribution
CTD turbulent dispersion force coefficient
(dP/dx) pressure drop
D pipe diameter
d drop diameter
dmax maximum drop diameter
d10 linear mean diameter
d32 (D32) Sauter mean diameter
d95 drop diameter of 95% percentile
DSD drop size distribution
E steady state entrainment fraction
Entl entrainment fraction at lower layer
Entu entrainment fraction at upper layer
Eo Eötvös number
ERT electrical resistance tomography
F force
FD drag force term
Flift lift force term
23
FM interphase force term
FTD turbulent dispersion force term
f friction factor
fDR friction factor with drag reduction used in Equation 7.8
Fr Froude number
g gravity
I turbulence intensity
K constant used in Equation 2.5.
k turbulent kinetic energy
k crowding factor in Equation 6.10
k1 coefficient used in Equation 2.11
k1 coefficient used in Table 2.5
k2 coefficient used in Equation 2.11
k2 coefficient used in Table 2.5
kD empirical deposition rate constant used in Equation 7.2
L turbulent length scale
N agitation speed
n number density
n coefficient used in Equation 2.11
O oil
OC oil continuous
O/W oil-in-water
ΔP pressure gradient
24
Pk shear production of turbulence
Q volumetric flowrate
QCV quick closing valve
r drop radius
Rb rate of drop break-up
Rc rate of drop coalescence
Rdep rate of deposition
Rent entrainment rate
Re Reynolds number
S perimeter
S source term
S slip ratio
Si length of interface
Sv slip ratio
T Reynolds stress tensor
Tcd Turbulence modulation terms used in Equation 2.30 and 2.31
t time
U velocity
U* frictional velocity used in Equation 7.3 and 7.4
Vcum cumulative volume fraction of drops
Vent volume of entrainment fraction
W water
WC water continuous
Wd mass flow rate of the dispersed phase used in Equation 7.2
We Weber number
25
WFcritical critical water fraction
W/O water-in-oil
Greek symbols
α volume fraction
Δ change
 energy dissipation rate
 viscosity
μφ effective viscosity used in Chapter 6
μr relative viscosity used in Table 6.1
 pi
 density
τ shear stress
 interfacial tension
λ wavelength
Φ phase fraction
Φ dispersed phase fraction in Equation 6.1
Φmax maximum packing factor
φ phase fraction
δxy Kronecker delta
θ contact angle
υ viscosity
26
Subscripts
c continuous phase
d dipsersed phase
I interfacial
i interfacial
l lower
m mixture
o oil
s superficial
t turbulent
u upper
w water
Superscripts
I inversion
27
Chapter 1: Introduction
1.1 MULTIPHASE FLOW IN PETROLEUM INDUSTRY
In an offshore oil production operation, crude oil is pumped from the reservoirs through
the wellhead and pipelines to the processing platform or FPSO (floating platform, storage
and offloading) unit where phase separation will take place. Along the transportation
pipeline, the crude oil stream will tend to contain a percentage of water due to natural
inclusion of groundwater, leakage along the long pipeline, or deliberate injection of water
into the reservoir to enhance oil recovery. Water content tends to increase as the well
ages and, at some extreme situation, wells are still in operation with a production stream
of 98% of water or more. Gas may also be present especially at the riser when the
production stream undergoes a pressure reduction from the reservoir. As such, a mixture
of fluids is generally flowing simultaneously in the pipeline. This type of flow stream is
commonly known as multiphase flow.
Figure 1.1 Offshore production pipeline schematic from well centres to production
platforms and FPSO.
28
As oil exploitation has moved further from shore, hostile climate and greater water depth
threaten the construction of oil production platforms. Many satellite wells have to be
connected and relayed to existing platforms through long transportation pipelines. A
complex network of pipeline is thus developed. Maintaining the consistency of the fluid
flow becomes a technical challenge. The situation is worsened with depleting well where
the reservoir pressure is no longer sufficient to drive the flow.
In order to combat the increasing operating costs for offshore production while
maintaining high efficiency and consistency for oil recovery, technological advances
have been ongoing which have led to various technology commercialisation. For
example, multiphase boosting technology was introduced offshore to ensure that oil
recovery will not be disrupted by insufficient reservoir pressure and, preferably, can also
be enhanced by artificial boosting. The first commercial subsea multiphase boosting
system is the Shell multiphase underwater booster system (Smubs). A review on the
deployment of multiphase technology in the North Sea is presented in Leporcher et. al
(2001) with a focused case study on the DUNBAR project.
Currently, R&D in multiphase technology is still on high demand to tackle the continual
challenge on the demand for crude oil. Understanding the nature of these multiphase
flows has been complex and this is especially the case with oil/water mixture flow.
Literatures on the simultaneous flow of oil/water mixture has become more transparent
over the recent years. These literatures have primarily presented the different flow
regimes that can be observed across a wide range of operating conditions. Some authors
have extended the scope to review the effect of flow regime changes on pressure gradient
across pipeline (e.g. Angeli & Hewitt, 1998; Ioannou, 2006; Trallero, 1995; Pal 1993).
1.2 FUNDAMENTALS OF MULTIPHASE FLOW
In this thesis, multiphase flow refers to fluid flow with two immiscible phases. In
particular, oil/water flow will be focused to extend the investigation on this system. One
of the most significant differences between two-phase flow and single phase flow is the
presence of flow regimes (i.e. how the two phases are distributed). The development of
29
these flow regimes are determined by the change of operational condition (e.g. the input
flow velocities of the phases and the way they are introduced). Other influencing factors,
e.g. fluid rheological properties and pipe material, can also significantly affect the spatial
distribution of the phases. The effect of some of these influencing factors will be
investigated and presented in subsequent chapters.
1.2.1 THE OCCURRENCE OF PHASE INVERSION
As the flow rate increases or the Reynolds number is sufficiently high, one of the phases
may be broken up into dispersed drops in the continuum of the other phase. If the
concentration of the dispersed phase is gradually increased, this phase will become
closely packed and, at some point, the drops coalesce and the phase continuity will
switch. The initial continuous phase will on the other hand become dispersed as a result.
The change of phase continuity is generally referred to as phase inversion. The
corresponding phase fraction at which this change occurs is called the phase inversion
point. The occurrence of a catastrophic inversion process is however arguable.
Phase inversion investigation begins from work in agitated vessels where two immiscible
liquids are mixed. Experimental results showed that a hysteresis effect occurs between
the inversions from either phase as the continuous phase (i.e. the dispersed phase tends to
remain dispersed). This results in the formation of an ambivalent region (i.e. a range of
phase fractions) over which either phase can be continuous (Selker and Sleicher, 1965;
Luhning and Sawistowski, 1971; Arashmid and Jefferys, 1980). Literature reviews that
the width of the ambivalent region is dependent on the initial setup condition, viscosity
ratio of the two phases as well as the vessel wall material.
The occurrence of phase inversion will lead to changes in the system with different
rheology. Understanding the phase inversion process is thus important as its occurrence
can be beneficial (e.g. polymerisation) or catastrophic (e.g. significant changes in
pressure gradient in pipeline). Failure to account for the occurrence of phase inversion
can lead to reduced pipeline capacity and lower oil productivity.
30
In addition to the experimental investigation on the behaviour of the mixture flow during
phase inversion, research has also been conducted to predict the critical phase fraction at
which phase inversion occurs (e.g. Yeh et al., 1964; Luhning & Sawistowski, 1971;
Arashmid & Jefferys, 1980; Brauner & Ullman, 2002). However without a well
developed understanding on the actual mechanism for phase inversion, the suggested
models still have a wide discrepancy between the prediction and the actual experimental
results.
1.3 OBJECTIVES OF STUDY
The work presented in this thesis summarizes the work conducted at the University
College London (UCL). The main aim of the work is to gain further understanding on the
phase inversion process and subsequently evaluate the effects of various influencing
conditions on the change in phase inversion occurrence as well as the corresponding
impact on the flow. Prediction models for phase inversion and simulation of highly
concentrated dispersion will aim to identify improvement to better account for the phase
distribution and inversion especially for scenarios where experimental investigation is not
permitted.
The objectives for the experimental and theoretical investigations can be summarized as
follows:
Objectives for experimental investigation:
(1) To understand the flow development and the formation of flow regimes which
will eventually lead to the onset of phase inversion as the operating conditions are
changed.
(2) To investigate the occurrence of phase inversion using various measurement
techniques and account for the spatial distribution during the phase inversion
process. In addition, measurement of pressure gradient will be made as the
indication on the energy requirement for the fluid flow during phase inversion.
31
(3) To investigate the effect of various influencing factors (e.g. fluid rheology, pipe
materials, etc) on phase inversion occurrence. It will also be made available the
corresponding effect on pressure gradient due to the changes.
(4) To apply the developed technology of dual impedance probe for the investigation
on the changes in drop size distribution during phase inversion. This will aid in
the learning on the mechanism of phase inversion based on the drop coalescence
and break-up.
(5) To obtain the necessary data in the development of prediction models as well as
input conditions necessary for computational simulation of dense dispersed flow.
Objectives for numerical modelling:
(1) To apply the mechanism of momentum balance of the fluids and predict the
critical phase fraction for phase inversion to occur.
(2) To predict the flow conditions during phase inversion based on the momentum
balance of the fluid flow.
(3) To establish an understanding on the fluid motion of densely dispersed two-phase
flow through computational simulations.
1.4 STRUCTURE OF THE THESIS
In order to achieve the thesis objectives, the investigations described in this thesis is
organised as follows. A literature review is conducted in Chapter 2 to cover past studies
on the experimental conditions at which phase inversion occurs and the associated
phenomena. It also covers the theoretical studies on the mechanism causing the
occurrence of inversion. Lastly, literatures covering the application of computational
fluid dynamics are also reviewed to apply the necessary techniques for simulation
studies.
Chapter 3 presents the experimental facilities at which the occurrence of phase inversion
will be investigated. The instrumentations used in the experimental investigations will be
described in detail. The use of these instrumentation and the presentation of how phase
inversion is developed in an oil/water system is presented in Chapter 4. The associated
32
change in the drop size distribution of the dispersed phase and pressure gradient will also
be discussed. In Chapter 5, the effect of changing the interfacial tension of the oil/water
system is tested by the addition of glycerol. The investigation will provide an
understanding on the corresponding changes to the phase distribution, drop size and
associated pressure gradient.
In Chapter 6, a prediction model is developed to estimate the critical phase fraction at

which phase inversion will occur based on the criteria of momentum balance. This model
development was initiated due to the observation of a peak in pressure gradient at near to
the phase inversion point. Chapter 7 extends on the development in Chapter 6 and other
models developed in UCL (e.g. the entrainment model by Al-wahaibi and Angeli (2009))
to predict the phase distribution and corresponding pressure gradient based on the
momentum balance. A selection criterion is thus developed to select either the
homogenous mixture model or the two-fluid model for the determination of the pressure
gradient.
In Chapter 8, an attempt is made to establish an understanding on the fluid motion of a

densely dispersed two-phase system through computational fluid dynamics. Various
interphase forces are studied to identify their significance in distributing the two phases.
Current limitations of CFD for dense dispersion will also be discussed.
Lastly, an overview on the investigation outcome will be presented in Chapter 9 and

future works are recommended for review.
33
Chapter 2: Literature Review
2.1 OVERVIEW
This chapter presents a review of both experimental and theoretical studies on phase
inversion during oil-water flows. In Section 2.2, the liquid-liquid flow regimes in
pipelines and the relevant flow pattern maps will be reviewed. Section 2.3 discusses the
occurrence of phase inversion together with the effect of various parameters both in
stirred vessels and pipelines. Prediction models for phase inversion are then presented in
Section 2.4 based on several proposed inversion mechanisms. Among the model
parameters, drop size of the dispersed phase and its distribution is regarded to be
important. Observation of various types of drop size distributions in experiments are
presented in Section 2.5 and commonly used characteristic diameters for drop size will
also be introduced. In addition, understanding of the interaction between the fluid phases
in a dispersed flow is important but complex. In Section 2.6, the use of computational
fluid dynamics (CFD) simulations to predict the interactions of the fluid phases in an oil-
water mixture pipe flow will be discussed. Various closure equations and interphase force
correlations will be presented in Section 2.7. Section 2.8 discusses on the conclusions
from the reviews.
2.2 FLOW DEVELOPMENT AND FLOW PATTERN MAPS
When two immiscible fluids flow simultaneously in a pipeline (e.g. oil-water flow), a
number of different flow regimes will appear depending on how the phases are
distributed in a pipe cross section.
The identification of flow regimes is usually studied through visual observations and
image recording with high speed cameras. However, visual techniques do not always
offer a clear indication of regime transitions and of the oil-water interface especially at
high velocities. As such, flow regime transitions are also determined by indirect methods
based on the associated flow characteristics (e.g. pressure drop, conductivity, etc).
Pressure gradient, for example, will show significant changes during flow regime
34
transitions. Al-Sarkhi and Soleimani (2004) attributed the pressure gradient changes,
during transition from smooth stratified to slug flow, to the formation of interfacial waves
that cause significant increase in interfacial and wall shear stresses. Conductivity is also
widely used as an indicator for phase distribution if one of the phases is non-conductive.
The conductivity methods for detecting the distribution of the two phases also set the
foundation for the development of commercialized instruments such as electrical
capacitance tomography (ECT) and electrical resistance tomography (ERT).
The experimental observations of these flow regimes are typically mapped with respect to
either the fluid superficial velocities or mixture velocities, or phase fraction. An example
is presented in Figure 2.1 from the work of Nädler and Mewes (1997). It can be seen that
different flow patterns can emerge depending on the variations of fluid flow rates or their
phase fraction. A number of different names have been given by various investigators on
the flow patterns (see Table 2.1).
(a) (b)
Figure 2.1: (a) Flow regime map that is based on the fluid superficial velocities (b) Flow
regime map that is based on input water fraction and mixture velocity.
35
Author Flow pattern classifications
Russell et al. (1959) Oil Bubbles in water
Stratified flow (SF)
Mixed Flow
Guzhov et al. (1973) Stratified flow
SF with mixing at the interface and a water lower layer
SF with mixing at the interface and a lower layer of oil/water dispersion
Water/oil and oil/water emulsions
w/o emulsion
oil/water emulsion and a water lower layer
oil/water emulsion and a lower layer of oil/water dispersion
o/w emulsion
Oglesby (1979) Segregated
Semi-segregated
Semi-mixed (oil dominant, water dominant)
Mixed (oil dominant, water dominant)
Annular or concentric core of one phase within the phase (oil dominant,
water dominant)
Slug: Phases alternatively occupying the pipe as a free phase or as a
dispersion phase (oil dominant, water dominant)
Semi-dispersed (oil dominant, water dominant)
Fully-dispersed: homogeneous mixture (oil dominant, water dominant)
Nädler & Mewes (1995) Stratified flow
SF with mixing at the interface and a water lower layer
SF oil/water dispersion and a water lower layer
o/w emulsion
SF water/oil dispersion, oil/water dispersion and a water lower layer
SF water/oil dispersion and a water lower layer w/o emulsion
Trallero (1995) Stratified flow
SF with mixing at the interface
o/w dispersion and free water layer
w/o dispersion and o/w dispersion
Full o/w emulsion
Full w/o emulsion
Brauner (2002) Stratified flow
SF with mixing at the interface
SF with a free liquid and a dispersion of another liquid (Do/w&w)
SF with a free liquid and a dispersion of another liquid (Dw/o&w)
w/o dispersion above o/w dispersion
w/o dispersion above o/w dispersion with pure oil at the top and pure
water at the bottom
Full o/w dispersion
Full w/o dispersion
Core-annular flow (viscous oil in core and water in annulus)
Core-annular flow (water in core and oil in annulus)
Core-annular flow (dispersion of w/o in core and water in annulus)
Core-annular flow (dispersion of o/w in core and oil in annulus)
Core-annular flow (dispersion of one phase in core and dispersion of
another in annulus)
Intermittent flow
Elongated or spherical bubbles of one phase in a continuum of another
phase
Table 2.1: Flow pattern classification suggested in various literature.
The main ones are stratified flow, annular flow, dual continuous flow and dispersed flow
and are described in more detail below:
36
a) Stratified flow (ST) – (Figure 2.2a): Stratified flows are generally occurring at
low flow rates. Each of the oil and water layers flows as a continuum with a
distinct liquid-liquid interface. At the low flow rates, the flow is dominated by the
gravitational effect rather than inertia. As such, the separation of the two layers is
based on the density difference between the two phases. The oil phase (i.e. the
lighter phase) occupies the upper layer while the water phase occupies the lower
layer. As the flow rate is increased, waves of various lengths and amplitudes will
form at the interface and droplets of one phase may be entrained into the other
leading to a transition into the dual continuous flow pattern. A description of this
transition is presented by Al-Wahaibi et al. (2007).
(a)
(b)
Figure 2.2: (a) Stratified smooth flow (b) stratified wavy flow with the oil phase on the
top layer while the water phase on the bottom layer. (Ngan et al., 2007)
b) Dual continuous flow (DC) – (Figure 2.3): Dual continuous flow generally
occurs at intermediate mixture velocities between stratified and fully dispersed
flows. The oil and water phases retain their continuity and there is a distinct
interface between them. However, one or both phases are entrained into the other
as drops. The degree of entrainment varies according to the fluid velocities.
37
The transition into a dual continuous flow generally begins with wave formation
from a stratified flow. The interfacial waves are initially long compared to the
pipe diameter. As the fluid velocities increase, these waves become shorter.
Guzhov et al. (1973) reported that the relative movement of the two liquid phases
causes vortex motions due to the shear forces that penetrate the interface boundary.
This causes the formation of small droplets of one phase into the other (and is
regarded as the onset of entrainment). Once the drops are entrained, the
distribution of these drops will depend on the balance between inertial and
gravitational forces. At low flow rates, the entrained drops are few and gravity
tends to keep them near to the fluid interface. This flow pattern is sometimes
referred to as stratified flow with mixing dispersion at the interface. As the flow
rate increases, the degree of entrainment increases as drops are more evenly
distributed in the opposite layer due to the increased importance of the inertial
forces.
Figure 2.3: Dual continuous flow with several oil drops entrained in the water
continuous layer. (Ngan et al., 2007)
c) Annular flow (AN) – (Figure 2.4): A core of one phase is surrounded by an

annulus of the other. This flow pattern is common when the two phases have
equal densities or when one of the phases has a very high viscosity (Russell and
Charles, 1959).
Figure 2.4: Annular flow with an oil core and a water annulus. Small amount of drops
are also observed within the oil core in the figure. (Ngan et al., 2007)
38
d) Dispersed flow (D) – (Figure 2.5): One of the phases loses its continuity and
forms drops in the continuum of the other. This pattern occurs at high velocities.
There is usually a distribution of drop sizes which depends on the balance of drop
break-up and coalescence events. Drops can also be deformed and the deviation
from their sphericity can have significant impact on the mixture flow. In addition,
gravitational forces affect the vertical concentration of these drops leading to a
spatial distribution.
(a)
(b)
Figure 2.5: (a) Dispersed flow of oil drops in water. Drops observed are polydispersed
and highly deformable. (b) Oil slug flow in water (Ngan et al., 2007).
2.3 PHASE INVERSION
Phase inversion is a commonly observed phenomenon in dispersed liquid-liquid mixtures

(e.g. in pipe flow or in stirred vessels) but its mechanism is still not well understood. Two
types of dispersions are generally found (i.e. oil-in-water and water-in-oil) according to
the phase fraction and initial conditions. Phase inversion is generally found when the
mixture undergoes changes in the phase distribution as the phase fraction reaches certain
critical values (Yeh et al., 1964; Arirachakaran et al., 1989; Pal, 1993; Pacek et al., 1994;
Elseth, 2001; Ioannou et al., 2005; Hu, 2005; Piela et al., 2008). The critical phase
fraction where inversion occurs is known as phase inversion point. Coalescence and
break-up of the dispersed phase occur continuously in a dispersion. At low dispersed
39
phase fractions, this dynamic process can reach equilibrium. As the dispersed phase
fraction increases, the process may be unbalanced and coalescence becomes more
prominent due to the proximity of the dispersed drops. Eventually, phase inversion will
occur when the two phases switch their continuity. The occurrence of phase inversion and
the changes in phase continuity can lead to substantial changes in the mixture rheology
causing large fluctuations in pressure gradient during pipe flow (Angeli, 1996; Nädler
and Mewes, 1995).
Arirachakaran et al. (1989) has presented the development of the phase inversion process
as the initial water dispersed phase fraction is increased (see Figure 2.6). Some authors
have suggested that the inversion process is rapid and catastrophic (Smith and Lim, 1990;
Tyrode et al., 2005; Vaessen et al., 1996). However, other investigations suggest a
gradual inversion process particularly during pipe flow (e.g. Liu, 2005; Piela et al., 2008)
when partial inversion occurs at a certain location before the entire mixture is inverted.
The details of these observations will be discussed in the subsequent sections on phase
inversion in stirred vessels and pipe flow.
Figure 2.6: A schematic on the proposed phase inversion mechanism by Arirachakaran

et al. (1989).
40
2.3.1 PHASE INVERSION IN STIRRED VESSELS
Investigations on phase inversion have mainly been carried out in stirred vessels and the
outcomes from these works can provide valuable information on the phenomenon in pipe
flow. Reviews of literature available on phase inversion in stirred vessels can be found in
Yeo et al. (2000), Liu (2005) and Hu (2005).
One of the interesting findings in stirred vessel is the hysteresis effect observed when
inversion is approached from a water continuous or from an oil continuous dispersion.
The term ambivalent region has been introduced to define the region of phase fractions
where either dispersion can exist. Thus, either type of dispersion can only be clearly
defined beyond the ambivalent region. Figure 2.7 presents experimentally found
ambivalent region (Noui-Mehidi et al., 2004).
Figure 2.7: The existence of ambivalent region between the aqueous and organic
continuous dispersion across the organic phase fraction (Φ o) (Noui-Mehidi et al., 2004).
Pacek et al. (1994) used video recording to capture the phase inversion process in a
stirred vessel and found that the drop size increased significantly near inversion while
secondary droplets (i.e. continuous phase drops within the dispersed phase) are formed
41
(Figure 2.8). Pacek et al. (1994) also found that secondary droplets only occur for
chlorobenzene in the dispersed glycerol/water phase but not in the opposite dispersion.
This difference in the formation of secondary droplets between the organic and the
aqueous continuous dispersions could be responsible for the appearance of the
ambivalent region. The formation of secondary droplet can have a significant widening in
the ambivalent region and hysteresis effect during phase inversion. According to them,
the coalescence of the dispersed phase is the most important mechanism controlling
phase inversion.
Figure 2.8: Droplet in drop for a glycerol/water and chlorobenzene system captured by
Sony video printer (Pacek et al., 1994).
Phase inversion and the ambivalent region are affected by many parameters. For example,
Selker and Sleicher (1965) found that the properties of the apparatus used (i.e. size and
materials of the vessel and impellers) and the operational conditions (i.e. agitation speed,
phase fraction) did not affect significantly the phase inversion. Deshpande & Kumar
(2003) also found that the limits of the ambivalent region reach asymptotic values at high
agitation speed and thus depend only on properties of the fluid system. According to
42
Selker and Sleicher (1965), fluid density would affect the ambivalent boundaries if the
agitation is slow and settling of the denser phase is prominent. The phase with higher
viscosity is also more likely to be the dispersed one. Viscosity also affects the ambivalent
region probably because of the lower coalescence rates caused by longer film drainage
time (Coulaloglou and Tavlarides, 1977). Interfacial tension has also been reported as an
important factor affecting phase inversion. Luhning & Sawistowski (1971) and Norato et
al. (1998), for example, showed that a decrease in interfacial tension can lead to a
widening of the ambivalent region.
Various empirical correlations have been developed from experiments to predict the
boundaries of the ambivalent region. The critical phase fraction of the organic phase, Φo,
was given by:
Luhning and Sawistowski (1971) for impeller speed between 600 to 1360 rpm:
(1) Φ for the upper inversion curve (w/o → o/w) (2.1)
(2) Φ for the lower inversion curve(o/w → w/o) (2.2)
Fakhr-Din (1973) for impeller speed below 680.85rpm:
(1) Φ for the upper curve (2.3)
(2) Φ for the lower curve (2.4)
where µc and µd are the continuous and dispersed phase viscosities (in Pa.s), Δ is the
density difference between phases (in kg/m3), FrI, ReI and WeI are the Froude number,
Reynolds number and Weber number at the impeller region.
The above equations demonstrate the importance of fluid properties on the width of the
ambivalent region. Fakhr-Din’s correlation also indicates the importance of the agitation
speed and impeller dimensions (in the Froude and Reynolds number terms). This may be
due to the low agitation speeds used where there may have been a separation of the
mixture.
43
Several mechanisms have been suggested by various investigators on the phase inversion
mechanism. According to Pacek et al. (1994), it is the imbalance between the break-up
and coalescence processes of the dispersed drop. This is also similarly suggested by
Arashmid & Jeffreys (1980) and Groeneweg et al. (1998). Phase inversion has also been
suggested to occur when the system free energy of the two possible dispersions (oil
continuous or water continuous) become equal (e.g. Luhning & Sawistowski, 1971;
Tidhar et al., 1986; Yeo, 2002). Yeh et al. (1964) have suggested that inversion occurs
when there is no shear between the two phases. However, it is difficult to measure
accurately the interfacial area close to and during inversion as well as the drop break-up
and coalescence rates. Further details about prediction of phase inversion based on these
mechanisms are presented in Section 2.4.
2.3.2 PHASE INVERSION IN PIPE FLOW
The investigation of phase inversion in pipelines can lead to better understanding on the
operating conditions and help to improve the pipe design to facilitate the transportation of
multiphase mixtures. This is particularly important as the presence of water is inevitable
and the mixture can only be separated after it has been transported over miles of pipeline
before processing. Inversion has been found to cause significant increase in pressure
gradient (Martinez et al., 1988; Angeli, 1996; Valle and Utvik, 1997; Nädler and Mewes,
1997 and Soleimani et al., 1997). A good understanding of the occurrence of phase
inversion is necessary to predict and control the pumping power required to transport the
mixture across and may lead to poor productivity. The exact relation between phase
inversion and pressure gradient is not well understood. For example, Ioannou (2006) has
shown that phase inversion occurs at the peak of the pressure gradient for Exxsol D80 but
not for Marcol 52 (see Figure 2.9). Exxsol D80 is less viscous (1.7 mPa.s) while Marcol
52 is more viscous (11mPa.s). Due to the high viscosity of Marcol 52, the high pressure
gradient caused during the oil continuous flow may have overshadowed the peak in
pressure gradient during phase inversion leading to an almost step change as a result.
44
2.0
Pressure gradient, kPa/m

Phase
inversion
1.5
1.0
0.5
0.0
0 20 40 60 80 100
Input oil fraction, %
Type I (o/w←w/o) Type II (o/w→w/o)
(a)
3.0
Phase inversion
Pressure gradient, kPa/m
2.5
initiation
2.0
1.5
1.0
0.5 Phase inversion

completion
0.0
0 20 40 60 80 100
Input oil fraction, %
Type I (o/w←w/o) Type II (o/w→w/o)
(b)
Figure 2.9: (a) Pressure gradient measurement of a Exxsol D80/water system in a 60mm
I.D. pipe (b) Pressure gradient measurement of a Marcol 52/water system in a 60mm I.D.
pipe (Ioannou, 2006).
45
While changes in pressure gradient are direct consequences of the phase inversion
process, these changes cannot yet be accurately attributed to the point of phase inversion
and thus cannot be used to detect phase inversion. Other system parameters have been
used to detect the inversion process, e.g. conductivity. Conductivity measurements are
applicable to the flow of oil-water mixtures particularly as one phase is conductive and
the other is not. During phase inversion, the change is prominent with sharp changes in
conductivity. Conductivity measurements can be used to detect phase continuity at
locations where visual observations are not possible (e.g. at the centre of a pipe or in
highly concentrated dispersions). Such measurements have indicated that inversion may
not occur simultaneously across the whole pipe cross section. Soleimani et al. (2000)
reported that the spatial distribution of the two phases can be inhomogeneous due to
wetting effects of the pipe wall causing the water phase to concentrate towards the wall
and the oil phase concentrate in the pipe core. This inhomogeneity implies that phase
inversion may occur locally. Using laser induced fluorescence, Liu et al. (2006) showed
that zones of oil and water continuous dispersion appear within the pipe cross section
during inversion. This gradual phase inversion process has led to a transitional region
where phase inversion begins and completes over a range of phase fraction. An example
of such gradual inversion can be seen in Figure 2.10 (Hu, 2005) where oil continuous
dispersion begins to appear at one phase fraction until water becomes completely
dispersed at a higher dispersed phase fraction. Secondary droplets may also appear at
certain regions (see Figure 2.11 for example). Piela et al. (2006, 2008) have also reported
the formation of multiple dispersions and gradual inversion that spread over the cross
section in their pipe flow experiments.
46
Transitional
Region
O/W W/O
Figure 2.10: Conductivity measurement conducted at the centre of a vertical pipe across
the phase inversion process (Hu, 2005).
Secondary droplets
Figure 2.11: Multiple dispersions of an oil/water system at a mixture velocity of 1.5m/s

and input oil fraction of 31%. The dark region represents the oil phase and the light
region represents the water phase. (Liu et al., 2006)
47
The effects of various parameters on phase inversion have been studied for pipe flow and
a summary of these parameters is presented in Table 2.2. It can be seen that fluid
properties, size and material of pipe used as well as operating conditions during the pipe
flow are of great importance. These parameters are similar to those observed in stirred
vessels.
Parameter investigated Author (Year)

1. Viscosity Arirachakaran et al.(1989); Luo et al. (1997); Nädler and
Mewes (1997)
2. Pressure drop Martinez et al.(1988); Angeli et. al (1996, 1998, 2000);
Valle and Utvik (1997); Luo et al. (1997); Nädler and
Mewes (1997); Soleimani (1999); Ioannou et al. (2004,
2005)
3. Velocity Luo et al.(1997); Angeli & Hewitt (1998 & 2000)
4. Phase Distribution Arirachakaran et al.(1989); Nädler and Mewes (1997);
Ioannou et al. (2004, 2005)
5. Pipe diameter & Material Arirachakaran et al.(1989); Ioannou et al.(2004, 2005)
6. Surfactant Pal (1993); Gillies et al. (2000)
7. Wettability Ioannou et al.(2004, 2005); Pettersen et al.(2001)
8. Conductivity Ioannou (2006); Hu (2005)
9. Drop size Liu et al. (2004) & Liu (2005); Hu (2005)
10. Interfacial tension Rodriguez and Bannwart (2006)
Table 2.2: Parameters affecting phase inversion in liquid-liquid pipeline flow.
2.3.3 PARAMETRIC STUDY ON PHASE INVERSION
The parameters which affect phase inversion in pipes are discussed in more detail here:
Agitation speed / Mixture velocity in pipe

The speed of agitation and the velocity of the mixture in the pipe flow have been reported
to enhance the mixing process and the dynamic processes of drop coalescence and break-
up. McClarey and Mansoori (1978) showed phase inversion occurs at a higher volume
fraction of the aqueous phase when the impeller speed is increased from 240RPM to
480RPM. However at high impeller speeds (beyond 480RPM), there is no effect of
48
agitation on the phase inversion hold up. Similar findings were also reported by Quinn
and Sigloh (1963), Luhning and Sawistowski (1971), and Deshpande and Kumar (2003).
Pacek et al. (1994) found that phase inversion will not occur under certain circumstances
if the agitation speed is very high but will occur if the agitation speed is reduced. On the
other hand, they also observed that phase inversion will not occur at very low speeds.
Efthimiadu and Moore (1994) suggested that the increase in agitation speed will increase
the shear rate and will lead to finer and more stable dispersions even at higher dispersed
phase fraction.
Similar experiments have been conducted in pipe flows by Arirachakaran et al. (1989)
who showed that mixture velocity has little effect on the occurrence of phase inversion.
However, Ioannou et al. (2005) found that an increase in mixture velocity increased the
fraction of oil continuous phase when inversion occurred.
Fluid Density
The effect of fluid density was previously found to be important in gas-liquid system
where the density difference is large. In liquid-liquid dispersed systems, Selker and
Sleicher (1965), McClarey and Mansoori (1978) and Norato et al. (1998) have shown that
the fluid densities have little effect on the ambivalent region in stirred vessels provided
that the difference in density is not large. However, large density difference can lead to
difficulty in maintaining the dispersion at low agitation speeds. Thus, more stirring power
is required to achieve an homogeneous dispersion. Other investigators have found that a
large density difference can increase the tendency for phase inversion (Rodger et al.,
1956; Kumar et al., 1991). Chiang and Chen (1994) found that inversion will occur at a
higher aqueous phase fraction when water is the dispersed phase for a fluid system with
large density difference.
Efthimiadu and Moore (1994) performed their phase inversion experiments in parallel
plates and found that the density of the aqueous continuous phase caused a small increase
in the volume fraction of the organic phase at inversion. However, these experiments
were conducted in conditions where effects of other fluid properties are coupled.
49
Fluid viscosity
Fluid viscosity has been reported in various investigations to be a significant factor
affecting phase inversion as well as the ambivalent region. Selker and Sleicher (1965)
reported that the viscosity ratio of the two phases in a stirred vessel can influence the
ambivalent boundaries. In addition, an increase in the viscosity of one phase will increase
its tendency to be the dispersed phase. However, Treybal (1963) suggested that the higher
viscosity of one phase will favour it to be the continuous phase. Efthimiadu and Moore
(1994) also found that the more viscous phase tends to become continuous in parallel
shearing plates (especially if the viscous phase tends to wet the surface). They attribute
this finding to be the increased dynamic stability of the dispersion when the viscous
phase is continuous. Similar results have also been reported by Norato et al. (1998), who
found that the upper and lower limit of the ambivalent region can be widened by
increasing the viscosity regardless of whether the phase is continuous or dispersed. The
widening behaviour is postulated to be caused by the lower coalescence rate due to longer
film drainage time which agrees with the findings from Groeneweg et al. (1998),
Coulaloglou and Tavlarides (1977) and Calabrese et al. (1986).
Arirachakaran et al. (1989) from literature data on inversion in pipelines with different
viscosity oils found that the more viscous oils tend to be dispersed. Similarly, Ioannou
(2006) showed that the more viscous Marcol 52 oil tended to invert at a higher oil
fraction. Nädler and Mewes (1997) however using oils with different viscosity (between
22 and 35 mPa.s) found no significant effect on inversion.
Interfacial tension
The effect of interfacial tension is the least well understood among the fluid properties on
phase inversion. According to Selker and Sleicher (1965), the magnitude of interfacial
tension is unlikely to influence the ambivalence region as the interfacial tension between
a pair of fluids is not a function of the interfacial curvature. Yeh et al. (1964) suggested
that interfacial tension plays a small role on phase inversion. In the absence of other
forces, the interfacial tension will cause inversion to occur at 50% volume fraction.
50
Luhning and Sawistowski (1971), on the other hand, found that interfacial tension can
affect the phase inversion since it affects drop size and interfacial area. Clarke and
Sawistowski (1978) later confirmed the change in interfacial area during phase inversion
and showed that the width of the ambivalent region is significantly affected by interfacial
tension. Lowering the interfacial tension will widen the width.
Norato et al. (1998) also found that decreasing the interfacial tension widens the
ambivalent region and makes it more difficult to cause an appearance of phase inversion.
According to Coulaloglou and Tavlarides (1977), decreasing the interfacial tension will
lead to an increase in drainage time for the film between drops leading to higher drop
breakage and lower drop coalescence and as such smaller drop size. A higher dispersed
phase fraction will thus be required for inversion to occur. This suggestion has been
confirmed by experiments. Efthimiadu and Moore (1994) found that lowering the
interfacial tension has a stabilising effect on the type of dispersion which depends also on
the other properties of the system. It was also experimentally found that the effect of
shear rate (i.e. rotational speed and gap width) is minimised with the decrease in
interfacial tension. The influence of the wettability of the contacting surfaces is also
reduced.
Surface wettability
The material of construction of the dispersion container (e.g. stirred vessel or pipe) has
been suggested to affect the type of dispersion formed and as a result phase inversion and
ambivalent region. McClarey and Mansoori (1978) suggested, through their experimental
outcome for an equal density and viscosity mixture, that the discrepancies in their
inversion curves can be attributed to the difference in wettability of the vessel surface
with the two liquids. However, the effect is only found to be significant at low agitation
speeds. Giulinger et al. (1988) changed the material used in the stirred vessel from water
to oil wetted and showed that the wetting phase was favoured to remain continuous. In
stirred vessels, drop break-up generally occurs within the impeller region and coalescence
at the other regions. However, Kumar et al. (1991) reported that the drops can coalesce
on the impeller and form a thin film if the impeller is preferably wetted by the dispersed
51
phase. The drops will subsequently be broken at the impeller or at its vicinity. The effects
on the coalescence and break-up process can significantly change the phase inversion
behaviour.
Efthimiadu and Moore (1994) investigated the effect of material on phase inversion of
liquid-liquid dispersions between parallel plates. Their experimental findings indicated
that the liquid which preferentially wets the surface of the plates tends to form the
continuous phase.
The effect of surface wettability is not well documented for pipeline flow. Angeli and
Hewitt (1998) conducted experiments in 24 mm I.D. pipe with Exxsol D80 as the oil
phase and water as the aqueous phase. They observed similar range of input water
fraction (at 37% - 40%) regardless of the material of the pipe (acrylic and stainless steel).
Pettersen et al. (2001) conducted experiments with the same type of oil in PVC and
stainless steel pipes of 100 mm I.D. and found that inversion is delayed to a higher oil
fraction in PVC compared to the steel one when experiments started from an oil
continuous mixture. Ioannou et al. (2005) found that acrylic is wetted more by the oil
phase than steel and thus phase inversion occurs at a higher oil fraction for acrylic pipe.
Inlet conditions
Inlet conditions are important as these determine how the two fluids are introduced into
the pipe and, coupled with other effects, can affect the phase inversion appearance. Piela
et al. (2006, 2008) discussed two approaches in introducing the phases into the test pipe.
In the continuous experiment, the continuous phase was introduced in the pipe loop and
wetted the pipe wall. The dispersed phase was then injected into the loop and a
corresponding volume of mixture was removed to ensure constant mixture velocity of
fluid in the system. The dispersed phase fraction was gradually increased with the
continuous injection of the dispersed phase until phase inversion occurred. In the direct
experiment, the two fluids were introduced as two separate continuous phases and
subsequently mixed within the test pipe. The input flow rates of the two phases were
adjusted to maintain a constant mixture velocity. From the experimental outcome, the
52
critical fraction of the dispersed phase for inversion can be significantly higher for the
continuous experiment than the direct one. The authors also suggested that the phase
inversion can be postponed or avoided by altering the inlet conditions.
Dispersion initialisation
The initialisation of the dispersion has been reported to have a strong effect on phase
inversion. For example, Selker and Sleicher (1965) found that the inversion point in
stirred vessels is strongly influenced by the initial type of dispersion (i.e. water
continuous or oil continuous) especially if the phase fraction is within the ambivalent
region.
The occurrence of ambivalent region as a direct effect of dispersion initialisation has

been seen mainly in stirred vessels but not studied systematically in pipes. Ioannou et al.
(2005) observed in their pipe flow experiments that the initial continuous phase will tend
to preserve its continuity leading to an hysteresis effect in the occurrence of phase
inversion according to the initial dispersion type. Hu and Angeli (2006), however,
observed no effect of dispersion initialisation in a vertical pipe flow based on their
experimental conditions studied.
2.4 PHASE INVERSION MECHANISMS AND PREDICTION

MODELS
Several types of prediction models based on different phase inversion mechanisms have
been reported in various literatures. These models will be discussed in detail in the
following sections.
2.4.1 MODELS BASED ON RATE OF DROP BREAK-UP AND

COALESCENCE
Phase inversion is commonly found in dispersion. Parameters affecting phase inversion

are known to also influence the drop coalescence and breakup (e.g. fluid properties,
agitation speed, wettability). This has led to various investigations to relate phase
53
inversion with the dynamic process of drop break-up and coalescence (Arashmid and
Jeffreys, 1980; Groeneweg et al., 1998). In a stable dispersion, e.g. O/W, the dynamic
process of drop break-up and coalescence is balanced. Increasing the dispersed phase (i.e.
oil) will lead to higher collision frequency and coalescence rate. Concurrently, more
drops of larger size will also be broken up. Thus, a new equilibrium of this dynamic
process will be formed.
According to Arashmid & Jeffreys (1980), collision of drops in a stirred vessel will lead
to some drops being coalesced. As the dispersed phase is increased under a constant
agitation speed, the chance of successful coalescence for each collision will increase. At
phase inversion, it is suggested that coalescence will occur for every collision. Thus, the
frequency of drop collision will be equal to the frequency of drop coalescence. Based on
Levich (1962)’s collision frequency model and Howarth (1967)’s coalescence frequency
model, the dispersed phase holdup (Φ dI) at inversion can be expressed as follows:
K
Φ dI  (2.5)
d N 0.48
2
d I
where K is a constant depending on the type of agitator, d d is the drop size and NI is the
agitation speed.
Vaessen et al. (1996) applied the idea from Arashmid & Jeffreys (1980) to determine
phase inversion. However, they suggested that drop break-up should be part of the
inversion process in addition to having coalescence occurring at each collision. At phase
inversion, drop break-up is dominated by the drop coalescence process and the
equilibrium drop size diverges.
To estimate the drop size at inversion, the following coalescence and break-up rate
equations are used.
54
Coalescence rate (based on Saffman and Turner, 1956):
R c (d)  6.87ε1/3d1/3n(d) 2 (2.6)
Break-up rate (based on Delichatsios and Probstein, 1976):
2 1/3 2/3 σ
R b (d)  1.37 ε d exp( )n(d) (2.7)
π 1.88ε 2/3d 5/3
where Rc(d) and Rb(d) are the rates of coalescence and break-up of drops with mean
diameter, d, ε is the turbulent energy dissipation rate, ζ is the interfacial tension, n(d) is
the number density of drops with diameter d.
It can be seen that the break-up process is proportional to the number of drops in a
volume while the coalescence process is proportional to the number of drops in a volume
squared. The coalescence rate will increase faster as the volume of dispersed phase
increases. This leads to phase inversion when the dynamic balance between the two rates
collapses.
Groeneweg et al. (1998) suggested that phase inversion is governed by the balance
between drop break-up and coalescence which can be altered in favour of coalescence by
the increase of the effective volume fraction of the dispersed phase on continuous stirring.
This results in the inclusion of the continuous phase into the drops of the dispersed phase
forming secondary droplets. The enclosed droplets can also be escaped back to the
continuous phase. The effective volume fraction of the dispersed phase will continue to
increase as long as the inclusion dominates over the escape. Inversion will start when the
effective volume fraction for inversion has been reached. At this point, the droplet size
increases significantly and inversion follows.
Desphande and Kumar (2003) proposed that drop break-up and coalescence do not occur
independently in high dispersed phase holdups. The ratio of the rates at which the two
events occur in turbulent flow field is determined by the average relative separation
between the drops and the physical properties of the fluids in the system. It is also
55
independent of the intensity of the turbulent flow field for sufficiently intense turbulent
flow.
Hu et al. (2004) applied the population balance model to predict the birth and death rates
of drops of specific size and account for the corresponding drop size distribution. The
birth and death of drops are associated with the drop break-up and coalescence. In order
to predict the phase inversion, the total volumetric break-up rate will be equal to the total
volumetric coalescence rate. With this approach, the ambivalent region can be found. It
agrees reasonably well with literature data except in high dispersed phase fraction. The
disagreement can be attributed to the coalescence efficiency correlations that are not
meant for dense dispersion.
2.4.1.1 SECONDARY DISPERSIONS
With the dynamic process of drop coalescence and break-up, secondary dispersions can
also arise in a dispersed flow as the continuous phase entrains in the dispersed phase (e.g.
O/W/O dispersion). The formation of secondary dispersions is found to be more likely
near the phase inversion point (Luhning and Sawistowski, 1971) and secondary
dispersions depend on the continuous phase. For example, Pacek and Nienow (1995)
have found that only O/W/O dispersion can exist in a water-kerosene mixture but not
W/O/W. Kumar (1996) suggested that secondary droplets are formed when successive
coalescence of drops leads to droplets entrained in them. However, two criteria must be
met (1) the coalescence efficiency should be high so that the drops can coalesce
simultaneously (2) the entrained droplets should be at a stable condition upon
entrainment. The absence of W/O/W dispersions could be due to the low coalescence
efficiency of the oil drops in water and the secondary water droplets, if any, will have
high coalescence efficiency leading to the water droplets escape to the water continuous
phase.
Liu et al. (2005) observed secondary droplets using laser induced florescence (LIF)
technique in pipe flow (Figure 2.12). The formation of secondary droplets will lead to a
higher effective dispersed phase fraction (i.e. the total volume fraction of the primary
56
drops and the secondary droplets) and would lead to an earlier occurrence of phase
inversion than that expected from the actual phase fractions of the two phases.
Secondary droplet
Primary drop
Figure 2.12: Visual observations of (a) O/W/O and (b) W/O/W dispersions using LIF
technique. The oil phase is shown in black while the water phase is shown in luminous
green (Liu et al., 2005).
2.4.2 MODELS BASED ON MINIMIZATION/EQUAL SYSTEM AND

INTERFACIAL ENERGY
Luhning and Sawistowski (1971) suggested that the total energy (including kinetic and
interfacial energy) of the system would reach a minimum value at phase inversion and the
change in total energy is caused by the fluid system itself. The measured impeller power
input in the stirred vessel was observed to remain constant throughout the inversion
process.
Other investigators (Tidhar et al., 1986; Decarre and Fabre, 1997; Yeo et al., 2002;
Brauner and Ullmann, 2002) have suggested that the surface energies (i.e. interfacial
energy and the surface energy of the fluids with the contacting wall) of the two possible
types of dispersion, O/W and W/O, should be balanced at phase inversion. To account for
the surface energy, the interfacial area can be calculated using Monte Carlo simulation
(Yeo et al., 2002) or Population Balance model (Hu, 2005). Brauner and Ullmann (2002)
expressed the interfacial area in terms of Sauter mean diameter and derived the following
equation for the oil volume fraction at phase inversion.
57
 / D 32 w / o  s  cos 
I
O  6 (2.8)
 / D 32 w / o   / D 32 o / w
where θ is the solid-liquid surface contact angle with 0   < 90 denoting a surface
preferentially wetted by water (hydrophilic surface) and 90 <   180 denoting a
surface wetted by oil (hydrophobic surface), s is the solid surface area per unit volume
where s=4/D and D is the pipe diameter, ζ is the interfacial tension and D 32 is the Sauter
mean drop diameter.
Equation 2.8 can be simplified as follows when the solid-liquid surface energy is
neglected (Yeo et al., 2002):
I
O D 32 O / W
 (2.9)
I
1  O D 32 W / O
While the correlations based on this approach does not account for the existence of
ambivalent region, Brauner and Ullmann (2002) suggested that ambivalent region can be
regarded as the change in interfacial tension when contaminant is present. Yeo et al.
(2002) use different correlations to calculate the Sauter mean drop diameter depending on
the dispersed phase fraction. The use of these correlations results in the presence of
ambivalent region according to the history of the system.
2.4.3 MODELS BASED ON ZERO SHEAR STRESS AT INTERFACE
Phase inversion has been previously mentioned to be influenced by the viscosities of the
liquid phases in the system. Yeh et al. (1964) developed a correlation for predicting the
critical phase fraction at inversion as a function of the viscosity ratio of the two phases
based on the assumption that the interfacial shear is zero at the point of the inversion (i.e.
no tendency to mix or create new surfaces). From the momentum balance of the two
phases, they suggested the following correlation for the phase inversion point which was
58
modified by replacing the viscosity of the continuous phase with the interfacial viscosity
( i) and was found to improve on the prediction:
1/ 2
 dI  
  d  (2.10)
1   dI   i 
where θdI is the dispersed phase fraction at phase inversion, d and i are the dispersed
phase and the interfacial viscosity respectively.
Nädler and Mewes (1995, 1997) also used the momentum balance of the phases and the
assumptions of zero interfacial shear and no slip between the two phases to develop the
following correlation for the initial water fraction at inversion.
1
 Iw  (2.11)
1/ k 2
 C  (1 n o )  n o 
1  k1  o o o
(DU m ) n w  n o 
(1 
 C w  w w  ww
n ) n 
where is the fluid density, subscripts o and w referring to the oil and water phases
respectively, D is the pipe diameter, and Um is the mixture velocity, and C and n are the
parameters used in the Blasius friction factor equation, CRe-n, k1 and k2 are the empirical
parameters depending on the type of dispersion (O/W or W/O) and flow conditions. If
both phases are in laminar flow, k1 and k2 can be taken as 1 and 2 respectively and
equation 2.11 is equivalent to equation 2.10.
2.5 DROP SIZE DISTRIBUTION AND CHARACTERISTIC

DIAMETER
As discussed in the previous section, the drop size of the dispersed phase and its
distribution are important elements in the models of phase inversion prediction (e.g.
based on the dynamic process of drop break-up and coalescence, equal and minimization
59
of system/interfacial energy). Drop size distributions in liquid-liquid dispersions and the
various ways of expressing them are discussed in this section.
2.5.1 TYPES OF DROP SIZE DISTRIBUTION
Coulaloglou and Tavlarides (1977) developed a model for drop size distribution in a
stirred vessel based on the effects of drop break-up and coalescence in an isotropic
turbulent field. A break-up model was derived by considering drop deformation, break-up
under influence of local pressure fluctuation and the time needed for a critically deformed
drop to break-up in a locally isotropic field. A coalescence model was also derived based
on the assumption that the coalescence rate is proportional to the collision rate and the
coalescence frequency in the kinetic regime of locally isotropic turbulent fields. They
used a Gaussian probability density function of drop sizes in the feed and found good
agreement between the model and experimental distribution in the stirred vessel (see
Figure 2.13 for comparison).
Figure 2.13: Comparison between calculated drop size distribution and experimental
drop size data. Dotted line represents the histogram at the circulation region and the solid
line represents the histogram at the impeller region. (Coulaloglou and Tavlarides, 1977).
Collins and Knudsen (1970) investigated experimentally dilute O/W dispersions (up to 10%
by volume) in pipe flow using 3 different organic phases and did not observe coalescence.
60
They suggested that the drop size distribution in the pipe is a result of two superimposed
distributions – (1) produced by the injection nozzle and (2) produced by the turbulence of
the flow. The drop size distribution is nearly log-normal at the injection nozzle but
deviates from the log-normal type along the flow downstream due to the action of
turbulence.
Karabelas (1978) re-examined some of the experimental data from Collins and Knudsen
(1970) and compared that with his own experimental results in pipe flow using two
different organic phases. He found that the drop size distribution can be fitted with a
Rosin-Rammler or an upper limit log-probability distribution. While both distribution
types yield similar result, Rosin-Rammler distribution in Equation 2.12 is simpler for use.
  d  
n
1  Vcum  exp  2.996   (2.12)
  d  
 95
where Vcum is the cumulative volume fraction of drops with diameter less than d, d is the
drop diameter and d95 is the drop size corresponding to 95% of drops smaller than d 95. n
is a fitting parameter of the distribution.
Angeli and Hewitt (1998) investigated the drop size distribution experimentally in an
oil/water pipe flow. They observed that the drop size distribution is strongly influenced
by the pipe material with drops smaller in steel pipe than acrylic pipe and can be
reasonably fitted with a Rosin-Rammler distribution with parameter, n, in Equation 2.12
ranging from 2.1 to 2.8.
Simmons and Azzopardi (2001) also examined the drop size distribution experimentally
in a pipe flow. The backscatter technique applied in their experiments allows drop size
distribution to be obtained even at highly dense dispersion. From their results, the drop
size distribution of the mixture flow fits well with an upper-limit log-normal distribution.
61
Vielma et al. (2008) compared their experimental results in an horizontal pipe flow of
water and mineral oil with various dispersion types – Normal, Log-normal and Rosin-
Rammler. They conclude that log-normal distribution is the best probability distribution
for all their experimental results regardless of the type of dispersion (O/W or W/O).
Figure 2.14 presents one of comparisons of the oil drop size distribution at an oil and
water flow velocity of 0.75m/s.
Figure 2.14: Comparison between experimental drop size data and selected drop size
distribution types in an horizontal pipe flow (Vso = 0.75m/s, Vsw=0.75m/s, oil drops).
(Vielma et al., 2008).
2.5.2 CHARACTERISTIC DIAMETERS OF DROP SIZE DISTRIBUTION
Knowing the changes in drop size distribution is useful to understand the underlying
dynamics in the drop break-up and coalescence process. However, the distributions are
difficult to compare across different studies. An easier way to characterize a drop size
distribution is by using the characteristic mean diameters despite that some information
may be neglected. Two common characteristic diameters are the linear mean diameter,
d10, and the Sauter mean diameter, d32. They are defined in Equation 2.13 and 2.14
62
respectively. In this thesis, Sauter mean diameter will be used in general as this diameter
is commonly reported in prediction of mean drop size for liquid-liquid dispersions.
(1) Linear mean diameter, d10, which defines the number-averaged drop size.
n
 di
d10  i 1 (2.13)
n
(2) Sauter mean diameter, d32, which defines the characteristic diameter that has the same
volume/area ratio as the distribution.
n
 d 3i
d 32  i 1 (2.14)
n
 d i2
i 1
where di is the diameter of the sample drop and n is the number of drops in the sample
distribution.
2.6 NUMERICAL SIMULATION ON DISPERSED FLOW
The occurrence of phase inversion is a direct consequence of the dynamic flow behaviour.
It involves the interaction of the fluid phases in the system and causes a spatial re-
distribution of the phases across the confined system (e.g. pipes). Phase inversion can be
influenced by many parameters and these parameters can themselves be inter-related. In
addition, empirical correlations on two-fluid flow usually provide global information on
the system. The important local information which can have great significance in
understanding dynamic phenomena such as phase inversion is neglected by such
correlations.
Information on the flow field would be important for understanding phase inversion. This
can be achieved by computational fluid dynamics (CFD) modelling. CFD is a
computational code for solving the continuity and Navier-Stokes equations using a
63
number of closure correlations. Various authors have attempted to improve the prediction
of multiphase flow and apply the development in simulating complex flows (Hill, 1998;
Rusche, 2002; Chesters and Issa, 2004; Liu, 2005; Krepper et al., 2005). While current
CFD simulation has been reported to achieve reasonable success in comparison with
experimental data, there are still a lot of developments required due to the uncertainty in
the actual mechanisms for the occurrence of fluid flow phenomenon (e.g. phase
inversion). The development of the CFD code should also be complemented with
improved experimental techniques for more detailed investigation on fluid flow.
While many process flows operate at high dispersed fraction, currently CFD handle dilute
dispersion with <10% dispersed fraction. This is due to the lack of information on terms
such as inter-phase momentum, turbulence effects as well as phenomenon such as phase
inversion at high dispersed phase fractions.
The common approaches used in CFD to simulate multiphase flows are (1) Eulerian-
Lagrangian, (2) Eulerian-Eulerian and (3) Volume of fluid (VOF). These approaches are
discussed below (more detailed discussion can be found in Hill (1998), Rusche (2002),
Ranade (2002)). In the Eulerian-Lagrangian (E-L) framework, the dispersed phase is
represented by individual dispersed elements (i.e. drops). The trajectories of these drops
are tracked by solving the appropriate momentum equations in a Lagrangian framework
(i.e. the dependent variables are functions of the properties of the tracked drop). At the
same time, the conservation equations for the continuous phase are expressed in the
Eulerian framework (i.e. the fluid properties are functions of space and time with respect
to the reference volume domain). This approach is good in capturing the flow dynamics
of the dispersed phase. The drop size distribution and inter-particle interactions can also
be accounted for. However, it is computational intensive especially at large numbers of
dispersed drops.
In the Eulerian-Eulerian (E-E) approach, the conservation equations of all the fluid
phases are expressed in the Eulerian framework. The fluid phases are assumed to share a
volume domain and interpenetrate during their motions. The presence of each phase is
64
accounted for by their velocity fields across the volume domain and their volume
fractions. Interphase transport models, such as drag force and lift force, are applied to
define the coupling effect between the phases. The Eulerian-Eulerian approach is more
efficient in computational resources as the dispersed phase is regarded as a single
continuum (and greatly reduces the number of conservation equations needed). This
approach is thus well suited for more dense dispersions (Sommerfelds et al., 2008).
However, the inter-particle interactions and drop size distribution are difficult to account
for. A schematic comparison between the E-E and E-L approach was presented by
Rusche (2002) and can be seen in Figure 2.15.
(a) (b)
Figure 2.15: Schematic representation on the modelling approaches for two-phase flow:
(a) Eulerian-Lagrangian approach and (b) Eulerian-Eulerian approach. The number next
to the blue arrow in (b) represents the volume fraction of the dispersed phase within the
control volume. (Rusche, 2002)
The Volume of Fluid (VOF) approach aims to simulate the motion of the fluid interface
by tracking the motion of the phases in a single set of conservation equations for the
mixture (Ranade, 2002). The VOF approach distinguishes the respective fluid phases
using a marker function (usually the volume fractions) within the control volume domain
in the system. If one phase occupies the entire control volume, its properties will be used
to avoid any discontinuity caused by the absence of the other phases. This approach is
65
suited for investigations where the shape and fluid motion at near the interface are of
interest. While it offers better resolution close to the interface, computational resources
increase significantly if there is a dense dispersed phase. This approach is thus commonly
used to investigate, for example, deformation of single drop in a flow field and the
breakup and binary drop coalescence processes (Stover et al., 1997; Delnoij, 1997;
Krishna and van Baten, 1999).
2.6.1 CONSERVATION EQUATIONS USED IN CFD
As phase inversion tends to occur in dense dispersions, the simulation of such dispersion
will be a valuable source of information in addition to the experimental data. It will also
ease the investigation on the effects of various parameters especially with conditions that
are difficult to match with experiments. As suggested by Sommerfelds et al. (2008), the
numerical simulations used for intermediate and dense dispersion should be based on the
Eulerian-Eulerian framework. The Eulerian framework is based on volume-averaged
mass and momentum transport equations for the respective fluid phases. For an
incompressible multiphase system (e.g. oil/water), the phases are assumed to share the
space proportionally to their volume fractions such that:
αd + αc = 1.0 (2.15)
where α is the volume fraction of the respective phases and d, c represent the dispersed
and continuous phases respectively.
The continuity equation is:

 k      k u k   0 (2.16)
t
The momentum transfer equation is:
k

t
 

 k u k   k   k u k u k    kp  k  k g     k k  FM (2.17)
66
From equations 2.16 and 2.17, the subscript k represents a specific phase (i.e. oil or water
in this study), is the mean fluid density, u is the velocity vector given by , u = [u, v, w],
g is the gravity, η is the phase shear stress tensor (η k=-µk(  uk + (  uk)T)) and FM is the
interphase force term.
For turbulent flow, the flow variables can be decomposed into two components: (1) a
steady mean, e.g. U, and (2) a fluctuating component that oscillates around the mean
value, u′(t). This is called the Reynolds decomposition (Hill, 1998; Elseth, 2001). The
flow variables of the turbulent flow can be expressed as follows:
u(t) = U + u′(t); v(t) = V + v′(t); w(t) = W + w′(t); p(t) = P + p′(t) (2.18)
By substituting equation 2.18 into equations 2.16 and 2.17, the effect of turbulence can be
accounted for in the conservation equations.
The continuity equation becomes:

α k     α k U k   0 (2.19)
t
The momentum equation becomes:


k k Uk     kk Uk Uk   kp  k k g    k  FM (2.20)
t
where U is the volume averaged velocity in a turbulent flow, and Τ is the Reynolds stress
tensor given by:
  xx  xy  xz 
 
k    yx  yy  yz 
  zy  zz 
(2.21)
 zx
67
The tensor creates six extra stress terms including three normal stresses
 xx   k  k u' 2  yy   k  k v' 2  zz   k  k w ' 2 (2.22)
and three shear stresses
 xy   yx   k  k u' v'  xz   zx   k  k u' w'  yz   zy   k  k v' w ' (2.23)
These six extra stresses are called Reynolds stresses. The inclusion of these terms in
equation 2.21 requires a turbulence model to predict the Reynolds stresses. The k-ε model,
which will be used in this thesis, is by far one of the most widely used and validated
turbulence models. It is based on the assumption that there is an analogy between the
action of the viscous stresses and the Reynolds stresses on the mean flow.
According to Boussinesq (1877), the Reynolds stresses can be expressed in the following
way:
 U U y  2
 xy  u ' x u ' y   k  t  x     k  k k xy
 x  3
 y x x  (2.24)
1 2
where k  (u '  v' 2  w ' 2 ) is the turbulence kinetic energy per unit mass, µt is known
2
as the turbulent or eddy viscosity. The Kronecker delta, δ xy, becomes 1 if x=y and 0 if
x≠y. The other Reynolds stress terms can be expressed in the same form as Equation 2.24.
FM is a source term that accounts for the overall effect of various body forces – buoyancy
force, lift force, drag & turbulent dispersion force, virtual/added mass effects. Virtual
mass effects can be ignored unless the high frequency fluctuations of the relative velocity
occur. These high frequency fluctuations are generally not resolved in the CFD code and
also tend to be lost in the averaging process (Drew, 1983; Chen et al., 2005).
The conservation equations can be further simplified for the steady-state and
incompressible flow with no mass transfer encountered. In addition, the modeling of drop
68
dynamics and drop size distribution are not included in this thesis to reduce complexity of
the simulations and the required run time. With these assumptions, the continuity
equation becomes:
  α k U k   0 (2.25)
and the momentum transfer equation becomes:

   k k U k U k
 
  k p 

k  k g
 
 

    k  k  Flift ,k  Fint,k (2.26)
 
Changeof momentum Pressureforce Gravitational force Viscousforce Lift force Drag &Turbulent
due to convection dispersionforce
In order to close the set of equations 2.26, appropriate models for the interphase drag, lift
and turbulent dispersion forces are needed.
2.6.2 MODELS FOR SIMULATION OF LIQUID-LIQUID DISPERSED

PIPE FLOW
Turbulence models
In order to achieve a balance between the level of accuracy expected from the CFD
simulations and the resources required to achieve it, turbulence models have been
proposed to simplify the numerical calculations needed to take into account the
turbulence. These models will be discussed with special focus on Ansys CFX which is
the CFD code used in this study.
Standard k-ε model

The k-ε model is a semi-empirical model based on the solution of two separate transport
equations for the mean turbulence kinetic energy, k, and the dissipation of this kinetic
energy, ε. The effect of turbulence is accounted for via the turbulence viscosity term (µt)
without modifying the effect of molecular viscosity (µk) in the conservation equations
(eddy viscosity hypothesis). Molecular viscosity refers to the viscosity effect caused by
the fluids while the turbulence viscosity is an imaginary viscosity effect caused by the
turbulence eddies proposed by Boussinesq (1877). The model assumes that the Reynolds
69
number of the flow is high and the turbulence across the domain is homogeneous. This is
valid for a fully developed turbulent flow in a pipe except at locations very near to the
wall.
The transport equations for k and ε are:
   
( cc k c )     cc U c k c       c t ,c k c   Sk (  ) (2.27)
t       
 Transport of k by convection 

k (

 ), c 
 Net rate of production/destruction of k
Rate of changeof k
Transport of k by diffusion
   
( cc  c )     cc U c  c       c t ,c  c   S (2.28)
t     ,c 
 Transport of  by convection 
  Net rate of production
/destructi
on of 
Rate of changeof 
Transport of  by diffusion
where µt,c is the turbulence viscosity of the continuous phase. Among the terms, there are
several empirical parameters that have been evaluated through data fitting over a wide
range of turbulent flows. The set of constants given by Launder and Spalding (1972) is
the most widely used and is the default set of constants for CFX (see Table 2.3).
Model parameter Suggested value

Cµ 0.09
C1ε 1.44
C2ε 1.92
ζk(ε),c 1.00
ζε,c 1.30
Table 2.3: Coefficients for k-ε turbulence model (Launder and Spalding, 1972)
70
The continuous phase turbulence viscosity can be expressed as follows:
k2
 t ,c   c C  c (2.29)
c
The Sk & Sε terms in equations 2.27 and 2.28 are the sources for the net rate of
production/destruction of k and ε. These two terms can be expressed as follows:
k ()
Sk ()  c (Pc  cc )  cd (2.30)
 
S  c c (C1 Pc  C2cc )  cd (2.31)
kc
In equations 2.30 and 2.31, the additional Tcd term accounts for the extra production and
destruction of turbulence due to the presence of the dispersed phase (i.e. turbulence
modulation).
The k-ε model reduces the number of equations for turbulence from 7 to 2, and
significantly reduces the computational resources required. At the same time, it has
reasonable accuracy and takes into account the interaction between the continuous and
dispersed phase (Liu, 2005; Krepper et al., 2007).
RNG k-ε model

The RNG k-ε model is based on the renormalization group analysis by Yakhot et al.
(1992). The renormalization group is applied to account for the effects of smaller scales
of motion. It has the advantage over the standard k-ε by accounting for the effect of
different turbulence length scale on the turbulent diffusion. The modified form of the
RNG k-ε model leads to different coefficients in the ‘ε’ equation (2.28 and 2.31).
71
Model parameter Suggested value
βRNG for eqn. 2.32 0.012

Cµ,RNG 0.085
C1ε,RNG 1.42-fη
C2ε,RNG 1.68
ζk(ε),c,RNG 0.7179
ζε,c,RNG 0.7179
Table 2.4: Coefficients for k-ε turbulence model (Adapted from Ansys CFX Manual).
For fη term in Table 2.4,
  
1  
f   4.38 
and  
Pk

1   RNG  
3
C  ,RNG 
(2.32)
2.7 INTERPHASE FORCES MODELS
In order to account for the interphase momentum interactions between the continuous and
dispersed phases in a turbulence model, various closure equations for the interphase
forces that contribute to the momentum balance are required. This section will discuss the
various interphase forces.
2.7.1 INTERPHASE DRAG

The interphase drag accounts for the resistance force encountered by one phase due to the
presence of the other phase. This force has always a direction opposite to the relative
velocity and the total drag force exerted on a single spherical drop can be given by:
72
1
FD  C D  c (U c  U d ) 2 A (2.33)
2
where FD, the total drag force, is related to the relative velocity of the dispersed (subscript:
d) and the continuous (subscript: c) phase; A is the projected area of a single drop in the
flow direction; CD is the drag coefficient which depends on the dispersed phase Reynolds
number. In laminar flow, the Stokes’ law is valid and CD is expressed as:
24
CD  if the dispersion Reynolds number, Re d 1 (2.34)
Re d
When the dispersed phase Reynolds number is sufficiently large and the inertial effects
dominate over the viscous effects, the drag coefficient becomes constant.
C D  0.44 if 1000  Re d  2  10 5 (2.35)
In the transitional range of Red (0.1<Red<1000), both inertial and viscous effects are
important and the drag coefficient is a complex function of Re d which needs to be
empirically determined.
Schiller and Naumann (1933) provide an empirical correlation for this transitional range.
CD 
24
Re d

1  0.15 Re 0d.687  (2.36)
In order to account for the total drag force exerted on the dispersed phase by the
continuous phase per unit volume, the sum of the drag forces exerted on all the drops is
taken.
2.7.1.1 DRAG FORCE IN DENSE DISPERSION
With the presence of adjacent drops in a dense dispersion, the drag force on a drop is
significantly different from that on a single drop in a continuous flow. With the change in
dispersed phase fraction, the mixture density also changes accordingly. As a result, a
small difference in the buoyancy between a specific drop and the surrounding mixture
73
Authors Proposed correlation for drag coefficient
Mixture viscosity:
2.5 max ( d  0.4 c ) ( d   c )
m  d 
 1  
 c   max 
where Φd and Φmax are the dispersed phase fraction and

maximum packing factor respectively.
In viscous regime where drops are approximately spherical

(similar to Schiller Naumann correlation):
CD 
24
Re m

1  0.15 Re 0m.687 
c U d  U c d d
where Re m 
m
In distorted fluid drop regime:

Ishii and Zuber
2
(1979) C D (ellipse)  Eo E( d )
3
(1  17.67f ( d ) 6 7
where E d   and
18.67f ( d )
c
f ( d )  1   d 1 2
m
In the regime where churn flow is dominant:

C D (cap )  1   d 2
8
3
Ansys CFX provides an automatic selection criterion for the

appropriate CD
C D  C D (sphere) if C D (sphere)  C D (ellipse)

C D  min( CD(ellipse) if C D (sphere)  C D (ellipse)
Kumar and Hartland
(1985)

C D   0.53 

24 
Re 
 0.73
 1  4.56 d 
C D  C D0 exp k1d    d 2 where k1 and k2 are empirical
k
Rusche and Issa (2000)
coefficient.
Table 2.5: Drag coefficient correlations for dense homogeneously distributed dispersions.
74
occurs. An increase in dispersed phase fraction will increase the mixture viscosity. This
will in turn increase the drag force exerted on the specific drop. In addition, the effect of
drop shape at the presence of adjacent drops will affect the drag force. Some of the
common models used for dense dispersions include the Ishii and Zuber (1979) model
which applies a mixture viscosity approach to calculate the drag force in a dense
dispersion, the Kumar and Hartland (1985) model which is an empirical correlation based
on 998 measurements across a range of Reynolds numbers, the Rusche and Issa (2000)
model which is also an empirical correlation between drag coefficient and dispersed
phase fraction. The correlations of the above models are listed in Table 2.5. In particular,
the Ishii and Zuber (1979) model will be applied in this work as it is the only default
model available for dense homogeneously distributed fluid dispersed flow within Ansys
CFX.
2.7.2 LIFT FORCE
Lift force is exerted on the dispersed drops when the flow field is non-uniform. This
generates a velocity gradient within the continuous phase leading to a force acting onto
the drops perpendicular to the direction of the relative motion of the two phases. In Ansys
CFX, a model for the shear-induced lift force acting on the dispersed phase is used based
on the vorticity of the continuous phase (Equation 2.37)
Flift ,c  Flift ,d  dcCL (Ud  Uc )  curl Uc (2.37)
where Φd is the dispersed phase fraction. Depending on the sign of the lift coefficient, CL,
as well as the direction of the relative velocity, the lift force can significantly affect the
radial phase distribution.
2.7.3 TURBULENT DISPERSION FORCE
The turbulent dispersion force accounts for the effect of the turbulent eddies in the
continuous phase acting on the drops and leads to the homogenization of the dispersion.
75
Two models are included in Ansys CFX – the Favre averaged drag model and the Lopez
de Bertodano model (1991).
2.7.3.1 FAVRE AVERAGED DRAG MODEL
Favre averaging is an averaging operation for the turbulent terms using a density
weighted average of the turbulence terms. This is significant mainly for compressible
flow. In the Favre averaged drag model, an additional drag term is added into the
momentum equation by expressing the time averaged drag in terms of Favre averaged
velocities. The model is implemented in Ansys CFX as follows:
 tc   d  c 
FTD ,c  FTD ,d  CTD CD,c    (2.38)
 tc   d  c 
where CD,c is the coefficient of the interphase drag force, tc is the kinetic eddy viscosity,
ζtc is the turbulent Schmidt number for continuous phase volume fraction (default value:
0.9). CTD is the turbulent dispersion coefficient that is defined by the user where the
default value for CFX as 1.0.
2.7.3.2 LOPEZ DE BERTODANO MODEL

The Lopez de Bertodano model (1991) is one of the first models developed for the
turbulent dispersion force. It aims to physically account for the turbulent diffusion based
on the fluctuating components of the turbulent forces acting on the dispersed drops. The
model can be presented as follows:
FTD,c  FTD,d  CTD ρc k cΦc (2.39)
Where kc is the turbulent kinetic energy and Ñ Φc is the concentration gradient of the
continuous phase. CTD values between 0.1 and 0.5 have been used successfully for bubbly
flow with bubble diameters in the order of a few millimeters. However, much higher
values have been required in some situations (Lopez de Bertodano (1998) and Moraga et
al. (2003)).
76
2.8 SUMMARY OF LITERATURE REVIEW
Previous works on two-phase oil-water flows and particularly the phase inversion
phenomenon have been reviewed. Flow regime development and phase inversion were
found to depend on the properties of the fluids, pipe configuration, and operational
conditions in a pipeline. Phase inversion has been previously studied mainly in stirred
vessels but also increasingly on pipe flow. Among the pipe flow investigations, changes
in pressure drop have been used as an indication of phase inversion occurrence. However,
it is also known that these changes differ across different fluid systems. In addition, the
changes in pressure drop have not been clearly associated with the corresponding changes
in phase continuity. Inversion has been reported to occur across a range of phase fraction
before completion due to the presence of local phase inversion at certain locations within
the system. Thus, experimental investigations to understand the changes in phase
continuity and distribution (i.e. changes in flow regime) corresponding to that in pressure
drop during inversion will be conducted with results presented in Chapter 4. Successful
outcome from this investigation will also allow better prediction of the changes in
pressure drops as flow regime changes during inversion. A prediction method for these
changes in pressure drop is presented in Chapter 7.
From the literatures reviewed, various models have been developed based on mechanisms
including the dynamic process of drop coalescence and break-up, minimization/equal
system/interfacial energy of the mixture and zero shear stresses at the liquid-liquid
interface. While there is no definite mechanism for the occurrence of phase inversion,
these models provide important information about phase inversion. For example,
interfacial tension is found to be an important factor on the Weber number and phase
inversion in many of the correlations. However, the effects of interfacial tension on phase
inversion have not been previously established and thus will be studied in this work
(Chapter 5).
Momentum balance is regarded as the underlying principle behind the mechanism of zero
shear stresses at the interface. Pressure drop across a pipe is the direct effect of the shear
stresses at the interfaces in the system and these shear stresses in turn depend on the fluid
77
viscosities. A prediction model on phase inversion occurrence will be developed based on
this relation between viscosity and pressure drop and is presented in Chapter 6.
Computational Fluid Dynamics (CFD) simulation can offer some insight on the behavior
of dispersed systems and will be used in this work. A review on numerical simulations
using CFD has been presented. In particular, the k-ε model has been proposed for use in
subsequent chapter due to its good estimation of the system without a high demand on
computational resources. This is especially important at high concentration of the
dispersed phase. Closure equations for the interphase interactions are also presented.
Some of the more important interactive forces include the drag force, the turbulent
dispersion force and the lift force. The effect of these forces on the spatial distribution of
the phases in dispersed systems will be investigated in Chapter 8.
78
Chapter 3: Facility and Instrumentation
3.1 OVERVIEW
This chapter will introduce the flow facility and instrumentation used for the
experimental investigations of phase inversion during oil-water pipe flow. The
experimental facility has been constructed in the Department of Chemical Engineering,
UCL. There are two test sections made of steel and acrylic respectively and two types of
inlets, one that ensures minimum mixing of the fluid and the other creates a dispersion at
the beginning of the test section. Various measuring probes are used to measure phase
continuity and distribution, pressure drop and drop size distribution. This allows detailed
investigations of the flow parameters as the system approaches phase inversion.
3.2 TEST FLUIDS
The fluids used for the experimental investigations of the phase inversion phenomenon
are Exxsol D140 and tap water. Table 3.1 presents the physical properties of the two
fluids.
Exxsol D140 Tap Water

Density (kg/m3) 828 1000
o
Viscosity (mPa.s) @25 C 5.5 1.0
Interfacial Tension (mN/m) @ 25oC 37.04
Table 3.1: Physical properties of working fluids
The density of the oil phase is based on the data provided by the supplier (Exxon
Chemicals). The oil viscosity was measured using a Contraves 155 Rheometer over a
range of temperature. The oil-water interfacial tension was measured using the drop
volume method in a Krüss Drop Shape Analysis System (DSA100).
79
3.3 EXPERIMENTAL FLOW FACILITY
The schematic diagram of the flow facility used in the experiments is shown in Figure
3.1. It consists of three main sections – fluid storage, test pipeline and fluid separation
units. The test pipeline has been designed to accommodate various instrumentations and
inlet to investigate the development of dispersed flow and the occurrence of phase
inversion. These will be discussed in detail below.
Fluid Storage Section

This section consists of two fibre-glass storage tanks with a capacity of 880 liters each
for oil and water respectively. There are two acrylic baffles in each tank to prevent
vortices that may allow air to entrain in the fluids and be transported into the test section.
The baffles are also positioned in such a way to allow extra time for separation of the
fluids in case one phase enters the storage tank of the other phase. The tanks are similarly
connected to the centrifugal pumps on one end and to the separator on the other end.
Fresh fluids are manually put into the tanks while the fluid level in each tank is kept at
least ¾ of the tank height. Fluids can be recycled from the separator. In addition, fluid
from the recycle line of the pump is returned into the tank to regulate the outflow
through the pumps into the test pipeline. At high flowrates, water may enter into the oil
storage tank. In this case, the oil tank is drained into a spare separation tank from which
the oil is recycled bank to its storage tank and water is discarded. Fresh water is
consistently used if possible to ensure clean water stream is pumped into the pipeline
during experiments. Fresh water is also used to maintain the temperature of the fluid
mixture in the test section at approximately 20oC (+/- 5oC).
80
Figure 3.1: Schematic of the pilot scale facility with adjustable sections in the test
pipeline.
81
The centrifugal pumps (Ingresoll-Dresser CPX200) are capable of generating a flow rate
of 240l/min. A recycle line back to the storage tank is used to regulate the flow rate of
each fluid in the flow meter. Good control over the regulation of the flow rate through
the pumps and the recycle streams is crucial in minimizing fluctuation of flow to the test
pipe section. This also avoids exceeding the designed capacity of the pumps and cause
overheating.
The flow rates of the respective fluids are monitored by two variable area flow meters
(ABB Instrumentation 10A5400) with maximum error of 1% of the full scale. The flow
meters measure a range between 20 – 240 l/min and are connected to a PC for data
logging.
Test Pipe Section

The test pipe has a total length of 16m with an internal diameter of 38mm and is split
into the front and back legs connected by an U-bend with a 150mm diameter curvature.
Two test pipes have been constructed from stainless steel and acrylic respectively. The
entire pipeline consists of sections of 1m and 2m long. This allows the various
instruments that are fitted in between two pipe sections to be installed at different axial
locations along the test pipe. The sections are either joined by tri-clamps (in the stainless
steel pipeline) or flanges (in the acrylic pipeline). These instruments are however fitted
away from the U-bend at the end of the first leg of the pipeline to avoid any influence on
the flow distribution. The probes can also be located at 7m (~184D) from the inlet when
the quick closing valves are not required.
82
Separator
Test Pipeline
Storage Tank
Figure 3.2: Photograph of the pilot scale facility in stainless steel.
Oil
Water
Figure 3.3: Stainless steel Y-inlet where the oil and water phases are joined at the end.
83
As observed in both Figures 3.2 and 3.3, oil and water are pumped through the inlet to
the test section. Two types of inlets were used. A Y-inlet (Figure 3.3 and 3.4) was used
to ensure that the fluids join with little mixing. The oil stream flows from the upper inlet
pipe and the water stream flows from the lower inlet pipe. The two pipes join with an
angle of approximately 15o. In addition in the acrylic inlet section, a splitting plate is
added where the two pipes join to further ensure minimum mixing of the two liquids (see
Figure 3.4).
Oil
Split Plate
Water
Figure 3.4: Acrylic Y-inlet with a split plate to enhance stratification of the oil and water.
A dispersed inlet (Figure 3.5) was also constructed to create a dispersion from the
beginning of the test section and influence the type of dispersion formed. The inlet is
made from a steel pipe with a total of 1056 nozzles with heights 2mm and 4mm around
its periphery. Nozzles of the two heights are staggered so that the fluid jets from the
nozzles have more space to develop and do not coalesce with each other.
84
Continuous
Phase
Dispersed
Phase
(a) (b)
Figure 3.5: Dispersion inlets with staggered nozzles. (a) The dispersed phase flows
within the stainless steel pipe while the continuous phase flows in the acrylic pipe. (b)
Dispersed phase is ejected from the nozzles.
Separation Section
A horizontal two phase separator is installed at the end of the test section (Figure 3.2)
with 800 litre in capacity. The separator contains a KnitMesh (DC 9201/SS/PPL)
coalescer to aid the separation process. The coalescer uses a composite of two materials
with different surface free energies (metallic and plastic). The junction effect from the
combination of the two materials increases the rate of coalescence when the droplets are
trapped and move towards the junction points of the coalescer as shown in Figure 3.6.
The separator is also used, at the end of each experimental session, for a more thorough
separation by gravity. This will generally take about 48 hours before the next
experimental run before the fluids are recycled into the storage tanks. In occasions when
there is water entrained in the oil storage tank, further separation will take place within
the tank and excess water will be removed.
85
Figure 3.6: Junction effect of the coalescer that enhances separation. (Information from
manufacturer - KnitMesh Technologies).
Instrumentation
(a) Visual Observation
Visual observation of the flow development can be captured throughout the acrylic
pipeline or through a viewbox in the steel pipeline. Visual imaging is facilitated with
a high speed video camera (Kodak HS 4540 MX) which has a capturing rate of 4500
fps and a total recording time of 1.8 second. A monitor is connected for real time
visual observation. Figure 3.7 shows the setup of the video camera. The camera is
connected to the PC for data capture and storage. Video images can be saved and
individual frames extracted for further analysis. Background lighting (Spot light unit)
is used to enhance the contrast between the oil and water phases and improve the
quality of the images.
86
Figure 3.7: Setup of the high speed video imaging unit, Kodak HS 4540 MX.
(b) Fluid hold-up

Average volume fraction is measured using two quick closing valves (QCVs). Two
ball valves are fitted on both ends of a transparent pipe section of 1m in length to trap
fluid mixture with a volume of 1.13 litre (Figure 3.8). The pipe section is located at
7m (~184D) from the inlet (and also the last section before the U-bend in the setup).
Once the pipe section is closed, the mixture can be drained into a measuring cylinder
for separation. The respective volume fraction of the two phases can then be
identified. A manual hand pump is used to minimize any retention of the oil phase in
the pipe. The measurements are repeated several times with 2% deviation across
measurements. In the experiments, the two valves are closed simultaneously with the
pumps. The trapped mixture is drained in a measuring cylinder where the volume of
each phase is identified and recorded.
87
Quick Closing Valves
View Box
Figure 3.8: Section of pipe with quick closing valves in the 38mm stainless steel
pipeline. A viewbox is included to aid visual observation and video capturing.
(c) Instrumentation for phase continuity measurement
1) Conductivity Ring Probe

The ring probe consists of two metallic rings embedded on the pipe perimeter,
flush with the wall, and in contact with the fluid mixture. Each ring is 4mm wide
and the rings are 38mm apart (Figure 3.9). The conductivity ring probe is
generally located at about 7.0m (~184D) from the inlet where the flow is
considered as fully developed. The ring probe is normally paired with the
conductivity wire probe to measure the conductivity at the test section and
provide information on phase continuity and distribution. High conductivity
values indicate that a conductive water continuous mixture is in contact with the
probes, while low values denote an oil continuous mixture.
88
Figure 3.9: Conductivity ring probe for phase continuity measurement at the pipe
periphery.
2) Conductivity Wire Probe

Each wire probe consists of a pair of bended copper wires with a diameter of
0.5mm (Figure 3.10). The wires are 10mm apart and this gap is assumed to be
greater than the largest droplet of the dispersed phase. This will ensure that any
change in conductivity signal is due to a change of the continuous phase. It can
detect phase continuity at different locations in a pipe cross section using a
traversing mechanism. The wires are bent (Figure 3.10b) in L-shape at opposite
directions and only the horizontal portions of the wires are exposed to the flow.
This provides a large contact surface at particular height in the pipe cross section.
During experiments, the ring probe and the conductivity wire probe are used primarily to
detect the phase continuity at the pipe periphery and the centre of the pipe cross section
respectively unless otherwise stated. These locations are found to be critical during phase
inversion from previous experiments (i.e Hu, 2005, Ioannou, 2006). The electrical
resistance tomography will also be used to supplement on the phase continuity
investigation on other locations (e.g. top of pipe cross section). The application of the
tomographic system will be discussed in the next section.
89
(a)
(b)
Figure 3.10: Local conductive wire probe for phase continuity measurement at the
localized position. (a) Photo of wire probe (b) Schematic of the interior (side view) of the
wire probe.
90
3) Electrical resistance tomography (ERT)
Electrical resistance tomography provides information on the distribution of the
phases in a mixture through a series of electrodes arranged at regular intervals
within the boundary of the vessel investigated. A commercial ERT system
(Model: M3000 by Industrial Tomography Systems, ITS, plc) is used in the
current work. The tomographic sensor consists of 16 electrodes equally
distributed around the periphery of an acrylic pipe (Figure 3.11a). Each electrode
is made of a 4mm diameter circular stainless steel piece embedded on the acrylic
pipe wall and in contact with the mixture inside the pipe. Care is taken to ensure
that each electrode is flush with the interior pipe wall. Thus, ERT is considered to
be non-intrusive. The ERT can provide a signal as long as there are at least 2
electrodes in contact with the conductive phase (i.e. water) where one electrode
acts as a transmitter and the other as a receiver. The tomographic sensor provides
a resolution of 316 pixels over the 38mm pipe cross section with each pixel
representing approximately 2mm x 2mm.
Electrical signal is transmitted from a transmitting electrode and received by the

remaining 15 electrodes. The sample data transmission is repeated in sequence
until all 16 electrodes have acted as the transmitting electrode. All the received
data will be processed through a reconstruction algorithm to generate the phase
distribution through a cross-sectional image. The sampling frequency in this
study is set to be 10 images per second. The algorithm used in this study is the
Linear Back Projection (LBP) which allows quicker and less computational
resource intensive for the reconstruction. This is important in capturing the rapid
changes during phase inversion despite the compromise on the resolution of each
image. A software (ITS toolsuite) is used to present the reconstructed images
(Figure 3.12) for visualization and data analysis. Localised conductivity value at
each pixel across the pipe cross section can also be exported to Excel for further
data processing.
91
(a)
(b)
Figure 3.11: (a) Stainless steel electrodes embedded across the pipe periphery. (b)
Experimental setup of the ERT system with a PC for data processing.
92
Figure 3.12: Sample ERT reconstructed image generated from ITS toolsuite. The colour
bar denotes the range of raw conductivity data with respective to the reference frame.
From the phase distribution data, the fraction of each phase in a pipe cross section can be
found. This can be compared with the results from QCVs and an average difference of
5.6% is observed between the two methods. Figure 3.13 presents the results of the
comparison. An average deviation of 5.6% from the diagonal line can be found from the
ERT measurement while an average deviation of 3.6% from the QCV measurement. It
can thus be assumed that the mixture flow across the wide range of phase fraction to
have no slip (i.e. the input volume fractions of the oil and water equals to the in-situ
phase volume fractions). The ratio between the volume fractions is defined as the slip
ratio (S) presented in Equation 3.1 and the slip ratio is 1 when there is no slip.
 /
S o w (3.1)
o /  w
where βo and βw are the input volume fractions of the oil and the water phase respectively,
and εo and εw are the in-situ volume fractions of the oil and water phase respectively.
93
100
80
In-Situ Oil Fraction [%]
60
40
20
Quick Closing Valve
Electrical resistance tomography
0
0 10 20 30 40 50 60 70 80 90 100
Input Oil Fraction [%]
Figure 3.13: In-situ liquid hold-up comparison between electrical resistance tomography
(ERT) and quick closing valves across a range of input phase fractions at a mixture flow
velocity of 3m/s.
(d) Drop velocity and drop size distribution

The measurement of drop velocity and drop size distribution is made using the dual
impedance probe developed by Lovick (2004) at University College London (see
Figure 3.14). The details on the development and operation principles can be viewed
in his thesis (Lovick, 2004).
As shown in Figure 3.14, the dual impedance probe consists of two coaxial wires
which can work independently as sensors detecting the contacting phase at each of
the wire tips. Each tip is tapered with insulation in between the coaxial wires as
shown in Figure 3.14(b). The coaxial wires are placed at the same housing as the
conductivity wire probe presented in Section 3.3 Figure 3.10. The probe tips can thus
be moved to measure at different locations in the pipe cross section. The distance
94
between the two wires is again 10mm and this is assumed that no drop is larger than
this distance within the experimental flow conditions used in this work.
(a)
(b)
Figure 3.14: (a) Photograph of the dual impedance probe setup. (b) Probe tip
configuration (Lovick, 2004).
Measurements using the dual impedance probe are conducted at 7m from the test
section inlet. The impedance wires are set at the same height inside the pipe
perpendicular to the flow direction. It is assumed that drops that pass through the
first probe will also pass through the second probe. A single frequency regulator
controls the frequencies of the two probes to ensure that they are measuring at the
same cycle. The sampling frequency is set to 45kHz. At this high frequency, each
sampling cycle has a time span of 22μs, and this is capable of measuring a minimum
95
drop size of 66μm at 3m/s mixture velocity or 88μm at 4m/s mixture velocity. At
this high frequency, it is expected that each dispersed drop will cross each probe tip
over a succession of sampling cycles (see Lovick, 2004; Hu, 2005; Ioannou, 2006).
120,000 data points are captured at each location. An example of the normalized
signal data from the dual impedance probe is presented in Figure 3.15.
1.0
Probe 1
0.8 Probe 2
Normalised Signal [-]
0.6
0.4
0.2
0.0
0 1 2 3 4 5
Time [ms]
Figure 3.15: A sample set of normalized signal from the two impedance wires at a
mixture velocity of 3m/s. Dispersed phase will be in contact with probe 1 before it
contacts with probe 2.
The time delay (see highest peak of Figure 3.16) for the drops to pass through the
two probes is found by cross correlating the signals of the two probes with a fixed
distance of 10mm. By combining with the known distance between the probes, the
drop velocity can be found (Lovick, 2004). The chord lengths of the measured drops
can be estimated from the signal of either probe by multiplying the drop velocity
with the time duration of each drop passage. The duration of each drop passage is
obtained by converting the raw signals of the probe into square waves based on the
level and slope threshold values (Lovick, 2004 and Hu, 2005). It should also be
noted that measurements of chord length through this method tend to be biased
96
towards the larger drops as they are more likely to pass through the probe tips than
the smaller drops.
0.35
0.30 Time Delay
0.25
Frequency [-]
0.20
0.15
0.10
0.05
0.00
-0.05
0.00 2.00 4.00 6.00 8.00 10.00
Time Delay [ms]
Figure 3.16: A sample frequency plot of the time delay between the two impedance
wires. The highest peak denotes the time delay where most dispersed drop take to cross
the two wires.
Drop size distribution is then estimated from the chord length measurements. An
algorithm by Hu et al. (2006) provides the relationship between the chord length
distribution (CLD) and drop size distribution (DSD). In order to apply this algorithm,
the measured drops are assumed to be spherical. Sauter mean diameter (d32) can then
be calculated from the drop size distribution.
(e) Phase distribution measurements

Phase distribution measurements can be made using the same dual impedance probe
as described in the previous section. Only one impedance wire is required (generally
the first one to avoid any obstruction) for the measurement. Local phase distribution
measurements are made across 20 locations at 2mm intervals vertically across the
pipe cross section and diagonally at 45o and 10 locations horizontally. Phase
distribution is assumed to be mirrored at the vertical plane and thus the phase
97
distribution data at the 135o plane is the same as the 45o plane. By interpolating the
local measurements over the whole pipe cross section, the in-situ time-averaged
volume fraction of the two phases can be found. Phase distribution contour plots can
also be created using Matlab and compared with the tomograms generated by the
ERT system. Table 3.2 presents the comparison between the impedance phase
distribution contour plots and the ERT tomograms.
Input
Phase 80% Water 54% Water 40% Water
Fraction
Impedance
phase
distribution
contour plot
ERT
tomogram
Table 3.2: Comparison of phase distribution contour plot using single impedance probe
(red for oil and blue for water) and ERT tomogram (green for oil and blue for water) at
different input phase fractions in a 3m/s oil/water mixture flow.
From the table, it can be observed that the phase distribution measured using both
methods produce similar phase distribution diagrams. However, the tedious process
of generating a single contour plot requires 30 minutes of measurement and
processing time while an ERT tomogram requires less than a second for each
tomogram. The ERT system will thus be able to provide more instantaneous
information about the phase distribution and this will be done so in subsequent
studies presented in this thesis.
98
(f) Pressure gradient measurements.
Pressure gradient is measured by a FP2000 wet/wet differential pressure transducer
(RDP Electronics Ltd). The pressure transducer is connected to the test section via
nylon tubing and quick connect couplings. The first pressure measuring port is
located on the test section at 3.8m (~100D) from the inlet and the second port is
located 1.5m apart. The male adaptor of the quick connect couplings is connected to
the port while the female adaptor is fitted to the nylon tubing as shown in Figure
3.17. These quick connect couplings offer the flexibility to change the location of
pressure gradient measurements. During connection and disconnection, the
couplings are automatically shut off to avoid any air entrainment or leakage which
would affect the pressure gradient measurements. The dimensions of the ports were
carefully designed with the bore diameter of the port opening at the test section to be
no more than 1/8th of the pipe diameter (Perry, 1997). In addition, the pressure
transducer is connected to a transducer conditioner (type E308) which is calibrated
for pressure gradient data logging. The transducer conditioner has a capability of
measuring a maximum pressure of 22kPa and an accuracy of 0.25% of the full scale.
(a) (b)
Figure 3.17: (a) Pressure measuring port at stainless steel pipeline. Similar ports are also
found in the acrylic pipeline (b) Differential pressure transducer and control valves.
99
3.4 SUMMARY
In this chapter, the experimental facilities used for the experimental investigation of
phase inversion have been discussed. Two test sections were made from stainless steel
and acrylic, with 38mm I.D. had been used, and two inlets, a Y-junction and a dispersed
inlet, were developed to study their effect on flow patterns and phase inversion.
A number of instruments were used to study the phase continuity, drop size distribution,
average holdup and pressure drop. From the various conductivity probes used to detect
phase continuity in a pipe cross section, electrical resistance tomography provides faster
measurements. However, it cannot detect thin water layers that do not cover at least two
sensor electrode and also, in the case of an annular flow with oil in the annulus, it cannot
give measurements on phase continuity in the pipe core. A dual impedance probe will
also be used to investigate drop size distributions under different conditions. The probe
can also provide phase distribution data which is comparable with the ERT results.
Nonetheless, ERT has the advantage of quick measurement and will be used for the
purpose of accounting the phase distribution in the pipe cross section in experimental
studies.
100
Chapter 4: The occurrence of phase inversion
4.1 OVERVIEW
Phase inversion has been investigated in horizontal pipelines (in stainless steel and
acrylic) for mixture velocities between 3.0m/s and 4.0m/s where dispersed flow was
observed in previous investigations (Hu, 2005; Ioannou, 2006). Investigations have been
conducted for two inversion routes by (1) starting from a water continuous dispersion or
pure water phase if possible and increasing the oil input fraction at a constant mixture
velocity (o/w → w/o) and (2) starting from an oil continuous dispersion or pure oil phase
if possible (w/o → o/w) and increasing the water fraction at a constant mixture velocity.
In Section 4.2, a brief description of the experimental procedure is discussed. The applied
procedure is similar to the direct experiments by Piela et al. (2008) where specific phase
fractions of oil and water are introduced in the pipe at a constant mixture velocity to
allow for flow development along the pipeline. Measurements are conducted at the end of
the pipeline where the flow is developed.
Phase inversion is first observed through high speed images and the visual observations
are summarized in Section 4.3. Based on the visual observations, the various conductivity
probes are positioned and the change in phase continuity during inversion is discussed in
Section 4.4. Conductivity ring and wire probes detect the phase continuity of the
contacting phase at the pipe periphery and the centre of the pipe respectively. Electrical
resistance tomography is also used to support the investigation by identifying the phase
fraction at other locations as well as to provide visual images on how the phases are
distributed during phase inversion. The use of the various probes is then applied to
investigate the effect of inversion route and mixture velocities on the development of
phase inversion (Section 4.5 and 4.6).
101
The associated changes in pressure gradient are presented in Section 4.7 whereby a
change in phase distribution can have an impact on the corresponding pressure gradient.
Similar changes effected by the inversion route or mixture velocity are also discussed in
the section.
As phase inversion leads to significant change in drop size, Section 4.8 aims to
investigate the drop size distribution of the dispersed phase as well as a mechanism for
phase inversion based on the dynamic balance between drop coalescence and drop break-
up.
Lastly, similar investigation has been conducted in a dispersed inlet. The effect of the
dispersed inlet is presented in Section 4.9. It is observed in the section that the inversion
process is significantly different from the split inlet where the two fluid phases are
allowed to mix along the pipeline. With the outcome from this chapter, the results will
form the basis for further investigations in the investigation on the effect of interfacial
tension on phase inversion in Chapter 5 as well as the data set for comparison with
mechanistic models and CFD simulations in Chapter 7 and 8 respectively.
4.2 EXPERIMENTAL PROCEDURE
The experiments for the investigation of phase inversion occurrence were conducted in
the pilot scale facility discussed in Chapter 3 with a 38 mm ID test section. An acrylic
test section was made to allow visual observations of the flow along the pipeline
especially at the initial length when the flow pattern is still developing. A modified inlet
with a splitting plate is used to minimise any turbulent mixing caused by the Y-junction
inlet. Thus, any mixing along the pipe is assumed to be due to the turbulent flow. High
speed images are taken at approximately 7m (~184D) downstream the inlet to visually
observe the change in phase distribution as the dispersed phase fraction is increased.
Conductivity measurements are also taken to determine phase continuity (See Section
4.4). The conductivity ring probe, mounted on the pipe perimeter, indicates which phase
is in contact with the pipe wall. A conductivity wire probe, that can transverse along a
102
pipe diameter, is used to detect the local phase continuity; the probe tip is set at the centre
of the pipe during the current experiments. The probes are located at 7m from the inlet.
An electrical resistance tomographic (ERT) system is also used to complement the
conductivity probe measurements. In the results reported below, the ERT system is
primarily used to detect changes in phase continuity near the top of the pipe. The ERT
system is located at approximately the same position as the conductivity probes. Pressure
drop is recorded via a pressure transducer which is connected with two measuring ports
in the pipe, 1.5m apart, with the first port located at 3.8m (~100D) downstream the inlet.
The mixture velocity varies between 3.0 and 4.0 m/s. Previous work showed that the flow
is fully dispersed at this velocity range (Ioannou, 2006). Exxsol D140 and tap water are
used as the oil and water phase respectively with their properties shown in Table 3.1.
Two phase inversion routes are followed (1) starting from an oil-in-water dispersion, or
single phase water if possible (2) starting from a water-in-oil dispersion, or single phase
oil if possible. In both cases, the initial dispersed phase fraction is gradually increased
while the continuous phase fraction is reduced until phase inversion is reached and
beyond while maintaining a constant mixture velocity. This approach is similar to the
‘direct’ experiments conducted by Piela et al. (2008). The intervals in the input phase
fractions for measurements are smaller close to the inversion point (i.e. 20% - 60% water
fraction) than away from it (ie. below 20% and beyond 60% water fraction). At each
phase fraction in a particular mixture velocity (i.e. a single data point), the flow is
allowed to establish for approximately 5 minutes before any measurement is taken.
4.3 VISUAL OBSERVATIONS OF PHASE INVERSION
The use of the high speed camera provides the first indication on how the two phases are
distributed as they flow together along the test section. These images facilitate the
positioning of the various probes for detecting changes in phase continuity. Figure 4.1
shows images from the stainless steel pipe for 3m/s mixture velocity starting with high
water fractions where the mixture is water continuous and gradually reducing the water
fraction until phase inversion occurs and an oil continuous dispersion is obtained
(Camera model: Vision Research Phantom V5.1). The oil phase tends to appear white in
103
the images and the water phase appears transparent. While the oil drops in an oil-in-water
dispersion reflects light from the foreground lighting causing a sparkling appearance, a
water-in-oil dispersion tends to be dull white.
(60%) (53%) (40%)
(33%) (20%) (13%)
Figure 4.1: Visual images are taken in a stainless steel pipe using an high speed camera
(Model: Vision Research Phantom V5.1) at 7m away from the inlet. Mixture velocity is
kept constant at 3m/s. The percentage in the bracket represents the input water fraction
which the image is taken.
At 60% input water fraction, an oil-in-water dispersion can be seen with the oil drops
evenly distributed across the pipe. As the water fraction is reduced to 53%, the oil drops
104
tend to move towards the centre of the pipe (i.e. milky white region). The region close to
the bottom pipe wall appears to remain dominated by the water phase with oil drops
flowing within. Further reduction in water fraction to 40% and 33% shows dense
dispersions especially at the upper part of the pipe due to buoyancy of the oil drops.
Water continuous dispersion can only be identified visually near the bottom pipe wall. At
about 20% of water fraction and beyond, the whole pipe is completely dominated by the
oil continuous dispersion. With an oil continuous dispersion, water drops cannot be easily
seen.
Similar experiments were also carried out in the acrylic test section. The resolution of the
images captured in this set of experiments is lower than in the steel pipe because a high
speed camera (Model: Kodak HS 4540) with lower resolution and weaker lighting units
was used. The difference in appearance between the oil and water phase was still possible
to detect.
Figure 4.2 shows the images taken using a cold light unit that illuminates the image from
the back. Since water is transparent, light will be able to penetrate through an oil-in-water
dispersion (with an appearance of a light grey image). No light can get through a water-
in-oil dispersion leading to the appearance of an opaque black image. For example, the
oil-in-water dispersion at 60% water fraction has a bright appearance. As more oil is
added (34%, 40% and 54% water fraction), the upper part of the pipe becomes opaque as
oil drops concentrate here due to buoyancy. The lower region remains bright and
complex oil structures can clearly be observed flowing within the continuous water
phase. This water continuous region becomes thinner and closer to the bottom pipe wall
as more oil is added. From approximately 20% input water fraction and below, the entire
pipe is completely opaque indicating that oil is the continuous phase throughout.
105
Homogeneous O/W dispersion Complex structure formed
(60%) (54%)
Complex structure formed Complex structure formed
(40%) (34%)
Homogeneous W/O Dispersion Homogeneous W/O Dispersion
(20%) (14%)
Figure 4.2: Visual images are taken in an acrylic pipe using an high speed camera
(Model: Kodak HS 4540) at 7m away from the inlet. Mixture velocity is kept constant at
3m/s. The percentage in the bracket represents the input water fraction at which the
image is taken. The arrows present samples of complex structures observed in images.
106
In both pipes, it can be seen that oil will tend to occupy the upper part of the pipe due to
buoyancy as the oil fraction is increased. Phase inversion seems to occur at this part of
the pipe first. As more oil is added, phase inversion is completed when the water layer
near the bottom is completely consumed. The pipe material does not appear to have a
significant effect on how phase inversion occurs. The phase inversion process was also
studied in detail with the use of conductivity probes and ERT. The results will be
discussed in the following section.
4.4 DETECTION OF PHASE INVERSION OCCURRENCE
The investigation on the occurrence of phase inversion is conducted in both the stainless
steel and acrylic pipes. With the acrylic configuration, the Y-junction inlet is modified
with a splitting plate to minimise any turbulent mixing at the inlet. The two phases enter
the test pipe as stratified flow. The detection of phase inversion is investigated with
conductivity probes which provide information on the phase continuity at specific
locations. The spatial difference in changes of phase continuity across the pipe cross
section determines the phase inversion transition. In addition, the availability of an
electrical resistance tomographic system in the acrylic pipe test section provides
information about the phase distribution which can substantiate on the visual
observations made in the previous section. These conductivity measurements are
recorded at 7m downstream from the inlet where previous investigations have found that
the flow is fully developed (Ioannou, 2006). The results of measurements for phase
inversion occurrence will be presented at a mixture velocity of 3m/s but similar approach
of detection is also conducted at a higher velocity (i.e. 4m/s) which will be presented in a
later section on the effect of increasing the mixture velocity.
4.4.1 IN-SITU PHASE DISTRIBUTION MEASUREMENTS
As visual images do not provide clear information on the phase distribution especially in
the pipe core, an electrical resistance tomographic (ERT) system is used to reconstruct
the phase distribution across pipe cross section. For each mixture velocity, the mixture
107
flow is adjusted to the required input phase fraction and left for approximately 5 min to
run before any recordings are made. While the ERT system receives signals from the
sensors continuously, selected raw data are used for analysis. After the initial 5 min of
stabilisation time, the first 50 – 60 frames are ignored. The subsequent 50 frames are
captured which span a duration of approximately 30 sec of run time. This procedure is
repeated for all the input phase fractions used in a particular mixture velocity.
Type of dispersion Tomographic images
Oil-in-water
dispersion
(100%) (80%) (60%)
Phase inversion
transition
(54%) (40%) (26%)
Water-in-oil
dispersion
(18%) (10%) (0%)
Figure 4.3: Phase distribution in an acrylic pipe cross section during the transition from a
water continuous to an oil continuous mixture at a mixture velocity of 3m/s. The
percentages in brackets represent the input water fraction.
108
The tomograms during the phase inversion transition from an oil-in-water dispersion to a
water-in-oil dispersion are presented in Figure 4.3. The water phase is represented in blue
while the oil phase is represented in green. In the region that is reported as water
continuous dispersion (i.e. between 60% - 100% water fraction), it can be observed that
the oil drops accumulate near the pipe centre. During the phase inversion transition (i.e.
between 20% - 54%), part of the pipe cross section turns to be oil continuous as reflected
by the near-zero normalized conductivity in localized positions (e.g. pipe centre or near
top wall). With further increase in oil fraction, the wall contacting area by the water
phase will be reduced. Phase inversion is completed when the entire tomogram turns
green (i.e. water fraction below 20%). Due to the limitation of electrical resistance
tomography on non-conductive continuous phase, the distribution of the dispersed water
phase cannot be shown.
4.4.2 CHANGES IN PHASE CONTINUITY (PARTIAL PHASE

INVERSION)
Based on both the visual images and tomograms, the oil drops tends to move towards the
core of the pipe as the oil phase fraction is increased at a constant mixture velocity. As oil
starts to accumulate at this location, it seems to be a likely location where a local phase
inversion may occur (i.e. if starting from an oil-in-water dispersion). As such, an
adjustable wire probe is placed at the centre of the pipe to detect the phase continuity at
this location. The measurement of local phase continuity will allow better accountability
on whether the pipe core continues to be water continuous with a dense concentration of
oil drops or an inverted oil continuous dispersion. Similarly, a conductivity ring probe is
used to track on the continuity of the contacting phase with the pipe wall as previous
observations have indicated that a water layer remains until the water fraction is greatly
reduced (i.e. approximately 20%). The ERT system (where available with the acrylic test
section) can also provide local conductivity information at near to the top wall. This
allows the differentiation between a dual continuous flow from an annular flow.
109
The combined conductivity measurements, at different locations, can be used to map out
the change in phase continuity during the phase inversion transition. Figure 4.4 presents
the combined data in an acrylic pipe at 3 specific locations – near the top wall (4mm from
the top), centre of pipe (19mm from the top) and around the pipe periphery. Averaged
values over several experimental runs for each probe are presented here.
Expt
Transition
O/W
W/O
Δ 20% × 40%
○ 54%
Figure 4.4: Normalized conductivity data of the oil/water system in an acrylic pipe at a
mixture velocity of 3m/s from the ring and wire probes as well as the ERT system. The
vertical lines limit the phase inversion transition region and are drawn at the first and last
near zero conductivity values recorded using the various probes. The percentages
represent the water fraction at which the different probes show near-zero conductivity
values.
From the figure starting from a pure water flow, the flow tends to be homogenous (at
about 90% water fraction) whereby the change in the normalized conductivity is linear to
the reduction in water fraction. As more oil is added, the value around the pipe periphery
is higher than at other locations. This is due to the formation of water layer at the bottom
110
as observed in the previous section. With further increase in oil fraction (at 60% water
fraction), the value at all three locations are different with the lowest value observed at
the centre of the pipe. This indicates that the oil drops are drawn towards the centre of the
pipe leaving other regions to have a higher presence of water. At 54% water fraction, the
conductivity at the centre of the pipe has reached a near-zero averaged value (i.e. less
than 3% deviation from zero). At this fraction, phase inversion has occurred at the centre
of the pipe. While the corresponding conductivity at the top is low, it still reflects more
than 5% of normalized conductivity. Thus, the flow pattern at 54% appears to resemble
an annular flow with an oil continuous core. The conductivity at the top falls to zero at
about 40% water fraction. From this phase fraction onwards, the flow pattern switches
into a dual continuous flow with an upper oil continuous layer and a lower water
continuous layer. This flow pattern persists until 20% water fraction. At 20% and beyond,
all three locations show a zero conductivity value. The phase inversion is completed with
a full water-in-oil dispersion. Based on the use of the different electrical techniques,
similar outcome have been found to observe the transitional stages during a phase
inversion process. This outcome agrees well with the observation from both the ERT and
visual images.
Similar combined data in stainless steel pipe is presented in Figure 4.5 without the
measurements from the ERT system. Nevertheless, previous measurements, by scanning
across the vertical pipe cross section using the conductivity wire probe, had found that
phase inversion occurs firstly from the centre of the pipe if starting from a water
continuous dispersion. As shown in Figure 4.5, the inversion at the centre occurs at 53%
water fraction while it completes at about 20% where the conductivity at the pipe
periphery falls to a near-zero value. By comparing between the two pipe materials, it is
observed that there is no significant difference between the limits on the phase inversion
transition and this finding supports what is observed from the visual images in Section
4.3.
111
Expt
Transition
O/W
W/O
Δ 20% ○ 53%
Figure 4.5: Normalized conductivity data of the oil/water system in a stainless steel pipe
at a mixture velocity of 3m/s from the ring and wire probes as well as the ERT system.
The vertical lines limit the phase inversion transition region and are drawn at the first and
last near zero conductivity values recorded using the various probes. The percentages
values.
4.5 EFFECT OF DISPERSION INITIALISATION ON PHASE

INVERSION
Previous literatures (e.g. Arashmid et al., 1980, Kumar, 1996, Ioannou, 2006) have
reviewed that the initial dispersion condition can affect the occurrence of phase inversion
leading to an ambivalent region. In order to find out the effect of dispersion initialization
on the phase inversion process, similar set of experiments as those presented in Section
4.4 is carried out starting from an oil continuous dispersion in an acrylic pipe.
112
Transition
O/W
W/O
Δ 20% ○ 54%
Figure 4.6: Normalized conductivity data for the oil/water system in an acrylic pipe at a
mixture velocity of 3m/s from the ring and the wire probes for the two inversion routes (a)
starting from oil continuous (OC) and (b) starting from water continuous (WC) mixture.
The vertical lines limit the phase inversion transition region and are drawn at the first and
last near zero conductivity values recorded from the two probes. The percentages
values.
From Figure 4.6, the results from the wire and ring probes are shown since they represent
the pipe centre and periphery respectively where the phase inversion begins and ends. It
can be seen that the conductivity values from the ring and wire probes are very close for
both inversion routes. Only slight deviation is observed between 20-40% water fractions.
This is possibly because of the difference in the mechanism of the phase inversion
process. Starting from a water continuous dispersion, the water layer at the bottom of the
pipe is likely to have oil drops within it and the conductivity value of that layer will be
lower than in the case of pure water. Starting from an oil continuous dispersion, on the
other hand, the water layer at the bottom of the pipe is formed by the coalescence and
deposition of water drops. As such, there would probably be no oil drops present and the
113
conductivity will be higher than in the water continuous route. Once the water layer has
reached a certain thickness, entrainment of oil drops can occur. In this case, it explains as
well the agreement of the data for the two routes beyond 40% water fractions. The wire
probe gives a higher conductivity value for the oil continuous route compared to the
water continuous route between 20% - 40% due to the opposite effect in the
concentration of water phase at the pipe centre.
4.6 EFFECT OF MIXTURE VELOCITY ON PHASE INVERSION
Phase inversion experiments were carried out at a mixture velocity of 4m/s starting from
a water continuous dispersion and the phase distribution at decreasing water fractions are
shown in Figure 4.7. At high water fraction when the flow is water continuous, the
dispersion appears more homogeneous than at 3m/s (comparing with Figure 4.3). At 40%
water fraction, the core of the pipe flow becomes oil continuous (green) while it remains
water continuous over the pipe annulus. A water layer still remains at the bottom of the
pipe but is thinner than the water layer at 3m/s with the same input water fraction
(compare 26% water fraction with Figure 4.3). Phase inversion is completed only when
this layer of water is also dispersed into the oil (i.e. at about 22% water fraction). Full oil
continuous dispersion is achieved at a water fraction below 22%.
The effect of mixture velocity on the changes in phase continuity in the pipe cross section
is shown in Figure 4.8. It can be observed that the mixture remains to be water
continuous in the pipe centre until a lower water fraction (>40%) at 4m/s compared to
54% at 3m/s. However, phase inversion is completed at approximately the same input
water fraction (20%) when the water layer at the bottom of the pipe is dispersed. Thus,
increasing the mixture velocity aids the homogenisation of the dispersion leading to a
narrower phase inversion transition.
114
Oil-in-water
dispersion
(80%) (70%) (60%)
Phase inversion
transition
(54%) (40%) (26%)
Water-in-oil
dispersion
(22%) (20%) (16%)
Figure 4.7: Phase distribution in a pipe cross section during the transition from a water
continuous to an oil continuous dispersion at a mixture velocity of 4m/s. The percentages
in brackets represent the input water fraction.
115
Expt
W/O Transition O/W
Figure 4.8: Comparison of normalized conductivity data for the oil/water mixture at a
mixture velocity of 3m/s and 4m/s using the ring probe and wire probe. The vertical lines
determine the phase inversion transition at which the first and last near zero conductivity
value are recorded using the various probes – dotted lines for 3m/s and solid lines for
4m/s.
4.7 PRESSURE GRADIENT DURING PHASE INVERSION
The pressure drop, across a 1.5m section of the pipeline, is also measured during the
conductivity measurements. Each data point at a specific phase fraction is an average of
3000 measurement points (taken in 30 seconds interval). The data points are then
normalized as pressure gradient. Figure 4.9 presents the average pressure gradient at 3m/s
mixture velocity starting from a water continuous dispersion across the phase inversion
transition.
Starting from 100% water fraction, an increase in the oil phase fraction will lead to a
reduction in pressure gradient. This can be attributed to the drag reduction effect due to
the presence of the dispersed oil drops in the water continuous phase and the effect has
been previously reported in literature (e.g. Arirachakaran et al., 1989, Pal, 1993, Nädler
and Mewes, 1997). The pressure gradient continues to reduce while water is still in
116
contact with the pipe wall. As the water fraction reaches about 40%, the pressure gradient
appears to fluctuate significantly especially within the transition region. The large
fluctuations are believed to be caused by the presence of both water continuous and oil
continuous dispersion co-existing as complex structures. Piela et al. (2008) has reported
similar findings where the friction factor is significantly affected by the changes in phase
continuity as the probes pass through the complex structures. As the presence of oil
continuous dispersion increases and upon the first contact of the oil phase with the pipe
wall, the pressure gradient starts to increase towards the single phase oil value as the
more viscous oil phase comes into contact with more of the pipe wall.
Expt
W/O Transition O/W
Figure 4.9: Experimental pressure gradients in an acrylic pipe at a mixture velocity of

3m/s. The error bars represent the standard deviation of the fluctuations at each specific
phase fraction. The arrow denotes the direction of experiment from water continuous to
oil continuous dispersion and the vertical lines represent the boundaries of the phase
inversion transition region.
117
Comparing between the water continuous and oil continuous dispersion, the drag
reduction is observed to be stronger in the oil continuous dispersion. Similar observation
has also been reported by both Ioannou, 2006 and Hu, 2005 when they conducted their
experiments in stainless steel pipes using Exxsol D140 as the oil phase. Pal (1993) has
also found that, in stainless steel pipe, unstable water-in-oil emulsions exhibit higher drag
reduction than oil-in-water emulsions. Ioannou (2006) has observed that the pipe material
has an effect on the drag reduction (when using Marcol 52 with viscosity at 11mPa s) of
which acrylic pipe will have a stronger drag reduction with oil-in-water dispersion.
However, this effect is not observed in the current experiments.
4.7.1 EFFECT OF INVERSION ROUTE ON PRESSURE GRADIENT
As the presence of oil at the pipe wall can have a significant effect on the changes in
pressure gradient, it is of interest to investigate whether the inversion route can influence
the pressure gradient by affecting the phase distribution. The pressure gradient results for
the two inversion routes are shown in Figure 4.10. In both cases, the minimum pressure
gradient falls within the phase inversion transition region. The inversion route does not
affect the pressure gradient significantly. This agrees with the phase continuity data that
showed changes in the continuity of phases in contact with the wall at similar phase
fractions for both inversion routes.
4.7.2 EFFECT OF MIXTURE VELOCITY ON PRESSURE GRADIENT
The results of the pressure gradient data with respect to the different mixture velocity are
presented in Figure 4.11 - 4.12. The pressure gradient increases with mixture velocity.
Similar to 3m/s, the pressure gradient at 4m/s initially decreases with decreasing water
fraction until about 40% and start increasing again towards the single phase oil value.
Large fluctuations are again observed between 26% and 54% water fraction. While the
absolute values for pressure gradient has increased across the whole range of phase
fraction, the difference in pressure gradient appear to be larger both below 20% and
beyond 60% water fraction. According to Ioannou (2006), drag reduction increases with
118
increasing velocity/turbulence as the increased occurrence of drop coalescence and
breakup event in the flow can cause a suppression of the velocity/turbulence in the
continuous phase. Thus, it is expected to have a wider difference in pressure gradient in a
dilute dispersion compared with a dense dispersion.
W/O Transition O/W
Figure 4.10: Comparison of pressure gradient for the oil/water system in an acrylic pipe
at a mixture velocity of 3m/s between the two inversion routes (a) starting from oil
continuous dispersion (OC) (b) starting from water continuous dispersion (WC). The
vertical lines represent the boundaries of the phase inversion transition region.
Similar trends in pressure gradient are also found for the stainless steel pipe. The
minimum in pressure gradient however appears at a lower water fraction indicating that
the water layer in the steel pipe persists till a lower water fraction than in the acrylic pipe.
Steel is a more hydrophilic pipe material than acrylic pipe (Ioannou, 2006) and this may
have aided the water layer to retain longer.
119
Expt
Figure 4.11: Comparison of pressure gradient measurement between mixture velocity of

3m/s and 4m/s in an acrylic pipe. The error bars represent the standard deviation of the
fluctuations at each phase fraction. The arrow denotes the direction of experiment from
water continuous dispersion to oil continuous dispersion. The vertical lines limit the
boundaries for phase inversion transition - dotted lines for 3m/s and solid lines for 4m/s.
Expt
Figure 4.12: Comparison of pressure gradient measurement between mixture velocity of

3m/s and 4m/s in a stainless pipe. The arrow denotes the direction of experiment from
water continuous dispersion to oil continuous dispersion.
120
4.8 DROP SIZE DISTRIBUTION IN OIL & WATER DISPERSION
In a liquid-liquid dispersion, the loss of balance between drop breakage and coalescence
rates has been considered as a possible mechanism for phase inversion (Coulaloglou &
Tavlarides, 1977; Kumar, 1996; Yeo et al., 2000). Breakage and coalescence rates
depend on drop size and dispersion phase fraction. Other criteria for phase inversion such
as equating the surface energy between water continuous dispersion and oil continuous
dispersion is also a relation with the drop size. In addition, the various forces that act on
the drops which determine their movements and placement in the flow field and
eventually the phase distribution also depend on the drop size. Thus, drop sizes and their
distribution are important in understanding phase inversion.
Drop size measurements are conducted with the dual impedance probe (see Chapter 3)
for a range of phase fractions. Measurements are taken for both inversion routes (starting
both from O/W and from W/O dispersions). Based on the conductivity probe experiments
(see Figure 4.4), it can be seen that the phase inversion region is within two boundaries (1)
the boundary at high water fraction beyond which the dispersion is water continuous and
(2) the boundary at low water fraction below which the dispersion is oil continuous. The
second boundary (2) involves an inversion of the water continuous layer at the bottom of
the pipe. Detection of drop size using the dual impedance probe can be difficult at that
location due to the possible interference of interfacial waves with the thin water layer.
Therefore, the investigation of drop size before and after inversion is focused on the first
inversion boundary (1) occurring at the pipe centre.
During the experiment, the drop velocity is calculated from the cross-correlation of
signals for the two dual impedance probe sensors (see Lovick, 2004 and Hu, 2005). The
calculated drop velocity together with the time duration of the drops in the signal of
either of the two impedance sensors will provide a drop chord length. The chord length
distribution from the measured drops can be further converted into drop size distribution
based on an algorithm developed by Hu et al. (2006).
121
4.8.1 DROP SIZE DISTRIBUTION MEASUREMENT
During the experiment, it was found that no cross correlation could be obtained at dilute
dispersions because of the small number of drops passing through the impedance sensors.
Therefore, drop size distributions are obtained from 60% water fraction across the first
inversion boundary and down to 40% water fraction at different locations along the
vertical pipe diameter. Drop size measurements are also conducted for the two inversion
routes (i.e. o/w → w/o and w/o → o/w).
Figure 4.13 represents a sample of chord length measurements an oil-in-water dispersion

(60% water fraction) at different locations along the vertical pipe diameter for a mixture
velocity of 3m/s and both inversion routes. Any chord length larger than 10mm has been
ignored since this exceeds the physical distance between the two impedance sensors. It
can be seen that the chord length distributions are similar for both inversion routes.
Larger drops can be seen at the centre of the pipe (see Figure 4.13 (c)) where, as
discussed before, the dispersed oil phase tends to accumulate. The higher fraction would
favour drop coalescence and increase the drop size. Data at other phase fractions are
given in Appendix A. As the dispersed oil fraction is increased, very few or no drop can
be captured at about 6mm from the top wall.
The probability density functions of an oil-in-water dispersion and a water-in-oil

dispersion are presented in Figure 4.14. For the same conditions as Figure 4.13 and at,
example, 14mm, it can be seen that a log-normal distribution fit the data well as this has
also been previously reported by other investigators for dispersed pipe flow (e.g.
Karabelas (1978), Simmons & Azzopardi (2001)). There is predominantly a large number
of small drops with approximately 1-2 mm in diameter. The long tail of large drops
(>5mm) may be caused by irregular drop structures but these large structures account for
less than 3% of the drop number. From the figure, the oil-in-water dispersion has a
narrower distribution than the water-in-oil dispersion which at the same time is shifted
towards smaller drop sizes. This has also been observed by Piela et al. (2008), Hu (2005)
and Liu et al. (2005). As suggested by Kumar (1996), the difference in drop size may be
122
(a) 6mm (b) 14mm
(c) 20mm (d) 26mm
(e) 32mm
Figure 4.13: Chord length distribution of oil drops in water (60% water fraction) at a
mixture velocity of 3m/s at (a) 6mm (b) 14mm (c) 20mm (d) 26mm (e) 32mm from the
top wall of the pipe. The distributions of the two inversion routes are given – starting
from a O/W dispersion (□) and from a W/O dispersion (■).
123
due to the different dielectric constants of the two phases. Oil drops in water experience
repulsive forces due to the overlapping of the electrical double layers forming in water
and thus have lower coalescence efficiencies whereas water drops in oil of low dielectric
constants (which applies to Exxsol D140) do not experience similar repulsion and
therefore have higher coalescence efficiencies. Coalescence efficiency also depends on
the continuous phase viscosity (Coulaloglou & Tavlarides, 1977; Tsouris & Tavlarides,
1994; Liu & Li, 1999). Based on the larger water drops in oil, it appears that the viscosity
of Exxsol D140 does not significantly affect the film drainage and lower the coalescence
efficiency in relation to other influencing factors.
The evolution of drop size during inversion can be seen in Figure 4.15. In terms of
average Sauter mean diameter, d32, for a mixture velocity of 3m/s and both inversion
routes. The results are from the core of the pipe (14mm from the top wall) where
inversion across boundary (1) occurs. It can be seen that the oil drops in water are smaller
than water drops in oil, regardless of the inversion route. The drop size, at 54% water
fraction, is higher than at either 40% or 60% water fraction. Based on previous
investigations, the large Sauter mean diameter is likely caused by the presence of
irregular complex structures as well as coalesced large drops especially at a location
where phase inversion is prominent. When comparing the effect of inversion route, no
significant difference can be seen on the drop size across the inversion boundary (1). This
may explain the reason for the same phase fraction when phase inversion occurs for
inversion boundary (1) reported in Figure 4.6.
124
(a) O/W
(b) W/O
Figure 4.14: Sample drop size distribution is recorded at 14mm from the top wall with a
mixture velocity of 3m/s (a) oil-in-water dispersion (60% water fraction). The
experimental data is compared with a log-normal distribution with a mean of 1.51mm and
a variance of 0.238. (b) water-in-oil dispersion (40% water fraction). The experimental
data is compared with a log-normal distribution with a mean of 2.07mm and a variance of
0.451.
125
Figure 4.15: Sauter mean diameter at different input water fraction across the phase
inversion boundary (1). The phase in brackets represents the initial continuous phase and
the number in each column represents the Sauter mean diameter at the specific sample.
4.9 EFFECT OF DISPERSION INLET ON PHASE INVERSION
With previous sections discussed on a split inlet where the flow is gradually developed,
phase inversion occurs across a transitional region due to partial inversion occurring at
different input water fraction. It is interesting to investigate the effect of phase inversion
if the two phases are introduced as a full dispersion straight from the inlet. Thus, a
dispersion inlet is developed and the design of the inlet was discussed in Chapter 3.
Figure 4.16 presents the conductivity data from the ring and wire probes. Measurements
are conducted at the centre of the pipe (with the conductivity wire probe) and at the
periphery of the pipe (with the ring probe) similar to the configuration as presented in
Section 4.4.2.
126
Figure 4.16: Normalized conductivity data for an oil/pure water mixture at a mixture
velocity of 3m/s using the ring probe and wire probe. The letter in bracket represents
initial dispersed phase – (W) water (O) oil. The arrow denotes the direction of experiment
and the vertical line denotes the first input water fraction where phase inversion occurs.
As can be seen in figure, the conductivity data from both probes are very close for all
input water fractions indicating that the flow is fully dispersed for all phase fractions at
the mixture velocity of 3m/s. As a result, the change from a water continuous dispersion
to an oil continuous dispersion (and vice versa) occurs throughout the pipe cross-section
and at a particular phase fraction rather than over a transitional region as in the case with
a split inlet. The decrease in conductivity is almost linearly proportional to the decrease
in the water phase. Phase inversion occurs at a single phase fraction of about 28% input
water fraction.
The effect of the dispersion initialization conditions is shown in Figure 4.17. There is
only very small difference on the phase inversion point between the two inversion routes,
28% input water fraction for an initial O/W dispersion and 26% water fraction for an
127
initial W/O dispersion. It should be noted that 2% is the minimum step change in phase
fraction that can be achieved in the experimental set up.
Figure 4.17: Normalized conductivity data for an oil/pure water mixture at a mixture
velocity of 3m/s using the ring probe and wire probe through two initialization conditions
– starting from water continuous dispersion (empty symbols) and starting from oil
continuous dispersion (solid symbols). The dotted vertical line represents the phase
inversion point starting from water continuous dispersion and the solid line represents the
phase inversion point starting from oil continuous dispersion.
The pressure gradient across the phase inversion is presented in Figure 4.18. It can be
observed that the gradual increase in the dispersed phase (oil) by reducing the input water
fraction at a constant mixture velocity leads to a gradual increase in pressure gradient (i.e.
between 40% and 100% water fraction). This is possibly due to the higher effective
viscosity of the mixture in the pipe flow. Starting from 36% water fraction and below, the
pressure gradient has a sudden drop which is likely due to the drop coalescence and
deformation. The deformable drops can fill up the interstitial spaces leading to higher
fluidity of the mixture flow (i.e. lower mixture viscosity & pressure gradient in a pipe
flow). The pressure gradient reaches a minimum at the phase inversion point (i.e. 26%)
and, beyond this point, the oil continuous phase will lead to a gradual increase possibly
128
because the deformed or coalesced drops are likely to be separated as the mixture is
diluted. This will result in a more narrowly distributed drop size and, thus, reducing the
fluidity of the mixture flow once again. The relatively higher pressure gradient by the oil
continuous dispersion is also due to the higher oil viscosity as the continuous phase.
Phase
inversion
Figure 4.18: Experimental pressure gradients for a dispersed inlet configuration at a

mixture velocity of 3m/s. The arrow denotes the direction of experiment from water
continuous to oil continuous dispersion and the vertical line represents the phase
inversion point.
In order to determine the change in drop size during the phase inversion process for the
dispersed inlet configuration, Figure 4.19 presents the chord length distribution at
different phase fractions. The chord length distribution can be seen to be narrow when the
input water fraction is higher than 36%. The distribution becomes broader with longer
chord length between 26% and 36% water fraction of which drop
deformation/coalescence was previously suggested to occur. The distribution is the
widest at about 26% where phase inversion is observed. At 20% water fraction (W/O
dispersion is formed), the distribution becomes narrower beyond the phase inversion
point. The corresponding drop size (by converting the chord length based on the method
by Hu (2005) of the W/O dispersion is also found to be higher than O/W dispersion.
129
Figure 4.19 Comparison of chord length distribution at different input water fraction
(fraction in bracket) across the phase inversion process at 3m/s mixture velocity.
4.10 CONCLUSION
The phase inversion transition of an oil-water dispersion was investigated in this chapter.
The ERT system revealed the distribution of the phases and the respective flow patterns
while the conductivity data from the ring and wire probes as well as the ERT provide
further information on the phase continuity in a pipe cross section. The following
mechanism for phase inversion is proposed based on the data from the split inlet
configuration.
130
For inversion starting from an oil-in-water dispersion:
(1) At low oil fraction, the added oil phase is initially dispersed in the continuous water
phase.
(2) As the oil fraction increases, the oil drops will accumulate at the pipe centre where
they coalesce and the mixture inverts.
(3) With the mixture also inverts at the top of the pipe, a dual continuous flow pattern is
formed with oil as the continuous phase at the top of the pipe and water as the continuous
phase at the bottom.
131
(4) At high oil fractions, the thin layer of water at the bottom of the pipe also disperses
and the flow becomes fully dispersed with oil as the continuous phase.
The change from a water continuous to an oil continuous dispersion occurs over a range
of volume fractions and is bounded by two lines: the first line is related to the inversion at
the pipe centre and the second line is related to the diminution/formation of the water
layer at the bottom of the pipe depending on the inversion route. However, no significant
difference was found in the phase inversion transition boundaries between the two
inversion routes. With increasing mixture velocity, the transitional region over which
phase inversion occurs reduces in size.
Change in phase distribution during the phase inversion transition is found to affect the
pressure gradient. Starting from a water continuous dispersion, the addition of dispersed
phase initially led to drag reduction. Pressure gradient continued to reduce even after
inversion has occurred at the centre of the pipe. When the water layer at the bottom of the
pipe eventually disperses and the viscous oil phase comes in contact with the whole pipe
wall, the pressure gradient starts to increase again. There is no significant difference in
pressure gradient between the two inversion routes. The V-shaped trend, which the
changes in pressure gradient exhibits, is similar regardless of mixture velocity and pipe
materials. The degree of changes is found to be higher at higher mixture velocity possibly
because of higher drag reduction. There is also observed, in the stainless pipe, a shift in
the minimum pressure gradient leftward with the increase in mixture velocity.
In addition to the change in flow conditions (e.g. mixture velocity, dispersion

initialization), the change in the inlet configuration (to a dispersed inlet) has led to a
132
significant change in the phase distribution in the mixture flow. This results in a
substantial difference in the phase inversion occurrence from a phase inversion transition
across a range of phase fraction into a phase inversion point. The pressure gradient
increases with the initial increase in the dispersed phase (i.e. oil) until drop
coalescence/deformation become prominent. During the rearrangement of the deformable
drops and until phase inversion has occurred, the pressure gradient has a sharp drop. The
higher oil viscosity will subsequently led the increase in pressure gradient beyond phase
inversion.
For the split inlet, drop size measurements are in accordance with the phase distribution
data. Larger drop sizes are found at the pipe core just before inversion from O/W to W/O
dispersion where the dispersed phase has concentrated. Differences are also seen between
the sizes of oil drops in water and water drops in oil regardless of the inlet configuration.
From the drop size measurement, it can also be concluded that drop coalescence and
deformation is a critical process during phase inversion.
The understanding on how phase inversion progresses during pipe flow will form the
basis of the investigation in the following chapters – effect of interfacial tension on phase
inversion (Chapter 5), prediction model of phase inversion based on flow regimes
(Chapter 7) and CFD simulation on the interaction of the dispersed and continuous phase
in a pipe flow (Chapter 8).
133
Chapter 5: Effect of interfacial tension on phase
inversion
5.1 OVERVIEW
The pipeline flow of two immiscible liquids, usually an aqueous and an organic, is
commonly encountered in process industries and during oil production. Depending on the
flow rates and phase fractions, dispersed flow can establish either the water (O/W) or the
oil (W/O) as being the continuous phase. At some critical operational conditions, phase
inversion occurs whereby the initial dispersed phase becomes continuous and the
continuous phase becomes dispersed. The change in phase continuity can have a
significant effect on the mixture rheology and on frictional pressure drop while the nature
of the continuous phase is related to pipeline corrosion and to the rate of the mixture
separation at the end of the process. It is important, therefore, to understand and be able
to predict the conditions under which inversion appears.
Many factors have been reported to affect phase inversion and the phase inversion point,
i.e. the volume fraction of the liquids at inversion. Experimental work has mainly been
carried out in stirred vessels (for a review see Yeo et. al, 2000) where it was found that
phase inversion is affected not only by the fluid properties such as viscosity, density and
interfacial tension, but also by the stirred vessel geometric configuration (e.g. type of
impeller), the vessel and impeller material wettability and the experimental set up (e.g.
the phase that the impeller is in at the beginning of the experiment). In contrast to stirred
vessels, there is less information available on the effect of various parameters on phase
inversion during pipeline flows. Apart from fluids properties, parameters such as mixture
velocity, rate at which the dispersed phase is added in the continuous, size of the pipe and
wettability of the pipe material have also been found to influence phase inversion (Angeli
and Hewitt, 2000; Ioannou et. al, 2005; Luo et. al, 1997; Piela et. al, 2008).
134
Among the fluid properties, viscosity is considered to have a significant impact on phase
inversion. Selker and Sleicher (1965) found that as the viscosity of one phase increases,
its tendency to be dispersed also increases, i.e. the minimum fraction that the phase can
be continuous will decrease and the maximum fraction that it can be dispersed will
increase. Notably, many models for the prediction of the phase inversion fraction are
based on the liquid viscosities (Arirachakaran et. al, 1989; Nädler and Mewes, 1995;
Ngan et al., 2009; Yeh et al., 1964). Density does not seem to affect inversion
significantly especially when the density difference between the two liquids is small but
it will influence the homogeneity of the dispersion (McClarey and Mansoori, 1978;
Selker and Sleicher, 1965). However, systems with large density difference between the
phases show an increased tendency to invert (Kumar et. al, 1991).
The effect of interfacial tension is the least understood. In agitated vessels, Selker and
Sleicher (1965) suggested that interfacial tension cannot affect the type of dispersion
since it is not a function of the sign of the interfacial curvature. However, the suggestion
was not further investigated. According to Yeh et. al (1964) interfacial tension will cause
inversion of a dispersion to occur at equal volumes of the two liquids in the absence of
other influencing parameters. Luhning and Sawistowski (1971), in contrast, stated that
interfacial tension is important and found that the limits of the ambivalent region were
correlated with the impeller Weber number (for Weber numbers between 350 and 4000).
A decrease in interfacial tension will make inversion more difficult to occur and will
widen the boundaries of the ambivalent region (Clarke and Sawistowski, 1978; Norato et.
al, 1998). Interfacial tension is also related to drop break up and coalescence
(Coulaloglou and Tavlarides, 1977; Luo and Svendsen, 1996; Prince and Blanch, 1990;
Sovová, 1981; Tsouris and Tavlarides, 1994) that define the size of the dispersed phase.
The evolution of the dispersed phase size and the departure from equilibrium between
break up and coalescence rates have been used to predict phase inversion in stirred
vessels (Arashmid and Jeffreys, 1980; Groeneweg et. al, 1998; Hu et. al, 2005), thus
indicating a strong dependence on interfacial tension. There is no reported work on the
effect of interfacial tension on phase inversion in pipeline flow of liquid-liquid
dispersions. This is not because interfacial tension was kept constant during
135
investigations of the fluid properties, rather that it was not monitored. It is actually quite
difficult to vary a liquid property without affecting at the same time to some extent the
interfacial tension. In this chapter, the interfacial tension of an oil-water mixture is varied
in a controlled way in order to investigate its effect on phase inversion during pipeline
flow while keeping all other fluid properties constant as shown in Table 5.1.
5.2. EXPERIMENTAL SETUP
The experimental investigations in this study were conducted in a pilot scale liquid-liquid
flow facility (See Chapter 3 for the detailed description of the experimental setup). A
38mm I.D acrylic test section will be used in all experiments during this study to
minimize any effect on phase inversion caused by the experimental configuration. In this
study, Exxsol D140 (density of 828 kgm-3, viscosity of 5.5 mPa.s) is used as the oil phase
while tap water or a glycerol solution is used as the aqueous phase. The two fluids are
joined at the end of a Y-junction inlet to ensure that the two fluids join with minimum
mixing at the inlet.
For each experiment, simultaneous measurements of pressure gradient and phase

continuity were carried out. Pressure gradient was measured with a differential pressure
transducer between two ports 1.5m apart, with the first port located at 3.8m (~100D)
from the inlet. Two types of conductivity probes were used to detect phase continuity and
the appearance of phase inversion, namely a ring and a wire probe. These probes were
located at 7m (~184D) from the inlet. High conductivity values indicate that a
water/glycerol solution continuous mixture is in contact with the probes, while low
values denote an oil continuous mixture. The ring probe consists of a pair of metallic
rings, 4mm thick and 38mm apart, embedded on the pipe perimeter, flush with the wall,
and in contact with the fluid mixture. This probe provides information on the continuity
of the mixture adjacent to the pipe wall. The wire probe is made of two conducting wires
with 0.5 mm diameter set 10 mm apart. It detects the phase continuity at different
locations in a pipe cross section using a traversing mechanism. In this work, it was placed
in the middle of the pipe.
136
An Electrical Resistance Tomographic (ERT) imaging system was also used. ERT
tomograms were taken at the same axial location (i.e. 7m from the inlet) as the
measurements of the two conductivity probes. In addition, the conductivity data from the
ERT system were measured at 4mm from the top pipe wall in conjunction with the data
from the conductivity probes to study changes in the mixture continuity in a pipe cross
section at different phase volume fractions. Conductivity data for all three sensors were
time averaged when the conductivity measurements were stabilized and over a period of
30 seconds. The data were normalized with respect to the values of single phase water. It
was previously found that, at the same pipe cross section location, the normalized
conductivities from the different sensors differ by 1% and can be regarded to have
sufficient precision to relate the conductivity data from the various systems across a pipe
cross section for a given phase fraction.
A dual impedance probe using alternating current was also implemented to determine the
drop size distribution in the dispersion. During the experiment, the probe is positioned at
7 m downstream of the inlet. Sampling rate is set at 45kHz and a total number of 120,000
data points is collected. This high rate ensures that each dispersed drop is represented by
more than one data points in the probe output signal. The probe measures chord lengths
from which drop diameters can be calculated (Hu et. al, 2006).
The experiments were carried out at 3m/s mixture velocity and phase inversion was
investigated starting from the aqueous phase as being continuous as well as from the
organic phase. Each experiment started with a single phase flow of an initial continuous
phase where the flow rate of the dispersed phase was gradually increased while that of
the continuous phase was reduced to maintain a constant mixture velocity until phase
inversion has occurred and beyond. The same experimental run was repeated at least
three times and the averaged results are reported in Section 5.3.
The effect of glycerol addition on the physical properties of water and on the oil-water
interfacial tension is shown in Table 5.1. These measurements were conducted at ambient
temperature (~20oC±0.4oC). Dilute glycerol solutions (0.5% and 1% v/v) were chosen to
137
ensure that the glycerol addition only affects interfacial tension and not the other
properties of the aqueous phase. The density of water and glycerol solutions was
measured using a weighing flask and the viscosity was measured using the Bohlin
Gemini rheometer by Malvern Instruments over a range of shear rates. For the interfacial
tension with ExxsolTM D140, a Krüss Contact Angle Measuring System (DSA100) was
used. The viscosity measurements also revealed that the glycerol solutions are Newtonian.
As shown in Table 5.1, while the density and viscosity of the aqueous phase do not
change significantly with the addition of glycerol at both concentrations, the oil-water
interfacial tension is reduced when glycerol is added. With the suggested concentrations
of glycerol, it can be regarded that interfacial tension will be the influencing factor on the
correlations for drop size (e.g. Brauner and Ullmann, 2002) as well as the critical phase
fraction for inversion (e.g Fakhr-Din, 1973) through the Froude number, Reynolds
number and Weber number terms.
Water with 0.5% Water with 1%

Pure water
glycerol glycerol
Density (kg/m3) 996 1004 (0.80%) 1007 (1.10%)
Viscosity (cp) 1.031 1.036 (0.48%) 1.024 (-0.68%)
Interfacial Tension with
48.14 44.84 (-6.86%) 41.75 (-13.27%)
ExxsolTM D140 (mN/m)
Table 5.1: Properties of the glycerol solutions. The number in bracket represents the
percentage change in the value of the respective physical property compared to that of
pure water.
5.3. PHASE INVERSION RESULTS
The changes in phase continuity and distribution were first studied in a water-oil system
starting from single phase water and gradually adding the oil phase while keeping the
mixture velocity constant at 3m/s. The mixture conductivity values, from the various
probes, which denote phase continuity at different locations in a pipe cross section, are
shown in Figure 5.1(a). Phase distribution tomograms by ERT at representative phase
fractions can be seen in Figure 5.1(b).
138
1.0
Normalized conductivity [-]
Pipe Periphery Expt

Pipe Centre
0.8 Pipe Top Wall
0.6 Transition
0.4 O/W
W/O
0.2 Δ 20% × 40%

○ 54%
0.0
0 20 40 60 80 100
Input water fraction [%]
Figure 5.1: (a) Normalized conductivity data of the oil/water system at a mixture velocity
of 3m/s from the ring and wire probes and the ERT system. The direction of the
experiment from water to oil continuous is shown by the arrow. The vertical lines denote
the boundaries of the phase inversion transition region and are drawn at the first and last
near zero conductivity values recorded using the various probes. The percentages
represent the water fraction where conductivity approaches zero at the various locations
in the pipe cross section.
139
Oil-in-water dispersion
(100%) (80%) (60%)
Phase inversion transition
(54%) (40%) (26%)
Water-in-oil dispersion
(20%)
Figure 5.1: (b) Phase distribution in a pipe cross section during the transition from a
water continuous to an oil continuous dispersion at a mixture velocity of 3m/s. The
percentages in brackets represent the input water fraction. Blue denotes the region of the
water phase and green denotes the region of the oil phase. Last tomogram (i.e. water-in-
oil dispersion) represents the completion of phase inversion upon which ERT
measurement fails.
Initially (at pure water and high water fractions) the conductivity values are high
throughout the pipe cross section which means that the mixture is water continuous
(shown in Figure 5.1(b) as oil-in-water, O/W, dispersion). At low oil (high water)
fractions, the mixture is almost a uniform dispersion. As the fraction of the non-
conductive dispersed oil phase increases, the conductivity values begin to decrease. The
phase distribution also indicates that the dispersion is not as uniform any more but there
is higher concentration of oil in the middle of the pipe (see 60% water fraction in Figure
5.1(b)). The mixture however remains water continuous. At an input water fraction of
54%, the conductivity value in the middle of the pipe reduces to near zero (Figure 5.1(a))
indicating that the continuous phase at this location has changed from water to oil.
However, the conductivity values at the top and the periphery of the pipe are still high
140
which means that a flow pattern has formed with an oil continuous region in the middle
of the pipe and a water continuous region as an annulus around it. The water continuous
layer is thinner at the top of the pipe and finally it inverts as well at approximately 40%
water fraction. The flow regime then becomes dual continuous with an oil continuous
region at the upper part of the pipe and water continuous region at the lower part. The
respective tomogram in Figure 5.1(b) also shows the segregation of the phases at this
fraction. This transitional regime persists with a further decrease in the water fraction
until about 20% where the conductivity from the pipe periphery also reduces to zero
(Figure 5.1(a)) indicating that an oil continuous mixture is now established throughout
the pipe cross section.
1.0
Normalised Conductivity [-]
Pipe Periphery Expt

Pipe Centre
0.8 Pipe Top Wall
0.6
Transition
W/O
0.4 O/W
0.2 Δ 10% × 26%

○ 54%
0.0
0 20 40 60 80 100
Input Water Fraction [%]
Figure 5.2: (a) Normalized conductivity data of the oil/1% glycerol solution system at a
mixture velocity of 3m/s from the ring and wire probes and the ERT system. The
direction of the experiment from water to oil continuous is shown by the arrow. The
vertical lines denote the boundaries of the phase inversion transition region and are drawn
at the first and last near zero conductivity values recorded using the various probes. The
percentages represent the water fraction where conductivity approaches zero at the
141
Oil-in-water dispersion
(100%) (80%) (60%)
Phase inversion transition
(54%) (40%) (26%)
Water-in-oil dispersion
(18%) (10%)
Figure 5.2: (b) Phase distribution in a pipe cross section during the transition from a 1%
aqueous continuous to an oil continuous dispersion at a mixture velocity of 3m/s. The
percentages in brackets represent the input water fraction. Blue denotes the region of the
water phase and green denotes the region of the oil phase. Last tomogram (i.e. water-in-
oil dispersion) represents the completion of phase inversion upon which ERT
measurement fails.
Similar conductivity curves are obtained when 1% glycerol solution is used as the
aqueous phase. As can be seen in Figure 5.2(a), the inversion in the middle of the pipe
appears at the same phase fraction, 54% water, as that in the oil/pure water mixture.
Interestingly, the inversion at the top of the pipe appears at an input water fraction of 26%
when glycerol is present, which is lower to that for pure water (40%). It seems that with
the addition of glycerol, oil tends to concentrate in the middle of the pipe more than when
only water is used while the aqueous phase is being driven to the outer part of the pipe.
This is supported by the conductivity values which in the middle of the pipe are slightly
lower and close to the pipe wall are higher for the glycerol solution than for pure water
(see Figure 5.3 for water fractions between 26% and 44%). It can also be seen from
Figures 5.2(a) and 5.3 that the water layer near the bottom of the pipe persists over lower
142
water fractions in the presence of glycerol and the inversion to fully dispersed oil
continuous flow occurs at 10% water fraction compared to the 20% of the pure water
system.
0.5
3m/s (Pure Water) - Pipe Periphery

3m/s (Pure Water) - Pipe Centre
0.4 3m/s (Glycerol 1%) - Pipe Periphery
3m/s Glycerol 1%) - Pipe Centre
Expt
0.3
0.2
0.1
0.0
0 10 20 30 40 50
Figure 5.3: A comparison of the normalised conductivities in the middle of the pipe and
the pipe periphery between the 1% glycerol solution and the pure water dispersions at a
mixture velocity of 3m/s.
Similar results are obtained with the 0.5% glycerol solution (Figure 5.4). As with the 1%
glycerol solution, the inversion in the middle of the pipe again occurs at 54%, the
inversion at the top occurs at 20% water fraction while fully dispersed oil continuous
flow establishes at 10% water fraction. The addition of glycerol seems to affect the
inversion behaviour of the oil-water mixture but, at least for the conditions tested, the
concentration of glycerol does not appear to play a significant role.
143
1.0

Pipe Periphery Expt
Pipe Centre
0.8 Pipe Top Wall
0.6 Transition
W/O O/W
0.4
0.2 × 20%
Δ 10% ○ 54%
0.0
0 20 40 60 80 100
Figure 5.4: Normalized conductivity data of the oil/0.5% glycerol solution system at a
mixture velocity of 3m/s from the ring and wire probes and the ERT system. The
direction of the experiment from water to oil continuous is shown by the arrow. The
vertical lines denote the boundaries of the phase inversion transition region and are drawn
at the first and last near zero conductivity values recorded using the various probes. The
percentages represent the water fraction where conductivity approaches zero at the
5.4. EFFECT OF GLYCEROL ADDITION ON PRESSURE

GRADIENT DURING PHASE INVERSION
The changes in phase distribution and flow pattern during the inversion of the oil/water
dispersion, which were discussed in the previous section, are expected to affect the
frictional pressure gradient. The variation in pressure gradient as the oil fraction increases
(inversion from O/W to W/O) can be seen in Figure 5.5 for a mixture velocity of 3m/s
and for the three aqueous phases used, namely pure water, 0.5% and 1% glycerol
solutions.
144
3.0
Transition
Pressure Gradient [kPa/m] 2.5 Expt
O/W
2.0 36%
1.5
1.0
W/O 26%
18%
Pure Water
0.5 0.5% glycerol
1% glycerol
0.0
0 20 40 60 80 100
Figure 5.5: Experimental pressure gradient for pure water, 0.5% 1% glycerol solutions at
different input water fractions at a mixture velocity of 3m/s. The vertical lines indicate
the boundaries of the phase inversion transition region for pure water (solid lines) and
both glycerol solutions (dotted lines). The percentages indicate the water fractions where
pressure gradient shows a minimum. The arrow denotes the direction of experiment from
aqueous to oil continuous dispersion.
Pressure gradient is found to be the same for the single phase flow of either water or
glycerol solutions indicating that the addition of glycerol at these concentrations does not
affect the frictional pressure gradient because it does not alter significantly the density or
the viscosity of the aqueous phase. As the oil fraction increases, pressure gradient
remains the same for all fluid pairs used, while a drag reduction phenomenon is observed,
where the presence of the dispersed phase causes a decrease from the single phase value
of the continuous phase at the same as the mixture velocity (Pal, 1993). The initial
inversion in the middle of the pipe at 54% water fraction is the same for all three systems
and does not affect the pressure gradient possibly because the continuity of the phase in
contact with the pipe wall has not changed. Pressure gradient for pure water starts to
145
increase at about 36% water fraction after the mixture at the top of the pipe has inverted
(see Figure 5.1 (a)). Beyond this fraction, the more viscous oil phase comes in contact
with part of the pipe wall. The increase in pressure gradient for the two glycerol solutions
occurs at lower water fractions of about 18% and 26% for the 0.5% and 1% glycerol
solutions respectively. These changes again happen after the mixture at the top of the
pipe inverts from water to oil continuous. With further increase in the oil fraction
(decrease in water fraction), more oil contacts the pipe wall and the pressure gradient
continues to increase as phase inversion spreads across the pipe until all three systems
reach the single oil phase value. Similar results were found for the opposite experiment
starting from an oil continuous mixture.
5.5. EFFECT OF GLYCEROL ON DROP SIZE DISTRIBUTION IN

OIL AND WATER CONTINUOUS DISPERSIONS
Drop size distributions were measured using the dual impedance probe. The
measurements were conducted for both dispersion initialization conditions, i.e. starting
from oil continuous and from water continuous flows, and for the three aqueous solutions
considered, i.e. pure water, 0.5% and 1% glycerol. It was found that drop size
measurements in dilute solutions could not be obtained as the signals of the two sensors
of the dual impedance probe did not cross-correlate well probably because of the small
number of drops present. The same happened in the inversion transition region just after
the first inversion in the middle of the pipe when glycerol was present in the water phase.
Based on the tomographic images, the reason for this is probably the better separation
between the two phases in the transition region when glycerol is added which leaves a
lower concentration of the dispersed water phase in the middle where measurements are
taken. Drop size measurements were therefore conducted for input water fractions
between 40% and 60%. Measurements were carried out in the middle of the pipe
(approximately 14mm from the top pipe wall). This location was chosen because the
inversion at 54% allows both oil and water continuous dispersions to be investigated
while it is also away from the interface formed during the dual continuous flow that may
affect the measurements if any waves are present. Results are shown below for two
146
indicative fractions, 60% O/W and 40% W/O dispersion. To compare the drop sizes the
average Sauter mean diameter, d32, is used.
The Sauter mean diameters calculated for systems with pure water (PW), 0.5% and 1%
glycerol solutions can be seen in Figure 5.6(a) for an initial water continuous dispersion
and in Figure 5.6(b) for an initial oil continuous dispersion. In general, oil drops (at 60%
water fraction) are smaller than water drops (at 40% water fraction), and this difference is
more obvious for the pure water system. The addition of glycerol does not seem to affect
the oil drop size significantly, which may explain why the initial inversion in the middle
of the pipe from water to oil continuous mixture occurs at the same input phase fraction
for all the three aqueous solutions. In contrast, the size of water drops is reduced when
glycerol is added although the glycerol concentration does not seem to play an important
role. It has been found that water drops in oil tend to coalesce faster when glycerol is
added (Wang et. al, 2009); this would have resulted in larger drops when glycerol is
present which seems to contradict the current result. However, it is possible that these
larger drops will also tend to precipitate in the water layer that is still present at the
bottom of the pipe at 40% input water fraction. This agrees with the findings of the ERT
system which showed reduced concentration of water in the middle of the pipe and an
increase in the water layer at 40% water fraction when glycerol is present compared to
pure water (see Figures 5.1b, 5.2b and also the comparison of the conductivity values in
Figure 5.3). Similar results were found for the opposite experiment starting from an oil
continuous mixture (Figure 5.6(b)).
147
3.5 3.14
3.0
2.59
2.5 2.33
d32 [mm]
2.42
2.0 2.41
1.5 2.15
1.0
0.5 40
0.0
60
PW (Water)
0.5% (Water)
1% (Water)
(a)
3.5 2.94
3.0
2.53
2.5
d32 [mm]
2.0 2.12
2.15 2.26
1.5 2.17
1.0
0.5 40
0.0
60
PW (Oil)
0.5% (Oil)
1% (Oil)
(b)
Figure 5.6 Sauter mean diameter for pure water, 0.5% and 1% glycerol solutions starting
from (a) a water continuous dispersion and (b) an oil continuous dispersion. The phase in
the bracket represents the initial continuous phase. The numbers on the columns represent
the d32 diameter of the respective sample in mm.
148
5.6 CONCLUSION
The effect of interfacial tension on the phase inversion procedure of an oil-water mixture
in pipe flow was investigated in this paper. By adding small concentrations of glycerol to
water, the interfacial tension was reduced but the density and viscosity of the aqueous
phase were not significantly affected.
For all fluid pairs tested initially at low oil fractions, the dispersion was water continuous
with a fairly uniform distribution of the dispersed drops, as revealed by tomograms. As
the oil fraction increased for the same mixture velocity, the mixture at the centre of the
pipe inverted to become oil continuous. With a further increase in the oil fraction,
inversion spread to the top and finally the bottom of the pipe and the mixture became oil
continuous. It was found that while the addition of glycerol in water did not affect the
phase fraction where the first inversion in the middle of the pipe occurred, it delayed the
overall inversion at the rest of the pipe to higher oil (lower water) fractions. Drop size
measurements in the middle of the pipe revealed that the size of the oil drops before
inversion was not significantly affected by the addition of glycerol but the size of water
drops after inversion was reduced. Although with the addition of glycerol, larger water
drops are expected to form through coalescence; it is possible that these large drops also
separate more easily due to gravity leaving only the smaller ones in the pipe centre to be
measured. This settling would also explain the higher water concentration at the pipe
periphery which delays the inversion in these regions when glycerol is added.
A drag reduction phenomenon was seen and pressure drop decreased as the oil fraction
increased. Only when the mixture at the top of the pipe inverted and an oil continuous
phase came in contact with the top pipe wall did the trend reverse and the pressure drop
started to increase with further increase in the oil fraction.
149
Chapter 6: Prediction of phase inversion through
fluid viscosities
6.1 OVERVIEW
Dispersions of two immiscible liquids, usually an organic and an aqueous, where one
phase is in the form of drops within the continuum of the other, are very common in oil
production and transportation. One of the characteristic but least understood phenomena
in dispersions is phase inversion, where the dispersed phase changes to become
continuous and vice versa. Knowing the phase fraction where inversion occurs (phase
inversion point) is important for the processing of dispersions since such change results
to a mixture with different properties (e.g. rheology, drop size). The significant
consequence to the transportation of dispersions is the observed change in pressure
gradient that accompanies phase inversion (Arirachakaran et al., 1989; Angeli and Hewitt,
1998; Ioannou et al., 2005) which can have implications for the flow rate in pipelines.
Previous phase inversion studies have mainly focused on dispersions formed in

mechanically agitated batch or semi–batch vessels (see review by Yeo et al., 2000), but
are significantly less for dispersions in pipeline flows. Investigations of oil-water pipe
flows (Arirachakaran et al., 1989; Pal, 1993; Nädler and Mewes, 1997; Soleimani, 1999;
Liu et al., 2005; Ioannou et al., 2004 & 2005; Chakrabarti et al. 2006; Hu and Angeli,
2006) revealed that the phase inversion point is affected by phase viscosity ratio, velocity,
flow orientation and pipe diameter and material. It was also suggested that the change in
the mixture continuity does not happen at one particular volume fraction but over a range,
during which the flow is transitional.
Several physical mechanisms for phase inversion have been suggested in the literature.
These include instability of the dispersed phase drop size where drop coalescence
dominates over drop break up (Vaessen et al., 1996); minimum system energy or equal
surface energy of the two possible dispersions (oil-in-water and water-in-oil) at phase
150
inversion (Tidhar et al., 1986; Brauner and Ullmann, 2002); zero interfacial shear stress
(Yeh et al., 1964; Nädler and Mewes, 1997). It is, however, difficult to test these models
mainly due to lack of appropriate data.
From the various parameters that affect phase inversion, the viscosities of the phases in
particular, and consequently the dispersion mixture viscosity, appear to play a significant
role. As such, the aim of the chapter is to derive a method to identify phase inversion
based on the fluid viscosities. Section 6.2 presents the proposed method for determining
the phase inversion point. The various mixture viscosity models are listed in Section 6.3
which will be selected as the best model using the proposed methodology to identify the
phase inversion point. The prediction method is tested against 3 different oil/water
system presented by Ioannou (2006) (Section 6.4) and also tested against Yeh et al. (1964)
systems in Section 6.5. Comparison with other literature correlation on phase inversion
prediction is also conducted in Section 6.5. In addition, a dispersed experimental setup is
constructed to produce a homogeneous dispersion and the experimental outcome is tested
against the prediction method (Section 6.6). Lastly, a conclusion on the applicability of
the prediction method will be made in Section 6.7.
6.2. SUGGESTED METHOD FOR DETERMINING THE PHASE

INVERSION POINT
The methodology for identifying the phase inversion point is demonstrated in Figure 6.1
for a dispersion of oil with viscosity 5.5 mPa s and water. The viscosities of the two types
of dispersions, water continuous and oil continuous, are calculated from the viscosity
model by, for example, Brinkman/Roscoe (1952) given by Equation (6.7) in Table 6.1
(other viscosity models are discussed in the following section).
151
20
Phase
Oil Continuous
Mixture Viscosity [mPa s]

Inversion
15 Point Water Continuous
10 W/O O/W
Favourite
5 route
34%
0
0 20 40 60 80 100
Water Fraction [%]
Figure 6.1: Prediction of phase inversion point for the Exxsol D140-water dispersion.
Phase inversion occurs at the phase fraction where the difference in viscosity between the
oil continuous and the water continuous dispersions becomes 0. The Brinkman/Roscoe
(1952) model is used for calculating the viscosity of the dispersions.
The mixture viscosities for oil and water continuous dispersions at dispersed phase
volume fractions from 0% to 100% are plotted together in Figure 6.1. The two viscosity
plots intercept at a point where the mixture viscosities of the two dispersions are the same.
Phase inversion is proposed to occur at this phase fraction where the two mixture
viscosities become equal. The mixture viscosity is related to the pressure gradient which
drives the dispersion. In the example of Figure 6.1 starting from an oil continuous
dispersion the mixture viscosity increases with the water fraction. If the mixture was to
remain oil continuous for water fractions above 34% then its viscosity (and in effect the
pressure drop) would be higher than if it inverted to water continuous. By inverting to
water continuous the mixture viscosity decreases with any further increase in the water
fraction. By adopting the continuous phase that results in the lower viscosity for a given
phase fraction therefore suggests that with inversion the mixture follows a more
favourable route. A similar change of viscosity of the dispersion with phase inversion
was found by Pal (1993).
152
The methodology for identifying the inversion point depends on the mixture viscosity
model used. There are a number of dispersion viscosity models in the literature and these
are briefly reviewed below and summarised in Table 6.1.
Author Model Remarks Eq.
Einstein μφ
= 1 + 2.5Φ Monodispersed system (6.2)
(1906, 1911) μc
μφ  0.4 + μ d /μ c 
Taylor (1932) = 1 + 2.5Φ 
 Monodispersed system (6.3)
μc  1 + μ d /μ c 
μφ
=
1 + 0.5Φ
Furuse (1972) Monodispersed system (6.4)
μc 1  Φ2
Yaron and Gal-Or
Cell model, spherical drops (6.5)
(1972)
Choi and - -
Cell model, spherical drops (6.6)
Schowalter (1975) - - - -
μφ
= 1  Φ 2.5
Brinkman (1952)/
Polydispersed system (6.7)
Roscoe (1952) μc
C  max
Krieger & μφ  Φ  Intrinsic viscosity,
Dougherty (1959) = 1   [C]=-2.5 (6.8)
μ c  Φ max  Monodispersed system
Phan-Thien and Polydispersed system,

(6.9)
Pham (1997) spherical drops
μφ  2.5Φ  Crowding factor.

Mooney (1951) = exp   1.35<k<1.91 (6.10)
μc  1  kΦ  Monodispersed system
Pal and Rhodes

Semi-empirical correlation (6.11)
(1989)
Pal Eq. 16 (2001) (6.12)
Pal Eq. 27 (2001) Based on Brinkman/Roscoe (6.13)

(1952)
Pal Eq. 29 (2001) Based on Frankel and Acrivos (6.14)

(1967)
Table 6.1: Literatures for dispersion viscosity models.
153
6.3. LITERATURE MIXTURE VISCOSITY MODELS
Various models for calculating the viscosity of liquid-liquid dispersions can generally be
grouped into three main categories, namely linear, exponential and power function
models. A linear dispersion viscosity model was first suggested by Einstein (1906, 1911)
for an infinitely dilute suspension of small solid spheres (Equation (6.2), Table 6.1).
However, this theory would not account for non-spherical particles at high concentrations
or for particle size distributions. Still for near spherical drops and infinite dilution Taylor
(1932) extended Einstein's equation to fluid dispersions (Equation (6.3), Table 6.1) by
considering the internal circulation in the drops caused by the tangential stresses on the
drop surface. Equation (6.3) can be reduced to Einstein's model for solid particles
when d>> c while a corresponding lower limit is for gas bubbles where μ d << μ c .
Attempts to extend Einstein's theory to high dispersion on entration, , fa ed many

challenges. Theoretical approaches based on a reflection technique led to power series
models, as generally described by Equation (6.1).
μφ
= 1 + K1Φ + K 2 Φ 2 + K 3Φ 3 +  (6.1)
μc
The contributions of the high order terms, though, in these models are extremely complex
to estimate and only the first order terms can be evaluated which result in equations
similar to that by Einstein (e.g. Guth and Gold, 1938). Furuse (1972) considered the
importance of the hydrodynamic effects of neighbouring particles in a concentrated
solution. By extending Einstein's model, a power law model was derived (Equation (6.4),
Table 6.1) to account for the higher order terms that were ignored by Einstein. An
alternative approach was the cell model (Yaron and Gal-Or, 1972, Equation (6.5), Table
6.1; Choi and Schowalter, 1975, Equation (6.6) Table 6.1) in which a certain number of
drops are confined within a representative cell to achieve a phase fraction that resembles
the concentration in the bulk surrounding the cell. However, Yaron & Gal-Or’s
expression failed to redu e to Taylor’s equation in the limit of → Pal,
154
Brinkman (1952) and Roscoe (1952) followed a slightly different approach by accounting
for the incremental change in viscosity caused by the addition of one extra solute particle
to a dispersion of known concentration (Equation (7), Table 1). As there is no assumption
on the shape and size of droplets, this model allows polydispersity but the interactions
between adjacent particles when they are closely packed are ignored. Krieger and
Dougherty (1959) applied Mooney's (1951) concept of crowding factor (which is
inversely proportional to the maximum drop packing fraction defined as the maximum
dispersed phase fraction in a confined space assuming monodispersed spherical drops) to
Brinkman's equation, in order to include the effect of close packing on the mixture
viscosity (Equation (6.8), Table 6.1). Pham-Thien and Pham 99 extended Brinkman’s
approach to evaluate the effect of a dispersed drop size distribution (Equation (6.9), Table
6.1).
In terms of exponential models, Vand (1948) developed a kinematic model by Couette

streaming of dispersion mixture. As the rate of shear is directly proportional to the
dispersion concentration, the mixture viscosity was found to be an exponential function
of the concentration. Mooney (1951) modified Vand's approach by including a crowding
factor (Equation (6.10), Table 6.1). Mooney's model was reported to correlate well with
various experimental data. This approach is similar to that by Arrhenius (1887). Barnea
and Mizrahi (1973) presented a generalised exponential equation, identical to that by
Mooney based on the empirical solution by Thomas (1965). The authors have
subsequently added a term to include the effect of internal drop circulation. From Vand to
Barnea and Mizrahi, these models have marked the development of exponential functions
for mixture viscosity.
Pal and Rhodes (1989) and Pal (2001) suggested a number of viscosity models (some are
given in Equations 6.11-6.14, Table 6.1). Pal (2001) evaluated various single parameter
viscosity models against experimental data and found that Yaron and Gal-Or (1972)
performed best for dispersed phase volume fractions up to about 55%. He also suggested
a number of new two parameter power and exponential function viscosity models but
found that, in overall, the model given by Equation 6.14 in Table 6.1 performed best.
155
Models used in the current study are presented in Table 6.1. Although all models were
tested using the proposed methodology for predicting phase inversion, only selected ones
are included in the graphs that follow. These were selected because (a) they are simple to
use (e.g. Brinkman/ Roscoe (1952)), (b) have been shown to give the best predictions of
dispersion viscosity (e.g. Yaron and Gal-Or (1972); Equation 6.14 by Pal (2001)) and (c)
represent a category of viscosity models (e.g. linear, exponential and power law).
6.4. COMPARISON BETWEEN PREDICTED PHASE INVERSION

POINTS AND EXPERIMENTAL RESULTS FROM LITERATURE
The phase inversion points predicted by the methodology described above are compared
against available experimental data from oil-water pipeline flows by Ioannou (2006).
Details of the properties of the three oils and of the test sections used can be found in
Table 6.2. Phase inversion was experimentally determined through conductivity
measurements obtained either at the flow perimeter with ring probes or at various
locations (e.g. pipe centre) within the pipe cross section with local conductivity wire
probes. At the operated range of mixture velocity, the liquid-liquid dispersion can be
inhomogeneous at high dispersed phase fraction close to phase inversion and phase
inversion may thus appear at certain parts of the pipe first before it spreads to the whole
pipe cross section. For example in an oil continuous dispersion in a horizontal pipe,
inversion would first occur at the bottom of the pipe where the dispersed water drops tend
to accumulate due to gravity (Ioannou et al., 2004). A hysteresis effect has also been
reported where the phase inversion point depends on the initial continuous phase (oil or
water) which results in an ambivalent range similar to that seen in stirred vessels
(Ioannou et. al., 2005). Some investigators have also found that inversion may not happen
at one particular volume fraction but is a transition occurring over a range of fractions
during which complex multiple dispersion structures form (Liu et al., 2005; Piela et. al.,
2008). Pipe material has also been found to affect the inversion point due to the
wettability of the fluid with the pipe wall. Thus for a given oil-water fluid pair in
dispersed pipe flow there is a range of phase volume fractions (phase inversion range)
156
within which inversion will occur that accounts for the effects of pipe material, size,
initial continuous phase and mixture velocity. Only outside the phase inversion range will
the dispersion be either oil or water continuous with certainty.
Density (kgm-3) Viscosity (mPa s) Pipe Property

Water 1000 1.0
Exxsol D80 796 1.7 60mm ID acrylic and steel pipe
Exxsol D140 828 5.5 38mm ID steel pipe
Marcol 52 835 11 60mm ID acrylic and steel pipe
Table 6.2: Physical properties of fluids and pipes used in the phase inversion
experiments.
The phase fractions where the mixture viscosities between the oil and the water
continuous dispersions become equal (as described schematically in Figure 6.1) for
selected viscosity models from Table 6.1 are shown in Figure 6.2 for the three oils
considered. In these figures (Figure 6.2(a)-(c)), the vertical lines denote the phase
inversion ranges found experimentally for each of these three oil-water pairs and include
data from different pipe materials, mixture velocities and initial continuous phase. In the
implementation of the viscosity models it was assumed that the dispersed drops are
spherical for the whole range of volume fractions, while the maximum packing factor,
max, was taken equal to 0.74 (and 0.637 for the model by Pal (2001)) for a
monodispersed mixture (Yeh et al., 1964) with a corresponding crowding factor, k, of
1.35 i e k / max). From the figures, it is apparent that most viscosity models can
predict the equal mixture viscosities point and hence the phase inversion point within the
experimental phase inversion range. From the models shown, Taylor’s linear model
(Equation 6.3) falls, as expected, outside the range. Additionally, the exponential model
by Mooney (Equation 6.10) always predicts the phase fraction to be 0.5 regardless of the
oil used. All the other viscosity correlations shown in Table 6.1 have also been tested and
it is found that power function models give better predictions. It can also be seen that the
experimental range of water volume fractions where inversion occurs shifts to lower
values as the oil viscosity increases which agrees with previous findings (Arirachakaran,
157
1989; Selker & Sleicher, 1965; Yeh et al., 1964); the predicted inversion points follow
the same trend.
The predicted phase inversion points for Marcol 52 are compared in Figures 6.3 - 6.5
against the phase inversion ranges obtained under different experimental conditions. The
effect of dispersion initialisation conditions is demonstrated in Figure 6.3, where it can be
seen that the inversion range depends on which phase is initially continuous. Mixture
velocity is also found to affect the boundaries of the phase inversion range and this is
demonstrated in Figure 6.4. With increasing flow velocity, the inversion region width
decreases as the dispersion becomes more homogeneous. While a transitional region in
the experimental outcome is maintained in the current operation conditions, a clear trend
of narrowing of this region to a specific phase fraction is observed if the mixture velocity
increases leading to higher turbulence and homogeneous dispersion (similar to Brauner
and Ullmann, 2002, and Chakrabarti et. al., 2006). Thus, the prediction of a phase
inversion point through the dispersion viscosity still valid as long as it falls within the
transitional region. The pipe material effect on the inversion range is shown in Figure. 6.5.
The phase inversion range is shifted to lower water fractions in the acrylic pipe compared
to the steel one. In all the above cases the predicted phase inversion points (equal mixture
viscosities), when the viscosity models by Brinkman/Roscoe (1952), Furuse (1972) and
Equation 6.14 by Pal (2001) are used, consistently fall within the phase inversion ranges.
Similar results were found for the other two oils investigated.
158
Figure 6.2: Predicted inversion points using different dispersion viscosity models against
the experimental phase inversion range for a dispersion of water and (a) Exxsol D80
(1.7mPas), (b) Exxsol D140 (5.5mPas), (c) Marcol 52 (11mPas) (data in stainless steel
pipe by Ioannou, 2006). Points in graph refer to the model predictions while the lines
represent the experimental phase inversion region.
159
the experimental phase inversion ranges in stainless steel pipe obtained starting from oil
continuous and from water continuous dispersions for a water-Marcol 52 mixture. Points
in graph refer to the model predictions while the lines represent the experimental phase
inversion region.
the experimental phase inversion ranges in stainless steel pipe obtained at different
mixture velocities for a water-Marcol 52 mixture. Points in graph refer to the model
predictions while the lines represent the experimental phase inversion range.
160
the experimental phase inversion ranges obtained with different test sections for a water-
Marcol 52 mixture. Points in graph refer to the model predictions while the lines
represent the experimental phase inversion range.
6.5. COMPARISON OF THE PROPOSED METHODOLOGY WITH

LITERATURE CORRELATIONS
The phase inversion points predicted with the equal mixture viscosities methodology for
the three oils used are compared against those found from the models by Yeh et al.
(1964), Arirachakaran et al. (1989) and Nädler and Mewes (1997) (see Table 6.3). From
Figure 6.6, it can be seen that only Yeh et al.'s model (Equation (6) in Yeh et al. (1964))
predicted phase inversion points fall within the phase inversion ranges for all oils studied.
In addition, the proposed methodology, using the Brinkman/Roscoe (1952) viscosity
model, is found to predi t well Yeh et al ’s own experimental data that included a large
range of oil viscosities (see Table 1 in Yeh et al. (1964)), with error ranging from 0.08%
(water-Chlorobenzene) to 8% (water-Nitrobenzene) and an average error of
approximately 3% across the different organic-aqueous fluid pairs. The deviations were
similar to those of Yeh et al ’s own orrelation Equation (6) in Yeh et al. (1964)), while,
interestingly, the inversion points found by the current methodology were closer to the
161
predictions of a modified equation suggested by Yeh et al. (Equation (6a) in Yeh et al.
(1964)) where the interfacial viscosity is used instead of the water phase viscosity. To
develop his model, Yeh et al. assumed that there are no interfacial shear stresses at the
point of phase inversion indicating a momentum balance at the fluid interface. The
proposed methodology can also predict well with a wide range of oil/water system
reported in Arirachakaran et al. (1989).
In Figure 6.7, the proposed method is compared with the experimental data from both
Yeh et al. (1964) and Arirachakaran et al. (1989). It is also compared with the suggested
models proposed by the respective authors. From the figure, it can be shown that the
models from Yeh et al. (1964) and Arirachakaran et al. (1989) can only predict limited
range of viscosity ratio with Yeh et al. predicting well with the lower viscosity range and
Arirachakaran et al. (1989)’s at the higher viscosity range. In comparison, the proposed
method can predict very well across the wide range of viscosity ratio from an organic
viscosity of 0.235mPa.s and about 1500mPa.s. It demonstrates that the phase inversion
point is not a linear function of the logarithmic relation with the viscosity ratio. In fact, it
appears to reach an asymptote of about 0.23 for water fraction as the viscosity ratio
increases. This may be because the organic phase in the lower viscosity pairs can be
deformed more easily than that of the higher viscosity. This leads to the prediction of Yeh
et al. (1964)’s model to a lower water fra tion for phase inversion as it extrapolates
across the range of viscosity ratio tested. However, the increase in viscosity ratio is likely
to affect the deformation of the dispersed phase leading to a maximum packing that the
dispersed phase can sustain and beyond which a phase inversion will occur.
162
Figure 6.6: Comparison between proposed methodology (suggested best models) and
models for critical phase fraction from literature.
Author Model Eq.
Arirachakaran et al. (1989) (6.15)
Nädler and Mewes (1997) (6.16)
Yeh et al. (1964) (6.17)
Yeh et al. (1964) (6.18)
Table 6.3: Critical phase fraction models for phase inversion from literature.
163
Figure 6.7: Application of proposed model on various systems from Yeh et al. (1964)
and Arirachakaran et al. (1989). The solid line represents the prediction model by Yeh et
al. (1964) (see Equation 6.17) and the dotted line represents the prediction model by
Arirachakaran et al. (1989) (see Equation 6.15).
6.6. REVIEW ON PREDICTION MODEL WITH AN

EXPERIMENTAL HOMOGENEOUS DISPERSED FLOW
The proposed method for phase inversion prediction is now tested against the phase
inversion data obtained in the experimental facility using the dispersed inlet. As shown in
Chapter 4 with this inlet setup, the flow is fully dispersed for all phase fractions at the
mixture velocity of 3m/s. As a result, the change from a water continuous dispersion to
an oil continuous dispersion (and vice versa) occurs throughout the pipe cross section at a
particular phase fraction rather than over a transitional region as in the case of a split inlet.
The experimental phase inversion points from the dispersed inlet are therefore well suited
164
for comparison with the phase inversion prediction model since the mixture viscosity
models used are derived from homogeneously dispersed mixtures.
In order to compare with the prediction model, the pressure gradient from the prediction
model is firstly al ulated using Brinkman 9 ’s mixture vis osity equation and apply
that to a single phase mixture equation for pressure gradient (Equation 6.19).
P 2f m U 2m
 (6.19)
l D
where ∆P/l is the pressure gradient a ross a pipe, f is the fri tion fa tor f 6Re m-0.2),
ρm is the mixture density, Um is the mixture velocity and D is the pipe diameter. The
calculated pressure gradient is then compared with the experimentally measured pressure
gradient together with the conductivity data on phase continuity switch.
From Figure 6.8, it can be observed that the experimental pressure gradient closely
resemble the trend deduced by the correlation (Equation 6.19) between 38% and 100%
water fraction. The difference between the experimental result and model outcome may
be due to the presence of drag reduction that is not accounted for in the correlation. There
is a sharper drop in pressure gradient for the experimental result between 26% and 38%
water fraction and subsequently an increase in pressure gradient below 26%.
Based on the prediction from the interception point in Figure 6.8, the predicted phase
inversion point (at 34%) falls in the phase fraction range where the pressure gradient is
decreasing. From the investigation outcome in Chapter 4, this lies at the range of phase
fraction where drop coalescence becomes prominent with likely formation of complex
structures. While the conductivity measurements detects a change in phase continuity at a
lower phase fraction (26%), the predicted phase inversion point can be considered to be
within close proximity because the process of drop coalescence can lead to the formation
of oil pockets (if starting from an O/W dispersion) as presented in Piela et al. 2008. These
oil pockets may not be easily detected by the conductivity wire and ring probe but
nonetheless an important stage in the onset of phase inversion. As such, the prediction
can be considered to have good prediction with the experiments conducted.
165
4.0
Pressure gradient [kPa/m] 3.5
3.0
2.5
2.0
1.5
1.0 Model (Oil continuous)
0.5 Model (Water continuous)
Experimental
0.0
0 20 40 60 80 100
Figure 6.8: Comparison of pressure gradient at different phase fraction between

experimental data □ and modeling out ome via Brinkman/Ros oe 9 ’s mixture
vis osity orrelation The solid arrow ← represents the dire tion for the experiment and
the vertical line denotes the predicted phase inversion point.
6.7. CONCLUSION
A methodology has been introduced to identify the inversion point during oil-water
dispersed flow. It is suggested that inversion will occur at the phase fraction where the
mixture viscosities of the oil and the water continuous dispersions become equal. The
approach was tested against extensive available experimental data on phase inversion
where oil viscosity, pipe material, mixture velocity and dispersion initialisation
conditions were varied. From the available literature dispersion viscosity models tested,
the power function correlations by Brinkman (1952)/Roscoe (1952), Furuse (1972) and
Equation (14) by Pal (2001) always predicted inversion within the experimental phase
inversion range. The proposed methodology was also tested against the predictions of
literature correlations on phase inversion point. Very good agreement was found across a
wide range of two phase systems with organic viscosity of 0.235mPa.s and about
1500mPa.s from Yeh et al. (1964) and Arirachakaran et al. (1989).
166
The suggested prediction method with Brinkman (1952)/Roscoe (1952) mixture viscosity
model was subsequently tested with experimental data of an homogeneous dispersed flow.
Similar increasing mixture viscosity can be observed as the dispersed phase fraction is
increased. The method shows closely at the phase fraction (i.e. 36% water fraction) where
the mixture viscosity is highest (and resulting in a corresponding maximum pressure
gradient). Below 34% water fraction, the dispersion appears to encounter drop
deformation/coalescence which leads to the complete change in phase continuity at 26%
water fraction. Pressure gradient of an inverted W/O system subsequently shows a
significant increase due to the higher viscosity by the continuous oil phase.
From this chapter, a conclusion can be drawn that the viscosity of the liquid-liquid
mixture is important for understanding phase inversion. The prediction method falls
within close proximity with the experimentally detected change in phase continuity.
Despite that there is a discrepancy within the two values, the predicted phase inversion
point highlights a point where drop coalescence/deformation becomes dominant in the
system and the coalescence process is a key factor for the inversion process. Further work
is needed to understand the changes in the mixture morphology as it approaches phase
inversion. Such studies will help to better describe the phase inversion phenomenon and
will also improve the model by including, for example, the drop packing and size
distribution.
167
Chapter 7: Prediction of pressure gradient
during phase inversion
7.1 OVERVIEW
Multiphase flow of two or more fluids is commonly encountered in the petroleum

industry during oil production and transportation. Water often appears in oil wells
particularly in the later years of production. At high velocities, the mixture of oil and
water is transported as a dispersion with either oil (water-in-oil, W/O) or water (oil-in-
water, O/W) as the continuous phase. The type of dispersion depends on the phase ratio,
rheological properties of the two fluids, wetting properties of the pipe material and the
operation initialisation conditions (Yeo et. al, 2000; Ioannou, 2006). Interestingly, the
continuous phase of the dispersion changes under certain conditions to become the
dispersed one and vice versa (phase inversion phenomenon). The phase fraction where
this change occurs is called the phase inversion point. During pipeline flow, however, it
has been found that inversion occurs over a range of input phase fractions. This
transitional range can be attributed to phenomena such as formation of multiple
dispersions (Liu et al., 2005) and partial or local inversion as the dispersed phase fraction
increases (Ioannou, 2006; Piela et al., 2008). Partial inversion results in stratification of
the flow with two layers that have different continuous phase.
The changes in the spatial distribution of the phases as their volume fractions change can
have significant impact on pressure drop and pose great challenges in its modelling. To
predict pressure drop in dispersed oil-water flows, the homogeneous model is often used
where the fluids are considered as one “pseudofluid” and single phase flow equations are
used with a weight averaged mixture density and an empirically determined mixture
viscosity (viscosity models by e.g. Brinkman, 1952; Yaron & Gal-Or, 1972; Choi &
Schowalter, 1975; Pal, 2001). However, the homogeneous model does not accurately
represent the phase distribution in horizontal pipelines where stratification due to gravity
can appear even at high mixture velocities. To account for the separation of the two fluids,
168
the two-fluid model (TFM), developed for stratified flows (e.g. Brauner & Moalem
Maron, 1989), can be applied.
In this chapter, experimental results on phase inversion and pressure drop during
horizontal oil-water dispersed flows are presented (Section 7.2). A methodology (see
Section 7.3) is developed based on a two layer flow configuration, in order to predict the
phase volume fractions where the mixture is completely dispersed or there is stratification
of the two fluids. The entrainment fractions in each layer are calculated from previously
developed correlations, while a phase inversion criterion is used to determine the
continuity of each continuous layer in the model. The outcome of the developed model
will be compared with experimental results in Section 7.4 and conclusions on the quality
of the model are made in Section 7.5.
7.2 EXPERIMENTAL INVESTIGATION
Experimental studies on phase inversion were conducted in the pilot scale liquid-liquid
flow facility (See Chapter 3). The test section is an acrylic pipe with 38mm ID. Exxsol
D140 oil (density of 828 kgm-3, viscosity of 5.5mPa.s) and tap water were used as test
fluids. The two fluids were pumped from storage tanks via variable area flow meters into
the test section inlet where they were joined in a modified Y-junction to ensure minimum
mixing. Experiments were conducted at mixture velocities 3m/s and 4m/s where previous
work indicated that the flow was in the dispersed regime. At each velocity the
experiments were started with single phase water. Oil was then introduced into the pipe.
The oil flowrate was gradually increased while at the same time that of the water
decreased in order to increase the oil fraction while keeping the mixture velocity constant.
The experiments continued until phase inversion was observed and beyond until about 80%
input oil fraction. Pressure drop was measured via a differential pressure transducer
(Validyne DP103) between two measuring ports 1.5m apart, with the first port located at
3.8m (~100D) from the inlet. A conductivity wire probe was used to identify phase
continuity at different locations (i.e. every 2mm intervals) along the vertical diameter in a
pipe cross section and thus to understand the development of phase inversion. The probe
was located at 7m (~184D) from the inlet.
169
3.5
3.0
Pressure gradient (kPa)
2.5
2.0
1.5 Full W/O Full O/W

Partial
Dispersion Dispersion
1.0 Inversion
0.5
0.0
0 20 40 60 80 100
Input water cut (%)
(a) 3m/s mixture velocity
6.0
5.0
Pressure gradient (kPa)
4.0
3.0
Full W/O Partial Full O/W
2.0 Dispersion Dispersion
Inversion
1.0
0.0
0 20 40 60 80 100
Input water cut (%)
(b) 4m/s mixture velocity
Figure 7.1: Pressure gradient at mixture velocity (a) 3m/s and (b) 4m/s measured
experimentally. The solid lines represent the boundaries of the transitional region
between water continuous and oil continuous fully dispersed flow. The arrow indicates
the direction of the experiment.
170
The pressure gradients measured at mixture velocities of 3m/s and 4m/s are depicted in
Figure 7.1 (a) & (b) respectively. At high water cuts the conductivity probe showed that
the mixture was water continuous in the whole pipe cross section. By gradually
increasing the oil fraction (as indicated by the direction of the arrow) the flow remains
fully dispersed with water as the continuous phase until about 60% water cut in both
mixture velocities, while the pressure gradient decreases. This indicates a drag reduction
phenomenon that has also been reported before (see Pal, 1993). Beyond this volume
fraction, partial inversion takes place at the top part of the pipe and the pattern is now
dual continuous with a water continuous layer at the bottom of the pipe and an oil
continuous layer at the top of the pipe, as indicated by the measurements of the
conductivity probe. At about 34% water cut, the pressure gradient reaches a minimum
and with further decrease in the water fraction it increases again. This increase coincides
with the change of the flow pattern from dual continuous to fully dispersed with oil as the
continuous phase. In these experiments inversion from the O/W to the W/O dispersion
occurs through a transitional region which is enclosed in Figure 7.1 between the solid
vertical lines.
7.3 DEVELOPMENT OF MODEL FOR THE PREDICITON OF

PRESSURE DROP
A methodology was developed to predict the flow configuration, dual continuous or fully
dispersed, as well as the pressure gradient during the flow of liquid-liquid mixtures.
7.3.1 PREDICTION OF FLOW CONFIGURATION
For each phase fraction it is assumed initially that the mixture has a dual continuous
configuration with a water continuous layer with oil drops dispersed in it at the bottom
and an oil continuous layer with water drops dispersed in it at the top of the pipe. A range
of interfacial heights is then considered with an interval of 0.1mm. At each interfacial
height, the dispersion in each layer is assumed to be homogeneously distributed. The
amount of dispersion of one phase into the other is calculated using the entrainment
model described in Section 7.3.2. By comparing the phase fractions in each layer with a
171
criterion for the phase inversion point, the continuous phase in that layer can be
determined. The phase inversion point (WFCritical) is calculated using the methodology
developed by Ngan et al. (2009) according to which phase inversion will occur at the
phase fraction where the viscosities of the two possible dispersions, oil continuous and
water continuous, are equal. For the Exxsol D140-water mixture used in this work, the
critical water fraction for phase inversion is found to be 34%. According to this, the
possible scenarios for the two layer mixture are:
Water entrainment in Oil entrainment in

Flow Pattern Valid
upper (W/O) layer (Entu) lower (O/W) layer (Entl)
upper lower
layer layer
Entl<(1- WFCritical) W/O O/W Yes
Entu<WFCritical
(1- WFCritical)<Entl<1 W/O W/O Yes
Entl<(1- WFCritical) O/W O/W Yes
WFCritical<Entu<1
(1- WFCritical)<Entl<1 O/W W/O No
Notes:
1) The entrainment fraction cannot be above 1.
2) The less dense oil cannot be continuous underneath the more dense water.
Table 7.1: Flow patterns resulting from the application of a critical water fraction for
phase inversion (WFcritical) in each of the upper and lower layers of a dual continuous
flow configuration.
A two-fluid model, as detailed in Section 7.3.3, is then applied for each interfacial height
only to the valid cases in Table 7.1. The model has a solution when the pressure gradients
of the upper and lower layers are the same within a 2% difference. If there is no solution
of the two fluid model for the entire range of interfacial heights for the particular phase
fraction, then the dual continuous flow configuration cannot exist and the mixture is fully
dispersed. This happens when the comparisons with the critical phase fraction have
indicated that both layers have the same continuity. In that case the homogeneous model
(see Section 7.3.4) is used to predict the pressure gradient.
172
7.3.2 ENTRAINMENT MODEL
To calculate the entrainment fraction in each layer of the dual continuous configuration,
the model developed by Al-Wahaibi and Angeli (2009) was used. According to the model,
drops will form from the crests of unstable waves at the oil-water interface when the drag
force on the waves exceeds the stabilising interfacial tension force. At steady state, the
rate of drop entrainment will be equal to the rate of drop deposition. According to Al-
Wahaibi and Angeli (2009), the rate of entrainment (Rent) can be calculated as follows:
V ρ U U
R  ent d w o (7.1)
ent 2
S λ
i
where ρd is the density of the entrained phase, Vent is the volume of the entrained phase
from each wave crest, λ, is the wave length, which, based on experimental observations
and for simplicity is taken equal to the pipe diameter and S i is the length of the interface.
The rate of deposition (Rdep) is equal to:
 W 
R k  d  (7.2)
dep D  Q S 
 v
where Wd is the mass flow rate of the dispersed phase in the layer, Q is the volumetric
flow rate of the mixture in a layer and S v is the slip ratio between the dispersed drops and
the continuous phase. In this work, no slip is assumed between the dispersed and
continuous phases (i.e. Sv=1). kD is an empirically determined deposition rate constant
which was found by Al-Wahaibi (2006) to be equal to
U 
k  0.051  so   U* (7.3)
D(O/W) U 
 m
173
U 
k  0.077   sw   U* (7.4)
D(W/O) U 
 m 
where Us is the superficial velocity and U* is the frictional velocity. Subscripts, o and w,
represents the oil and water phases.
By equating the rates of entrainment and deposition, the entrainment fraction (W/W) in a
layer can be found as follows:
V ρ U U
ent d w 
o  k  E (1  E)  ρ  E  ρ
D c d
 (7.5)
S  λ2
i
where E is the steady state entrainment fraction, ρ c and ρd are the densities of the
continuous and the dispersed phases respectively. The calculated entrainment fraction is
in w/w and for the application in the two-fluid model it is converted to v/v.
7.3.3 TWO FLUID MODEL
For a steady state, fully developed dual continuous flow configuration with an oil
continuous dispersed layer at the top and a water continuous dispersed layer at the bottom
the one-dimensional momentum equations for each layer are given by:
 dP  S   S 
    τ u   u   τ  i
 i A


(7.6)
 dx  u A
 u   u 
 dP  S  S 
    τ   l   τ   i  (7.7)
 dx  l l  A  i A 
 l  l
174
where τ is the shear stress, S is the perimeter covered by the respective layer and A is the
pipe cross sectional area. In the above equations, subscripts “u” and “l” denotes the upper
and lower layers and subscript “i” denote the interface between the two layers.
The shear stresses in Equations (7.6) and (7.7) can be found as follows:
ρ U2 ρ U2 ρ (U  U ) U  U
τ f u u τ f l l τ f f u l u l
u u 2 l l 2 i i 2
U is the in-situ velocity of the respective layer, ρ denotes the weight-averaged mixture
density, f is the frictional factor and ρf denotes the density of the faster layer. In all cases
investigated both upper and lower layers were in turbulent flow and the friction factors
were calculated from 0.046Re -0.2, where Re is the Reynolds number of the respective
layer. The Reynolds number in each layer is calculated based on the mixture fluid
properties and velocity of that layer. The interfacial Reynolds number is calculated based
on the faster phase properties while the interfacial frictional factor is the maximum
between 0.014 and 0.046Rei-0.2. To account for the drag reduction observed during fully
dispersed flow (dispersion pressure gradient less than that of the single phases at the same
mixture velocity) the following equation is used for calculating a modified friction factor
(Rozentsvaig, 1982).
f
original
f  (7.8)
DR 1  nε
where foriginal is the original friction factor (foriginal = 0.046Re-0.2) and fDR is the modified
friction factor due to drag reduction, n is an empirical coefficient with a suggested value
of 1.125 and ε is the fraction of the dispersed phase in the layer.
175
The mixture density in each layer is equal to the weight averaged densities of the two
phases while for the calculation of the mixture viscosity Brinkman’s (1952) correlation is
used.
ρ  ε  ρ  (1  ε)  ρ (7.9)
m d c
μ  μ  (1  ε)  2.5 (7.10)
m c
7.3.4 HOMOGENEOUS MODEL
According to the homogeneous model the pressure gradient is found from:
2
dP 2fρ m U m
 (7.11)
dx D
where f is the friction factor (0.046Re-0.2), ρm is a weight-averaged density (Equation

(7.9)), Um is the mixture velocity and D is the pipe diameter. Drag reduction is similarly
accounted for by Rozentsvaig’s equation (Equation (7.8)).
7.4. COMPARISON OF MODEL PREDICTIONS WITH

EXPERIMENTAL DATA
The model predictions on the flow pattern that establishes in the pipe and on pressure
drop for the different volume fractions at mixture velocities 3m/s and 4m/s are shown in
Figure 7.2 against the experimental results. Experimentally it was found that the flow at
3m/s is fully dispersed with water as the continuous phase for input water fraction above
62% and with oil as the continuous phase for input water fraction below 28%. In between
28-62% input water fraction there is stratification of the flow with an oil continuous
dispersed layer at the top of the pipe and a water continuous dispersed layer at the bottom,
as found by the conductivity probe. According to the model, flow stratification occurs
between 34-66% while below and above these fractions the mixture is oil and water
176
continuous respectively. Similarly, at 4m/s mixture velocity the experimental transitional
region between the two types of dispersion was between 32-60% input water fraction
while that predicted one was between 33-70% water fraction. The transitional regions for
phase inversion in both experimental and predicted results are within close range.
It can also be seen from Figure 7.2 that the predicted pressure gradients, calculated from
the two-fluid model in the transitional region and from the homogeneous model in the
fully dispersed regions, are in good agreement with the experimental data. The
experimental drag reduction is captured well by the model. The model is also able to
predict the increase in pressure gradient after the mixture becomes fully oil continuous.
Discrepancies between the experimental and predicted data appear at the oil continuous
region and particularly for the lower mixture velocity. It is believed that the discrepancies
are due to lack of uniform drop distribution as assumed in the homogeneous model,
because under the effect of gravity water drops would tend to accumulate closer to the
pipe bottom compared to the top. Dispersion inhomogeneity is expected to be more
pronounced in the oil continuous dispersion compared to the water one because the oil
viscosity is higher and the turbulent mixing will be less intense. In addition, it is more
apparent at the low mixture velocity tested compared to the high one where again mixing
is less intense.
177
4
Pressure Gradient (kPa)

3
Partial Full O/W
Inversion Dispersion
2 Full W/O
Dispersion
1 Experiment
Prediction
0
0 20 40 60 80 100
Input Water Fraction (%)
(a) 3m/s mixture velocity
4
Full O/W
Dispersion
Pressure Gradient (kPa)
2 Full W/O Partial

Dispersion Prediction
Inversion
Experiment
1
0
0 20 40 60 80 100
Input Water Fraction (%)
(b) 4m/s mixture velocity
Figure 7.2: Experimental and predicted pressure gradient at (a) 3m/s and (b) 4m/s
predicted from two-fluid model. The lines represent the boundaries of the transitional
region between water continuous and oil continuous fully dispersed flow, solid for
experimental and dotted for predicted fractions. The arrow indicates the direction of the
experiment.
178
7.5. CONCLUSION
Phase inversion during horizontal oil-water flow was investigated experimentally with
the use of a local conductivity probe that identifies phase continuity. It was found that for
the mixture velocities studied there was a region of flow stratification between the fully
dispersed oil and water continuous flows. A methodology was suggested that predicted
the range of volume fractions where flow was separated and the pressure gradient. This
was based on the assumption of a two layer flow configuration, where the entrainment in
each layer was calculated and the continuity of the layer was decided based on a phase
inversion criterion. There was good agreement between the model predictions and the
experimental data apart from the oil continuous dispersed flow at the low mixture
velocity. This discrepancy was attributed to the inhomogeneous distribution of the
dispersed water drops in the pipe cross section.
179
Chapter 8: CFD simulation of horizontal two-
phase pipe flow
8.1 OVERVIEW
It was shown in the previous chapters that the spatial distribution of oil and water can be
significantly changed according to the input phase fractions. This consequently leads to a
change in flow regime (e.g. from a fully oil-in-water dispersion to a dual continuous
mixture). In this chapter, computational fluid dynamics (CFD) simulations will be
developed to predict phase distribution and flow characteristics for some of the
conditions used in experiments in Chapter 4. The results will be compared against the
experimental data. Use of CFD can provide information on flow characteristics that are
not easy in many cases to measure experimentally and thus offer insight on flow behavior
in combination with experiments. Validated CFD models can also be used for prediction
without the need to resort to experiments. Various commercial CFD codes have been
developed with multiphase flow capability (e.g. CFX, Fluent, Star-CD). In the current
study, CFX will be used. The outcome of the chapter will aim to have a better insight on
the effect of the different interphase force terms on the distribution of the mixture. The
chapter is organised in the following sections: description of CFX software package
(Section 8.2); description of the geometry and conditions simulated (Section 8.3);
comparison between the simulation and the experimental results (Section 8.4). In the
conclusions section, criteria in running a simulation of dense two-phase pipe flow will be
presented (Section 8.5).
8.2 DESCRIPTION OF ANSYS CFX SOFTWARE PACKAGE
The chosen CFD software, Ansys CFX, by Ansys Inc., is a commercial computational
fluid dynamics software package that has been widely used in solving a wide range of
fluid flow problems. The code has been applied to a number of multiphase flow systems
such as Krepper et al., 2005, Lucas et al., 2007 and Vallée et al., 2007. Many models
180
have been implemented to predict the various momentum transport terms allowing users
to predict the respective influence by the phases especially during dispersed flow.
Ansys CFX consists of four main modules:
 Geometry construction and Meshing Module (via Ansys ICEM CFD)

 Pre-processing Module (CFX-Pre)
 Solver Module (CFX-Solver)
 Post processing Module (CFX-Post)
The construction of the geometry (i.e. the pipe) and the creation of a computational grid
to define the points for computation (commonly known as meshing) are done via the
Ansys ICEM CFD module. The geometry is constructed in scale with the actual
construction process similar to the way a CAD package is used. Once the geometry is
constructed, the structure is subsequently divided into a grid with specified grid points
where the computation of the fluid motion is conducted. As the whole process is operated
interactively via the Graphical User Interface (GUI), the mesh can easily be refined at
specific locations (e.g. close to pipe wall) to improve predictions.
Through the CFX-Pre interface, the boundary and operation conditions of the simulation
are defined on the meshed structure (e.g. fluid properties, input phase fraction, phase
continuity, fluid velocities, wall roughness, and turbulence). A definition file with the
specified conditions and the system geometry will then be saved for the subsequent
calculations.
The solver module involves integrating the governing equations over the defined mesh
grid points using the finite volume method. This permits the mass, momentum and
energy terms to be conserved within each control volume. In Ansys CFX, a co-located
(non-staggered) grid layout is used whereby the control volumes for the pressure and
velocity transport equations are identical. Figure 8.1 presents the iterative method for
solving the governing equations for both transient and steady-state simulations.
181
Figure 8.1: Schematic diagram on the iterative method for solving the conservation
equations (Ansys CFX Manual).
182
The post processing module provides the environment to evaluate visually the results of
the simulation through vector, contour or streamline plots. Raw numerical data can also
be plotted into graphs (e.g. velocity profile, phase distribution profile). Numerical data
can be exported (as a whole domain or within specific region) to Excel for further
processing.
8.3 MODEL GEOMETRY AND SIMULATION CONDITIONS
Model Geometry
Due to the high computational effort for multiphase flow simulations, the geometry
adopted in this chapter is 2-dimensional. Due to the limitation of Ansys CFX in true 2D
simulation, the geometry used in this work is a rectangular slab with a width equal to 1%
of the pipe diameter. The final geometry of the 2D pipe is 3.0m in length and 19mm in
height (i.e. equivalent hydraulic diameter for a 38mm 3D pipe as according to Russell
and Charles (1959) and Agtersloot et al. (1996)). A length of 3.0m was chosen as
previous simulations for the same flow velocity showed that the flow is fully developed
beyond 2.5m. At the inlet, a splitting plate is used and located at the centre of the pipe
cross section with 100mm in length. This resembles the inlet configuration used in the
experimental setup. The grid points along the radial direction are non-uniform with more
grid points along the pipe wall. The shortest space between two grid points is 0.5mm and
the longest space is approximately 0.75mm (at the centre of the pipe). Figure 8.2 presents
the schematic of the pipe section with the boundaries used in the simulations. A total grid
size of 183000 is used after a grid sensitivity study (see Appendix C) was conducted by
adjusting the number of cells on the vertical height of the slab.
183
(a)
Oil
Outlet
Water
(b)
Figure 8.2: (a) Schematic diagram of the grid structure for the 2-D pipe section for the
CFD simulations. (b) Part of the mesh of the 2-D pipe section with the thick lines
representing the smooth walls in the simulation.
184
Boundary conditions
The following boundary conditions are used to define the solution domain:
1. Inlet: Two inlet sections are set up in the geometry separated by the plate – oil enters
the pipe from the top section and water from the bottom. The fluid velocity is set at the
boundary according to the relevant input phase fraction required. The inlet velocity is
uniform over the inlet cross section and normal to the inlet boundary. Equation 8.1
presents the inlet velocity of phase k (i.e. either oil or water).
U k  α k  Qm  A k (8.1)
where U is the input velocity, α is the phase fraction and A is the occupied area of the
pipe by phase k. Qm is the total volumetric flow rate of the mixture.
2. Outlet: A pressure specific boundary condition is used in the simulation. A reference

pressure is set equal to the atmospheric pressure.
3. Wall: The wall boundary conditions include the wall surface of the pipe and the
splitting plate at the inlet. No slip conditions are set at the walls. The wall is also assumed
to be smooth.
4. Symmetry planes: The two side planes are assumed to be symmetric. This imposes
constraints on the conditions of flow on either side of the plane.
Inlet and initial conditions
The properties of the two fluids introduced used in the simulation are the same as those
used in the experiments (i.e. Exxsol D140 and water). The two fluids are regarded as
incompressible with constant density. The inlet phase fraction is set by adjusting the flow
velocities of the phases while maintaining the mixture velocity to be constant at 3m/s. As
an initial condition, the decision for the phase continuity is based on the phase inversion
point calculated in Chapter 6 (i.e. 34% water fraction). Below 34% water fraction, water
is regarded as the dispersed phase and vice versa. The dispersed phase will be introduced
185
into the simulation inlet despite that it is flown as a single phase through the inlet section
of the slab. The drop size for the dispersed phase is taken to be 1.0 mm (unless otherwise
stated) as it has the highest frequency in a drop size distribution based on previous
experiments (see Chapter 4). It is also assumed that the dispersed phase drops are
spherical with uniform size for simplicity in this study. In addition, all simulations are run
at a constant temperature and no thermal effect on the fluid will be considered.
The initial turbulence level is set by estimating the turbulence intensity. In all simulations,
the intensity is taken to be 5% (i.e. the default value in CFX) which is also close to the
calculated value for experimental results according to the following equation.
I  0.16 Re m1 / 8 (8.2)
From the initial turbulence intensity, the initial turbulent kinetic energy (k) and the
energy dissipation (ε) can be found.
2
3 u  k3 2
k   c I    C 3 4
2  c  and L (8.3)
where L is the turbulent length scale that can be estimated by L=0.07D.
Convergence of solution
Convergence of the simulation is ensured by monitoring the residuals (i.e. the difference
in solution between the current step and the previous step) of the solutions of the
hydrodynamic equations. Convergence has been reached when the residuals are 10-6 or
less. The time step for each iteration is set to at least below 0.002s. With the current setup,
this ensures that the Courant number to be within 1 or less. This is suggested to be the
optimal configuration for numerical calculation with the current grid size.
186
8.4 RESULTS AND DISCUSSION
CFD simulations were carried out for a mixture velocity of 3m/s and three input water
fractions – 40% (dual continuous dispersion), 60% (dense O/W dispersion) and 80%
(dilute O/W dispersion). Results on phase distribution will be compared against the
experimental data (see Chapter 4). The phase distribution data along a vertical pipe
diameter for these water fractions obtained by the electrical resistance tomographic (ERT)
system are shown in Figure 8.3.
At 80% input water fraction, the dispersion is almost homogeneously distributed along
the whole vertical diameter. With the decrease in water fraction to 60%, inhomogenity
arises as the dispersed oil drops start accumulating at the core of the pipe. Further
decrease in water fraction (40%) leads to dual continuous.
1.0
Normalised distance from pipe centre
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6 80%
60%
-0.8
40%
-1.0
0 20 40 60 80 100
Oil Fraction [%]
Figure 8.3: Experimental oil phase distribution along a vertical pipe diameter at a
mixture velocity of 3m/s. Data is extracted from ERT measurements at 7m from the inlet
where flow is fully developed.
187
Base case simulation
The following simulation is set as the base case before the effects of various parameters
are explored in subsequent sections. A mixture velocity of 3m/s is used and an input
water fraction of 80% is set while oil is the dispersed phase. The drop size is set at 1.0mm.
No lift force is applied. The Lopez de Bertodano model is chosen for the turbulence
dispersion and the Ishii Zuber model is chosen for the drag coefficient. Figure 8.4
presents the result of the base case against the experimental data.
1.0
0.8
Normalised distance from Pipe Centre
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
Experiment
-0.8
Simulation
-1.0
0.00 0.20 0.40 0.60 0.80 1.00
Oil Fraction [-]
Figure 8.4: Comparison between simulated (base) and experimental (ERT) dispersed
phase fractions at an input water fraction of 80% and mixture velocity of 3m/s. The
simulated data is extracted at 3m from the inlet and the experimental data at 7m where
the flow is fully developed.
As can be seen, the simulated phase distribution shows more stratification of the flow
than the experimental one. There is a higher concentration of oil drops at the top of the
pipe. In the following sections, the effect that the various interphase forces have on the
phase distribution will be explored. As mentioned in Chapter 2 that the Ishii and Zuber
188
(1979) model is the only model for dense dispersion within CFX, no investigation on the
effect of drag force will be conducted.
Effect of lift force coefficient on phase distribution
The effect of lift force on phase distribution will be investigated here. The lift force is
proportional to the cross product between the relative velocity and the curl of the
continuous phase velocity (Equation 2.37). A positive cross product will lead to an
upward lift force by the continuous phase on the dispersed drops and a negative cross
product will lead to a lift force in the opposite direction. Figure 8.5 presents the effect of
the lift force coefficient on phase distribution. With a negative lift force coefficient (LF_-
0.5), a stratification of the phases can be observed. However, the difference between the
case with no lift force (LF_0) and a positive lift force coefficient (LF_0.5) is not
significant except at the lower part of the pipe where some homogenization of the
dispersion can be observed.
1.0
Normalised distance from pipe centre [-]
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
LF_-0.5
-0.6 LF_0
-0.8 LF_0.5
Experiment
-1.0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
Oil Fraction [-]
Figure 8.5: Effect of lift force on phase distribution at mixture velocity of 3m/s and input
water fraction of 80%. LF in the legend represents the lift coefficient applied to Equation
2.37.
189
To further understand the impact of lift force on phase distribution, Figure 8.6 presents
the lift force at different locations along the vertical diameter. The lift force is calculated
from Equation 2.37 assuming that the continuous phase is flowing only in the axial
direction. Other velocity components of the continuous phase are negligible and will not
affect the curl Uc term. A negative lift force coefficient (LF_-0.5) leads to a positive lift
force throughout the pipe. As a result, there is an upward force acting on the drops, which
is stronger near the walls that explains the stratification of the phases seen in Figure 8.5.
With a positive lift force coefficient (LF_0.5), the lift force is negative apart from the
region close to the lower pipe wall. This implies an inward drive of the dispersed drops
away from the bottom wall. The fact that the lift force acts on the opposite direction at the
lower part of the pipe can explain the greater homogenization of the flow at this region
(Figure 8.5).
1.0
0.8
0.6
0.4
0.2
0.0
-0.2 ds_1_LF_-0.5
-0.4 ds_1_LF_0
-0.6 ds_1_LF_0.5
-0.8
-1.0
-0.4 -0.2 0 0.2 0.4 0.6 0.8
Lift force [N]
Figure 8.6: Effect of lift force coefficient on the lift force term at a mixture velocity of
3m/s and an input water fraction of 80%. ds represents the drop size in mm and LF
represents the lift force coefficient applied to Equation 2.37.
190
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6 Experiment
ds_0.5
-0.8 ds_1.0
ds_2.0
-1.0
0.00 0.10 0.20 0.30 0.40
Oil fraction [-]
Figure 8.7: Effect of drop size on phase distribution at a constant lift coefficient (LF=0.5)
at a mixture velocity of 3m/s and input water fraction of 80%. ds represents the drop size
in mm.
According to Tomiyama et al., 1995, drop size is used to determine the Eötvös number
and influence the lift force coefficient. The effect of drop size is presented in Figure 8.7.
Interestingly, it is observed that in general the effect of drop size is small and that a
dispersion with larger drop size (i.e. ds_2.0) is only slightly more homogenous than the
small size dispersions. The effect is more distinctive at the regions between
0.4R<R<0.8R and -0.8R<R<-0.4R. No noticeable difference is seen at the centre of the
pipe (i.e. -0.4R<R<0.4R). The lift force term is plotted for different drop sizes in
Figure 8.8. It can be seen that the magnitude of lift force increases with drop size while
its direction of motion remains the same for all three cases i.e. negative along the whole
vertical pipe diameter apart from the region close to the bottom wall where it is positive.
This implies that the lift force will be more effective in pushing the large oil drops away
from the top wall where they would tend to concentrate due to gravity and help the
homogenization of the mixture. However, the opposite direction of the force close to the
191
lower wall also prevents the drops from accumulating in that area. In addition, the
volume of a large 2mm drop is 64 times the volume of a small 0.5mm drop. This implies
that any movement of a large drop will lead to a much larger change in phase fraction
compared to a small drop. The large change in dispersed phase fraction will likely lead to
counteracting forces that are related to concentration gradient (such as the turbulent
dispersion forces) tend to keep the dispersion mixed.
1.0
Normalised distance from the pipe centre [-]
0.8
ds_0.5_LF_0.5
0.6 ds_1_LF_0.5
0.4 ds_2.0_LF_0.5
0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0
-0.4 -0.2 0 0.2 0.4
Lift force [N]
Figure 8.8: Effect of drop size on the lift force of a constant lift coefficient (LF=0.5) at a
mixture velocity of 3m/s and input water fraction of 80%. ds represents the drop size in
mm.
192
Effect of turbulent dispersion force on phase distribution
The changes in the lift force coefficient did not seem to homogenize the mixture to the
extent that it matches the experimental data. The effect of the turbulent dispersion force is
investigated as it can also act to homogenize the mixture.
Figure 8.9 presents the effect of increasing the turbulent dispersion force coefficient on
the phase distribution for a lift coefficient of 0.5 which gave the best results compared to
experimental data. As the coefficient is doubled, the distribution becomes more
homogeneous which agrees better with the experimental result. It is possible that the
deviations from the experimental data are due to the polydispersity of the real system
compared to the single drop size used in the simulations.
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6 LB_1.0
LB_2.0
-0.8
Experiment
-1.0
0.00 0.20 0.40 0.60 0.80 1.00
Oil Fraction [-]
Figure 8.9: Effect of turbulent dispersion force coefficient on phase distribution at a

mixture velocity of 3m/s and input water fraction of 80%. LB represents the coefficient
for the Lopez de Bertodano model in Equation 2.39.
193
Effect of lift and turbulent dispersion force on dispersion of different phase fraction
The force coefficients that were found in previous sections to give the best agreement
against the experimental data (i.e. lift force coefficient of 0.5 and turbulent dispersion
force coefficient of 2) are used here to simulate the flow for different input water
fractions at 40% and 60%. The results can be seen in Figure 8.10 against the
experimental data. The polydispersity of the experimental mixture may have accounted
for the deviation of the simulated results from experimental. Krepper et al. (2004)
demonstrated that a bimodal distribution can have a different phase fraction profile
compared to either mono-sized distribution.
The discrepancy is larger for 40% water fraction where the simulations show that the
flow is highly stratified while the experiments reveal a large concentration of oil drops at
the middle of the pipe. The high concentration of oil drops in the middle of the pipe
resulted in experiments is due to phase inversion (see Section 4.4). However, no
accounting of phase inversion is available in CFX. The available models in CFX are also
limited in simulating a very dense dispersion coupled with the occurrence of phase
inversion. As a result, the phase fraction distribution cannot be correctly modeled at this
fraction.
194
1.0
Simulation (40%)
0.8 Simulation (60%)
0.6 Experiment (40%)

Experiment (60%)
0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0
0.00 0.20 0.40 0.60 0.80 1.00
Oil Fraction [-]
Figure 8.10: Comparison between experimental and simulated phase fraction

distributions for 40% and 60% input water fractions at a mixture velocity of 3m/s and
drop size of 1mm. Oil is set to be the dispersed phase in the simulation. The lift force
coefficient is 0.5 and the turbulent dispersion force coefficient is 2.
8.5 CONCLUSION
It was shown in previous chapters that the dispersed drops migrate towards the centre of
the pipe. The CFD simulations in this chapter offer some insight on the forces acting on
the dispersed drops and on how they influence drop movements. Lift and turbulent
dispersion forces can counteract the upward buoyancy acting on the dispersed oil drops.
Lift force depends on the velocity profile of the continuous phase and the relative
velocity of the dispersed phase. The turbulent dispersion force is driven by the
concentration gradient and tends to homogenize the dispersion.
195
For dilute dispersions (80% water fraction), the results showed that a positive lift force
coefficient (negative lift force) drives the drops away from the pipe wall, while the
magnitude of the force is larger for larger dispersed drops. This suggests that the larger
drops are more likely to be driven towards the centre of the pipe. An increase in the
turbulent dispersion force was also shown to improve the homogenization of the mixture.
Higher dispersed phase fractions (i.e. 40% and 60% water fraction) were also simulated.
There is a reasonable agreement between experimental and simulations for 20% and 40%
dispersed fraction. This may be further improved if a drop size distribution is taken into
account in the simulations (e.g. by using the MUSIG code) since it was found that the
magnitudes of the forces depend on drop size. However, the simulations fail to predict the
experimental results at 60%. This is probably because phase inversion has occurred
experimentally at this fraction which cannot be accounted for in CFX.
196
Chapter 9: Conclusions and Recommendations
9.1 OVERVIEW
Phase inversion in a dispersed pipe flow has been investigated experimentally using
various types of instrumentation. The effect of interfacial tension has also been studied as
it has been the least understood fluid property influencing the inversion process. Based on
the experimental outcome, prediction models have been developed to estimate the phase
fraction at which inversion will occur as well as the pressure gradient during the change
in phase fraction across an inversion process. Lastly, computational fluid dynamic
simulations were studied to understand the interactions between the fluid phases using
established models in Ansys CFX.
A summary of the main conclusions from the experimental and theoretical studies will be
given in Section 9.2, followed by recommendations for future studies in Section 9.3.
9.2 CONCLUSIONS
9.2.1 CONCLUSIONS FROM EXPERIMENTAL STUDIES
The occurrence of phase inversion

 ERT tomography and other conductivity measurements reveal that the dispersion
in the pipe undergoes a change in flow regime from a fully dispersed flow to a
dual continuous flow during phase inversion. This transition is due to the
gravitational stratification of the fluid phases.
 Phase inversion is also observed to be a gradual process across a range of phase
fraction. The range of phase fraction during inversion is regarded as the
transitional region. This transitional region narrows as the mixture velocity of the
fluids increases.
 Phase inversion is observed to occur firstly at the centre of the pipe (if starting
from an O/W dispersion). The accumulation of oil continuous dispersion will
move towards the top of the pipe forming a dual continuous flow. Finally, phase
197
inversion is completed as the lower water layer is being dispersed into the oil
phase. The reverse process will occur starting from a W/O dispersion.
 Pressure gradient along the pipe is found to be closely related to the distribution
of the phases contacting with the pipe wall. Drag reduction is first observed as the
dispersed phase is increased in an initial dispersion (e.g. O/W). As phase
inversion occurs at the centre of the pipe, no significant increase in pressure
gradient is observed. However, inversion will begin to increase once the oil
continuous dispersion is in contact with the pipe wall and before phase inversion
is completed across the pipe cross section.
 The introduction of a dispersed inlet produces an homogeneous dispersion and
phase inversion occurs at a specific phase fraction (spontaneous inversion).
 Pressure gradient is found to differ with an homogeneous dispersion. Pressure
gradient increases with the introduction of the dispersed phase. This is possibly
because of the increase in effective mixture viscosity. There is a suddenly drop at
36% water fraction likely due to drop coalescence and deformation of the closely
packed drops. Pressure gradient will fall to a minimum at the phase inversion
point before it gradually increases beyond inversion.
 The drop size of the dispersion is found to fit well with a log-normal distribution.
It is also found that the O/W dispersion has a smaller drop size than the W/O
dispersion. The dispersed inlet also produces a narrower dispersion than the split
inlet.
 The drop size distribution during phase inversion becomes wider due to the
dominance of coalescence of the dispersed phase.
Effect of interfacial tension on phase inversion

 The addition of glycerol to the water phase has shown significant decrease in
interfacial tension while it does not significantly affect the density and viscosity
of the solution.
 The presence of glycerol does not appear to affect the inversion of the dispersed
oil phase to become continuous. However, it retains the continuity of the water
phase before the water phase is inverted at a lower water fraction.
198
 Changes on pressure gradient are similar to pure water system with an initial
decrease due to drag reduction and increases again once the more viscous oil
phase is in contact with the pipe wall.
 Drop size measurements in the middle of the pipe reveals that the size of oil drops
before inversion is not significantly affected by the addition of glycerol while the
size of the water drops after inversion is reduced. While larger water drops are
expected with the addition of glycerol, the smaller size is possibly due to better
coalescence of the water drop and the formation of the water layer at the bottom
of the pipe.
9.2.2 CONCLUSIONS FROM THEORETICAL STUDIES
Prediction of phase inversion through fluid viscosities

 Pressure gradient was previously reported to reach a peak at phase inversion. The
change of pressure gradient is primarily caused by the shear stress of the mixture
with the pipe wall which is related to the fluid viscosities. A prediction model for
phase inversion is thus proposed by assuming equal mixture viscosities of the oil
and water continuous dispersions during inversion.
 Various correlations for mixture viscosities from literatures have been applied.
Brinkman (1952)/Roscoe (1952) is found to have the best agreement with
experimental data of oil/water systems of different oil viscosities regardless of its
mixture velocity, dispersion initialisation and pipe material.
 The proposed model with Brinkman (1952)/Roscoe (1952) also agrees with
experimental data from Yeh et al. (1964) and Arirachakaran et al. (1989).
 While the predicted phase fraction using the proposed model deviate from the
inversion point from the homogeneous dispersed flow, the predicted phase
fraction falls closely with the fraction at which maximum pressure gradient is
observed. The deviation with the inversion point maybe due to drop deformation
and coalescence which cannot be accounted by the applied mixture viscosity
correlations.
199
Prediction of pressure gradient during phase inversion
 Based on the experimental observation during inversion, the change in flow
regime (from fully dispersed flow to dual continuous flow) forms the basis for the
development of a prediction model for pressure gradient during phase inversion.
 The main difference in these two flow regimes is the presence of the interfacial
height during dual continuous flow. This is the selection criteria to choose
between the two layered flow model and the homogeneous flow model for the
prediction of pressure gradient. A phase inversion criteria is also set based on the
developed prediction model from fluid viscosities.
 The proposed prediction model for pressure gradient agrees well with the
experimental data at two different mixture velocities apart from the oil continuous
dispersed flow at low mixture velocity. This discrepancy is attributed to the
inhomogenous distribution of the dispersed water drops in the pipe cross section.
CFD simulation of horizontal two-phase pipe flow

 Lift force and turbulent dispersion force are found to be important to homogenize
the dispersion. The magnitude of the lift force is larger for larger dispersed drops
resulting larger drops move faster towards the centre of the pipe. An increase in
the magnitude of the turbulent dispersion force will also improve the
homogenization of the mixture.
 Good agreement of the CFD simulation with experimental results can be
observed at lower dispersed phase fraction (i.e. 20% and 40%). However, the
simulation fails to predict at 60% dispersed phase fraction. This is probably due
to the limitation of the CFD code to account for the occurrence of phase
inversion.
9.3 RECOMMENDATIONS
The experimental and theoretical investigations in this work help towards a better
understanding on aspects affecting the phase inversion process in an horizontal pipe flow.
200
However, further work would be recommended to allow further depth into the existing
investigations.
Recommendations for experimental studies

 At present, only one type of oil (Exxsol D140) is used throughout the whole
study. With existing instrumentation for phase inversion detection, it will be
beneficial to re-examine the oil/water system conducted by Angeli (1996) and
Ioannou (2006) using Exxsol D80 and Marcol 52. This will have a broader
understanding of the inversion process by having information about the changes
in phase distribution of different fluid mixtures during inversion using the ERT
tomography.
 Investigation conducted at a different pipe size will lead to significant difference
in changes of flow regimes. Smaller pipes, for example, will lead to relatively
higher influence of the surface wettability at the pipe wall on the fluid
distribution. Thus, difference in the occurrence of phase inversion will be
expected.
 More drop size results should be obtained at different location across the pipe
cross section to have a better understanding of the local drop size distributions
during inversion.
 Measurement technique should be developed to account for the drop size of the
secondary droplets as the presence of secondary droplets can significantly affect
the occurrence of phase inversion.
 Use of an oil-based surfactant to influence the interfacial tension and compare the
similarity/difference in the occurrence of phase inversion with the current use of
glycerol to the water phase.
Recommendations for theoretical studies

 Mixture viscosity correlations should be modified to include dispersion packing
and drop size distribution of the dispersed phase. This aims to account for drop
deformation and polydispersity in a dense dispersion.
201
 Pressure gradient prediction model should include the selection criteria from fully
dispersed flow to annular flow (when local inversion begins to appear at the
centre of the pipe) and from annular flow to dual continuous flow (due to the
gravitational stratification of the two continuous dispersions (O/W and W/O)).
 Apply the pressure gradient prediction model to a wider range of experimental
conditions. This can help in improving the robustness of the model by using more
appropriate correlations.
 Apply Multiple Size Group (MUSIG) model in CFD simulation to account for
the drop size distribution of the dispersed phase.
 3-dimensional (3D) geometry for CFD simulation should be used to account for
the lateral forces exerted from the sides of the pipe. This may improve the
prediction of phase distribution between the dispersed and continuous phases in a
pipe cross section.
202
Reference
 Agtersloot, R., Vuik, C., Zijlema, M. (1996). A comparison of two CFD packages
and engineering formulae for fluid flow problems. Report 96-93, Faculty of
Technical Mathematics and Informatics, Delft Univ. of Tech.
 Al-Sarkhi, Soleimani, A. (2004). Effect of drag reducing polymer on two-phase
gas-liquid flows in a horizontal pipe. Trans IChemE., 82(Part A): 1583-1588.
 Al-Wahaibi, T. (2006). Investigations on the transition between stratified and
non-stratified horizontal oil-water flows. Ph.D. thesis, UCL, UK.
 Al-Wahaibi, T., Smith, M., Angeli, P. (2007). Effect of drag-reducing polymers
on horizontal oil-water flows. J. of Petroleum Sci. and Eng., 57(3-4): 334-346.
 Al-Wahaibi, T., Angeli, P. (2009). Predictive model of the entrained fraction in
horizontal oil-water flows. Chem. Eng. Sci., 64(12): 2817-2825.
 Ansys CFX reference guide Version 11.0. . (2006).
 Angeli, P. (1996). Liquid-liquid dispersed flows in horizontal pipes. Ph.D. thesis,
Imperial College London, UK.
 Angeli, P., Hewitt, G. F. (1998). Pressure gradient in horizontal liquid-liquid
flows. Int. J. Multiphase Flow, 24(7): 1183-1203.
 Angeli, P., Hewitt, G. F. (2000). Flow structure in horizontal oil-water flow, Int. J.
Multiphase Flow, 26: 1117-1140.
 Arashmid, M., Jeffreys, G. V. (1980). Analysis of phase inversion characteristics
of liquid-liquid dispersions. AIChE J., 26(1): 51-55.
 Arirachakaran, S., Oglesby, K. D., Shoulam, O., Brill, J. P. (1989). An
investigation of oil water flow phenomena in horizontal pipes, Paper SPE 18836,
Society of Petroleum Engineers.
 Arrhenius (1887). Zeits. f. physic. Chemie 1: 285.
 Barnea, E., and Mizrahi, J. (1973). A generalized approach to the fluid dynamics
of particulate systems: Part 1. General correlation for fluidization and
sedimentation in solid multiparticle systems. Chem Eng. J., 5: 171-189.
203
 Batchelor, G. K. (1956). The theory of homogeneous turbulence. Cambridge
University Press, Cambridge.
 Boussinesq, J. (1877). Théorie de l’Écoulement Tourbillant. Mem. Présentés par
Divers Savants Acad. Sci. Inst. Fr., 23: 46-50.
 Brauner, N., Moalem Maron, D. (1989). Two-phase liquid–liquid stratified flow.
Physico-Chem. Hydrodynam. 11, 487–506.
 Brauner, N. (2002). Liquid-liquid two-phase flow systems. Modelling and control
of two-phase phenomena. V Bertola (Ed.), CISM Center, Udine, Italy.
 Brauner, N., Ullmann, A. (2002). Modelling of phase inversion phenomenon in
two-phase pipe flow. Int. J. Multiphase Flow. 28(1): 55-68.
 Brinkman, H. C. (1952). The viscosity of concentrated suspensions and solutions.
J. of Chem. Phy., 20(4): 571.
 Calabrese, R. V., Chang, T. P. K., Dang, P. T. (1986). Drop breakup in turbulent
stirred-tank contactors. Part I: Effect of dispersed-phase viscosity. AIChE J.,
32(4): 667-676.
 Chakrabarti, D. P., Das, G., Das, P. K. (2006). Transition from water continuous
to oil continuous flow pattern. AIChE J., 52(11): 3668-3678.
 Chen, P., Dudukovic, M.P., Sanyal, J. (2005). Three-dimensional simulation of
bubble column flows with bubble coalescence and breakup. AIChE J., 51: 696-
712.
 Chesters, A. K. (1991). The modeling of coalescence processes in fluid-liquid
dispersions: A review of current understanding. Chemical Engineering Research
& Design, Transaction IChemE (Part A), 69: 259-270.
 Chesters, A. K., Issa, R. (2004). A framework for the modeling of phase inversion
in liquid-liquid systems. Proc. 5th Int. Conf. on Multiphase Flow.
 Chiang, C. L., Chen, C. M. (1994). Characteristics of phase inversion in a turbine-
type agitator. J. Chinese Inst. Chem. Eng., 25(6): 367-374.
 Choi, S. J., Schowalter, W. R. (1975). Rheological properties of nondilute
suspensions of deformable particles. Physics of Fluids, 18(4): 420-427.
 Clarke, S. I., Sawistowski, H. (1978). Phase inversion of stirred liquid/liquid
dispersions under mass transfer conditions, Trans. IChemE, 56: 50-55.
204
 Collins, S. B., Knudsen, J. G. (1970). Drop-size distributions produced by
turbulent pipe flow of immiscible liquids. AIChE J., 16(6): 1072–1080.
 Coulaloglou, C. A., Tavlarides, L. L. (1977). Description of interactions processes
in agitated liquid-liquid dispersions, Chem. Eng. Sci., 32: 1289-1297.
 Delnoij, E., Kuipers, J. A. M., van Swaaij, W. P. M. (1997). Computational fluid
dynamics applied to gas-liquid contactors. Chem. Eng. Sci. 52: 3623-3638.
 Delichatsios, M.A., Probstein, R.F. (1976). The effect of coalescence on the
average drop size in liquid–liquid dispersions. Ind. Eng. Chem. Fund., 14: 134-
138.
 Deshphande, K. B., Kumar, S. (2003). A new characteristic of liquid-liquid
systems-inversion holdup of intensely agitated dispersions. Chem. Eng. Sci., 58:
3829-3925.
 Drew, D.A (1983). Mathematical modeling of two phase flow. Annual review of
fluid mechanics. 15: 261 – 291.
 Efthimiadu, I., Moore, I. P. T. (1994). Phase inversion of liquid-liquid dispersions
produced between parallel shearing plates. Chem. Eng. Sci., 49: 1439-1449.
 Einstein, A. (1906). Eine neue bestimmung der moleküledimensionen, Annalen
Phys., 19: 289-306.
 Einstein, A. (1911). Berichtigung zu meiner Arbeit: Eine neue bstimmung der
moleküledimensionen. Annalen Phys., 34: 591-592.
 Elseth, G. (2001). An experimental study of oil/water flow in horizontal pipes.
Ph.D. thesis, NTNU Trondheim, Norway.
 Fakhr-Din, S. M. (1973). Phase inversion and drop size measurements in agitated
liquid-liquid systems. Ph. D. thesis, University of Manchester, UK.
 Furuse, H. (1972). Viscosity of concentrated solution. Jap. J. of Applied Physics,
11(10): 1537.
 Gillies, R. G., Sun, R., Shook, C. A. (2000). Laboratory investigation of inversion
of heavy oil emulsion. Can. J. Chem. Eng., 78: 757-763.
 Groeneweg, F., Agterof, W. G. M., Jaeger, P., Janssen, J. J. M., Wieringa, J. A.,
Klahn, J. K. (1998). On the mechanism of the inversion of emulsions. Trans
IChemE, 76 (Part A).
205
 Guilinger, T. R., Grislingas, A. K., Erga, O. (1988). Phase inversion behavior of
water-kerosene dispersions. Ind. & Eng. Chem. Res., 27(6): 978-982.
 Guth, E. and Gold, O. (1938). On the hydrodynamical theory of the viscosity of
suspensions. Phys. Rev. 53: 322.
 Guzhov, A. I., Grishin, A. D., Medredev, V. F., Medredeva, O. P. (1973).
Emulsion formation during the flow of two liquids in a pipe. Neft Hhoz (in
Russian). 8: 58-61.
 Hill, D. (1998). The computer simulation of dispersed two-phase flows. Ph.D.
thesis, Imperial College London, UK.
 Hu, B. (2005). Experimental and theoretical investigation of phase inversion in
liquid-liquid dispersions. Ph.D. thesis, UCL, UK.
 Hu, B., Matar, O.K., Hewitt, G.F., Angeli, P. (2005). Prediction of phase
inversion in agitated vessels using a two-region model. Chem. Eng. Sci., 60:
3487-3495.
 Hu, B., Angeli, P. (2006). Phase inversion and associated phenomena in oil-water
vertical pipeline flow. Can. J. Chem. Eng., 84(1): 94-107.
 Hu, B., Angeli, P., Matar, O. K., Lawrence, C. J., Hewitt, G. F. (2006). Evaluation
of drop size distribution from chord length measurements. AIChE J., 52(3): 931-
939.
 Ioannou, K., Bin, H., Matar, O. K., Hewitt, G. F., Angeli, P. (2004). Phase
inversion in dispersed liquid-liquid pipe flows, Proc. 5th Int. Conf. on Multiphase
Flow.
 Ioannou, K., Nydal, O. J., Angeli, P. (2005). Phase inversion in dispersed liquid-
liquid flows. Experimental thermal and fluid science. 29(3): 331-339.
 Ioannou, K. (2006). Phase inversion phenomenon in horizontal dispersed
oil/water pipeline flows. Ph.D. thesis, UCL, UK.
 Ishii, M., Zuber, N. (1979). Drag coefficient and relative velocity in bubbly,
droplet or particulate flows. AIChE J., 25: 843-855.
 Karabelas, A. J. (1978). Droplet size spectra generated in turbulent pipe flow of
dilute liquid/liquid dispersions. AIChE J., 24: 170–180.
 Kennard, E. H. (1938). Kinetic theory of gases. McGraw-Hill, New York.
206
 Krepper, E., Lucas, D., Prasser, H. M. (2005). On the modelling of bubbly flow in
vertical pipes. Nuclear Engineering and Design, 235: 597–611.
 Krepper, E., Vanga, B. N. R., Zaruba, A., Prasser, H. M., Lopez de Bertodano, M.
A. (2007). Experimental and numerical studies of void fraction distribution in
rectangular bubble columns. Nuclear Engineering and Design. 237: 399-408.
 Krieger, I. M., and Dougherty, T. J., 1959. A mechanism for non-Newtonian flow
in suspensions of rigid spheres. J. of Rheology, 3(1): 137-152.
 Krishna, R., van Baten, J. M. (1999). Rise characteristics of gas bubbles in a 2D
rectangular column - VOF simulations vs Experiments. Int. Comm. in Heat and
Mass Transfer. 26: 965-974.
 Kumar, A., Hartland, S. (1985). Gravity settling in Liquid/Liquid Dispersions.
Can. J. Chem. Eng., 63: 368-376.
 Kumar, S., Kumar, R., Gandhi, K. S. (1991). Influence of the wetting
characteristics of the impeller on phase inversion. Chem. Eng. Sci., 46(9): 2365-
2367.
 Kumar, S. (1996). On phase inversion characteristics of stirred dispersions. Chem.
Eng. Sci., 51(5): 831-834.
 Launder, B. E., Spalding, D. B. (1972). Lectures in mathematical models of
turbulence. Academic Press, London, England.
 Leporcher, E., Delaytermoz, A., Renault, J. F., Gerbier, A., Burger, O. (2001).
Deployment of multiphase pumps on a North Sea Field. Paper SPE 71536,
Society of Petroleum Engineers.
 Levich, V. G. (1962). Physicochemical Hydrodynamics. Prentice-Hall,
Englewood Cliffs, New Jersey.
 Liu, L. (2005). Optical and computational studies of liquid-liquid flows. Ph.D.
thesis, University of London, UK.
 Liu, L., Matar, O.K., Perez de Ortiz, E.S., Hewitt, G.F. (2005). Experimental
investigation of phase inversion in a stirred vessel using LIF. Chem. Eng. Sci.,
60(1): 85-94.
207
 Liu, L., Matar, O., Lawrence, C., Hewitt, G. (2006). Laser-induced fluorescence
(LIF) studies of liquid-liquid flows. Part I: Flow structures and phase inversion.
Chem. Eng. Sci., 61: 4007-4021.
 Liu, S., Li, D. (1999). Drop coalescence in turbulent dispersions. Chem. Eng. Sci.,
54(23): 5667–5675.
 Lopez de Bertodano, M., (1991). Turbulent Bubbly Two Phase Flow in a
Triangular Duct. Ph. D. Thesis, Rensselaer Polytechnic Institute, New York.
 Lopez de Bertodano, M., (1998). Two fluid model for two-phase turbulent jet.
Nucl. Eng. Des., 179: 65 - 74.
 Lovick, J. (2004). Horizontal oil-water flows in the dual continuous flow regime.
Ph.D. thesis, UCL, UK.
 Lucas, D., Krepper, E., Prasser, H. M. (2007). Use of models for lift, wall and
turbulent dispersion forces acting on bubbles for poly-disperse flows. Chem. Eng.
Sci., 62(15): 4146 – 4157.
 Luhning, R. W., Sawistowski, H. (1971). Phase inversion in stirred liquid-liquid
systems. Proceeding of the International Solvent Extraction Conference, The
Hague, 873-887, Society of Chemical Industry, London.
 Luo, H., Svendsen, H. F. (1996). Theoretical model for drop and bubble breakup
in turbulent dispersions. AIChE J., 42(5): 1225-1233.
 Luo, Y. S., Chen, T. K., Cai, J. Y. (1997). Frictional pressure loss and phase
inversion point for oil-water emulsion in vertical tube. Int. Symp on multiphase
fluid, non Newtonian fluid & physicochemical fluid flows. 3: 53-58.
 Martinez, A. E., Intevep, S. A., Arirachakaran, S., Shoham, O., Brill, J. P. (1988).
Prediction of dispersion viscosity of oil/water mixture flow in horizontal pipes.
Paper SPE 18221: 427-438, Society of Petroleum Engineers.
 McClarey, M., Mansoori, G. (1978). Factors affecting the phase inversion of
dispersed immiscible liquid-liquid mixtures. AIChE J., 74: 134-139.
 Mooney, M. (1951). The viscosity of a concentrated suspension of spherical
particles. J. Colloid Sci, 6: 162.
208
 Moraga, J. F., Larreteguy, A. E., Drew, D. A., Lahey, R. T. (2003). Assessment of
turbulent dispersion models for bubbly flows in the low Stokes number limit. Int.
J. multiphase flow, 29: 655.
 Nädler, M., Mewes, D. (1995). Intermittent three-phase flow of oil, water and gas
in horizontal pipes. Proc. 5th Int. Offshore and Polar Eng. Conf.
 Nädler, M., Mewes, D. (1997). Flow induced emulsification in the flow of two
immiscible liquids in horizontal pipes. Int. J. Multiphase Flow, 23(1): 55-68.
 Ngan, K. H., Linnenberg, S., Al-Wahaibi, T., Angeli, P. (2007). Effect of Drag
Reducing Polymers on Oil-Water Flows. Int. Conf. on Multiphase Flow, Germany.
 Ngan, K.H., Ioannou, K., Rhyne, L.D., Wang, W., Angeli, P. (2009). A
methodology for predicting phase inversion during liquid-liquid dispersed
pipeline flow. Chem. Eng. Res. Des., 87:318-324.
 Norato, M. A., Tsouris, C., Tavlarides, L. L. (1998). Phase inversion studies in
liquid-liquid dispersions, Can. J. Chem. Eng., 76: 486-494.
 Noui-Mehidi, M., Wu, J., Zhu, Y. (2004). A few new findings on phase inversion
in a liquid/liquid system. AIChE J., 50: 3281-3283.
 Oglesby, K. D. (1979) An experimental study on the effects of oil viscosity,
mixture velocity, and water fraction on horizontal oil-water flow. M.S. thesis,
University of Tulsa, Oklahoma.
 Pacek, A. W., Moore, I. P. T., Nienow, A. W. (1994). Video technique for
measuring dynamics of liquid-liquid dispersion during phase inversion. AIChE J.,
40(12): 1940-1949.
 Pacek, A. W., Nienow, A. W. (1995). A problem for the description of turbulent
dispersed liquid-liquid systems. Int. J. Multiphase Flow, 21(2): 323–328.
 Pal, R., Rhodes, E. (1989). Viscosity/Concentration relationships for emulsions. J.
Rheol., 33(7): 1021-1045.
 Pal, R. (1993). Pipeline flow of unstable and surfactant-stabilized emulsions.
AIChE J., 39 (11): 1754-1764.
 Pal, R. (2001). Single-parameter and two-parameter rheological equations of state
for nondilute emulsions, Ind. Eng. Chem. Res., 40: 5666-5674.
209
 Perry, R. H., Green, D. W. (1997). Perry’s Chemical Engineers’ Handbook. 7th Ed.
McGraw-Hill, New York.
 Pettersen, B.H., Langsholt, M., and Ødegård, J. (2001). Pressure drop and wall
wetting: the effect of pipe wall material in multiphase flow. Proc. 10th
International Conference of Multiphase Flow, Cannes, France.
 Phan-Thein, N., Pham, D. C. (1997). Differential multiphase models for
polydispersed suspensions and particulate solids. J. Non-Newtonian Fluid Mech.,
72: 305.
 Piela, K., Delfos, R., Ooms, G., Westerweel, J., Oliemans, R.V.A., Mudde, R.F.
(2006). Experimental investigation of phase inversion in an oil-water flow
through a horizontal pipe loop. Int. J. Multiphase Flow, 32(9): 1087-1099.
 Piela, K., Delfos, R., Ooms, G., Westerweel, J., Oliemans, R.V.A. (2008). On the
phase inversion process in an oil-water pipe flow. Int. J. Multiphase Flow, 34(7):
665-677.
 Prince, M., Blanch, H. W. (1990). Bubble coalescence and break-up in air-
sparged bubble columns. AIChE J., 36(10): 1485–1499.
 Quinn, J. A., Sigloh, D. B. (1963). Phase inversion in the mixing of immiscible
liquids. Can. J. Chem. Eng., 41: 15-18.
 Ranade, V. (2002). Computational flow modeling for chemical reactor
engineering. Process Systems Engineering Vol. 5. Academic Press, California.
 Rodger, W. A., Trice, Jr., V. G., Rushton, J. H. (1956). The effect of fluid motion
on interfacial area dispersions. Chem. Eng. Prog., 52(12): 515-520.
 Rodriguez, O. M. H., Bannwart, A. C. (2006). Analytical model for interfacial
waves in vertical core flow. J. of Petroleum Sci. and Eng., 54 (3-4).
 Roscoe, R. (1952). The viscosity of suspensions of rigid spheres. British journal
of applied physics, 267-269.
 Rotta, J. C. (1972). Turbulente Strömungen. B. G. Teubner, Stuttgart.
 Rozentsvaig, A. K. (1982). Turublent pipe flow of concentrated emulsions with a
nonequilibrium disperse phase. Inzh.-Fiz. Zh., 42(3), 366-372.
210
 Rusche, H., Issa, R. I. (2000). The Effect on voidage on the drag force on particles
in dispersed two-phase flow. Japanese-European two phase flow meeting,
Tsukuba, Japan.
 Rusche, H. (2002). Computational fluid dynamics of dispersed two-phase flows at
high phase fractions. Ph.D. thesis, Imperial College London, UK.
 Russell, T., Hodgsen, G., Govier, G. (1959). Horizontal pipeline flow of mixtures
of oil and water. Can. J. Chem. Eng., 37: 9-17.
 Russell, T.W.F., Charles, M.E., (1959). The effect of the less viscous liquid in the
laminar flow of two immiscible liquids. Can. J. Chem. Eng. 37: 18–24.
 Saffman, P. G., Turner, J. S. (1956). On the collision of drops in turbulent clouds.
J. Fluid Mech., 1: 16-30.
 Schiller, L., Naumann, Z. (1933). Über die grundlegenden Berechungen bei der
Schwerkraftbereitung. Z. Ver. Deutsch. Ing., 77: 318 -320.
 Schwartzberg, H.G., Treybal, R. E. (1968). Fluid and particle motion in turbulent
stirred tanks. Ind. & Eng. Chem. Res. Fund., 7: 1-6.
 Seklker, A. H., Sleicher, Jr., C. A. (1965). Factors affecting which phase will
disperse when immiscible liquids are stirred together. Can. J. Chem. Eng.,
43:298-301.
 Simmons, M. J. H., Azzopardi, B. J. (2001). Drop size distributions in dispersed
liquid-liquid pipe flow. Int. J. Multiphase Flow, 27: 843-859.
 Smith, D. H., Lim, K. (1990). An experimental test of catastrophe and critical-
scaling theories of emulsion inversion. Langmuir, 6: 1071-1077.
 Soleimani, A., Lawrence, C. J., Hewitt, G. F. (1997). Effect of mixers on flow
pattern and pressure drop on horizontal oil-water pipe flow. Int. Symp. on liquid-
liquid two phase flow and transportation phenomena, Turkey.
 Soleimani, A. (1999). Phase distribution and associated phenomena in oil-water
flows in horizontal tubes. Ph.D. thesis, Imperial College London, UK.
 Soleimani, A., Lawrence, C. J., Hewitt, G. F. (2000). Spatial distribution of oil
and water in horizontal pipe flow. SPE Journal, 5(4): 394-401.
211
 Sommerfelds, M., Wachem, B.V., Oliemans, R.V.A (2008). Best practice
guidelines for computational turbulence dispersed multiphase flows. Seminar
presentation, SIAMUF, Chalmers University.
 Sovová, H. (1981). Breakage and coalescence of drops in a batched stirred vessel
– II. Comparison of model and experiments. Chem. Eng. Sci., 36(9): 1567-1573.
 Stover, R. L., Tobías, C. W., Denn, M. M. (1997). Bubble coalescence dynamics.
AIChE J., 43: 2385-2392.
 Taylor, G. I. (1932). The viscosity of a fluid containing small drops of another
fluid. Proc. R. Soc., 138: 41.
 Thomas, D. G. (1965). Transport characteristics of suspensions: VII. A note on
the viscosity of Newtonian suspensions of uniform spherical particles. J. Colloid
Sci., 20: 267-277.
 Tidhar, M., Merchuk, J. C., Sembira, A. N., Wolf, D. (1986). Characteristics of a
motionless mixer for dispersion of immiscible fluids- Phase inversion of liquid-
liquid systems. Chem. Eng. Sci., 41(3): 457-462.
 Tomiyama, A., Zun, I., Kanami, N., Sakaguchi, T. (1995). Effects of Eötvös
number and dimensionless liquid volumetric flux on lateral motion of a bubble in
a laminar duct flow. Adv. Multiphase Flow, 3-15.
 Trallero, J. L. (1995). Oil water flow patterns in horizontal pipes. Ph.D. thesis,
University of Tulsa, Oklahoma.
 Treybal, R. E. (1963). Liquid Extraction. 2nd Edition, McGraw-Hill, New York.
 Tsouris, C., Tavlarides, L. L. (1994). Breakup and coalescence models for drops
in turbulent dispersion. AIChE J., 40(3): 395-406.
 Tyrode, E., Allouche, J., Choplin, L., Salager, J. (2005). Emulsion catastrophic
inversion from abnormal to normal morphology. 4. Following the emulsion
viscosity during three inversion protocols and extending the critical dispersed-
phase concept. Ind. & Eng. Chem. Res., 44: 67-74.
 Vaessen, G. E. J., Visschers, M., Stein, H. N. (1996). Predicting catastrophic
phase inversion on the basis of droplet coalescence kinetics. Langmuir, 12: 875-
882.
212
 Valentas, K.J., Bilous, O., Amundson, N.R. (1966). Analysis of breakage in
dispersed phase systems. Ind. Eng. Chem. Fundamentals, 5: 271–279.
 Valle, A., Utvik, O. H. (1997). Pressure drop, flow pattern and slip for two phase
crude oil-water flow. Int. Symp. on liquid-liquid two phase flow and
transportation phenomena, Turkey
 Vallée, C., Höhne, T., Prasser, H.M., Sühnel, T. (2007). Experimental
investigation and CFD simulation of slug flow in horizontal channels.
Forschungszentrum Dresden-Rossendorf, Report FZD-485, Dresden, Germany.
 Vand, V. (1948). Viscosity of solutions and suspensions. I – Theory. J Phys Coll
Chem, 52: 277–299.
 Vielma, M., Atmaca, S., Sarica, C., Zhang, H. (2008). Characterization of
Oil/Water Flows in Horizontal Pipes. Paper SPE 109591, Society of Petroleum
Engineers.
 Wang, W., Gong, J., Ngan, K. H., Angeli, P. (2009). Effect of glycerol on the
binary coalescence of water drops in stagnant oil phase, Chem. Eng. Res. Des.,
87(12): 1640-1648.
 Wilcox, D.C. (1986) Turbulence Modelling for CFD. DCW Industrie.
 Yakhot, V., Smith, L.M. (1992). The renormalization group, the ε-expansion and
the derivation of turbulence models. J. Sci. Comp, 7: 35-61.
 Yaron, I., Gal-Or, B. (1972). On viscous flow and effective viscosity of
concentrated suspensions and emulsions – Effect of particle concentration and
surfactant impurities. Rheologica Acta, Band 11, Heft 3/4: 245 – 252.
 Yeh, G. C., Haynie Jr., F. H., Moses, R. A. (1964). Phase-volume relationship at
the point of phase inversion in liquid dispersions. AIChE J., 10(2): 260-265.
 Yeo, L. Y., Matar, O. K., Perez de Ortiz, E. S., Hewitt, G. F. (2000). Phase
inversion and associated phenomena. Multiphase Science and Technology, 12(1):
51-116.
 Yeo, L. Y. (2002). Modelling of phase inversion and associated phenomena in
liquid-liquid systems. Ph.D. thesis, University of London, UK.
213
Appendix A – Chord Length Measurement during
Phase Inversion
Chord length distribution during the phase inversion is measured at various vertical
locations across the pipe cross section using a dual impedance probe discussed in Section
4.8. To supplement the data in Figure 4.13, the chord length distribution between 40%
and 54% water fraction are presented below.
0.6 0.3
Water Water
0.5
Oil
Frequency
oil
Frequency
0.4 0.2
0.3
0.2 0.1
0.1
0 0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Chord Length (mm) Chord Length (mm)
(a) 14mm (b) 20mm
0.3 0.3
Water
Water
Oil
Frequency
Frequency
0.2 Oil
0.2
0.1 0.1
0 0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
(c) 26mm (d) 32mm

Figure A-1: Chord length distribution of oil drops in water (54% water fraction) at
various locations from the top wall of the pipe with a mixture velocity of 3m/s. The
distributions of the two inversion routes are given – starting from a O/W dispersion (□)
214
0.6 0.6
Water Water
0.5 0.5
Oil Oil
Frequency
0.4 0.4
Frequency
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
(a) 20mm (b) 26mm
0.5
Water
0.4
Oil
Frequency
0.3
0.2
0.1
0
0 1 2 3 4 5 6 7 8 9 10
Chord Length (mm)
(c) 32mm
215
0.5 0.5
Water
Water
0.4 0.4
Oil
Oil
Frequency
Frequency
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
(a) 26mm (b) 32mm

From the above figures, only detectable chord length distributions are presented. At 54%
water fraction, it can be observed that the distribution becomes wider at 20mm (i.e. centre
of pipe) where it is previously found to be the beginning of phase inversion. As the water
fraction decreases, it can also be observed that the distribution becomes wide across all
the measured locations and the phenomenon spreads towards the bottom of the pipe cross
section. This supports the observation found using the ERT system in Figure 4.3 where
the phase inversion process has the same downward shift.
216
Appendix B – Matlab Code for Pressure Gradient
Prediction
A. MAIN EXECUTION CODE
%%%%%%%%%%%%%%%%%%
%Fluid properties%
%%%%%%%%%%%%%%%%%%
%density of oil (denso), density of water (densw)

denso=828; densw=1000;
%viscosity of oil (visco), viscosity of water (viscw)
visco=5.5*0.001; viscw=1.0*0.001;
%Interfacial tension (ST)
ST=39.6*0.001;
%%%%%%%%%%%%%%%%%
%Operation Input%
%%%%%%%%%%%%%%%%%
%gravity (g), pipe diameter (D)

g=9.81; D=0.038;
%Pipe area (A)
A=pi*D^2/4;
%Curvature (0-Flat Interface, 1-Curved Interface)
curve=0;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%Critical amplitude at the onset of entrainment (Talal 2006)%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
a_oil=D/10;
a_water=D/10;
hwstart=min(a_oil,a_water);
%%%%%%%%%%%%%%%%%%%
%Flow rate (l/min) %
%%%%%%%%%%%%%%%%%%%
Flow_rates=[ ];
217
%%%%%%%%%%%%%%%
%flow rate conversion (m3/s)%
%%%%%%%%%%%%%%%
conversion=0.0000166667;
Q=conversion*Flow_rates;
%Friction factor mode (refer to fric.m for detail)

fric_no=2;
%Critical water fraction for phase inversion

wfc=0.34;
%Initialisation (Do Not Change)

N_hw=(D-2*hwstart)/0.0001;
N_fr=size(Q,1);
output=[];
output1=[];
output2=[];
%Execution Script (Do Not Change)

for m=1:N_fr
Qo=Q(m,2);
Qw=Q(m,1);
hwnew=hwstart;
fr=Qw/(Qw+Qo)*100
olddiff=6000;
count=0;
for i = 1:N_hw
if hwnew<D-hwstart
[Ento_w,Entw_o]=entrain1(Qo,Qw,hwnew,D,denso, densw, visco, viscw, ST,
a_oil, a_water, curve, fric_no);
if (Entw_o >=1)||(Ento_w>=1)||(imag(Ento_w) || imag(Entw_o))
elseif (Entw_o>=wfc)&&(Ento_w>=(1-wfc))
else
count=count+1;
[dPu,dPl, dPa,UL,LL,Entw,Ento, Reu, Rel, Au,Al]=dP1(hwnew, Qo, Qw,
denso, densw,Entw_o,Ento_w, visco, viscw, D, curve, fric_no, wfc);
if abs((dPu-dPl)/dPa*100)>=2 || ((UL==1)&&(LL==1))||((UL==2)&&(LL==2))
else
output=[output;fr hwnew UL LL dPa/1000 abs((dPu-dPl)/dPa*100)];
end
end
hwnew=hwnew+0.0001;
else
break
218
end
end
OP=isempty(output);
if (OP==1)
[dPa]=homo(Qo, Qw, denso, densw, visco, viscw, D, curve, fric_no);
output1=[output1;fr 0 0 0 dPa/1000 0];
else
output2=sortrows(output,[1,6]);
idx=find(min((output2(:,6))));
output1=[output1;output2(idx,:)];
output=[];
end
end
csvwrite('output.csv', output1)
B. ENTRAINMENT MODULE
function [Ento_w1,Entw_o1]=entrain(Qo,Qw,hwnew,D,denso, densw, visco, viscw, ST,
a_oil, a_water, curve, fric_no)
%Calculating entrainment fraction

%Entw_o is the entrainment fraction of water in oil
%Ento_w is the entrainment fraction of oil in water
%Um=mixture velocity
Entw_o=0;
Ento_w=0;
[Si,So,Sw,A,Ao,Aw,Hw,Ho]=geo(hwnew,D,curve);
Um=(Qo+Qw)/A;
Usw=Qw/A;
Uso=Qo/A;
hwinput=hwnew;
[Tw,To,Ti]=hwF(hwinput, Usw, Uso, D, denso, densw, visco, viscw, curve, fric_no);
Uw=Usw/Hw;
Uo=Uso/Ho;
if Uo/Uw>=0.98 && Uo/Uw<1.05

Dw=4*Aw/(Sw); Do=4*Ao/(So);
elseif Uo>Uw
Dw=4*Aw/(Sw); Do=4*Ao/(So+Si);
else
Dw=4*Aw/(Sw+Si); Do=4*Ao/(So);
end
219
%Reynolds number
Rew=Dw*Uw*densw/viscw;
Reo=Do*Uo*denso/visco;
%Drag coefficient
Cd=2.7e-7*Reo^(0.77)*Rew^(0.86);
%Wavelength (lamda)
lamda=1*D;
%Prediction of entrainment fraction of oil in water (Ento_w)

%Calculating the critical amplitude of oil waves at the new conditions
%amplitude (al)
al=0.00;
k=real((D-hwnew)/0.000001);
for m=1:k
al=al+0.000001;
hap=hwnew+al;
[Siap,Soap,Swap,A,Aoap,Awap,Hwap,Hoap]=geo(hap,D,curve);
Soapeak=Soap;
Siapeak=Siap;
Aoapeak=Aoap;
Awapeak=Awap;
hat=hwnew-al;
[Siat,Soat,Swat,A,Aoat,Awat,Hwat,Hoat]=geo(hat,D,curve);
Siatrough=Siat;
Soatrough=Soat;
Aoatrough=Aoat;
Awatrough=Awat;
Aw1=Awapeak;
Aw2=Awatrough;
Ao1=Aoapeak;
Ao2=Aoatrough;
Uo2=Uo*(Ao/Ao2);
Uw2=Uw*(Aw/Aw2);
%Drag force (Fd)

Fd=0.5*densw*(Aw-Aw2)*Cd*(Uw2-Uo2)^2;
%Surface tension force (Fst)

L=(al^2+(lamda/2)^2)^0.5;
Fst=2*Si*ST*(lamda/2)/L;
Total=Fd-Fst;
220
if Total>=0
break
end
end
%Amount of oil transfer to the water phase

Vent_o=0.5*((a_oil-al)*lamda*Si/(pi));
vdfo=0.051*(Uso/Um)^-2.08*(Tw/densw)^0.5;
%Calculation of the entrainment fraction of oil in water (Ento_w)

C_oil=Vent_o*denso*abs(Uw2-Uo2)/(Si*lamda^2*vdfo);
Ento_w=((densw)-((densw)^2-4*(densw-denso)*C_oil)^0.5)/(2*(densw-denso));
%Calculating the critical amplitude of water waves at the new condition

%amplitude (au)
au=0.00;
k=real((D-hwnew)/0.000001);
for m=1:k
au=au+0.000001;
hap2=hwnew+au;
[Siap2,Soap2,Swap2,A,Aoap2,Awap2,Hwap2,Hoap2]=geo(hap2,D,curve);
Soapeak=Soap2;
Siapeak=Siap2;
Aoapeak=Aoap2;
Awapeak=Awap2;
hat2=hwnew-au;
[Siat2,Soat2,Swat2,A,Aoat2,Awat2,Hwat2,Hoat2]=geo(hat2,D,curve);
Siatrough=Siat2;
Soatrough=Soat2;
Aoatrough=Aoat2;
Awatrough=Awat2;
Aw1=Awapeak;
Aw2=Awatrough;
Ao1=Aoapeak;
Ao2=Aoatrough;
Uo1=Uo*(Ao/Ao1);
Uw1=Uw*(Aw/Aw1);
%Drag force (Fd)

Fd=0.5*densw*(Aw1-Aw)*Cd*(Uw1-Uo1)^2;
%Surface tension force (Fst)

L=(au^2+(lamda/2)^2)^0.5;
221
Fst=2*Si*ST*(lamda/2)/L;
Total=Fd-Fst;
if Total>=0
break
end
end
%Amount of water transfer to the oil phase
Vent_w=0.5*((a_water-au)*lamda*Si/(pi));
vdfw=0.077*(Usw/Um)^-1.8*(To/denso)^0.5;
%Calculation of the entrainment fraction of oil in water (Ento_w)

C_water=Vent_w*densw*abs(Uw1-Uo1)/(Si*lamda^2*vdfw);
Entw_o=((denso)-((denso)^2-4*(denso-densw)*C_water)^0.5)/(2*(denso-densw));
Ento_w1=(Ento_w)/(1-Ento_w+densw/denso);
Entw_o1=(Entw_o)/(1-Entw_o+denso/densw);
C. SHEAR STRESS MODULE
function [Tw,To,Ti]=hwF(hwinput, Usw, Uso, D, denso, densw, visco, viscw, curve,

fric_no)
[Si,So,Sw,A,Ao,Aw,Hw,Ho]=geo(hwinput,D,curve);
Uw=Usw/Hw;
Uo=Uso/Ho;
if Uo/Uw>+0.98 && Uo/Uw<1.05

Dw=4*Aw/(Sw); Do=4*Ao/(So);
elseif Uo>Uw
Dw=4*Aw/(Sw); Do=4*Ao/(So+Si);
else
Dw=4*Aw/(Sw+Si); Do=4*Ao/So;
end
%Reynolds number
Rew=Dw*Uw*densw/viscw;
Reo=Do*Uo*denso/visco;
%Reynolds number of interface

if Uo>Uw
Rei=Si*Uo*denso/(pi*visco);
densf=denso;
else
Rei=Si*Uw*densw/(pi*viscw);
densf=densw;
end
222
%friction factors
[fw,fo,fi]=fric(Rew,Reo,Rei,Ho,Hw,Do,Dw,fric_no);
%shear stress
Tw=fw*densw*Uw*abs(Uw)/2;
To=fo*denso*Uo*abs(Uo)/2;
if Uo/Uw >= 0.98 && Uo/Uw<1.05
Ti=0;
else
Ti=fi*densf*(Uo-Uw)*abs(Uo-Uw)/2;
end
%Two Fluid Model

%Fhw=To*So/Ao+Ti*Si*(1/Aw+1/Ao)-Tw*Sw/Aw;
D. DIFFERENTIAL PRESSURE MODULE (LAYERED FLOW)

function [dPu,dPl, dPa,UL,LL,Entw,Ento, Reu, Rel, Au,Al]=dP1(hwinput, Qo, Qw,
denso, densw,Entw_o,Ento_w, visco, viscw, D, curve, fric_no, wfc)
%Geometry
[Si,Su,Sl,A,Au,Al,Hw,Ho]=geo(hwinput,D,curve);
%Error test for Input Entrainment

if (imag(Ento_w) | imag(Entw_o))~=0
Entw_o=NaN;
Ento_w=NaN;
end
%Error test for entrainment

if Ento_w==NaN
Ento1=NaN;
return
end
if Entw_o==NaN
Entw1=NaN;
return
end
%Phase Inversion
%Flow Pattern: 1-W/O, 2-O/W, 3-Pure oil, 4-Pure water
if Entw_o<wfc
if Ento_w<(1-wfc);
Entw=Entw_o;
223
viscu=visco;
Uo=Qo/(Au);
Ento=Ento_w;
viscl=viscw;
Uw=Qw/(Al);
UL=1;
LL=2;
elseif Ento_w<1
Entw=Entw_o;
viscu=visco;
Uo=Qo/(Au);
Ento=(1-Ento_w);
viscl=visco;
Uw=Qw/(Al);
UL=1;
LL=1;
else
%Invalid Scenario
Entw=Qw/(Qo+Qw);
viscu=visco;
Uo=Qo/(Au);
Ento=0;
viscl=visco;
Uw=Qw/(Al);
UL=1;
LL=3;
end
elseif Entw_o<1
if Ento_w<(1-wfc);
Entw=(1-Entw_o);
viscu=viscw;
Uo=Qo/(Au);
Ento=Ento_w;
viscl=viscw;
Uw=Qw/(Al);
UL=2;
LL=2;
elseif Ento_w<1
%Invalid Scenario
Entw=(1-Entw_o);
viscu=viscw;
Uo=Qo/(Au);
Ento=(1-Ento_w);
viscl=visco;
Uw=Qw/(Al);
UL=2;
224
LL=1;
else
%Invalid Scenario
Entw=(1-Entw_o);
viscu=visco;
Uo=Qo/(Au);
Ento=0;
viscl=visco;
Uw=Qw/(Al);
UL=2;
LL=3;
end
else
if Ento_w<(1-wfc);
%Invalid Scenario
Entw=0;
viscu=viscw;
Uo=Qo/(Au);
Ento=Ento_w;
viscl=viscw;
Uw=Qw/(Al);
UL=4;
LL=2;
elseif Ento_w<1
%Invalid Scenario
Entw=0;
viscu=viscw;
Uo=Qo/(Au);
Ento=(1-Ento_w);
viscl=viscw;
Uw=Qw/(Al);
UL=4;
LL=1;
else
%Invalid Scenario
Entw=0;
viscu=viscw;
Uo=Qo/(Au);
Ento=0;
viscl=viscw;
Uw=Qw/(Al);
UL=4;
LL=3;
end
end
225
%Mixture Density
densu=(Entw)*densw+(1-Entw)*denso;
densl=(Ento)*denso+(1-Ento)*densw;
%Mixture Viscosity
muu=viscu*(1-Entw)^-2.5;
mul=viscl*(1-Ento)^-2.5;
if Uo/Uw>+0.98 && Uo/Uw<1.05

Dw=4*Al/(Sl); Do=4*Au/(Su);
elseif Uo>Uw
Dw=4*Al/(Sl); Do=4*Au/(Su+Si);
else
Dw=4*Al/(Sl+Si); Do=4*Au/Su;
end
%Reynolds number
Rel=Dw*Uw*densl/mul;
Reu=Do*Uo*densu/muu;
%Reynolds number of interface

if Uo>Uw
Rei=Si*Uo*densu/(pi*muu); densf=densu;
else
Rei=Si*Uw*densl/(pi*mul);densf=densl;
end
%friction factors
[fw,fo,fi]=fric(Rel,Reu,Rei,Ho,Hw,Do,Dw,fric_no);
if LL==1
fl=fw/(1+1.125*Ento);
else
fl=fw/(1+1.125*Ento);
end
if UL==1
fu=fo/(1+1.125*Entw);
else
fu=fo/(1+1.125*Entw);
end
fi=fi;
%shear stress
Tl=fl*densl*Uw*abs(Uw)/2;
Tu=fl*densu*Uo*abs(Uo)/2;
226
if Uo/Uw >= 0.98 && Uo/Uw<1.05
Ti=0;
else
Ti=fi*densf*(Uo-Uw)*abs(Uo-Uw)/2;
end
%Two Fluid Model

if Uo>Uw
dPu=Tu*Su/Au-Ti*Si*(1/Au);
dPl=Tl*Sl/Al+Ti*Si*(1/Al);
else
dPu=Tu*Su/Au+Ti*Si*(1/Au);
dPl=Tl*Sl/Al-Ti*Si*(1/Al);
end
dPa=(dPu+dPl)/2;
Ento1=100*Ento;
Entw1=100*Entw;
E. PRESSURE GRADIENT MODULE (HOMOGENEOUS MODEL)
function [dPa]=homo(Qo, Qw, denso, densw, visco, viscw, D, curve, fric_no)
A=pi*D^2/4;
fr=Qo/(Qw+Qo)*100;
WF=Qw/(Qo+Qw);
OF=Qo/(Qo+Qw);
densm=denso*OF+densw*WF;
muw=visco*(1-WF)^-2.5;
muo=viscw*(1-OF)^-2.5;
if muw<=muo
mum=muw;
else
mum=muo;
end
Um=(Qo+Qw)/A;
Reh=densm*D*Um/mum;
[fh]=hfric(Reh,fric_no);
if muw<=muo
fh1=fh/(1+1.25*WF);
else
fh1=fh/(1+1.25*OF);
end
dPu=0;
227
dPl=0;
dPa=(2*fh1*densm*Um^2)/D;
F. GEOMETRY MODULE
function [Si,So,Sw,A,Ao,Aw,Hw,Ho]=geo(hwinput,D,curve)
if curve==0
%disp('flat interface');
hw=hwinput;
h=2*(hw/D)-1;
Si=D*((1-h^2)^(1/2));
So=D*acos(h);
Sw=D*pi-So;
A=pi*D^2/4;
Ao=D*((So)-(Si*h))/4;
Aw=A-Ao;
Hw=Aw/A;
Ho=Ao/A;
elseif curve==1
%disp('curved interface')
hw=hwinput;
X=(hw^2+4*(D/2)*hw)/(2*(D/2)+2*hw);
teta1=2*acos(((D/2)-X)/(D/2));
teta2=2*acos((2*(D/2)-(X-hw))/(2*(D/2)));
A=pi*D^2/4;
Aw=(0.5*((D/2)^2)*(teta1-sin(teta1)))-(0.5*(2*(D/2)^2)*(teta2-sin(teta2)));
Ao=A-Aw;
Sw=(D/2)*teta2;
So=pi*D-Sw;
Si=2*(D/2)*teta2;
Hw=Aw/A;
Ho=Ao/A;
end
G. FRICTION FACTOR MODULE (LAYERED FLOW)

function [fw,fo,fi]=fric(Rel,Reu,Rei,Hu,Hl,Do,Dw,fric_no)
%Friction factor for Upper Layer

%if Reu<4000
%fo=(1/(-
2*log10((0.015/3.70)+((4.518*log10(1/7*Reu))/(Reu*(1+1/29*Reu^0.52*(0.015)^0.7))))
))^2;
%fo=7e-10*(Reu)^2; %(S.W. Churchill (1973), AIChE J, 19(2):375)
228
%fo=sqrt(1/(-2*log10((0.0015/3.7065)-(5.0272/Reu)*log10((0.0015/3.827)-
(4.567/Reu)*log10((0.0015/7.7918)^0.9924+(5.3326/(208.815+Reu))^0.9345)))));
%fo=16*Reu^-1;
%else
if fric_no==1
fo=0.079*Reu^-0.25;
elseif fric_no==2
fo=0.046*Reu^-0.2; % Taitel & Dukler (1976)
elseif fric_no==3
fo=0.764*(Hu*Reu)^(-0.562); %Srichai (1994)
elseif fric_no==4
fo=(-3.6*log10((6.9/Reu)+(1*10^(-5)/(3.7*Do))^1.1))^(-2); %Haaland (1983);
Hadzlabdic & Oliemans (2007)
elseif fric_no==5
fo=(-4*log10((1*10^(-5)/(3.7*Do))-(5.02/Reu)*log10((1*10^(-5)/(3.7*Do))-
(5.02/Reu)*log10((1*10^(-5)/(3.7*Do))+(13/Reu)))))^(-2); %Zigrang & Sylvester (1982)
else
return
end
%end
H. FRICTION FACTOR MODULE (HOMOGENEOUS MODEL)
function [fh]=hfric(Reh,fric_no)
if fric_no==1
fh=0.079*Reh^-0.25;
elseif fric_no==2
fh=0.046*Reh^-0.2;
else
disp('No friction factor model chosen')
return
end
229
Appendix C – Grid Sensitivity Test for CFD
Simulation
A grid sensitivity test was conducted to define the optimal grid structure for the CFD
simulations conducted in Chapter 8. The optimal grid structure will be used to simulate a
pipe section of 3m in length and 38mm in internal diameter. The test was done by
changing the number of cells in the vertical height of a rectangular slab (see Figure C-1).
Three structures of different grid sizes are chosen – 40 cells (grid 40), 60 cells (grid 60)
and 80 cells (grid 80). The length of the slab is kept at 3100 cells including 100 cells to
represent the splitting inlet as introduced in Chapter 3. Table C-1 presents the total
number of grid points and elements in each structure.
Figure C-1: Sample schematics of simulation domain with a vertical height consisting of
60 cells.
The simulation was tested by running an oil/water flow at a mixture velocity of 3m/s. Oil
was inputted as the dispersed phase while the water being the continuous phase. Drop
size was set at 0.5mm (smallest drop size group found in experiment). An input dispersed
phase fraction of 0.2 was chosen. k-ε turbulence model was used. Ishii and Zuber (1979)
model was applied as the drag force for the dispersion. Lift force with coefficient of 0.5
230
was used together with Lopez de Bertodano model for the turbulent dispersion force
(coefficient: 1.0).
The calculations achieved convergence of less than 10-6 in approximately 2 hours (grid
40), 48 hours (grid 60) and 120 hours (grid 80) using a workstation having 4GB RAM
and 3.0GHz Pentium Core2Duo CPU processor.
Case Name Grid 40 Grid 60 Grid 80

Vertical Cell Size 40 60 80
Number of Grid Points 250282 372322 494362
Number of Cell Elements 122000 183000 244000
Table C-1: Number of grid points and cell elements in structures used in the grid
sensitivity test.
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
grid 40
-0.8 grid 60
grid 80
-1.0
0.00 0.20 0.40 0.60 0.80 1.00
Oil Fraction [-]
Figure C-2: Comparison of results on the dispersed phase fraction along a vertical
diameter at the end of the pipe with 3m in length. An input dispersed phase fraction of
0.2 is chosen for all 3 cases.
231
Figure C-2 presents the comparison of oil fraction at the end of the grid structure (i.e. 3m
from end of split plate). While all 3 cases have shown mixing of the two phases, grid 40
deviates from the other cases. The results from grid 60 and grid 80 are comparable with
an average difference of 3.5%. A possible reason for the deviation of grid 40 is because
the grid size may be too large and fails to simulate the motion of the 0.5 mm drops.
In conclusion, grid 60 will be used for subsequent simulations in Chapter 8 as it offers the
optimal grid structure with a relatively short runtime compared with grid 80.
232

Phase Inversion in Dispersed Liquid-Liquid Pipe Flow: Kwun Ho Ngan

Uploaded by

Copyright:

Available Formats

Phase Inversion in Dispersed Liquid-Liquid Pipe Flow: Kwun Ho Ngan

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Phase Inversion in Dispersed Liquid-Liquid Pipe Flow: Kwun Ho Ngan

Uploaded by

Copyright:

Available Formats

Phase Inversion in dispersed liquid-liquid pipe flow

A thesis submitted to University College London in fulfillment of the requirements

for the degree of Doctor of Philosophy

Department of Chemical Engineering

University College London

London WC1E 7JE

Experimental investigation on phase inversion and associated phenomena were

A mechanistic model is developed to predict the flow characteristics as well as the

Chapter 2: Literature Review

Chapter 3: Facility and Instrumentation

Chapter 4: The Occurrence of Phase Inversion

Chapter 5: Effect of Interfacial Tension on Phase Inversion

Chapter 6: Prediction of Phase Inversion through Fluid Viscosities

Chapter 7: Prediction of Pressure Gradient during Phase Inversion

Chapter 8: CFD Simulation of Horizontal Two-Phase Pipe Flow

Chapter 9: Thesis Conclusion and Future Work

Appendix A – Chord Length Measurement during Phase Inversion 214

Dual continuous flow with several oil drops entrained in the

Annular flow with an oil core and a water annulus. Small

(a) Dispersed flow of oil drops in water. Drops observed are

A schematic on the proposed phase inversion mechanism by

The existence of ambivalent region between the aqueous and

Droplet in drop for a glycerol/water and chlorobenzene system

(a) Pressure gradient measurement of a Exxsol D80/water

Multiple dispersion of an oil/water system at a mixture velocity

Visual observations of (a) O/W/O and (b) W/O/W dispersions

Comparison between calculated drop size distribution and

Figure 3.2 Photograph of the pilot scale facility in stainless steel. 83

Acrylic Y-inlet with a split plate to enhance stratification of the

Dispersion inlets with staggered nozzles. (a) The dispersed

Setup of the high speed video imaging unit, Kodak HS 4540

Section of pipe with quick closing valves in the 38mm stainless

Conductivity ring probe for phase continuity measurement at

Local conductive wire probe for phase continuity measurement

(a) Stainless steel electrodes embedded across the pipe

Sample ERT reconstructed image generated from ITS toolsuite.

In-situ liquid hold-up comparison between electrical resistance

A sample set of normalized signal from the two impedance

A sample frequency plot of the time delay between the two

Pressure measuring port at stainless steel pipeline. Similar ports

Visual images are taken in an acrylic pipe using an high speed

Phase distribution in an acrylic pipe cross section during the

Normalized conductivity data of the oil/water system in an

Normalized conductivity data of the oil/water system in a

Phase distribution in a pipe cross section during the transition

Comparison of normalized conductivity data for the oil/water

Experimental pressure gradients in an acrylic pipe at a mixture

Comparison of pressure gradient for the oil/water system in an

Comparison of pressure gradient measurement between mixture

Chord length distribution of oil drops in water (40% water

Sample drop size distribution is recorded at 14mm from the top

Sauter mean diameter at different input water fraction across

Normalized conductivity data for an oil/pure water mixture at a

Normalized conductivity data for an oil/pure water mixture at a

Comparison of chord length distribution at different input water

(a) Normalized conductivity data of the oil/1% glycerol solution

A comparison of the normalised conductivities in the middle of

Normalized conductivity data of the oil/0.5% glycerol solution

Experimental pressure gradient for pure water, 0.5% 1%