Michigan Technological University

Digital Commons @ Michigan

Tech
Dissertations, Master's Theses and Master's Reports
Dissertations, Master's Theses and Master's Reports
- Open

2015

DYNAMIC MESHING AROUND FLUID-

FLUID INTERFACES WITH APPLICATIONS
TO DROPLET TRACKING IN
CONTRACTION GEOMETRIES
Ahmad Baniabedalruhman
Michigan Technological University

Copyright 2015 Ahmad Baniabedalruhman

Recommended Citation
Baniabedalruhman, Ahmad, "DYNAMIC MESHING AROUND FLUID-FLUID INTERFACES WITH APPLICATIONS TO
DROPLET TRACKING IN CONTRACTION GEOMETRIES", Dissertation, Michigan Technological University, 2015.
http://digitalcommons.mtu.edu/etds/1005

Follow this and additional works at: http://digitalcommons.mtu.edu/etds

Part of the Applied Mathematics Commons

 DYNAMIC MESHING AROUND FLUID-FLUID INTERFACES WITH

APPLICATIONS TO DROPLET TRACKING IN CONTRACTION GEOMETRIES

By

Ahmad Baniabedalruhman

A DISSERTATION

Submitted in partial fulfillment of the requirements for the degree of

DOCTOR OF PHILOSOPHY

In Mathematical Sciences

MICHIGAN TECHNOLOGICAL UNIVERSITY

2015


c 2015 Ahmad Baniabedalruhman
This dissertation has been approved in partial fulfillment of the requirements for the Degree

of DOCTOR OF PHILOSOPHY in Mathematical Sciences.

Department of Mathematical Sciences

Dissertation Advisor: Prof. Feigl, Kathleen A

Committee Member: Prof. Tanner, Franz X

Committee Member: Prof. Xu, Zhengfu

Committee Member: Prof. Yang, Song L

Department Chair: Prof. Gockenbach, Mark S

Contents

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxi

ABBREVIATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxii

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxv

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Mathematical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 Equations to Describe Interface . . . . . . . . . . . . . . . . . . . . . . . . 13

2.2.1 Level Set Method (LSM) . . . . . . . . . . . . . . . . . . . . . . . 14

2.2.2 Volume of Fluid Method (VOF) . . . . . . . . . . . . . . . . . . . 15

2.3 Rheology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.3.1 Power Law Model . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.3.2 Carreau-Yasuda Model . . . . . . . . . . . . . . . . . . . . . . . . 20

v
 2.4 Numerical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.4.1 Finite Volume Method Discretization . . . . . . . . . . . . . . . . 21

2.4.1.1 Discretization of Convection Term . . . . . . . . . . . . . 25

2.4.1.2 Discretization of Diffusion Term . . . . . . . . . . . . . . 29

2.4.1.3 Discretization of Source Term . . . . . . . . . . . . . . . 33

2.4.1.4 Temporal Discretization . . . . . . . . . . . . . . . . . 34

2.4.2 Pressure-Velocity Coupling . . . . . . . . . . . . . . . . . . . . . . 37

2.4.2.1 The Semi-Implicit Method for Pressure-Linked Equation

(SIMPLE) algorithm . . . . . . . . . . . . . . . . . . . . 40

2.4.2.2 Pressure Implicit with Splitting of Operators (PISO)

algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.4.2.3 Merged PISO-SIMPLE (PIMPLE) algorithm . . . . . . . 43

2.4.3 Linear Solvers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3 Dynamic Meshing For Two-Phase Flows . . . . . . . . . . . . . . . . . . . . . 51

3.1 Discretization of Volume Fraction Equation . . . . . . . . . . . . . . . . . 52

3.2 Description of interFoam Solver . . . . . . . . . . . . . . . . . . . . . . . 54

3.3 Dynamic Mesh Refinement in interDyMFoam Solver . . . . . . . . . . . . 58

3.4 Test of interFoam and interDyMFoam in 3D . . . . . . . . . . . . . . . . . 64

3.4.1 Mesh Independence Study . . . . . . . . . . . . . . . . . . . . . . 68

3.4.2 Comparison Between interFoam And interDyMFoam Using Serial

Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

vi
 3.4.3 Effect of Parallelization on Efficiency of interDyMFoam . . . . . . 74

3.5 Modifications to the interDyMFoam Solver . . . . . . . . . . . . . . . . . 75

3.6 Test of interFoam and interDyMFoam in 2D Planar Geometry . . . . . . . 81

3.6.1 Drop Deformation and Break Up in Simple Shear Flow . . . . . . . 81

3.6.1.1 Test Case Using Ca = 0.3 . . . . . . . . . . . . . . . . . 82

3.6.1.2 Test Case Using Ca = 0.4 . . . . . . . . . . . . . . . . . 88

3.6.2 A Drop Detachment From a Micro T-channel . . . . . . . . . . . . 92

3.6.2.1 Mesh Independence Study . . . . . . . . . . . . . . . . . 99

3.6.2.2 dynamicMeshDict Parameters Study . . . . . . . . . . . 100

3.7 Test of interFoam and interDyMFoam in 2D Axisymmetric Geometry . . . 104

3.8 Summary and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

4 Break up Conditions Inside a Spray Nozzle . . . . . . . . . . . . . . . . . . . 113

4.1 Problem Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

4.2 Single Phase Flow Calculations . . . . . . . . . . . . . . . . . . . . . . . . 119

4.3 Drop Tracking Along Streamlines . . . . . . . . . . . . . . . . . . . . . . 124

4.4 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 136

5 Summary and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

A interFoam and interDyMFoam solvers . . . . . . . . . . . . . . . . . . . . . . 151

vii
B Modifications to interDyMFoam . . . . . . . . . . . . . . . . . . . . . . . . . 159

C nozzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

viii
List of Figures

2.1 Schematic diagram of a two-phase flow . . . . . . . . . . . . . . . . . . . 10

2.2 Viscosity vs shear rate for different fluids. . . . . . . . . . . . . . . . . . . 18

2.3 Viscosity vs shear rate for a Carreau-Yasuda model. . . . . . . . . . . . . 20

2.4 Arbitrary control volume. . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.5 Face interpolation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2.6 Vectors S and d on a non-orthogonal mesh. . . . . . . . . . . . . . . . . . 29

2.7 Vectors Δ and K in the minimum correction approach. . . . . . . . . . . 31

2.8 Vectors Δ and K in the over-relaxed approach. . . . . . . . . . . . . . . . 32

2.9 v-cycle and V-cycle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.1 A hexahedral cell with a point in the middle . . . . . . . . . . . . . . . . . 60

3.2 A point in the middle of a face . . . . . . . . . . . . . . . . . . . . . . . . 61

3.3 A points in the middle of an edges . . . . . . . . . . . . . . . . . . . . . . 61

3.4 Divide a face into four faces . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.5 Internal face added to the cell . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.6 Example of a dynamicMeshDict file. . . . . . . . . . . . . . . . . . . . . . 64

ix
3.7 Schematic diagram of a drop of radius 1 mm centered in a channel. The

x-axis and y-axis are horizontally and vertically, respectively, and the

positive z-axis point out of the paper. . . . . . . . . . . . . . . . . . . . . . 65

3.8 A droplet at steady-state for the 3D drop in shear flow test case (Ca = 0.3). . 66

3.9 Velocity at steady-state along line y = 0 using interFoam for the 3D drop in

shear flow test case Ca = 0.3. The vertical lines indicate the boundary of

the drop along y = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

3.10 Pressure at steady-state along line y = 0 using interFoam for the 3D drop

in shear flow test case Ca = 0.3. The vertical lines indicate the boundary of

the drop along y = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3.11 Velocity at steady-state along line y = 0 using interDyMFoam for the 3D

drop in shear flow test case Ca = 0.3. The vertical lines indicate the

boundary of the drop along y = 0. . . . . . . . . . . . . . . . . . . . . . . 72

3.12 Pressure at steady-state along line y = 0 using interDyMFoam for the 3D

drop in shear flow test case Ca = 0.3. The vertical lines indicate the

boundary of the drop along y = 0. . . . . . . . . . . . . . . . . . . . . . . 73

3.13 The points in the center of the faces and edges . . . . . . . . . . . . . . . . 76

3.14 The faces are divided into four new faces . . . . . . . . . . . . . . . . . . . 76

3.15 The faces are divided into two new faces . . . . . . . . . . . . . . . . . . . 77

3.16 Internal faces are added to the cell . . . . . . . . . . . . . . . . . . . . . . 78

3.17 A cell with divided face and two internal faces added in axisymmetric case . 79

x
3.18 Example of a dynamicMeshDict for 2D simulations . . . . . . . . . . . . . 80

3.19 Dynamic refinement in 2D at t = 0 s (top), t = 0.005 s (middle), and t =

0.99 s (bottom) for the 2D drop in a shear flow test case . . . . . . . . . . . 83

3.20 Dynamic refinement in 2D at t = 0.005 s for the 2D drop in a shear flow

test case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

3.21 Drop at steady-state t = 0.99 s for the 2D drop in a shear flow test case . . . 84

3.22 Velocity using interFoam on three different meshes Ca = 0.3 for the 2D

drop in a shear flow test case . . . . . . . . . . . . . . . . . . . . . . . . . 85

3.23 Velocity using interDyMFoam on three different meshes Ca = 0.3 for the

2D drop in a shear flow test case . . . . . . . . . . . . . . . . . . . . . . . 85

3.24 Pressure using interFoam on three different meshes Ca = 0.3 for the 2D

drop in a shear flow test case . . . . . . . . . . . . . . . . . . . . . . . . . 86

3.25 Pressure using interDyMFoam on three different meshes Ca = 0.3 for the

2D drop in a shear flow test case . . . . . . . . . . . . . . . . . . . . . . . 87

3.26 Drop breakup in 2D using Ca = 0.4 at t = 0.99 s using interFoam (top) and

interDyMFoam (bottom) for the 2D drop in a shear flow test case . . . . . . 89

3.27 Dynamic refinement in 2D with Ca = 0.4 at t = 0.99 s for the 2D drop in a

shear flow test case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

3.28 Dynamic refinement in 2D with Ca = 0.4 at t = 0.99 s for the 2D drop in a

shear flow test case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

xi
3.29 Geometry sketch for a drop detachment from a micro T-channel test case

where the units are in micrometers . . . . . . . . . . . . . . . . . . . . . . 94

3.30 Drop deformation and detachment at t = 0.01, 0.012, and 0.014 s using

interDyMFoam for a drop detachment from a micro T-channel test case . . 96

3.31 Drop deformation and detachment at t = 0.01, 0.012, and 0.014 s using

interFoam for a drop detachment from a micro T-channel test case . . . . . 97

3.32 Dynamic refinement at t = 0, 0.004, and 0.014 s for a drop detachment from

a micro T-channel test case . . . . . . . . . . . . . . . . . . . . . . . . . . 98

3.33 Pressure using interFoam for a drop detachment from a micro T-channel

test case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

3.34 Pressure using interDyMFoam for a drop detachment from a micro

T-channel test case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

3.35 Geometry sketch of a bubble rising in a water axisymmetric case . . . . . . 105

3.36 Drop deformation and detachment at t = 0.335, 0.34, and 0.35 s using

interDyMFoam for a bubble rising in water test case . . . . . . . . . . . . . 107

3.37 Drop deformation and detachment at t = 0.35, 0.355, and 0.365 s using

interFoam for a bubble rising in water test case . . . . . . . . . . . . . . . 108

3.38 Dynamic refinement around the interface for a bubble rising in water test case109

3.39 Pressure using interFoam for a bubble rising in a water axisymmetric test

case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

xii
3.40 Pressure using interDyMFoam for a bubble rising in a water axisymmetric

test case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

4.1 Schematic diagram of the nozzle geometry used in experiments (left) and

the computational domain used in the simulations (right). . . . . . . . . . . 118

4.2 Viscosity vs shear rate of the non-Newtonian continuous phase fluid

predicted by the Bird-Carreau model. . . . . . . . . . . . . . . . . . . . . . 119

4.3 Computational domain and number of blocks for the nozzle. . . . . . . . . 121

4.4 Residual using Newtonian (solid curves) and non-Newtonian (dashed

curves) fluids for the refined mesh for the nozzle. . . . . . . . . . . . . . . 122

4.5 Velocity along the centerline for the single phase calculations using

Newtonian (solid curves) and non-Newtonian (dashed curves) for the

nozzle. The vertical dashed lines at x = 0 and x = 5.5 mm indicate the

contracting part of the domain. . . . . . . . . . . . . . . . . . . . . . . . . 122

4.6 Pressure along the centerline for the single phase calculations using

Newtonian (solid curves) and non-Newtonian (dashed curves) for the

nozzle. The vertical dashed lines at x = 0 and x = 5.5 mm indicate the

contracting part of the domain. . . . . . . . . . . . . . . . . . . . . . . . . 123

4.7 Nozzle streamlines at y equal to 0.75, 1, 1.5, 2, and 2.5 millimeter. . . . . . 126

xiii
4.8 Shear rates as a function of transit time along a set of streamlines for the

Newtonian (solid curves) and non-Newtonian (dashed curves) continuous

phase fluid. Along each streamline, t = 0 corresponds to beginning of the

contraction at x = 1 mm. . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

4.9 Mesh around the drop interface in the low-shear-rate upstream (top) and

high-shear-rate downstream (bottom) portions of the domain for the nozzle. 128

4.10 Drop deformation at t = 0.01, 0.08, and 0.13 s for the nozzle. . . . . . . . . 129

4.11 Drop deformation and breakup for streamline y = 1.5 at t = 0.02, 0.42, and

0.5 s for the non-Newtonian continuous phase for the nozzle. . . . . . . . . 130

4.12 Critical drop sizes as a function of the streamline position (top) and the

downstream shear rate (bottom) for the Newtonian and non-Newtonian

continuous phase for the nozzle. . . . . . . . . . . . . . . . . . . . . . . . 131

4.13 Critical Capillary number as a function of the streamline position for the

Newtonian and non-Newtonian continuous phase for the nozzle. . . . . . . 132

4.14 Breakup position of a drop along a given streamline in the Newtonian

continuous phase (top) and non-Newtonian continuous phase (bottom) for

the nozzle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

4.15 Critical capillary number vs viscosity ratio (Grace curve). . . . . . . . . . . 135

xiv
4.16 Critical capillary number as a function of viscosity ratio along two

streamlines in the Newtonian continuous phase for the nozzle. The dashed

vertical lines represent the range of viscosity ratios encountered for the

original drop viscosity (see Table 4.1). . . . . . . . . . . . . . . . . . . . . 136

xv
xvi
List of Tables

3.1 Boundary conditions for the 3D drop in shear flow test case, where Ca is

the capillary number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.2 DynamicMeshDict parameters for the 3D drop in shear flow test case . . . . 68

3.3 Initial mesh and number of cells using interFoam and interDyMFoam for

the 3D drop in a shear flow test case . . . . . . . . . . . . . . . . . . . . . 69

3.4 CPU time and cell size around the interface using interFoam and

interDyMFoam for the 3D drop in a shear flow test case (Ca = 0.3) . . . . . 73

3.5 interDyMFoam in parallel for the 3D drop in a shear flow test case (Ca = 0.3) 74

3.6 Boundary conditions for the 2D drop in a shear flow test case where Ca is

the capillary number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

3.7 Initial mesh and number of cells for the 2D drop in a shear flow test case . 82

3.8 CPU time, cell size around the interface, and relative change in radius for

the 2D drop in a shear flow test case (Ca = 0.3) . . . . . . . . . . . . . . . 87

3.9 CPU time, cell size around the interface, and relative change in radius for

the 2D drop in a shear flow test case (Ca = 0.4) . . . . . . . . . . . . . . . 90

xvii
3.10 Number of cells in each block for a drop detachment from a micro

T-channel test case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

3.11 Boundary conditions for a drop detachment from a micro T-channel test case 95

3.12 CPU time and ratio of a drop radius to the pore radius using maximum

refinement 1 for a drop detachment from a micro T-channel test case . . . . 99

3.13 CPU time and Ratio of a Droplet to the radius of the Pore with Different

Refine Interval Numbers using interDyMFoam solver with max. refinement

equal to 3 and buffer layer equal to one for a drop detachment from a micro

T-channel test case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

3.14 CPU time and Ratio of a Droplet to the radius of the Pore with Different

Maximum Refinement Numbers using refine interval equal and buffer layer

equal to one for a drop detachment from a micro T-channel test case . . . . 102

3.15 CPU time and Ratio of a Droplet to the radius of the Pore with Different

Number of Buffer Layers using refine interval equal to one and max.

refinement equal to 2 and 4 for a drop detachment from a micro T-channel

test case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

3.16 Number of cells in each block for the standard mesh of a bubble rising in a

water axisymmetric case . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

3.17 Boundary conditions of a bubble rising in a water axisymmetric test case . . 106

3.18 CPU time, bubble radius, and relative change using interFoam and

interDyMFoam for a bubble rising in a water axisymmetric test case . . . . 110

xviii
4.1 Fluid parameters used in the simulations. The parameters for the

non-Newtonian fluid correspond to the Bird-Carreau viscosity model,

Eq. (4.1). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

4.2 Boundary conditions for the nozzle. . . . . . . . . . . . . . . . . . . . . . 119

4.3 Number of cells in each block for the nozzle. . . . . . . . . . . . . . . . . . 121

4.4 DynamicMeshDict parameters for the nozzle. . . . . . . . . . . . . . . . . 124

4.5 Drops breakup location for streamlines y = 1.5 and y = 2 in the

non-Newtonian continuous phase for the nozzle. . . . . . . . . . . . . . . . 134

C.1 Critical breakup radius, capillary number, viscosity ratio, and the breakup

position in the nozzle for the non-Newtonian fluid . . . . . . . . . . . . . . 191

C.2 Critical breakup radius, capillary number, viscosity ratio, and the breakup

position in the nozzle for the Newtonian fluid . . . . . . . . . . . . . . . . 192

C.3 Critical breakup radius, capillary number, viscosity ratio, and the breakup

position in the nozzle for the Newtonian fluid at stream line y = 1 . . . . . . 192

C.4 Critical breakup radius, capillary number, viscosity ratio, and the breakup

position in the nozzle for the Newtonian fluid at stream line y = 2 . . . . . . 193

xix
Acknowledgments

I wish to express my gratitude to my advisors, Prof. Kathleen Feigl and Prof. Franz Tanner,

for their continuous support, understanding, guidance, and encouragement throughout my

research and writing the thesis. Also, I would like to thank the other committee members,

Prof. Song Yang and Prof. Zhengfu Xu, for their valuable comments and taking time to

serve on my committee.

Furthermore, I would like to acknowledge the collaboration and interesting discussion

with my friends and colleagues in the CFD group: Dr. Abdallah Al-Habahbeh, William

Case, Chao Liang, Samer Alokaily, and Olabanji Shonibare. Also, the department of

mathematical sciences at Michigan technological university is gratefully acknowledged for

their financial support.

Finally, I would like to thank my father, mother, brothers, and sisters for their support

and encouragement. My special thanks to my wife Ruba for her assistance, patience,

motivation, and encouragement. I appreciate her help and comments throughout writing

the thesis and I will always remember her.

xxi
ABBREVIATIONS

Acronyms

CV Control Volume

FVM Finite Volume Method

Greek symbols

T Cauchy stress tensor

τ viscous tensor

γ̇ shear rate

η0 viscosity at zero shear rate

η∞ viscosity at infinity shear rate

xxii
ρ Density

Roman symbols

g Body force per unit mass

P Pressure

v Fluid velocity

xxiii
Abstract

The dynamic meshing procedure in an open source three-dimensional solver for calculating

immiscible two-phase flow is modified to allow for simulations in two-dimensional planar

and axisymmetric geometries. Specifically, the dynamic mesh refinement procedure, which

functions only for the partitioning of three-dimensional hexahedral cells, is modified for

the partitioning of cells in two-dimensional planar and axisymmetric flow simulations.

Moreover, the procedure is modified to allow for computing the deformation and breakup

of drops or bubbles that are very small relative to the mesh of the flow domain. This is

necessary to avoid mass loss when tracking small drops or bubbles through flow fields.

Three test cases are used to validate the modifications: the deformation and breakup of

a two-dimensional drop in a linear shear field; the formation and detachment of drops

in a two-dimensional micro T-junction channel; and an axisymmetric bubble rising from

a pore into a static liquid. The tests show that the modified code performs very well,

giving accurate results for much less computational time when compared to corresponding

simulations without dynamic meshing.

The modified code is then applied to study drop breakup conditions inside a spray nozzle

when an emulsion is sprayed to produce a powder. This is done by tracking droplets of

various sizes through the flow field within the nozzle and determining conditions under

which they break up. The particular interest is in determining the largest drop sizes for

which breakup does not occur. The effects of viscosity ratio, capillary number, shear rate,

xxv
and fluid rheology on the critical drop sizes are determined.

Although the code modifications performed for this research were implemented for

dynamic mesh refinement of cells close to fluid-fluid interfaces, they may be adapted to

other regions in the domain and for other types of flow problems.

xxvi
Chapter 1

Introduction

Two-phase flow is a flow of a fluid system composed of two different kinds of matter,

e.g., solid particles in a gas or liquid, gas bubbles in a liquid or liquid droplets in a

gas stream or another immiscible liquid. An interface is a surface separating the two

phases of the fluid system. The study of two-phase flows is very important because of

their widespread applications in industry. Their applications includes lubrication, spray

processes, fluid-particle transport, food stuff processing (emulsions, foams), nuclear reactor

cooling and material manufacturing.

An emulsion is a mixture of two immiscible liquids of which one is dispersed in another.

The dispersed phase can be either droplets of a single fluid, in which case the fluid

system is called a simple emulsion, or the dispersed phase can itself be an emulsion, in

which case the fluid system is called double emulsion or multiple emulsion. The most

1
common types of emulsions are water-in-oil (w-o) and oil-in-water (o-w). Hydrophilic

or lipophilic surfactants are encapsulated to produce stable emulsions by reducing the

interfacial tension between the phases. Emulsions are inherently unstable due in part to

coalescence and compositional ripening [1], [2], [3]. Coalescence is the process by which

droplets merge with each other to form larger droplets, whereas compositional ripening

occurs by diffusion and/or permeation of the surfactants components across the disperse

phase. Hence their stability phenomena and the production of stable emulsions are studied

by many researchers [4, 5, 6, 7, 8]. Producing powders from emulsions reduces the problem

of stability and increases the shelf life. Many researchers have worked on spraying of

emulsions where they mostly studied simple emulsions rather than multiple emulsions

[9, 10, 11, 12]. Producing powders by spraying multiple emulsions is more complex in

terms of preserving its structure [13]. Many studies have looked at the effect of the spraying

process and the viscosity ratio on the spray drop size [14, 15, 16, 17, 18, 19]. A uniform

drop size distribution is desirable with drop radii on the order of microns.

The processing of emulsions can be studied computationally by numerically solving a

two-phase flow problem in which the location of the fluid-fluid interface must be computed

along with the flow variables, such as velocity and pressure. There are different numerical

approaches to solve two-phase problems. Two popular approaches are interface tracking

and interface capturing methods. In the interface tracking method, a mesh to track the

interface is needed and mesh points lie on the interface. In this method, the interface is

explicitly described by the computational mesh and the mesh is updated if the interface is

2
moved so that mesh points remain on the interface [20]. On the other side, the interface

capturing methods implicitly describe the interface by an artificial function where the mesh

is fixed. The most popular interface capturing methods are the level set [21, 22] and

volume-of-fluid [23, 24] methods. In the level set method, the signed distance function

φ is used to describe the interface, where φ is zero at the interface, positive in the dispersed

phase, and negative in the continuous phase. In addition, the level set function is smooth,

allowing for an accurate calculation for the curvature κ . The volume-of-fluid method uses

a discontinuous volume fraction function α instead of the level set function. The volume

fraction function α is one in the dispersed phase and zero in the continuous phase. Once

a mesh is introduced, the value of α in a cell is the volume fraction of the dispersed phase

in the cell. Therefore, 0 < α < 1 in cells that contains the interface. The volume-of-fluid

approach has much better mass conserving properties than the level set approach, but a

major challenge is accurately calculating the curvature κ . In order to obtain accurate

two-phase flow calculations, a sufficiently refined mesh around the interface is required.

Instead of refining the mesh throughout the whole domain, dynamic mesh refinement can

be used.

Dynamic mesh refinement allows an accurate solution with low costs by having high

mesh resolution in specific regions, for example, around the interface in two-phase flow

problems. It reduces the costs in terms of computational time and storage compared to a

refined uniform mesh. Dynamic mesh refinement was studied on structured grids by Berger

et al. [25]. In general, there are two methods for the adaptive mesh refinement. The first is

3
r-refinement in which the number of grid points and cells are fixed, and the grid points are

redistributed on the mesh to produce high resolution (i.e. small cells) in particular places.

The second method is h-refinement in which new points and cells are added to the mesh

in order to have sufficient resolution in desired regions. The h-refinement is achieved by

dividing a set of cells into smaller cells. Many scholars have worked on the dynamic mesh

refinement, including Mavriplis [26, 27] who formulated an adaptive mesh refinement for

an unstructured mesh; Pizadeh [28] who developed an unstructured grid adaption using

different adaptive techniques; and Anderson [29] who developed an algorithm to solve

Euler equations by combining the staggered grid arbitrary Lagrangian-Eulerian techniques

with structured local adaptive mesh refinement. In addition, Coirier [30] developed an

adaptively-refined, Cartesian, cell-based scheme for the Euler and Navier-Stokes equations,

while Hunt [31] developed a code to solve three dimensional equations using adaptive

refinement; and Qingluan [32] developed an adaptive mesh refinement algorithm for

engine spray simulations where the refinement is required in the spray region. Also,

the interDyMFoam solver for two-phase flow in OpenFOAM

R
uses a dynamic mesh

refinement for three dimensional hexahedral meshes.

OpenFOAM
R
stands for Open Field Operation And Manipulation. It is an open source

code using C++ libraries and serves as a modeling and computational fluid dynamic (CFD)

platform for the research community. The mesh generation, equations discretization, and

matrix manipulations can be accomplished using applications of source codes and libraries

in OpenFOAM
R
[33]. The software contains solvers for many computational fluid

4
dynamics problems ranging from fluid flow including heat transfer, chemical reactions,

and turbulence to solid dynamics and electromagnetic. New solvers or modifications to

existing solvers or libraries can be constructed by the user to meet the needs of his/her

specific application.

OpenFOAM
R
has the ability to study multi-phase flows, specifically through the

interFoam and interDyMFoam solvers. In this thesis, a modification of the

interDyMFoam solver is achieved for application to 2D planar and axisymmetric

flows. Two cases are studied to validate the modification for the 2D planar simulations,

specifically (1) a droplet in a planar linear shear flow, and (2) droplets detaching from a

pore into a shear flow field. Furthermore, a bubble rising in water is used to validate the

axisymmetric simulations.

The modified interDyMFoam is then used to study breakup conditions of drops inside a

nozzle when an emulsion is sprayed to produce a powder. Dynamic meshing around the

interface is necessary in this application since the drop sizes can be very small relative to

the geometry. Moreover, due to the large number of drops that need to be tracked, the

simulations are performed in two dimensions. The effect of capillary number, viscosity

ratio, shear rate, and fluid rheology on the critical break up radius is studied.

Contributions of this thesis

This dissertation makes several contributions to the field of Computational Fluid Dynamics

and the understanding of drop breakup conditions in complex geometries. The major

contributions are:

5
1. The dynamic meshing capabilities of a popular open source CFD software package

used in the research community has been improved.

2. The dynamic meshing in a two-phase flow solver has been modified to allow

for dynamic meshing around fluid-fluid interfaces in two-dimensional planar and

axisymmetric geometries. The modification to these geometries serves to reduce

computational time and allows for application to problems in which many small

drops or bubbles must be tracked.

3. The modified dynamic meshing code was applied to three test problems:

(a) Drop deformation and breakup in linear shear flow (two-dimensional planar).

(b) Drop formation and detachment from micro T-channels (two-dimensional

planar).

(c) Bubble rising from a pore into static liquid (axisymmetric).

The performance of the modified code was evaluated on these three test problems

in terms of computational time, mesh independence, and mass conservation.

Comparisons were made with the two-phase flow solver without dynamic meshing.

4. The effect of the dynamic meshing parameters on the results was determined for the

micro T-channel test problem.

5. The dynamic meshing code was further modified to improve the refinement around

droplets as they move in a flow field. This was necessary to prevent mass loss when

6
 tracking droplets that are very small relative to the flow domain length scale and

corresponding mesh.

6. The modified dynamic meshing code was used to track droplets in a contraction

geometry representing a spraying nozzle. From this, in-nozzle drop breakup

conditions were investigated when spraying an emulsion. Of particular interest to

spray engineers are critical drop sizes, that is, the largest drops that will not break up

within the nozzle.

7. The effect of shear rate, rheology of the continuous phase fluid, nozzle length,

capillary number, and viscosity ratio (i.e., drop viscosity relative to continuous phase

viscosity) on critical drop sizes was determined.

8. Grace curves for this geometry, which give the critical capillary number as a function

of viscosity ratio, were determined. Results of the above parameter study were

interpreted in terms of these Grace curves.

9. The modified dynamic meshing procedure may be used for other two-dimensional

and axisymmetric problems solved with OpenFOAM

R
.

7
Chapter 2

Mathematical Model

Fluid dynamics is the science which studies the motion of liquids and gases and how

they interact with the environment. It has applications in many fields and its uses include

calculating forces and moments on aircraft, determining the mass flow rate of petroleum

through pipelines and predicting weather patterns. The solution of a fluid dynamics

problem involves calculating various properties of the fluid, such as velocity, density,

pressure and temperature, as a function of space and time. Continuum mechanics treats

the material as a continuous mass instead of discrete particles. The basic conservation

laws of continuum mechanics are the conservation of mass, conservation of momentum

and conservation of energy. From these conservation laws we can derive the differential

equations that describe the properties of the fluid and flow.

Two-phase flow is best described as the flow of a fluid system composed of two different

9
 Fluid 2 S

nS

Fluid 1

Figure 2.1: Schematic diagram of a two-phase flow

kinds of matter, e.g., solid particles in a gas or liquid, gas bubbles in a liquid, or liquid

droplets in a gas stream or another immiscible liquid. An interface is a surface separating

the two phases of the fluid system. Two-phase flow has many industrial applications such

as lubrication, spray processes, fluid-particle transport, food stuff processing (emulsions,

foams), nuclear reactor cooling and material manufacturing.

2.1 Governing Equations

Two-phase flow is a flow of two fluids separated from each other by interface S as shown

in Figure 2.1, where nS is the unit normal vector on the interface S directed to fluid 2. In

this thesis, we are primarily interested in the two-phase flow where both fluids are liquids.

10
From the conservation of mass and momentum principles, the differential form of the

continuity equation is

∂ρ
+ ∇ · (ρ v) = 0, (2.1)
∂t

and the momentum equation is

∂ (ρ v)
+ ∇ · (ρ vv) = ∇ · T + ρ g + f (2.2)
∂t

where ρ is the density, v is the velocity, T is the Cauchy stress tensor, g is the gravity and

f is a force per unit volume. The stress tensor T can be expressed as:

T = −PI + τ (2.3)

where P is the pressure and τ is the viscous stress tensor. Using Eq. (2.3), the momentum

equation becomes

∂ (ρ v)
+ ∇ · (ρ vv) = −∇P + ∇ · τ + ρ g + f . (2.4)
∂t

If changes in ρ are negligible, as in the case of incompressible flow, the equations become

∇ · v = 0, (2.5)

11
 ∂v
ρ[ + (v · ∇)v] = −∇P + ∇ · τ + ρ g + f . (2.6)
∂t

An equation is also needed to describe the evolution of the interface. This equation depends

on the numerical approach that is used to solve the two-phase flow problem, and will be

described later. Interface conditions must also be specified to describe the behavior at

the fluid-fluid interface. There are two conditions on the interface, given below, where

[m]S denote a jump across the interface S of a function m, i.e., [m]s = m1 − m2 , where the

subscripts denote the fluids.

1. Continuous velocity: [v]S = 0 (there is no jump in the velocity across the interface).

2. Jump in surface traction: [T · n]S = Δf , where Δf represents a surface force due

to interfacial tension. A constitutive equation is needed for Δf . A common one is

Δf = σ κ nS , where κ = ∇ · nS is the local mean curvature, nS is the unit normal

vector on the interface, and σ is the interfacial tension. This is a generalization of

the Young-Laplace equation [34] that gives the capillary pressure difference across

the interface S between two static fluids.

There are several techniques for calculating the surface tension force. Some of those

methods are the Continuum Surface Stress method (CSS) [35, 36], ghost fluid method

(GFM) [37, 38], Meier’s method [39, 40] and The Continuum Surface Force method (CSF)

[41]. In the CSF method, these conditions, in particular, the jump in T · n is accounted for

12
in the momentum equation as

fS = σ κ nS δ (x − xS ) (2.7)

where fS is the volumetric surface tension force, κ = ∇ · nS is the local mean curvature,

nS is the unit normal vector on the interface, δ (x − xs ) is the Dirac delta function, and σ

is the interfacial tension. The calculation of the normal vector and curvature is discussed in

Section 2.2.

2.2 Equations to Describe Interface

There are two main approaches for describing the evolution of the interface: Interface

tracking methods and interface capturing methods. In interface tracking methods, the

moving interface is explicitly described by the nodes of the computational mesh. The mesh

must be adjusted so that the nodes lie on the interface. In interface capturing methods,

the location of the moving interface is implicitly described by a scalar function. These

methods are Eulerian in which the mesh is stationary or moving in a given manner. The

most popular interface capturing methods are the level set method (LSM), volume of fluid

method (VOF), and coupled level set-volume-of-fluid (CLSVOF) method. The basic LSM

and VOF method are described below.

13
2.2.1 Level Set Method (LSM)

The level set method captures the motion of an interface by embedding the interface as the

zero level set of the level set function ϕ [22]. The level set function is defined as:

ϕ (x,t) < 0 if x is in fluid 1,

ϕ (x,t) = 0 if x is on the interface and

ϕ (x,t) > 0 if x is in fluid 2

The level set equation is

∂ϕ
+ v · ∇ϕ = 0. (2.8)
∂t

Physically, this equation means that the value ϕ does not change with time along a particle

path since the left hand side of this equation is the material derivative. That is, the interface

is convected with the flow fluid. For example, on the interface the value of ϕ will be the

same at each time. Initially, ϕ (x, 0) is the signed distance function to the interface. The

momentum equation can be written as:

∂v
ρ (ϕ )( + v · ∇v) = −∇P + ∇ · τ + ρ (ϕ )g + σ κ (ϕ )nS (ϕ )δ (ϕ ) (2.9)
∂t

14
where

ρ (ϕ ) = ρ2 + (ρ1 − ρ2 )H(ϕ )

μ (ϕ ) = μ2 + (μ1 − μ2 )H(ϕ )

κ = ∇ · nS and
∇ϕ
nS = |∇ϕ |

where H is the Heaviside function and the subscript in ρ and μ indicate the fluid phase.

Also, μ is the dynamic viscosity, which is used in the constitutive equation for τ as

discussed in Section 2.3.

2.2.2 Volume of Fluid Method (VOF)

The volume-of-fluid method uses a volume fraction function α , instead of level set

function, to describe the location of the interface [23]. The volume fraction function α

is a discontinuous function such that α = 0 in the continuous fluid (fluid 2) and α = 1 in

the dispersed fluid (fluid 1). On a computational mesh, this discontinuity is smoothed by

letting

α = 0 in cells that contain only the fluid 2,

α = 1 in cells that contain only the fluid 1 and

VFluid1
0 < α < 1 in cells where the interface passes such that α = V , where VFluid1 is the

volume of fluid 1 in the cell and V is the volume of the cell.

15
The volume fraction equation is

∂α
+ ∇ · (α v) = 0. (2.10)
∂t

As in the level set method, we can write the surface force term as: fS = σ κδ (x−xS )nS . In

VOF, the CSF fS = σ κδ (x − xS )nS is reformulated as fS = σ κ ∇α [23]. The momentum

equation can be written as:

∂v
ρ (α )( + v · ∇v) = −∇P + ∇ · τ + ρ (α )g + σ κ (α )∇α (2.11)
∂t

where

ρ (α ) = αρ1 + (1 − α )ρ2 ,

μ (α ) = α μ1 + (1 − α )μ2 , and
∇α
κ = −∇ · |∇α | and

where the subscript in ρ and μ indicate the fluid phase.

The level set function is continuous, making the calculation of the unit normal and curvature

accurate, however the method does not guarantee the conservation of mass. On the

other hand, the volume-of-fluid is mass conserving but the volume fraction function is

discontinuous, making it less effective in calculating the unit normal and curvature as

surface tension force increases. In this thesis, we use the VOF method as implemented

in interFoam.

16
2.3 Rheology

To close the system of the governing equations in Section 2.1, an expression for the

stress tensor τ must be specified. This stress depends on the deformation and strain rate

experienced by the fluid.

Rheology is the science that studies the flow and deformation of materials. It describes

the relationship between stress and deformation (strain). The mathematical form of this

relationship is called the constitutive equation.

The common rheology terms are stress τ which is the force acting on an area divided by

that area, strain rate γ̇ which is the rate of change in shape of a deformed material with

respect to time and mathematically defined as the rate-of-strain tensor γ̇ = ∇v + (∇v)T

where v is the velocity, shear rate γ̇ which is the magnitude of the rate-of-strain tensor, and

viscosity which is the quantity that describes a fluid’s resistance to flow.

A Newtonian fluid is a fluid in which the stress is linear in the rate-of-strain tensor. The

constitutive equation for an incompressible Newtonian fluid is given by

τ = μ γ̇ = μ [∇v + (∇v)T ], (2.12)

where μ is the constant dynamic viscosity.

A non-Newtonian fluid is a fluid whose stress is not linear in γ̇. There are two types

of non-Newtonian fluids: time-dependent fluids in which the relation between stress and

17
 shear thickening
viscosity

Newtonian

shear thinning

shear rate
Figure 2.2: Viscosity vs shear rate for different fluids.

strain depends on how long the shear has been active, and time-independent fluids in which

the relation does not depend on the time. The time-independent non-Newtonian fluids are

the most popular and in this thesis only this type of non-Newtonian fluid is considered.

This type of fluid is classified into shear-thinning or pseudo-plastic fluids in which the

viscosity decreases when the shear rate increases, shear-thickening or dilatant fluids in

which the viscosity increases when the shear rate increases, and yield stress fluids in which

a minimum stress is required before the material will flow. Figure 2.2 shows a graph of

viscosity and shear rate for shear thinning, shear thickening and Newtonian fluids.

Generalized Newtonian fluid models assume a simple constitutive equation like the one

for the Newtonian fluid but here the viscosity is a function of the shear rate. The general

18
form of the constitutive equation for the generalized Newtonian fluid models is

τ = η (γ̇ )γ̇ (2.13)

where η (γ̇ ) is the viscosity function. The resulting values of the viscosity at very low and

high shear rates are known as the zero-shear-rate viscosity η0 and the infinite-shear-rate

viscosity η∞ respectively. Two popular models for generalized Newtonian fluid will be

described here.

2.3.1 Power Law Model

The model describes a power-law relation between the viscosity η and shear rate γ̇ , and is

given by:

η = K γ̇ n−1 (2.14)

where K is the consistency coefficient (units of Pa · sn ) which reflects the vertical shift in the

viscosity curve on a log-log plot, and the dimensionless n is the power-law index such that

n − 1 represents the slope of the viscosity curve on a log-log plot and reflects how close the

fluid is to Newtonian. For a Newtonian fluid, n = 1 and the consistency index K is equal

to the viscosity of the fluid. If n < 1, then the fluid is shear-thinning and if n > 1, then the

fluid is shear-thickening. This model is popular because most fluids have a linear relation

19
 η0
log(η) transition region

power law region

η∞

.
log(γ)
Figure 2.3: Viscosity vs shear rate for a Carreau-Yasuda model.

in some region of the log-log graph but it does not describe the zero- and infinite-shear-rate

viscosities.

2.3.2 Carreau-Yasuda Model

A model which does describe the upper and lower shear rate regions is the Carreau-Yasuda

model. The relationship between viscosity and shear rate is given by:

η − η∞ n−1
= [1 + (mγ̇ )a ] a (2.15)
η0 − η ∞

1
The parameter m is a constant with units of time, where m is the critical shear rate at

which viscosity begins to decrease or increase, a is a dimensionless constant which affects

20
the shape of the transition region (e.g., increasing a sharpens the transition), and n is

a dimensionless constant which describes the slope in the power law region as show in

Figure 2.3. The Bird-Carreau model is given by a = 2 in Eq. (2.15), therefore the equation

becomes

η − η∞ n−1
= [1 + (mγ̇ )2 ] 2 . (2.16)
η0 − η∞

2.4 Numerical Methods

The governing system of equations for two-phase flow involving Newtonian and

non-Newtonian fluids is solved using the finite volume method (FVM) described in

this section. Since the momentum equation is a time-dependent convection-diffusion

equation, we first describe the finite volume method for a general time-dependent

convection-diffusion equation, we then discuss its application to incompressible flow

problems.

2.4.1 Finite Volume Method Discretization

The finite volume method (FVM) is a popular method to solve numerically the governing

equations of fluid dynamics. In FVM, the computational domain is divided into a finite

21
number of control volumes (CVs) and the governing equations are integrated on the CV to

get the integral form for the equations. The description of the basic FVM below follows in

part that given in the thesis of Jasak [42].

The boundary of a CV contains a number of faces and, it is assumed here that each face in

the domain share at most two CVs. Figure 2.4 shows a CV, where VP is the volume of the

CV, P is a computational point at the centroid of the CV, f is a computational point at the

center of a face, S f is the area of the face f , n f is the face outward unit normal vector, N is

a computational point of a neighboring CV, d f is the vector between P and N and rP is the

vector between the origin and P.

Figure 2.4: Arbitrary control volume.

22
The coordinates of the centroid of the CV, xP , and the face, x f , are given by:


(x − xP )dV = 0, (2.17)
VP


(x − x f )dS = 0. (2.18)
f

The general convection-diffusion equation in fluid dynamics has the form:

∂ (ρ φ)
+ ∇ · (ρ vφ) = ∇ · (ρΓφ ∇φ) + qφ (φ) (2.19)
 ∂t          
convective term diffusion term source term
temporal derivative

where φ is a general property and Γφ is the diffusion coefficient. The key step of FVM is

the integration of Eq. (2.19) over a CV yielding

   
∂ (ρ φ)
dV + ∇ · (ρ vφ)dV = ∇ · (ρΓφ ∇φ)dV + qφ (φ)dV, (2.20)
VP ∂t VP VP VP

By applying the Gauss divergence theorem, Eq. (2.20) can be written as follows:

   
∂ (ρ φ)
dV + n · (ρ vφ)dS = n · (ρΓφ ∇φ)dS + qφ (φ)dV, (2.21)
VP ∂t ∂ VP ∂ VP VP

23
where n is the outward-pointing unit normal vector. By integrating Eq. (2.21) with respect

to time t over a small interval, we get the most general form

 t+δ t   
∂ (ρ φ)
dV + n · (ρ vφ)dS − n · (ρΓφ ∇φ)dS dt
t VP ∂t ∂ VP ∂ VP
 t+δ t 
= qφ (φ)dV dt. (2.22)
t VP

In the FVM discretization, the linear variation is used to approximate the function φ around

the point P. The approximation is a second-order accurate and it is given by:

φ(x) = φP + (x − xP ) · (∇φ)P , (2.23)

where φP = φ(xP ). In the below sections some of the discretization methods are described.

24
2.4.1.1 Discretization of Convection Term

Since each CV is bounded by a number of faces, then


 
(ρ vφ · n)dS = ∑ (ρ vφ · n f )dS . (2.24)
∂ VP f f

By using the assumption of linear variation for φ around the point f , the term ρ vφ is

written as:

ρ vφ(x) = (ρ vφ) f + (x − x f ) · (∇(ρ vφ)) f (2.25)

Therefore, the integral inside the sum above is approximated as following:

  
(ρ vφ · n f )dS = (ρ vφ) f · n f dS + (∇(ρ vφ)) f : (x − x f )n f dS, (2.26)
f f f

where g f stands for the value of g at the center of the face f . Assuming n f is constant on

face f (i.e., that face is a plane surface) and using Eq. (2.18), Eq. (2.26) becomes

 
(ρ vφ · n)dS = (ρ vφ) f · n f dS = (ρ vφ) f · S, (2.27)
f f

25
where S = S f n f is the outward area vector of a face and S f is the face area. The right-hand

side of Eq. (2.24) can be approximated using Eq. (2.27), so that Eq. (2.24) becomes

∂ VP
(ρ vφ · n)dS = ∑(ρ vφ) f · S
f

= ∑ S · (ρ v) f φ f
f

= ∑ Fφ f , (2.28)
f

where

F = S · (ρ v) f (2.29)

is the convective mass flux through the face f . To find F, the values of ρ and v have to be

found at the faces by interpolating from the values at the centroids. A weighted average is

used to estimate φ at the face as in the equation below:

φ f = b f φP + (1 − b f )φN (2.30)

as illustrated in Figure 2.5. Different values for b f gives different methods. Three basic

methods are presented below.

26
 φN
φf
φP

P f N

Figure 2.5: Face interpolation.

1. Central Differencing CD

In CD, b f in Eq. (2.30) is defined as:

fN
bf = (2.31)
PN

where f N is the distance between the face and the computational point N and PN is

the distance between the computational points P and N as shown shown in Figure 2.5.

If the mesh is uniform then b f = 12 . The method is second-order but unphysical

oscillations appear in the solution for convection-dominated problems, which often

makes the solution unbounded. More details are found in Chapter 14 of [43] and

Chapter 4 of [44].

27
2. Upwind Differencing UD

In UD, b f in Eq. (2.30) is defined as:

⎧
⎪
⎪
⎪
⎨1 if F ≥ 0
bf =
⎪
⎪
⎪
⎩0 if F < 0.

where F = S · (ρ v) f is the flux. The unphysical oscillations are removed in this

method because it depends on the flux direction. Also, it is bounded and stable but

it is a first-order accurate because it uses the first-order backward differencing, (see

[42]).

3. Blended Differencing BD

The BD is a combination between CD and UD and defined as:

φ f = (1 − k f )(φ f )UD + k f (φ f )CD (2.32)

where (φ f )UD is the value from the UD, (φ f )CD is the value from CD and k f is a

blending factor between 0 and 1. This method attempts to preserve the accuracy and

boundedness.

28
 S

.P d
.
f
>.N

Figure 2.6: Vectors S and d on a non-orthogonal mesh.

2.4.1.2 Discretization of Diffusion Term

Following the approach used for the convection term, we get the approximation for the

diffusion term

∂ VP
(ρΓφ ∇φ) · ndS = ∑(ρΓφ∇φ) f · S
f

= ∑(ρΓφ) f S · (∇φ) f . (2.33)

f

If the mesh is orthogonal then the estimation for S · (∇φ) f can be defined as:

φN − φ P
S · (∇φ) f =| S | . (2.34)
|d|

29
If the mesh is non-orthogonal, as in Figure 2.6, then S · (∇φ) f can be written as:

S · (∇φ) f = Δ · (∇φ) f + K · (∇φ) f , (2.35)

     
orthogonal contribution non-orthogonal contribution

where Δ is parallel to the vector d and S = Δ + K. The estimation in Eq. (2.34) can be

used to approximate the orthogonal contribution and the non-orthogonal contribution can

be approximated by approximating (∇φ) f using the weighted average as:

(∇φ) f = b f (∇φ)P + (1 − b f )(∇φ)N (2.36)

where b f is the same as in Eq. (2.31) and (∇φ)P can be approximated using the

second-order approximation to the Gauss divergence theorem as follows

 
∇φdV = φ · ndS (2.37)
VP ∂ VP

 
(∇φ)PVP = ∑ f
φ · n f dS (2.38)
f
1
VP ∑
(∇φ)P = Sφ f . (2.39)
f

The integral on the left in Eq. (2.37) is approximated by multiplying the value of the

function at the centroid of the CV by its volume. The integral on the face f is approximated

by using the linear variation of φ on the interface φ(x) = φ f + (x − x f ) · (∇φ) f . There

30
 S ^
K
.P .f Δ
> >.N
d

Figure 2.7: Vectors Δ and K in the minimum correction approach.

are many ways to find the Δ and K vectors. Here, two methods are described.

1. Minimum correction approach

In this method, we choose K to be orthogonal to the vector Δ to keep the

non-orthogonal contribution as small as possible as shown in Figure 2.7. Also, Δ

can be written as:

d·S
Δ= d. (2.40)
d·d

2. Over-relaxed approach

In this method Δ is defined as:

d S ·S
Δ= | S |2 = d. (2.41)
d·S d·S

31
 S ^
K
.P .f Δ
> >.N
d

Figure 2.8: Vectors Δ and K in the over-relaxed approach.

Figure 2.8 shows the vectors Δ and K in the over-relaxed approach. In this

approach, the importance of the term in φP and φN is caused to increase with the

increase in non-orthogonality Δ.

The non-orthogonal correction possibly creates unboundedness, especially when

the non-orthogonality is high. Therefore, if the boundedness is more important

than accuracy, then the non-orthogonal correction has to be reduced or discarded.

However this will result in reducing the order of accuracy [42]. According to Jasak

[42], the over-relaxed approach is the best approach that treats the non-orthogonality

from the aspect of stability, convergence, and computational time. The converged

solution is obtained, even if the non-orthogonality is severe, when the other

approaches cause divergence.

32
2.4.1.3 Discretization of Source Term

The source term qφ (φ) can be a function of φ and it is approximated by the linear

expression

qφ (φ) = qu + q p φ, (2.42)

where qu and q p can also depend on φ. This allows the implicit treatment of the source

term. The integral form of the source term can be approximated as follows


qφ (φ)dV = (qu + q p φ)PVP (2.43)
V

= quVP + q pVP φP . (2.44)

33
2.4.1.4 Temporal Discretization

Using the previous discretization for the convection, diffusion, and source term, Eq. (2.22)

can be written as

⎡ ⎤
 t+δ t ⎢

⎢ ∂ ρφ
 ⎥
⎥  t+δ t
⎢ VP + ∑ Fφ f ⎥
− ∑(ρΓφ ) f S · (∇φ) f ⎥ dt = (quVP + q pVP φP ) dt
⎢ ∂t   
t ⎣ P f f ⎦ t
      source term
convection term diffusion term
(2.45)

 ∂ ρφ
where VP ∂ t dV is approximated by the one-point centroid such as in the other terms. To

finish the discretization the following approximations are used


∂ ρφ ρPn φnP − ρPo φoP
= (2.46)
∂t P δt
 t+δ t
φ(t)dt = (wφo + (1 − w)φn )δ t, (2.47)
t

where φn = φ(t + δ t), φo = φ(t) and w is a constant.

By using the previous equations, assuming the density and diffusivity do not change over

34
time and dividing by δ t, Eq. (2.45) becomes

ρP φnP − ρP φoP  
δt
VP + ∑ (1 − w)Fφnf + wFφof
f
 
− ∑ (1 − w)(ρΓφ ) f S · (∇φ)nf + w(ρΓφ ) f S · (∇φ)of
f

= quVP + (1 − w)q pVP φnP + wq pVP φoP . (2.48)

For different w, various time integration methods can be obtained. For example, the

first-order explicit Euler method is obtained if w = 1, the first-order bounded Euler method

is obtained if w = 0, and the second-order Crank-Nicholson method is obtained if w = 12 .

The values of φ f and (∇φ) f depend on the values of φ in the neighboring cells, therefore

for any CV whose centroid is xP , Eq. (2.48) can be written as

aP φnP + ∑ aN φnN = RP . (2.49)

N

The summation in Eq. (2.49) is over the neighboring cells of the cell with centroid P. From

Eq. (2.48), the coefficients a f and aN result from the coefficients of φP and φN and the

coefficients of these functions resulting from approximating φ f and ∇φ f using φP and

φN . For the whole geometry, this produces a linear system of algebraic equations of the

35
form

By = R (2.50)

where B is a sparse matrix with coefficients aP on the diagonal and aN off the diagonal, y

is the vector with the unknown values of φ on all CVs, and R is the source vector which

contains firstly, the values of the constant part of the source term and secondly, the parts

of convection term, diffusion term and temporal derivative at the old time level. Numerical

approaches to solve the resulting equations, will be discussed later.

The momentum equation is a convection-diffusion equation with the pressure gradient as

a source term. Therefore, it can be discretized using the same methods as in Sections

2.4.1.1–2.4.1.4. However, there are some additional complexities that must be addressed.

Some of those are: (1) there are multiple equations and multiple unknowns such as vx , vy ,

vz , and pressure P; (2) there are nonlinear terms such as the convection ∇ · (vv) and the

viscous stress tensor τ for a non-Newtonian fluid; and (3) the equations are coupled and

a specific treatment is required in order to handle the pressure-velocity coupling. This is

discussed in the next section.

36
2.4.2 Pressure-Velocity Coupling

For incompressible non-Newtonian fluids, using the generalized Newtonian models (see

Section 2.3) the mass and momentum equations have the form

∇·v = 0 (2.51)

∂ (ρ v)
+ ∇ · (ρ vv) − ∇ · η (γ̇ )(∇v + ∇v T ) = −∇P (2.52)
∂t

where η is the viscosity and γ̇ is the shear rate. In the momentum equation, the nonlinear

term ∇ · (vv) appears. This issue is solved by solving a non-linear system, or by

linearization which is the chosen option to reduce the computational time. The non-linear

term is linearized as follows

 
∇ · (ρ vv)dV = (ρ vv · n)dS
VP ∂ VP

= ∑ v f (ρ v)of · S
f

= ∑ F ov f
f

= aP vP + ∑ aN vN , (2.53)
N

37
where v o is the velocity from the previous time step and F o = S · (ρ v)of is the flux from

the previous time step.

Another issue here is the incompressibility where the continuity equation does not involve

density because it is constant which thus results in no explicit equation for pressure. In

this system we have the same number of unknowns and equations. In the spirit of the

Rhie and Chow procedure [45], a pressure equation can be derived from the continuity

and momentum equations as follows. As in Section (2.4.1), the continuity equation can be

discretized as

 
0= ∇ · vdV = v · ndS = ∑ S · v f (2.54)
VP ∂ VP f

and the momentum equation as

aP vP = H(v) − ∇P (2.55)

where

vo
H(v) = − ∑ aN vN + . (2.56)
N δt

From Eq. (2.55), we have

H(v) 1
vP = − ∇P (2.57)
aP aP

38
and using interpolation, v f can be written as

H(v) 1
vf = ( ) f − ( ∇P) f . (2.58)
aP aP

Substituting Eq. (2.58) into Eq. (2.54) yields


H(v) 1 H(v) 1
0 = ∑S · ( ) f − ( ∇P) f = ∑ S · ( ) f − ∑ S · ( ∇P) f (2.59)
f aP aP f aP f aP

and hence

1 H(v)
∑ S · ( aP ∇P) f = ∑ S · ( aP
)f . (2.60)
f f

The pressure gradient can be found by interpolating the pressure field to the cell faces and

Eq. (2.55) can be written as

aP vP = H(v) − ∑ q(P) f . (2.61)

f

Note that the flux F can be calculated as following


H(v) 1
F = S · (ρ ) f − (ρ ∇P) f . (2.62)
aP aP

Eqs. (2.60) and (2.61) are the discrete pressure and velocity equations. Both equations

have two unknowns (P, v) and the following three predictor-corrector methods are the

39
most commonly used to solve them.

2.4.2.1 The Semi-Implicit Method for Pressure-Linked Equation (SIMPLE)

algorithm

The SIMPLE [46] algorithm is a predictor-corrector procedure for calculating the pressure

P and velocity v. It is a solver for steady-state (no time derivative) incompressible single

phase fluid. The (implicit) under-relaxed form of the momentum equation, Eq. (2.61), can

be written as

aP n 1 − αv
vP + ∑ aN vNn = RP + aP vPo , (2.63)
αv N αv

where αv is the velocity under-relaxation factor (0 < αv ≤ 1) and v o is the velocity from

the previous iteration.

The SIMPLE algorithm can be outlined as following:

1. Start with a guessed value of pressure P∗ in the first step and afterwards the resulted

pressure from previous step is then used.

2. Solve the under-relaxed momentum Eq. (2.63) to find the velocity v ∗ by using the

guessed pressure P∗ to find RP . This step is called the momentum predictor.

 ∗)

3. Compute the mass fluxes at the cells faces F ∗ = S · (ρ H(v
aP ) f , which is needed in

40
 the right hand side of the pressure Eq. (2.60).

4. Solve Eq. (2.60) to find the new value for the pressure P∗∗ .

5. Correct the mass fluxes at the cells faces Eq. (2.62) using the new value for the

pressure P∗∗ , F = F ∗ − (ρ a1P ∇P∗∗ ) f · S.

6. Apply some explicit pressure under-relaxation factor 0 < αP ≤ 1 to find the new

pressure Pnew = P∗ + αP (P∗∗ − P∗ ).

7. Calculate the corrected velocity v new using Eq. (2.57) and the new pressure value

Pnew .

8. For a non-Newtonian fluid, update the viscosity from a generalized Newtonian

constitutive equation using the corrected velocity.

9. Test for convergence and repeat the steps from step 2 assuming the new pressure Pnew

as the guessed pressure P∗ if not converged.

If there are nonorthogonal cells in a mesh, then it may be desired to repeat

step 4 for a specific number of iterations. In OpenFOAM

R
, this number

is called nNonOrthogonalCorrectors and is specified in the file

<case>\system\fvSolution. If this number is zero, then step 4 is performed

one time. The recommended values for the under-relaxation factors according to [47]

are αP = 0.2 and αv = 0.8. In OpenFOAM

R
[33], the default values are αP = 0.3 and

αv = 0.7. The convergence is checked by the residual values of the velocity and pressure.

41
If each residual is below a specific tolerance then the solver will stop. In OpenFOAM
R
,

the SIMPLE algorithm residuals are found in the file <case>\system\fvSolution

under the name of residualControl.

2.4.2.2 Pressure Implicit with Splitting of Operators (PISO) algorithm

The PISO [48] algorithm was developed originally for a non-iterative computation of

unsteady compressible flows, but it was further developed for steady calculation and for

incompressible flow. The algorithm uses more than one corrector rather than one like in

SIMPLE. In each time step, the algorithm can be described as follows:

1. Start with a guessed value of pressure P∗ in the first step and afterwards the resulted

pressure from previous step is then used.

2. Obtain an approximation for the velocity by solving the momentum equation

Eq. (2.61), using the pressure from the previous time step.

 ∗)

3. Approximate the mass fluxes at the cell faces F ∗ = S · (ρ H(v
aP ) f , which is needed

in the right hand side of the pressure Eq. (2.60).

4. Using the approximated velocity, solve the pressure equation Eq. (2.60).

5. Find the final flux correcting the approximated flux by the pressure effect using

42
 Eq. (2.62).

6. Correct the velocity using the new pressure value, where this is an explicit correction

and is achieved using Eq. (2.57).

7. For a non-Newtonian fluid, update the viscosity from a generalized Newtonian

constitutive equation using the corrected velocity.

The last five steps (3-7) are iterated a fixed number of times before moving to the next

time step. In OpenFOAM

R
, this number is called nCorrectors and is specified

in the file <case>\system\fvSolution. Also, as in the SIMPLE algorithm, the

nNonOrthogonalCorrectors should be defined. This value determines how many

times step 4 should be repeated. Note that no under-relaxation is performed for pressure or

velocity, and there are no residual controls.

2.4.2.3 Merged PISO-SIMPLE (PIMPLE) algorithm

The PIMPLE algorithm uses the SIMPLE and PISO algorithms combined. It is a good

algorithm to use for transient calculations. At each time step, the algorithm combines the

SIMPLE algorithm and then uses the PISO algorithm to adjust the pressure correction. In

each time step, the PIMPLE algorithm can be summarized as:

1. Calculate the velocity v ∗ using Eq. (2.63) and pressure P∗ from previous time step.

43
  ∗)

2. Approximate the face flux F ∗ = S · (ρ H(v
aP ) f , which is needed in the right hand

side of the pressure Eq. (2.60).

3. Calculate the corrected pressure P∗∗ using Eq. (2.60) and the approximated flux .

4. Correct the face fluxes using the new pressure value P∗∗ via Eq. (2.62).

5. Apply an explicit under-relaxation for the pressure as in the SIMPLE algorithm.

Pnew = P∗ + αP (P∗∗ − P∗ ).

6. Correct the velocity from the new pressure value Pnew using Eq. (2.57).

7. Repeat steps 2-6 nCorrectors more times.

8. For a non-Newtonian fluid, update the viscosity from a generalized Newtonian

constitutive equation using the corrected velocity.

9. Test for convergence using residual controls. If satisfied, move to next time step. If

not, then repeat steps 1-8 at most nOuterCorrectors more times.

The convergence is controlled by residualControl as in the SIMPLE algorithm. If

the nOuterCorrectors is equal to one, then the PIMPLE algorithm will be operating

in PISO mode. The nCorrectors, nOuterCorrectors, and residualControl

numbers are defined in the file <case>/system/fvSolution. The next section

describes some of the iterative methods used to solve the linear systems encountered in

the solution algorithms.

44
2.4.3 Linear Solvers

Each discretized momentum equation, pressure equation, and pressure correction equation

results in a linear system of the form By = R. There are many methods to solve these

linear systems. In this section, some of those methods are discussed.

1. Generalized Geometric-Algebraic Multi-Grid (GAMG) Method

The multi-grid method is a fast method. The idea is to accelerate the convergence of

an iterative method by correcting the solution from time to time. If the approximated

solution to the linear system is yh then the error is e = y − yh and the residual is

r = R − Byh . The error e satisfies

Be = B(y − yh ) (2.64)

= By − Byh (2.65)

= r. (2.66)

The multi-grid method solves the equation Be = r on a coarser grid and then

interpolates the solution to the fine grid. Then it adds the approximated error to the

approximated solution. The method is achieved by defining the following matrices:

• A restriction matrix T which transfers a vector from the fine grid to the coarse

45
 grid

• An interpolation (prolongation) matrix P which returns the vector to the fine

grid

The method can be summarized as follows:

(a) Solve the system By = R by a few iterations to find yh

(b) Find the residual on the coarse grid by rc = T r

(c) Solve the system Bc ec = rc

(d) Interpolate the error to the fine grid via eh = P ec

(e) Add the error to the approximated solution ynew = yh + eh

(f) Repeat steps (a)-(e) until convergence is reached

In Step (c), Bc = T BP and the subscript c represents the coarse grid and the cell

size in c is twice of the cell size in the original grid. The above multi-grid method is

called v-cycle, meaning that we have only two grids, as shown in Figure 2.9 on the

left, where f denotes the fine grid. The multi-grid method can be applied on more

than two grids in the same way as in the v-cycle. Figure 2.9 on the right shows a

multi-grid method on four grids. Multi-grid method can be used to find a good initial

guess by finding the solution on the coarsest grid and interpolating it to the fine grid.

The geometric multi-grid method uses the geometry to find the restriction matrix T

and prolongation matrix P . Algebraic multi-grid constructs the matrices from the

46
 f . .
c . .
f . . 2c
. .
.c .3c
Figure 2.9: v-cycle and V-cycle.

matrix B and does not use the geometry. For that reason, it is a good choice for

unstructured grids. The multi-grid method can be used as a preconditioner.

2. Gauss Seidel Method

The Gauss Seidel method uses the decomposition of the matrix B = D − U − L,

where B is a symmetric positive-definite matrix, D is a diagonal matrix containing

the diagonal entries of B, −U is the upper triangular part of B, and −L is the lower

triangular part of B. The linear system can be written as

(D − L)y = U y + R. (2.67)

47
 The Gauss Seidel solves this linear system by using the value of y from the previous

iteration on the right hand side of Eq. (2.67). The new value for y can be written as:

yk+1 = (D − L)−1 U yk + (D − L)−1 R. (2.68)

3. Conjugate Gradient (CG) Method

If the matrix B is symmetric positive-definite, then minimizing the quadratic

function f (y) = 12 y T By − RT y is equivalent to solving the linear system By = R.

Also, note that r = R − By = −∇ f (y). The solution is updated iteratively via

yk+1 = yk + αk pk . (2.69)

The idea here is to start with an initial guess y0 and then, at each step, walk in

a direction such that f (yk+1 ) < f (yk ). The conjugate gradient method chooses

the set of search direction vectors {p0 , p1 , ..., pn } such that the set is B-conjugate

(pTi Bp j = 0, ∀i = j). The step length αk and search direction pk are defined as

rkT rk
αk = T (2.70)
pk Bpk
pk = rk + βk pk−1 (2.71)

rkT rk
where βk = rk−1 rk−1
T . The conjugate gradient method starts with an initial residual

r0 = R − By0 and calculates the initial guess for the search direction p0 = r0 . It

48
 then repeats the following steps from k = 0 until the residual gets below a specified

tolerance:

rkT rk
(a) Calculate step length αk = pTk Bpk

(b) Calculate yk+1 = yk + αk pk

(c) Calculate the new residual rk+1 = rk − αk Bpk

rk+1
T r
(d) Calculate βk = k+1
rkT rk

(e) Calculate the new direction pk+1 = rk+1 + βk pk .

4. Bi-Conjugate Gradient (BiCG) method

The bi-conjugate gradient method is applicable for non-symmetric matrices. It uses

both matrices B and B T . The method makes the two sets of search direction vectors

{p0 , p1 , ..., pn } for B and {q0 , q1 , ..., qn } for B T mutually orthogonal (qiT Bpi = 0).

As in the conjugate gradient method, the BiCG method starts with an initial guess

R − By0 = r0 = p0 = s0 = q0 and then repeat the following steps from k = 0 until

convergence:

sTk rk
(a) Calculate step length αk = qkT Bpk

(b) Calculate yk+1 = yk + αk pk

(c) Calculate the new residual of B as rk+1 = rk − αk Bpk

(d) Calculate the new residual of B T as sk+1 = sk − αk B T qk

sTk+1 rk+1
(e) Calculate βk = sTk rk

49
 (f) Calculate the new direction pk+1 = rk+1 + βk pk

(g) Calculate the new direction qk+1 = sk+1 + βk qk .

Preconditioners

If the matrix M is nonsingular, then By = R and M −1 By = M −1 R have the same

solution. The preconditioner M of the matrix B is a matrix such that M −1 B has a smaller

condition number than B, where cond(B) = BB −1 .

Diagonal Incomplete-Cholesky (DIC)

This method can be used to find the preconditioner matrix M by using diagonal incomplete

Cholesky decomposition. For the matrix B, the diagonal incomplete Cholesky of a matrix

B has the form LDLT where L is a lower triangular matrix and D is a diagonal matrix.

The preconditioner matrix M defined to be M = LDLT .

Diagonal Incomplete Lower Upper (DILU)

The choice of the preconditioner matrix M in this method is defined to be M = (L +

D)D −1 (D + U ) where L is the lower part of the matrix B, U is the upper part of the

matrix B, and D is a diagonal matrix such that diagonal of M equal to the diagonal of B.

50
Chapter 3

Dynamic Meshing For Two-Phase Flows

This chapter describes the two-phase (VOF) flow solvers in OpenFOAM

R
, namely

interFoam and interDyMFoam. We start with some comments on the meshes in

OpenFOAM
R
. All meshes in OpenFOAM
R
are 3D Cartesian meshes, even for 2D

simulations. In 2D simulations, the computational domain in one of the coordinate

directions is always one cell thick. In OpenFOAM

R
the projection of the computational

domain boundary in the other two directions are called empty patches. However, in the

axisymmetric calculations, they are called wedge patches. Each cell in the mesh is assigned

a designated number which contains a number of points and faces. There are two kinds of

faces: an internal face that connects two cells, an owner and a neighbor cell, and a boundary

face that belonging to one owner cell. Each face is also assigned a designated number where

an internal face has an owner cell with the lower number and a neighbor cell with the higher

51
number.

Dynamic mesh refinement allows us to refine the cells in a coarse mesh at specific

regions that requires smaller cells. It is a good way to get accurate results with reduced

computational time in comparison to a refined static mesh. The coarse mesh should be

refined enough to give an accurate result outside the region that requires small cells. In

two-phase flow, the refinement should be on the interface between the fluids because of the

steep gradients in the volume fraction function and potentially the material properties of the

fluid system. In the following section, the discretization of the volume fraction equation is

described since it is needed in the two-phase flow calculation.

3.1 Discretization of Volume Fraction Equation

In this section, the discretization of the volume fraction Eq. (2.10) is described. Because the

conservation of the phase fraction is important to give accurate physical properties, such as

density, especially for the fluids with high density ratio, an artificial compression term is

added to Eq. (2.10) and the volume fraction equation becomes [49]

∂α
+ ∇ · (α v) + ∇ · ((α (1 − α )vr ) = 0 (3.1)
∂t

52
where vr = v2 − v1 and the subscripts represent the fluid phase. Note that the artificial

compression term is nonzero only in a thin region around the interface due to the factor

of α (1 − α ); therefore it does not affect the solution outside the interface region. The

compression term reduces the numerical diffusion, thus allowing sharp interface resolution.

The integral form of Eq. (3.1) is

 t+δ t   t+δ t   t+δ t 

∂α
dV dt + ∇ · (α v)dV dt + ∇ · ((α (1 − α )vr )dV dt = 0.
t VP ∂ t t VP t VP

(3.2)

By applying the Gauss divergence theorem the equation become

 t+δ t   
∂α
dV + n · (α v)dS + n · ((α (1 − α )vr )dS dt = 0. (3.3)
t VP ∂ t ∂ VP ∂ VP

The discretization of the terms in Eq. (3.3) is done in the same way as the discretization

of the terms in the general convection-diffusion equation in Section 2.4.1. The discretized

equation has the form

αPn − αPo  
δt
VP + ∑ (1 − w)ψα nf + wψα of
f
 
+ ∑ (1 − w)Ψ(α (1 − α ))nf + wΨ(α (1 − α ))of = 0 (3.4)
f

53
where S f is the face area, S = S f n f is the outward area vector, ψ = S · v f is the face flux

of the linear term, and Ψ is the face flux of the non-linear term and is calculated based on

the maximum velocity magnitude at the interface region and its direction as:



|φ| |φ|
Ψ = nf min Cα , max , (3.5)
|S| |S|

where φ = S · v f is face volume flux and Cα is an adjustable coefficient which

determines the magnitude of the compression. In OpenFOAM

R
, Cα is defined in the

<case>/system/fvSolution file. As in the previous section, different values of w

give different numerical methods. In OpenFOAM

R
, the explicit method is used (w = 1).

The fluxes are calculated from the previous time step.

In the next two sections, the implementation of the governing equations for two-phase flow

problems is described for the interFoam and interDyMFoam solvers in OpenFOAM.

After that, a brief comparison between them is presented for a 3D test problems before

describing the modifications for 2D and axisymmetric flows.

3.2 Description of interFoam Solver

The interFoam solver for two immiscible incompressible fluids uses a VOF (volume

of fluid) phase-fraction based interface capturing approach. It uses an adaptive time step
|v|δ t
depending on the Courant number Co = δx . To choose the new time step, a maximum

54
Courant number Coo is calculated from the flow conditions, using v and δ t from the

previous time step. The new time step δ t n is then calculated using the following expression

[50]

 
Comax o Comax o
δ t = min
n
δ t ; (1 + 0.1 )δ t ; 1.2δ t o ; δ tmax (3.6)
Coo Coo

where δ t o is the old time step, Comax is the pre-set maximum Courant number, and δ tmax

is the pre-set maximum time step. The values of Comax and δ tmax are specified in the

<case>/system/controlDict file.

It is critical to have the volume fraction α value accurate because it affects other physical

properties such as the density and viscosity, as well as the interface curvature. Therefore,

the volume fraction equation is solved in sub-cycles within each time step. The new

sub-cycle time step is calculated from the time step for the flow

δt
δ tsc = , (3.7)
nAl phaSubCycles

where nAlphaSubCycles is the number of sub-cycles defined in the file

<case>/system/fvSolution. The flux Fsc = S · (ρ v) f is calculated at each δ tsc

and the total flux, which is needed in the momentum equation, is calculated as F =
nAl phaSubCycles δ tsc
∑1 δ t Fsc . The use of sub-cycles speeds up the calculations by allowing

smaller time steps for the evolution of α , while retaining larger steps for solving the other

equations.

55
The specification of the pressure boundary conditions is simplified if the modified pressure

P is used which is obtained by removing the hydrostatic pressure from the pressure P. It

defined as

P = P − ρ g · x, (3.8)

where ρ is the density, g is the gravity, and x is the cell center. Using Eq. (3.8), the pressure

gradient is

∇P = ∇P + ρ g + g · x∇ρ . (3.9)

In addition to the advantage of a simpler specification of the pressure boundary condition,

this treatment enables efficient numerical treatment of the steep density jump at the

interface.

The pressure gradient in the momentum equation, Eq. (2.11), is replaced by the pressure

gradient in Eq. (3.9). Hence, the momentum equation becomes

∂v
ρ (α )( + v · ∇v) = −∇P − g · x∇ρ + ∇ · η (γ̇ )(∇v + ∇v T ) + σ κ (α )∇α . (3.10)
∂t

Using the same technique as in Section 2.4.2, the pressure-velocity coupling equations are:

the momentum equation

aP vP = H(v) − ∇P − g · x∇ρ + σ κ ∇α , (3.11)

56
the pressure equation


1  H(v) g · x∇ρ σ κ ∇α
∑ S · ( aP ∇P) f = ∑ S · ( aP ) f − ( aP ) f + ( aP ) f , (3.12)
f f

and the flux equation


H(v) g · x∇ρ σ κ ∇α 1 
F =S· ( )f −( )f +( ) f − ( ∇P) f . (3.13)
aP aP aP aP

The interFoam was validated on several benchmarks tests [23]. The

solver initiates the variables and then starts the time loop. Refer to

the source code of interFoam solver which is located in the folder

OpenFOAM/OpenFOAM2-1-0/application/solvers/multiphase/

interFoam. In each time step, the solver is outlined as follows:

1. Calculate the Courant number by calling the CourantNo.H library and adjust the

time step by calling the setDeltaT.H library.

2. Correct the phase properties, such as density and viscosity, using the new volume

fraction α by calling the function twoPhaseProperties.correct().

3. Solve the volume fraction equation (Eq. 3.4) as described in Section 3.1 to find α

using the fluxes from the previous time step by calling the alphaEqnSubCycle.H

library. The α value is iteratively corrected via Eq. (3.4) a number of times equal

to nAlphaCorr, which is specified in the <case>/system/fvSolution file.

57
 The cAlpha keyword specified in the file <case>/system/fvSolution is

a factor that controls the compression of the interface, where 0 corresponds to no

compression. After solving the volume fluid equation, the density and viscosity are

modified using the new values of α .

4. Start the PIMPLE loop to solve for the pressure and velocity as described in

Section 2.4.2.3. In this step, the volume fraction function α from the previous step is

used to calculate the CSF term σ κ ∇α .

The solver repeats these steps until a pre-set time which is specified in the

<case>/system/controlDict file under the name endTime.

3.3 Dynamic Mesh Refinement in interDyMFoam Solver

The interDyMFoam solver is the same as the interFoam solver but with the ability

of mesh motion and dynamic mesh refinement. In this study, we will concentrate on

the dynamic mesh refinement. The interDyMFoam solver can do the refinement only

for the 3D hexahedral cells by partitioning the cells equally in all three directions .

The mesh refinement in interDyMFoam is achieved by using the dynamicFvMesh

and dynamicMesh libraries. The mesh refinement is initialized by calling the

function mesh.update() (refer to the source code of interDyMFoam solver). This

function is defined in the dynamicFvMesh library and can be found in the file

58
OpenFOAM/OpenFOAM2-1-0/src/dynamicFvMesh/dynamicRefineFvMesh

/dynamicRefineFvMesh.C. This function can carries out the refinement as follows:

1. Reads the dynamicMeshDict file which is located in the constant folder of the

case directory. This file has some values that are needed to conduct the refinement,

which are further explained at the end of this section.

2. Determines the candidate cells that can be refined by calling the

function selectRefineCandidates() which is located in

dynamicRefineFvMesh.C file. The cells are chosen based on three bases.

First, a field is specified, which can be the magnitude of the velocity or the volume

fraction function alpha1. Second, a maximum number of refinement is specified.

The first and second bases are determined in the dynamicMeshDict file. Finally,

the cell must have an nAnchors value of 8. The nAnchors is defined in the

dynamicRefineFvMesh.C file. If the value of nAnchors for the cell is not 8,

then the cell can not be refined. To find the value of nAnchors for a given cell, a

loop is taken over all the points of the cell and if the pointLevel is less than or

equal to cellLevel for a point, then this point is added to the nAnchors. In the

dynamicRefineFvMesh file, the cellLevel and pointLevel are defined

such that each cell has a cellLevel starting with 0 for the original cell and if the

cell is refined once then this number becomes 1 for each new cell and so on. The

value of pointLevel is similarly defined.

59
3. Selects a subset of candidate cells for refinement by calling the function

selectRefineCells() which is located in the dynamicRefineFvMesh.C

file. The subset is chosen based on the maximum number of cells allowed, which is

defined in the <case>/constant/dynamicMeshDict file.

4. Perform the refinement by calling the function refine()

which is located in the dynamicFvMesh.C file. The function

refine() calls the function setRefinement() which is

defined in the dynamicMesh library and can be found in the file

OpenFOAM/OpenFOAM2-1-0/src/dynamicMesh/polyTopoChange/

polyTopoChange/hexRef8.C. This function can do the refinement as follows:

(a) For each cell to be refined a point is added in the center of the cell as shown in

Figure 3.1.

Figure 3.1: A hexahedral cell with a point in the middle

(b) For each cell to be refined, a point is then added in the center of each face of

the cell, one of which is illustrated in Figure 3.2.

60
 Figure 3.2: A point in the middle of a face

(c) For each cell to be refined, a point is added in the middle of each edge, as

illustrated in Figure 3.3.

Figure 3.3: A points in the middle of an edges

(d) Each face is divided into four new faces as shown in Figure 3.4 and each new

face assigned an owner and a neighbor cell.

Figure 3.4: Divide a face into four faces

61
 (e) Internal faces are added to the cell by connecting the points in the center of two

neighboring faces, the point in the center of the edge that connects the faces,

and the point in the center of the cell, as shown in Figure 3.5. Therefore, a

neighbor and an owner cell are assigned to each face.

Figure 3.5: Internal face added to the cell

(f) The fields are mapped from the old mesh to the new mesh as an initial condition

to speed up the computational process. The field value at the centroid of a cell

in the original mesh is transfered to the new cells by assigning them the same

values.

5. Determines the points that can be unrefined by calling the

function selectUnrefinePoints() which is located in the

dynamicRefineFvMesh.C file. The points are chosen based on the

PointLevel and nBufferLayers numbers which are defined in the file

<case>/constant/dynamicMeshDict. If the PointLevel is greater than

0, then the point can be unrefined. The nBufferLayers number will be discussed

later in this section.

62
 6. The unrefinement can be done by calling the function unrefine()

located in the dynamicFvMesh.C file. The unrefinement is

achieved by calling the function setUnrefinement() defined

in the dynamicMesh library and can be found in the file

OpenFOAM/OpenFOAM2-1-0/src/dynamicMesh/polyTopoChange/

polyTopoChange/hexRef8.C. This function removes the points selected by

the function selectUnrefinePoints() and their connected faces and points.

7. Finally, the fields are mapped from the old to the new mesh. The values at the

centroid are mapped by taking the average for the small cells.

After the refinement is done, the fluxes are corrected in the solver using the new values of

the velocity. Figure 3.6 shows the dynamicMeshDict file. In this file, the refine interval

should be one or greater. This number indicates the time step occurrence of the refinement.

For example, if the refine interval is two, the refinement will be operated every second

time step. The field that is used to determine the cells requiring refinement is specified

to be alpha1. The lower and upper refine levels determine the range of the field such

that each cell having the field value in this range will be a candidate for the refinement.

The unrefinement level is the number that controls the points which can be unrefined, and

the number of buffer layers is used to find the buffer layers that should be extended for

unrefinement. Each cell can be refined up to the maximum refinement number and if the

total cells exceed the maximum number of cells, the refinement stops. The fluxes that needs

63
to be corrected is defined in this file as well. Finally, the dumpLevel is true to write the

refinement level.

   
 
     
 

     
   

        

  
   

        

       

      

!       

"     #

"     #

      " 
$
$    %  &
$   '(  %&
$   '( ) %)&

&

     
*

Figure 3.6: Example of a dynamicMeshDict file.

3.4 Test of interFoam and interDyMFoam in 3D

In this section, we will compare the interFoam and interDyMFoam solvers from the

aspects of CPU time, mesh independence, and cell size around the interface. The test case

is a three-dimensional liquid drop in a linear shear flow. The computational domain is a

64
 top wall
3 mm

inlet outlet
(0,0)

-3 mm
-15 mm bottom wall 15 mm

Figure 3.7: Schematic diagram of a drop of radius 1 mm centered in a

channel. The x-axis and y-axis are horizontally and vertically, respectively,
and the positive z-axis point out of the paper.

channel of length 30 mm, height 6 mm, and depth 6 mm. The origin of the coordinate

system is placed at the center of the domain, so that −15 ≤ x ≤ 15, −3 ≤ y ≤ 3, and

−3 ≤ z ≤ 3. The drop is a sphere of radius 1 mm and center at (0, 0, 0). At time t = 0, the

upper wall (at y = 3 mm) moves at a constant speed of u in the positive x-direction, and the

lower wall (at y = −3 mm) moves at a constant speed of u in the negative x-direction. The

side walls (at z = ±3 mm) remain stationary. The geometry is illustrated in Figure 3.7. The

continuous and disperse phases are taken to be Newtonian fluids. The transport properties

are as follows:

μc = μd = 1.06 × 10−1 Pas

ρc = ρd = 103 kg/m3

σ = 0.0415N/m

where μ is the dynamic viscosity, ρ is the density, and σ is the interfacial tension. The

65
 Figure 3.8: A droplet at steady-state for the 3D drop in shear flow test case
(Ca = 0.3).

μd ρd
viscosity ratio and density ratio are λ = μc = 1 and ρc = 1.

The capillary number is a dimensionless number that characterizes the ratio of viscous

forces to interfacial tension forces. It is defined as

aμc γ̇
Ca = , (3.14)
σ

where γ̇ is the shear rate, and a is the radius of the undeformed droplet. The subscripts,

c and d, denote the continuous and disperse phases respectively. The critical capillary

number Cacrit is the value of Ca above which drop breakup occurs (Ca > Cacrit) and below

which breakup does not occur (Ca < Cacrit). The critical capillary number in a given type

of flow depends on the viscosity ratio.

Table 3.1 shows the boundary conditions for the test case. The zeroGradient condition

means that the derivative normal to the boundary is zero. For example, the condition of the

66
 Table 3.1
Boundary conditions for the 3D drop in shear flow test case, where Ca is
the capillary number

boundary velocity p− ρ gh alpha (α )

inlet zeroGradient zeroGradient zeroGradient
outlet zeroGradient 0 zeroGradient
bottom (-1.175(m/s) ∗Ca 0 0) zeroGradient zeroGradient
top (1.175(m/s) ∗Ca 0 0) zeroGradient zeroGradient
front and back (0 0 0) zeroGradient zeroGradient

∂v
velocity on the inlet boundary means ∂x = 0 since x-direction is the normal for the inlet.

The velocities in the x-direction for the upper and lower wall boundaries are given in terms

of the capillary number. This dependence comes from the formula of the capillary number

above and the formula of the shear rate γ̇ = u

H, where H is the half distance between the

bottom and top (here H = 3 mm). We take Ca = 0.3 which is a sub-critical capillary number
μd
for a viscosity ratio of λ = μc = 1. The end time for all cases is 1 s. The parameters used in

the interDyMFoam solver, defined in the <case>/constant/dynamicMeshDict

file, are shown in Table 3.2.

For this test case, steady-state is reached. The stationary drop shape (at time t = 0.99 s)

is shown in Figure 3.8. The steady-state results in the following sections are represented,

in part, by graphing data along the horizontal line that passes through the origin (y = 0) as

shown in Figure 3.7.

67
 Table 3.2
DynamicMeshDict parameters for the 3D drop in shear flow test case

refineInterval field lowerRefInterval upperRefInterval

1 alpha (α ) 0.1 0.9
unrefineLevel nBufferLayers maxRefinement maxCells
10 1 2 400000

3.4.1 Mesh Independence Study

This section studies the mesh independence for the interFoam and interDyMFoam

solvers. The study is achieved by comparing the solutions of the velocity and pressure

on three different meshes, described in Table 3.3. Figure 3.9 shows the velocity along

the centerline (y = 0) for the three different meshes using interFoam. As the mesh

is refined, the difference between the velocity curves becomes smaller, and the velocity

curves become smoother. Similar results were found for the pressure shown in Figure 3.10.

Figures 3.11 and 3.12 show the velocity and pressure for the interDyMFoam solver on

the three different meshes and, for comparison, for the interFoam solver on the fine

mesh. The graphs for the velocity and pressure show almost identical results on the three

meshes where the difference between the curves is negligible. The interFoam results

on the fine mesh agree better with those from interDyMFoam in terms of velocity than

pressure.

The pressure graphs in Figures 3.10 and 3.12 show a jump in pressure within the droplet.

The jump in pressure is almost the same for the three meshes using interDyMFoam

68
solver. It is clear that the coarse mesh used with interFoam is insufficient since the

solution curves have a lot of oscillation and they are inaccurate compare to the solution

curves from the standard and fine meshes. Figures 3.9–3.12 show that interDyMFoam

performs better than interFoam in terms of producing smooth mesh independence

results for this test case. As seen in Table 3.4, this is due in large part to the cell size around

the interface. The cell size around the interface using the coarse mesh of interDyMFoam

is smaller than the cell size using interFoam with fine mesh. More refinement around

the interface is needed to improve the results of interFoam. This would require a more

global refinement of the mesh compared to interDyMFoam, and would therefore increase

the CPU time.

Table 3.3
Initial mesh and number of cells using interFoam and interDyMFoam for
the 3D drop in a shear flow test case

Solver Initial mesh Number of cells

interFoam(Coarse) 125×25×25 78125
interFoam(Standard) 165×33×33 179685
interFoam(Fine) 218×43×43 403082
interFoam(non-uniform) 50,65,50×33×33 179685
interDyMFoam(Coarse) 76× 15×15 17100 − 27000
interDyMFoam(Standard) 100× 20×20 40000 − 53000
interDyMFoam(Fine) 132× 26×26 89232 − 115000

69
 Coarse Mesh
0.02 Standard Mesh
Fine Mesh
0.01
Velocity [m/s]

-0.01

-0.02

-0.03
-0.004 -0.002 0 0.002 0.004
x-Coordinate [m]
Figure 3.9: Velocity at steady-state along line y = 0 using interFoam for
the 3D drop in shear flow test case Ca = 0.3. The vertical lines indicate the
boundary of the drop along y = 0.

3.4.2 Comparison Between interFoam And interDyMFoam Using

Serial Calculations

This section compares the two solvers from the aspects of the CPU time, the number of

cells, and cell size around the interface. For the interFoam solver there are two cases.

In the first case, a uniform mesh is considered and in the second case (non-uniform),

the computational domain is divided into three blocks in the x-direction to have smaller

cells around the interface. The minimum and maximum x-coordinate for the blocks are

−15, −3; −3, 3; and 3, 15 started from block one to block three respectively. Table 3.4

shows the difference in CPU time, number of cells, and cell size around the interface. The

70
 Coarse Mesh

Dynamic Pressure [kg/m.s ]

Standard Mesh

2
90
Fine Mesh

60

30

0
-0.004 -0.002 0 0.002 0.004
x-Coordinate [m]
Figure 3.10: Pressure at steady-state along line y = 0 using interFoam for
the 3D drop in shear flow test case Ca = 0.3. The vertical lines indicate the
boundary of the drop along y = 0.

interDyMFoam has a lower overall CPU time, even when comparing its fine mesh with

interFoam’s standard and non-uniform meshes, keeping in mind that the cell size around

the interface is smaller for interDyMFoam. This is due in part to the fewer number of

cells in the fine mesh of interDyMFoam (maximum of 115000 cells) compared with the

standard and non-uniform meshes of interFoam (179685 cells). Note that although the

standard uniform mesh and non-uniform mesh of interFoam have the same number of

cells, the CPU time is almost doubled when using the non-uniform mesh. This is because

the cell size in the x-direction for the non-uniform mesh is δ = 0.09 mm which is half

of the cell size in the uniform mesh (0.18 mm), making the time step size smaller in the
|v|δ t
non-uniform mesh since it depends on the Courant number Co = δx . Specifically, if the

cell size (δ x) is decreased then the time step will decrease to have the Courant number

71
 Coarse Mesh
0.02 Standard Mesh
Fine Mesh
interFoam Fine Mesh
0.01
Velocity [m/s]

-0.01

-0.02

-0.03
-0.004 -0.002 0 0.002 0.004
x-Coordinate [m]
Figure 3.11: Velocity at steady-state along line y = 0 using interDyMFoam
for the 3D drop in shear flow test case Ca = 0.3. The vertical lines indicate
the boundary of the drop along y = 0.

below a specific number defined in the <case>/system/controlDict file. Indeed,

the typical time step was nearly doubled on the standard uniform mesh (2.7 × 10−4 s)

compared to the non-uniform mesh (1.4 × 10−4 s). The smaller cells on the non-uniform

mesh were located in the center part of the domain, where we know the drop remains. In

general, the location of the interface is unknown as time passes. Thus, when using a static

mesh, it is hard to refine the relevant region of the computational domain. Therefore, in

most cases, by using a static mesh, the whole mesh has to be refined, causing an increase in

the CPU time dramatically. Therefore, interDyMFoam will be a better option to do the

refinement around the interface with a much reduced CPU time.

72
 80

Dynamic Pressure [kg/m.s ]

2
60 Coarse Mesh
Standard Mesh
Fine Mesh
interFoam Fine Mesh
40

20

0
-0.004 -0.002 0 0.002 0.004
x-Coordinate [m]
Figure 3.12: Pressure at steady-state along line y = 0 using interDyMFoam
for the 3D drop in shear flow test case Ca = 0.3. The vertical lines indicate
the boundary of the drop along y = 0.

Table 3.4
CPU time and cell size around the interface using interFoam and
interDyMFoam for the 3D drop in a shear flow test case (Ca = 0.3)

Cell size around

Solver CPU Time (s) Number of cells
the interface(mm)
interFoam(Coarse) 8773 78125 0.24× 0.24×0.24
interFoam(Standard) 23423 179685 0.18× 0.18×0.18
interFoam(Fine) 62367 403082 0.14× 0.14×0.14
interFoam(non-uniform) 45232 179685 0.09× 0.18×0.18
interDyMFoam(Coarse) 3151 17100 − 27000 0.10× 0.10×0.10
interDyMFoam(Standard) 8671 40000 − 53000 0.075× 0.075×0.075
interDyMFoam(Fine) 23254 89232 − 115000 0.057× 0.057×0.057

73
 Table 3.5
interDyMFoam in parallel for the 3D drop in a shear flow test case
(Ca = 0.3)

Cell size around

Solver CPU Time (s) Initial Mesh
the interface(mm)
interDyMFoam (8 1 1) 3059 100×20×20 0.075× 0.075×0.075
interDyMFoam (4 2 1) 2802 100×20×20 0.075× 0.075×0.075
interDyMFoam (2 2 2) 2917 100×20×20 0.075× 0.075×0.075
interDyMFoam (Scotch) 3068 100×20×20 0.075× 0.075×0.075
interDyMFoam(Serial) 8671 100× 20×20 0.075× 0.075×0.075

3.4.3 Effect of Parallelization on Efficiency of interDyMFoam

In general, to run a case in parallel, the decomposePar has to be run before running the

solver directly. When running the decomposePar, it chooses the number of processors

and the method of how to split the computational domain. For more information, refer

to the decomposeParDict file, located in the system folder of the case. Table 3.5

summarizes the CPU time and the cell size around the interface using interDyMFoam

solver in parallel on 8 processors and in serial on the standard mesh. There are four different

parallel cases using two different methods. In the first three cases, the simple method is

used, whereas, in the fourth case the scotch method is used. In the simple method, the

number of sub-domains in each direction must be specified. For example, (4 2 1) means that

the domain is divided into four sub-domains in x direction, two subdomains in y direction,

and one subdomain in z direction. In the scotch method, the solver chooses how to split

the domain in the best way to minimize the number of processor boundaries. As shown

in the table in our case, the best method to use is the simple method with directions (4

74
2 1) since the CPU time is lowest. This method divides the domain around the interface

into four small subdomains. Thus, the refined regions of the domain can be distributed onto

more processors while keeping the number of cells on the processor boundaries acceptable.

In general, if the interface is unknown, the scotch method will be the best option to use.

By comparing between the serial and parallel calculations, the CPU time increases by a

factor ranging 2.8 − 3.1 when the serial calculation is used. Therefore, using the parallel

calculation will be a good option to reduce the CPU time and get the same solution.

3.5 Modifications to the interDyMFoam Solver

In this section, we will describe the modifications done for interDyMFoam to allow

dynamic mesh refinement in 2D simulations. The 2D interDyMFoam uses the same

functions that the 3D interDyMFoam uses. However, a cell is divided into four cells

instead of eight cells as the case in 3D. For the selected cells to be refined, the dynamic

mesh refinement in 2D simulations works as follows:

1. A point is added to the center of each face that belongs to empty patches as shown in

Figure 3.13. In 2D, no point should be added to cell centroid because in one direction

the number of cells must remain one. Furthermore, no point is added to the center of

other faces for the same reason.

75
2. A point is added to the center of each edge that lies on the empty patches as shown

in Figure 3.13. The points should not be added to the other edges since this is a 2D

simulation.

Figure 3.13: The points in the center of the faces and edges

3. Each face that has a point added to its center is divided into four new faces as shown

in Figure 3.14. Each new face is assigned a new owner and neighbor cell. Therefore,

a face of an adjacent unrefined cell has four neighboring cells.

Figure 3.14: The faces are divided into four new faces

76
 4. Each face that does not have a point added to its center is divided into two new faces

as shown in Figure 3.15.

Figure 3.15: The faces are divided into two new faces

5. Finally, four new internal faces are added to the cell by connecting the points that

were added to the center of the faces and the points that were added to the center of

the edges as shown in Figure 3.16.

The rest of the process is similar to the interDyMFoam solver. The interDyMFoam

for axisymmetric simulations works similarly as the 2D interDyMFoam by replacing

the empty patches with wedge patches. However, the cells on the center line should be

treated separately since they are wedges instead of hexahedral. Those cells have faces that

contains three vertices instead of four. Such triangular faces are divided into two faces, one

77
 Figure 3.16: Internal faces are added to the cell

containing four vertices and the other containing three vertices. Also, we should add an

internal face that contains three vertices, where one of the vertices is the middle point of

the center line edge. The partition of a wedge is illustrated in Figure 3.17.

In these solvers, we added a new values to the dynamicMeshDict file, axis,

axisVal, and nBufferLayersR (see Figure 3.18). The axis and axisVal are the

numbers that controls the points which can be unrefined. If the empty or wedge faces

are perpendicular on the x-axis, then the axis number should be 0 and the axisVal

number should be between the minimum and maximum values of the x-component in

the geometry. If the empty or wedge faces are perpendicular on the y-axis, then the

axis number should be 1 and the axisVal number should be between the minimum

and maximum values of the y-component in the geometry. If the empty or wedge faces

are perpendicular on the z-axis, then the axis number should be 2 and the axisVal

number should be between the minimum and maximum values of the z-component in the

geometry. The nBufferLayersR value works the same as nBufferLayers, which

78
 Figure 3.17: A cell with divided face and two internal faces added in
axisymmetric case

find the number of buffer layers that should be extended for the unrefinement, but for the

refinement instead of unrefinement. This value was added to extend the number of buffer

layers around the interface during the refinement step, since it was found that additional

resolution is often needed in this step, particularly when tracking small drops in a flow

field. If the resolution around the interface is extended, then the mass will be conserved for

a small droplet in a large geometry, whereas, without extending the resolution, the droplet

loses mass and may even vanish. The modifications are made via libraries, specifically,

the dynamicRefineFvMesh and polyTopoChange libraries. The modifications

are further explained in the Appendix B. To validate the modifications, three cases are

presented in the next sections.

79
Figure 3.18: Example of a dynamicMeshDict for 2D simulations

80
3.6 Test of interFoam and interDyMFoam in 2D Planar

Geometry

In order to validate the modifications of the code for 2D planar simulations, the solutions

from the modified code are compared with the solutions from the interFoam solver for

two test problems: drop deformation and breakup in simple shear flow, and drop formation

and detachment from a micro T-channel.

3.6.1 Drop Deformation and Break Up in Simple Shear Flow

In this section, we test interDyMFoam in 2D planar geometry for the test case of a

two-dimensional liquid drop in a simple shear flow. This is the two-dimensional version

of the 3D test case considered in Section 3.4. The computational domain is a channel of

length 30 mm and height 6 mm. The origin of the coordinates system is placed in the center

of the domain, such that −15 ≤ x ≤ 15 and −3 ≤ y ≤ 3. The drop is a circle of radius 1 mm

and center (0, 0) (see Figure 3.7). The boundary conditions are shown in Table 3.6. The

fluids are taken to be Newtonian and the material properties are μc = μd = 1.06×10−1 Pa.s,

ρc = ρd = 103 kg/m3 , and σ = 0.0415N/m. The coarse and fine meshes are produced by

decreasing and increasing the number of cells in the standard mesh by a factor of 1.5 in

each direction respectively (see Table 3.7). We first consider the case of Ca = 0.3, and then

81
the case of Ca = 0.4.

Table 3.6
Boundary conditions for the 2D drop in a shear flow test case where Ca is
the capillary number

boundary velocity p− ρ gh alpha (α )

inlet zeroGradient zeroGradient zeroGradient
outlet zeroGradient 0 zeroGradient
bottom (-1.175(m/s) ∗Ca 0 0) zeroGradient zeroGradient
top (1.175(m/s) ∗Ca 0 0) zeroGradient zeroGradient
front and back empty empty empty

Table 3.7
Initial mesh and number of cells for the 2D drop in a shear flow test case
Solver Initial Mesh Number of Cells
interFoam (Coarse) 300×60 18000
interFoam (Standard) 450×90 40500
interFoam (Fine) 675×135 91125
interDyMFoam (Coarse) 80×16 1280 − 2100
interDyMFoam (Standard) 120×24 2880 − 3900
interDyMFoam (Fine) 180×36 6480 − 8000

3.6.1.1 Test Case Using Ca = 0.3

For Ca = 0.3, the drop reaches a stationary shape after a while and no break up occurs.

Figures 3.19 and 3.20 show the refinement around the interface. Figure 3.19 illustrates the

coarsest mesh (top figure), the refinement in the x, y directions at t = 0.005 s (middle

figure), and the refinement in the x, y directions at steady-state t = 0.99 s (bottom

figure). Figure 3.20 shows the number of cells in the z-direction equal one. Note that

82
the refinements are shown as diagonals, although the cells are actually partitioned into

rectangles.

Figure 3.19: Dynamic refinement in 2D at t = 0 s (top), t = 0.005 s

(middle), and t = 0.99 s (bottom) for the 2D drop in a shear flow test case

83
Figure 3.20: Dynamic refinement in 2D at t = 0.005 s for the 2D drop in a
shear flow test case

Figure 3.21: Drop at steady-state t = 0.99 s for the 2D drop in a shear flow
test case

84
 Coarse Mesh
0.02 Standard Mesh
Fine Mesh

Ux [m/s] 0.01

-0.01

-0.02

-0.03
-0.01 -0.005 0 0.005 0.01
x-Coordinate [m]
Figure 3.22: Velocity using interFoam on three different meshes Ca = 0.3
for the 2D drop in a shear flow test case

Coarse Mesh
0.02 Standard Mesh
Fine Mesh
interFoam Fine Mesh
0.01
Ux [m/s]

-0.01

-0.02

-0.03
-0.01 -0.005 0 0.005 0.01
x-Coordintae [m]
Figure 3.23: Velocity using interDyMFoam on three different meshes Ca =
0.3 for the 2D drop in a shear flow test case

85
 40
Coarse Mesh
Dynamic Pressure [kg/m.s ]
2 Standard Mesh
30 Fine Mesh

20

10

0
-0.01 -0.005 0 0.005 0.01
x-Coordinate [m]
Figure 3.24: Pressure using interFoam on three different meshes Ca = 0.3
for the 2D drop in a shear flow test case

The velocity and pressure from interFoam and interDyMFoam for this 2D problem

are compared along the horizontal line y = 0 at time t = 0.99 s, when a stationary drop

shape is reached (see Figure 3.21). Figure 3.22 shows the velocity graphs for the same

case with three different meshes: coarse, standard, and fine mesh using the interFoam

solver. The vertical lines indicate the boundary of the drop. The solutions are almost

identical outside the droplet and behave similarly inside the droplet. Also, the graphs show

a zero velocity outside the droplet which is expected since the top and bottom walls are

moving at the same speed in the x-component with opposite directions. Inside the droplet,

the velocity increased and then decreased. Similarly, in Figure 3.23, the same behavior

is found using the interDyMFoam solver. The velocities agree with the interFoam

solution on the refined mesh. Similarly, the pressure graphs, Figures 3.24 and 3.25, show

86
 40

Dynamic Pressure [kg/m.s ]

2
30

Coarse Mesh
20 Standard Mesh
Fine Mesh
interFoam Fine Mesh

10

0
-0.01 -0.005 0 0.005 0.01
x-Coordinate [m]
Figure 3.25: Pressure using interDyMFoam on three different meshes Ca =
0.3 for the 2D drop in a shear flow test case

the same convergence. They have the same behavior overall, as the jump in pressure is

decreasing by increasing the mesh resolution using interFoam solver. However, the

solutions from interDyMFoam solver are accurate even on the coarse mesh. Table 3.8

Table 3.8
CPU time, cell size around the interface, and relative change in radius for
the 2D drop in a shear flow test case (Ca = 0.3)
Cell CPU Initial Cell size around relative change
Solver
Time (s) Mesh the interface(mm) in R
interFoam (Coarse) 1164 300×60 0.1 × 0.1 0.034
interFoam (Standard) 6409 450×90 0.067 × 0.067 0.008
interFoam (Fine) 21611 675×135 0.044× 0.044 0.002
interDyMFoam (Coarse) 107 80×16 0.09× 0.09 0.030
interDyMFoam (Standard) 355 120×24 0.063× 0.063 0.008
interDyMFoam (Fine) 918 180×36 0.042× 0.042 0.003

shows the difference in CPU time, number of cells, cell size around the interface, and

87
the relative change in radius. Here, the radius is calculated by considering the drop as a

circle using the α = 0.05 contour, so if the volume fraction in a cell less than 0.05, then

the cell does not contribute to the calculation of the radius. The interDyMFoam has a

much lower overall CPU time, even when comparing its fine mesh with the interFoam’s

coarse mesh as the CPU time increased by a factor of 1.2, keeping in mind that the cell

size around the interface is almost the same between the two solvers. The relative change
R0 −R0.99
in the radius is calculated using the formula R0 , where R0 is the radius at t = 0 s and

R0.99 is the radius at t = 0.99 s. The relative change decreases with decreasing cell size

around the interface, however it is almost identical in comparison between the two solvers.

Furthermore, the numbers are small in all cases especially in the fine mesh which indicates

that the mass inaccuracy due to the mesh is negligible. As a result, the two solvers give

similar results with a big difference in the CPU time.

3.6.1.2 Test Case Using Ca = 0.4

This section compares the two solvers on the 2D drop in shear flow for Ca = 0.4. This is a

super-critical Ca, where break up occurs. In the previous case using Ca = 0.3, the break up

does not occur. In this case, however, the droplet breaks up into two daughter droplets using

both solvers, interFoam and interDyMFoam. Figure 3.26 shows the daughter droplets

using interFoam (top) and interDyMFoam (bottom). The dynamic refinement in the

x, y directions is shown in Figure 3.27. In the figure, the refinement appears around the

88
 Figure 3.26: Drop breakup in 2D using Ca = 0.4 at t = 0.99 s using
interFoam (top) and interDyMFoam (bottom) for the 2D drop in a shear
flow test case

interface and there is no refinement between the droplets. That is, after the initial droplet

break up and the droplets move apart, the unrefinement occurs for the refined cells far from

the interfaces. The number of cells in the z-direction is one as shown in Figure 3.28.

Table 3.9 summarizes the CPU time, cell size around the interface, the initial radius R0 ,

the radius of the first daughter droplet R1 , the radius of the second daughter droplet R2 , the

89
 Table 3.9
CPU time, cell size around the interface, and relative change in radius for
the 2D drop in a shear flow test case (Ca = 0.4)
CPU Cell size around break up
R20 −(R21 +R22 )
Solver R0 (mm) R1 (mm) R2 (mm) R20
Time (s) the interface(mm) time (s)
interFoam 1578 0.1 × 0.1 1.003 0.667 0.688 0.09 0.115
interFoam 6811 0.067 × 0.067 1.006 0.715 0.688 0.03 0.21
interFoam 26068 0.044 × 0.044 1.002 0.703 0.708 0.008 0.36
interDyMFoam 242 0.09 × 0.09 1.036 0.695 0.694 0.1 0.15
interDyMFoam 661 0.063 × 0.063 1.017 0.699 0.719 0.03 0.225
interDyMFoam 1438 0.042 × 0.042 1.030 0.722 0.728 0.008 0.43

Figure 3.27: Dynamic refinement in 2D with Ca = 0.4 at t = 0.99 s for the

2D drop in a shear flow test case

relative change in mass, and the breakup time. The efficiency of interDyMFoam relative

to interFoam increases with mesh refinement, i.e. with decreased cell size around the

interface. For the three interDyMFoam meshes, the CPU time for interDyMFoam

decreases by a factor of 6.5, 10.3, and 18.1 relative to the corresponding interFoam

mesh. The droplet is broken into two droplets, thus a refinement is needed around the

interface for each droplet, (see Figures 3.27 and 3.28). The radius of each droplet depends

90
 Figure 3.28: Dynamic refinement in 2D with Ca = 0.4 at t = 0.99 s for the
2D drop in a shear flow test case

of the cell size around the interface and the initial radius of the droplet. The initial radius

depends on the initial mesh, especially for the interDyMFoam solver since the solver

starts with a coarse mesh compared to interFoam solver. As expected, each droplet

is broken into two daughter droplets with almost identical radius particularly on the fine

mesh. In principle, due to mass conservation, the area of the initial droplet must equal the

sum of the areas of the daughter drops. Mathematically, this can be written as

π R20 = π R21 + π R22 , (3.15)

where R0 is the radius of the initial drop. From Eq. (3.15), we get R20 = R21 + R22 . Therefore,
R20 −(R21 +R22 )
the relative change in R can be calculated using the formula R20
. In the table, the

91
relative change is small for all cases especially on the refined mesh. Also, the relative

change is almost the same if we compare between the solvers. The break up time is

dependent on the mesh resolution as the droplet breaks up on the coarse mesh faster than on

the finer mesh. Physically, the breakup time should not differ. However, in the simulations,

breakup time depends on the cell size, thus the difference in breakup time using different

meshes.

In conclusion, the modified interDyMFoam solver performs well compared with

interFoam on this 2D planar test problem. It produces similar results at much lower

CPU times.

3.6.2 A Drop Detachment From a Micro T-channel

In this section, the modified interDyMFoam is evaluated for the problem of a 2D

micro T-channel. In particular, the performance of the modified code is compared with

interFoam, and the effect of the parameters in dynamicMeshDict is investigated.

The micro T-channel flow problem consists of a disperse phase fluid which is transported

through a pore and into a gap containing a continuous phase fluid. Drops of the disperse

phase are then detached by the shear flow field of the continuous phase. The domain

used for the pore is a channel of length 25 micrometers and height of 200 micrometers.

The domain used for the gap is a channel of length 2250 micrometers and height of 500

micrometers The origin of the coordinate system is placed such that −250 ≤ x ≤ 2000

92
and 0 ≤ y ≤ 500 for the gap and −25 ≤ x ≤ 25, −200 ≤ y ≤ 0 for the pore. The

geometry is divided into four blocks as shown in Figure 3.29. Table 3.10 summarizes

the number of cells in each block for the three meshes in both solvers. The coarse and fine

meshes are obtained from the standard mesh by decreasing and increasing the number of

cells in each direction by a factor of 1.4, respectively. Table 3.11 shows the boundary

conditions used in the simulations. The fixedFluxPressure boundary means the

pressure gradient is adjusted such that the flux is specified using the velocity boundary

condition. The balancing of interfacial tension forces σls (liquid-solid), σ f s (fluid-solid),

and σl f (liquid-fluid) produces the equilibrium or static contact angle. The balance of these

surface tension can express in Young’s relation [51], σ f s − σls − σl f cos(θ ) = 0, where

θ is the angle between the tangent line of the liquid at the triple point, where the three

phases meet, and the solid from the liquid side. Wetting is the ability of liquid to maintain

contact with a solid surface, resulting from intermolecular interaction when the two are

brought together. Contact angle θ = 0 is a perfectly wetting case and θ = 180o is a perfect

non-wetting case. If the contact angle is larger than 90o then the surface is non-wetting.

On the other hand, if the angle is below than 90o the material is wetting the surface. In

the simulations, a non-wetting behavior is assumed by taking the static contact angle to be

180o . The end time for all cases is 0.1 s. The continuous and disperse phases are taken

to be Newtonian fluids. The velocity set to be 0.3m/s directed to the positive x-axis on the

inlet of the shear flow channel and about 0.01 m/s directed to the positive y-axis on the inlet

of the pore. The transport properties are as follows:

93
 (-250,500) top wall

inlet 1 block 2
block 1 outlet
block 4

(2000,0)
(-250,0)

block 3 bottom walls

(-25,-200) pore walls

inlet 2 (25,-200)

Figure 3.29: Geometry sketch for a drop detachment from a micro

T-channel test case where the units are in micrometers

μc = 1.056 × 10−1 Pa.s

μd = 1 × 10−3 Pa.s

ρc = 960kg/m3

ρd = 103 kg/m3

σ = 0.0415kg/s2

Table 3.10
Number of cells in each block for a drop detachment from a micro
T-channel test case

number
Solver block 1 block 2 block 3 block 4
of cells
interFoam (Coarse) 56 × 125 13 × 125 13 × 50 494 × 125 71081
interFoam (Standard) 78 × 175 18 × 175 18 × 70 692 × 175 139160
interFoam (Fine) 109 × 245 25 × 245 25 × 98 969 × 245 272685
interDyMFoam (Coarse) 19 × 42 5 × 42 5 × 17 165 × 42 8023
interDyMFoam (Standard) 27 × 59 7 × 59 7 × 24 231 × 59 15803
interDyMFoam (Fine) 38 × 82 10 × 82 10 × 33 323 × 82 30752

94
 Table 3.11
Boundary conditions for a drop detachment from a micro T-channel test
case

boundary velocity p− ρ gh alpha1

inlet 1 (0.3, 0, 0) zeroGradient zero
inlet 2 (0, 0.011, 0) zeroGradient 1
outlet zeroGradient 0 zeroGradient
bottom walls (0, 0, 0) fixedFluxPressure constantAlphaContactAngle(180o )
top wall (0, 0, 0) fixedFluxPressure constantAlphaContactAngle(180o )
pore wall (0, 0, 0) fixedFluxPressure constantAlphaContactAngle(180o )
front and back empty empty empty

where μ is the dynamic viscosity, ρ is the density, σ is the interfacial tension, and the

subscripts c and d stands for continuous and disperse phases respectively. The capillary
μc u
number Ca = σ , where u is the characteristic velocity, is about 0.76 and viscosity ratio
μd
λ= μc is about 0.0095. The Reynolds number is the ratio of inertial forces to viscous
ρ uL
forces and mathematically defined as Re = μ , where L is the characteristic length and u

is the characteristic velocity. The Reynolds number for the continuous phase Rec is about

1.37 and for the disperse phase Red is about 0.55. Figures 3.30 and 3.31 show the first

drop detachment using interDyMFoam and interFoam solver, respectively. In the

simulations, several droplets are produced of nearly uniform size. The drop size reported

in the following is the average of those sizes. In both cases, the radius of the detached drops

are larger than the pore radius. The coarsest mesh and refinement around the interface after

first drop detach are show in Figure 3.32.

95
Figure 3.30: Drop deformation and detachment at t = 0.01, 0.012, and
0.014 s using interDyMFoam for a drop detachment from a micro T-channel
test case

96
Figure 3.31: Drop deformation and detachment at t = 0.01, 0.012, and
0.014 s using interFoam for a drop detachment from a micro T-channel test
case

97
Figure 3.32: Dynamic refinement at t = 0, 0.004, and 0.014 s for a drop
detachment from a micro T-channel test case

98
3.6.2.1 Mesh Independence Study

This section compare the predictions of interFoam and the modified interDyMFoam,

each on three different meshes. In particular, we compare the pressure curves and drop

sizes. Figure 3.33 shows the pressure curves along a horizontal line through the center of

a detached droplet for three different meshes using interFoam. The graphs are almost

identical outside the droplet. The graphs indicate mesh independence for the standard

and fine meshes. The same behavior is noted in Figure 3.34, where the interDyMFoam

solver is used. The interDyMFoam appears to perform better than interFoam for this

case because the pressure does not undershoot close to the interface as observed in the

interFoam solution.

Next, we compare the detached drop size and the CPU time for the two solvers on the

Table 3.12
CPU time and ratio of a drop radius to the pore radius using maximum
refinement 1 for a drop detachment from a micro T-channel test case

Solver CPUTime (s) R/PR Relative Change

interFoam (Coarse ) 9056 1.54826
interFoam (Standard) 41817 1.6903 0.0840312
interFoam (Fine) 90305 1.77856 0.0496268
interDyMFoam (Coarse) 3798 1.62489
interDyMFoam (Standard) 9457 1.70793 0.0486203
interDyMFoam (Fine) 26159 1.77832 0.0395823

different meshes. This is summerized in Table 3.12. The drop size is given as the ratio of

the radius of the detached drop to the pore radius. The table shows that for both solvers the

99
 3000

Coarse Mesh
Standard Mesh
Fine Mesh
2500
Pressure [kg/m.s ]
2

2000

1500

1000
0.0003 0.0006 0.0009 0.0012
x-Coordinate [m]
Figure 3.33: Pressure using interFoam for a drop detachment from a micro
T-channel test case

drop size increases as the cell size around the interface decreases. However, the relative

change in drop size decreases with mesh refinement, and is smaller in interDyMFoam.

Moreover, the standard mesh of each solver produces nearly the same drop sizes (1.6903

vs 1.70793), as does the fine mesh of each solver (1.77856 vs 1.77832). In addition, the

CPU time using interDyMFoam is much smaller than the CPU time using interFoam

by a factor of 2.4 and 3.4 for the coarse and fine mesh respectively.

3.6.2.2 dynamicMeshDict Parameters Study

This section investigates the effect of the dynamicMeshDict parameters,

refineInterval, maxRefinement, and nBufferLayers, on the CPU

100
 3000

2500
Pressure [kg/m.s ]
2

2000

1500 Coarse Mesh

Standard Mesh
Fine Mesh
interFoam Fine Mesh
1000
0.0003 0.0006 0.0009 0.0012
x-Coordinate [m]
Figure 3.34: Pressure using interDyMFoam for a drop detachment from a
micro T-channel test case

times and drop sizes when using the interDyMFoam solver. The CPU time and

Table 3.13
CPU time and Ratio of a Droplet to the radius of the Pore with Different
Refine Interval Numbers using interDyMFoam solver with max.
refinement equal to 3 and buffer layer equal to one for a drop detachment
from a micro T-channel test case

refine interval CPUTime (s) CPUTime Ratio to 1 R/PR Relative Change to 1

1 10954 1.75554
3 5930 0.54 1.75498 0.000312
5 3718 0.34 1.75595 0.000233
7 3387 0.31 1.75047 0.002887
9 3231 0.30 1.73890 0.009481

normalized drop radius for different values of refineInterval is shown in Table 3.13.

Recall that the refine interval specifies how often the mesh should be refined, so that

refineInterval = n means that the mesh is dynamically refined every n time steps.

101
Assuming the solution is most accurate using refine interval equal to one, it will be used

as a standard to compare with other refine interval numbers. By comparing the refine

interval equal to 1 with the other refine interval numbers, the relative change in drop size

is negligible. Although there is a significant difference in the relative change in drop size

between refine interval equal to 7 and 9 compared to other refine interval numbers, the

relative change is still negligible. On the other hand, the time ratio is almost the same

using refine interval equal to 5, 7, and 9 which is around 0.3 but it is 0.5 when using refine

interval equal to 3. Thus, we suggest the refine interval to be between 5 and 9 because it

saves CPU time without compromising the accuracy of the results.

In Table 3.14, the effect of the maximum refinement number on the drop size and the

Table 3.14
CPU time and Ratio of a Droplet to the radius of the Pore with Different
Maximum Refinement Numbers using refine interval equal and buffer layer
equal to one for a drop detachment from a micro T-channel test case

max. refinement CPUTime (s) CPUTime Ratio R/PR Relative Change

2 3798 1.62489
3 10954 2.9 1.75554 0.074
4 33500 3.0 1.85210 0.052

CPU time is shown. Recall that the maximum refinement parameter, maxRefinement,

indicates how many times a given cell may be refined. The table shows that the CPU

time increases by a factor of 3 when the maximum refinement number is increased by 1.

However, the relative change in drop size is reduced from 0.074 to 0.052, which means

that the maximum refinement number has an effect on the CPU time and the drop size.

Table 3.15 outlines the effect of the number of buffer layers on the CPU time and drop

102
 Table 3.15
CPU time and Ratio of a Droplet to the radius of the Pore with Different
Number of Buffer Layers using refine interval equal to one and max.
refinement equal to 2 and 4 for a drop detachment from a micro T-channel
test case

Solver CPU time (s) R/PR

interDyMFoam (max. 2, Buf1) 3798 1.6249
interDyMFoam (max. 2, Buf3) 2888 1.6293
interDyMFoam (Max. 4, Buf1) 33500 1.8521
interDyMFoam (Max. 4, Buf3) 28260 1.8566

size for two different cases, one with maximum refinement 2 and the other with maximum

refinement 4. Recall that the number of buffer layers is used to find the buffer layers that

should be extended for unrefinement. Here, we take the number of buffer layers equal to 1

and 3. The table shows that for each value of maxRefinement (2 or 4), the CPU time

decreases slightly as the number of buffer layers increases from 1 to 3, while the drop sizes

remain essentially the same. Therefore, the results imply that the number of buffer layers

has little or insignificant effect on the CPU time and drop size.

103
3.7 Test of interFoam and interDyMFoam in 2D

Axisymmetric Geometry

A bubble rising in water is analyzed to test the modified interDyMFoam solver in

axisymmetric geometry. Figure 3.35 shows the geometry used for the simulations. The

computational domain is a small wedge channel open into a large wedge channel. The

small channel has width of 1 mm and height of 5 mm and contains the disperse phase (air).

The large channel has width of 40 mm and height of 80 mm and contains the continuous

phase (water). The origin is placed such that 0 ≤ x ≤ 1, −5 ≤ y ≤ 0, −0.01 ≤ z ≤ 0.01 for

the small channel and 0 ≤ x ≤ 40, 0 ≤ y ≤ 80, −0.4 ≤ z ≤ 0.4 for the large channel. The

center line of the axisymmetric geometry is placed such that −5 ≤ y ≤ 80, x = 0, and z = 0.

The number of cells in each block for the all cases is outlined in Table 3.16. The coarse

and fine meshes are produced by decreasing and increasing the standard mesh by a factor

of 1.5, respectively . Also, the boundary conditions are summarized in Table 3.17. The end

time for all cases is 0.5 s. The two fluids in the simulation are Newtonian. The transport

properties are as follows:

μc = 1 × 10−3 Pa.s

μd = 1.8 × 10−5 Pa.s

ρc = 1000Kg/m3

ρd = 1Kg/m3

104
 side upper
80mm
block 1

right wall
y-axis block 3

0mm 1mm 40mm

disperse phase channel wall
block 2 pore wall
-5mm x-axis
inlet

Figure 3.35: Geometry sketch of a bubble rising in a water axisymmetric

case

Table 3.16
Number of cells in each block for the standard mesh of a bubble rising in a
water axisymmetric case

Solver block 1 block 2 block 3 number of cells

interFoam (Coarse) 5 × 400 5 × 25 200 × 400 82125
interFoam (Standard) 8 × 600 8 × 38 300 × 600 185104
interFoam (Fine) 12 × 900 12 × 57 450 × 900 416484
interDyMFoam (Coarse) 1 × 130 1 × 08 66 × 130 8718
interDyMFoam (Standard) 2 × 195 2 × 12 99 × 195 19719
interDyMFoam (Fine) 3 × 293 3 × 18 149 × 293 44590

σ = 0.072Kg/s2

where μ is the dynamic viscosity, ρ is the density, σ is the interfacial tension, and the

subscripts c and d stands for continuous and disperse phases respectively. The capillary

number Ca is 0.375 × 10−3 , viscosity ratio λ is 0.018, and Reynolds number Re for

the disperse phase is 1.5. Figures 3.36 and 3.37 show the bubble deformation and

detachment using interDyMFoam and interFoam solver, respectively. The disperse

105
 Table 3.17
Boundary conditions of a bubble rising in a water axisymmetric test case

boundary velocity p− ρ gh alpha1

inlet (0, 0.0265258, 0) zeroGradient 1
side upper zeroGradient zeroGradient zeroGradient
right wall (0, 0, 0) 0 zeroGradient
channel wall (0, 0, 0) fixedFluxPressure constant angle (110o )
pore wall (0, 0, 0) zeroGradient zeroGradient
center line empty empty empty
front and back wedge wedge wedge

phase produced a bubble that is larger in size than the small channel length. The refinement

around the interface and unrefinement between the bubble and the inlet are shown in

Figure 3.38.

106
Figure 3.36: Drop deformation and detachment at t = 0.335, 0.34, and 0.35
s using interDyMFoam for a bubble rising in water test case

107
Figure 3.37: Drop deformation and detachment at t = 0.35, 0.355, and
0.365 s using interFoam for a bubble rising in water test case

108
Figure 3.38: Dynamic refinement around the interface for a bubble rising
in water test case

109
Figure 3.39 shows the pressure along the centerline predicted by interFoam. The jump

in the pressure decreases with mesh refinement, but the difference in pressure jump between

consecutive meshes decreases as well. This indicates the jump in pressure is converging

with mesh refinement. The mesh independence of the centerline pressure predicted by

the modified interDyMFoam is shown in Figure 3.40. This figure also shows that the

pressure jump predicted by both solvers agrees on the fine meshes.

Finally, Table 3.18 presents the CPU time, bubble radius, and relative change in the

radius for the two solvers. The CPU time is much less using interDyMFoam. For

example, on the fine mesh, using interFoam, the CPU time increased by a factor of

4.6 compared to interDyMFoam. The table also shows that the bubble radius increases

by increasing the mesh resolution but the difference becomes smaller for both solvers.

Moreover, the drop sizes predicted by the two solvers compare well. As a result, the

modified interDyMFoam solver gives results similar to interFoam but for much less

CPU time.

Table 3.18
CPU time, bubble radius, and relative change using interFoam and
interDyMFoam for a bubble rising in a water axisymmetric test case

Solver CPU Time (s) R relative change

interFoam (Coarse) 9932 1.767
interFoam (Standard) 30025 1.837 0.038
interFoam (Fine) 122298 1.865 0.015
interDyMFoam (Coarse) 1767 1.785
interDyMFoam (Standard) 7624 1.830 0.025
interDyMFoam (Fine) 26507 1.845 0.008

110
 -150
Coarse Mesh
Standard Mesh
Fine Mesh

Pressure [kg/m.s ]
2 -200

-250

-300
0.018 0.021 0.024 0.027 0.03
y-coordinate [m]
Figure 3.39: Pressure using interFoam for a bubble rising in a water
axisymmetric test case

3.8 Summary and Conclusion

In summary, the interDyMFoam solver for 3D geometry is modified to work for 2D

planar and axisymmetric geometries. Also, the solver is modified to allow for computing

the deformation and breakup of drops or bubbles that are very small relative to the mesh

of the flow domain. To validate the modified interDyMFoam, three test problems are

considered: two in 2D planar geometry and one in axisymmetric geometry. The solutions

from the modified interDyMFoam are compared with the solutions from interFoam

from the aspects of mass accuracy, CPU time, and cell size around the interface. The

modified interDyMFoam gives accurate solutions compared to interFoam with much

111
 -150
Coarse Mesh
Standard Mesh
Fine Mesh
interFoam Fine Mesh
Pressure [kg/m.s ]
2

-200

-250

-300
0.018 0.021 0.024 0.027 0.03
y-Coordinate [m]
Figure 3.40: Pressure using interDyMFoam for a bubble rising in a water
axisymmetric test case

less CPU time.

112
Chapter 4

Break up Conditions Inside a Spray

Nozzle

An emulsion is a mixture of two immiscible phases of which one is dispersed in

another. If the disperse phase itself contains drops of another liquid, then the emulsion

is called a double emulsion. The most common types of double emulsions are the

water-in-oil-in-water (W-O-W) and oil-in-water-in-oil (O-W-O). Their applications are

found in many industries, including the food and pharmaceutical industries. Hydrophilic

and lipophilic surfactants are encapsulated to produce a stable emulsion since they reduce

the interfacial tension between phases. It is desirable for emulsions and double emulsions

to have mechanical strength so they can stand against the mechanical forces generated by

the manufacturing process such as mixing and pumping. After production, they should be

113
weak enough to release their encapsulations in the desired manner, such as the controlled

release time during digestion and the release rate of flavor by chewing.

Double emulsions are inherently unstable due in part to coalescence and compositional

ripening [1], [2], [3]. Coalescence is the process by which droplets merge with each

other to form larger droplets, whereas, compositional ripening occurs by diffusion and/or

permeation of the surfactants components across the disperse phase. The solid powder

of emulsions produced by spraying is more stable than the liquids emulsions since the

coalescence and ripening are reduced [11], [52]. Therefore, emulsion powders tend to

have a longer shelf life which is desirable for economical reasons. In addition to these

advantages, powder emulsions reduces the amount of the stabilizer needed.

It is important for double emulsions to retain their structure during and after the spray

processing. For example, drops should not break up since this would change the structure,

and hence properties of emulsions. Therefore, it is important to study droplet break up

conditions inside and outside the nozzle.

In spraying emulsions, the atomization in a controlled manner is necessary to maintain its

structure such as droplet size and droplet distribution. In their experiments, Dubey et al.

[53] produced solid particles by spraying and studied the influences of the spray process

parameters on the structure of double emulsions. Uddin et al. [54] investigated the effect of

insoluble surfactants on the breakup of rotating liquid jets. Drop breakup during spraying of

emulsions was also investigated by other researchers experimentally and computationally

[55, 56] and emulsions with a good particle size distributions were produced.

114
In this chapter, the droplet breakup conditions inside a spray nozzle is analyzed for a

simple emulsion. The study is achieved by the simulation of two-phase flow where the

droplet is a Newtonian fluid and the outer phase is either Newtonian or non-Newtonian.

Specifically, drops of different sizes are tracked through the flow field of the continuous

phase fluid. The goal is to determine the effects of shear rate, capillary numbers, viscosity

ratio, and fluid rheology (Newtonian or non-Newtonian) on the droplet breakup. Of

particular interest is the critical drop size, that is, the largest drop that does not break up

within the nozzle. To determine these critical drop sizes, many drops of different sizes

must be tracked along different particle tracks for a given fluid system. Since this would

be computationally very expensive in three dimensions, even with the use of dynamic

meshing, the simulations are performed in two dimensions. This allows us to study

the qualitative behavior and functional relationships, and will help determine appropriate

three-dimensional simulations.

A first step, single phase flow calculations are performed to study the mesh independence

for the outer phase fluid. This allows us to determine a mesh suitable to describe the outer

phase flow field. Then, the two-phase flow is solved using dynamic refinement mesh to

have accurate resolution around the interface. Using dynamic meshing around the drop

interface as it moves through the flow field is necessary since the drops are most often very

small relative to the dimensions of the geometry. The refined mesh from the single phase

calculations is used as a basic mesh for the two-phase flow calculations.

115
4.1 Problem Description

The geometry and material properties in the simulations are taken from the experiments

of Dubey et al. [11]. A cylindrical tapered die geometry is used in their experiments, see

Figure 4.1. The upstream cylinder had radius Ru = 3 mm and length Lu = 22.5 mm, and

the downstream cylinder had radius Rd = 0.5 mm and length Ld = 1.5 mm, giving the

contraction ratio, Ru /Rd , of 6:1. Also, the emulsion (which is non-Newtonian) is produced

from two Newtonian fluids. The transport properties of the fluids used to produce the

emulsion are μd = 0.0634Pa.s, ρd = 918Kg/m3 , μc = 0.0209Pa.s, ρc = 1018Kg/m3 , and

interfacial tension is σ = 0.00575Kg/s2 . The computational domain is taken to be the

two dimensional version of the actual nozzle geometry, with symmetry assumed along

the centerplane. Moreover, in the simulations, the length of the downstream channel is

extended from 1.5 mm to 20 mm, which allows us to consider the effect of nozzle channel

length on drop breakup. The computational domain and coordinate system are also given

in Figure 4.1. The transport properties for the droplets are taken to be the same as above:

ηd = 0.0634 Pa.s ρ = 918Kg/m3 σ = 0.00575Kg/s2 .

The continuous phase is taken to be either the (non-Newtonian) emulsion or a Newtonian

fluid. The non-Newtonian fluid has a slightly shear-thinning behavior, whose viscosity

function followed the Bird-Carreau model

η − η∞ n−1
= [1 + (mγ̇ )2 ] 2 (4.1)
η0 − η ∞

116
with parameters η0 = 0.113Pa.s, η∞ = 0.08Pa.s, ρ = 977Kg/m3 , m = 0.0049 s, and n =

0.01323. In this model, η0 and η∞ represent the zero-shear-rate and infinite-shear-rate

viscosities, respectively, m is a time constant whose reciprocal gives the shear rate at which

the fluid begins to shear thin, and n is the dimensionless power-law index which controls

the rate at which the fluid shears thin. The values of ρ and η0 are used for the density

and viscosity of the Newtonian continuous phase fluid. Table 4.1 summarizes the material

parameters used in the simulations. The relation between the viscosity and shear rate for

the non-Newtonian continuous phase fluid, given by the Bird-Carreau model, is shown in

Figure 4.2. The viscosity ratio λ = ηd /ηc corresponding to the Newtonian continuous

phase was λ = 0.56, while λ ranged from 0.56 to 0.79 for the non-Newtonian continuous

phase. The boundary conditions are outlined in Table 4.2. The centerline boundary (y = 0)

Table 4.1
Fluid parameters used in the simulations. The parameters for the
non-Newtonian fluid correspond to the Bird-Carreau viscosity model,
Eq. (4.1).

Fluid Density Viscosity

Phase ρ [kg/m3 ] η [Pas]
Dispersed ρd = 918 ηd = 0.0634

Newtonian
Continuous ρc = 977 ηc = 0.113

Non-Newtonian
Continuous ρc = 977 η0 = 0.113
η∞ = 0.08
m = 0.0049 s
n = 0.01323 [-]

117
 -22.5
6 mm
22.5 mm

5 mm
3
1 mm 0 y [mm]
2.5
1
4.5 mm

5.5
1.5 mm

25.5 0.5
1 mm
x [mm]
Figure 4.1: Schematic diagram of the nozzle geometry used in experiments
(left) and the computational domain used in the simulations (right).
.
is specified as symmetry boundary. The velocity is set to be zero on the walls, zero gradient

on the outlet, and constant u ∼

= (0.03m/s, 0, 0) at the inlet, where this velocity corresponds

to one flow rate used in the experiments. The boundary conditions for the volume fraction

function α are taken to be zero gradient, where this function is used in the two-phase

calculations. Finally, the pressure is taken to be zero gradient along the inlet and walls, and

zero on the outlet. The Reynolds number for the Newtonian flow is Re = 0.76 and for the

non-Newtonian flow varies in the range 0.76 ≤ Re ≤ 1.1.

118
 0.14

0.12

Shear Viscosity [Pa.s]

0.1

0.08

0.06

0 1 2 3 4 5 6 7
10 10 10 10 10 10 10 10
Shear Rate [1/s]
Figure 4.2: Viscosity vs shear rate of the non-Newtonian continuous phase
fluid predicted by the Bird-Carreau model.
.

Table 4.2
Boundary conditions for the nozzle.

boundary velocity p alpha1

inlet (0.0294731, 0, 0) zeroGradient zeroGradient
outlet zeroGradient 0 zeroGradient
walls (0, 0, 0) zeroGradient zeroGradient
centerline symmetryPlane symmetryPlane symmetryPlane
front and back empty empty empty

4.2 Single Phase Flow Calculations

The computational domain is divided into five blocks as shown in Figure 4.3. Table 4.3

summarizes the number of cells in each block for the coarse, standard, and fine mesh. Note

119
that all meshes are non-uniform. The coarse and fine mesh are obtained by decreasing and

increasing the number of cells in the standard mesh by a factor of 1.5, respectively. The

single phase simulations are achieved using the simpleFoam solver of OpenFOAM
R
. In

this section, the mesh independence and the convergence are discussed for both continuous

phase fluids. In all graphs in this section, the dashed curves represent the non-Newtonian

fluid and the solid ones represent the Newtonian fluid. The residual convergence for the

fine mesh is illustrated in Figure 4.4 where the residualControl are 10−4 and 10−5

for the pressure and velocity respectively. That is, the velocity-pressure iterations terminate

when the current solution to each discrete system of equations for velocity components and

pressure produces a residual that meet these criteria simultaneous. The number of iterations

needed for convergent is a little less in the case of Newtonian fluid.

The velocity along the center line is shown in Figure 4.5. The figure shows mesh

independence for both fluids. By comparing the Newtonian and non-Newtonian graphs, the

figure also shows almost identical velocities until x = 5.5 mm and then in the small channel

(x ≥ 5.5 mm), the velocity is greater in the Newtonian case. This is due to the shear-thinning

behavior of the non-Newtonian fluid. The calculated pressure field for each fluid is also

mesh independent. This is illustrated in Figure 4.6 where the centerline pressure is plotted.

For both fluids, pressure decreases very slightly in the large channel, while the pressure

gradient is much larger in magnitude in the small channel. This is consistent with the

analytical expression for the constant pressure gradient in fully-developed pressure-driven

dp
flow, where it is seen that dx is inversely proportional to a power of channel height

120
 (0,3)

block 2 (1,2.5)

inlet
block 3
block 1

block 4
(5.5,0.5)
block 5 outlet

(-22.5,0) (0.0) (1,0) (5.5,0) (25.5,0)

centerline

Figure 4.3: Computational domain and number of blocks for the nozzle.

dp −3 Q
H. Actually, for a Newtonian fluid dx = 2 μ H3
where Q is the flow rate, while for a
 n
dp Q 1 1n +1 1
power-law, where η (γ̇ ) = K γ̇ n−1 , dx = −κ 2H ( H ) ( n + 2) . Figure 4.6 also shows

that the non-Newtonian fluid has lower pressure and lower pressure gradients. This is due

to shear-thinning behavior of the fluid.

We choose to perform the two-phase calculations using the fine mesh which is the most

suitable for tracking small drops as described in the next section.

Table 4.3
Number of cells in each block for the nozzle.

mesh block 1 block 2 block 3 block 4 block 5 number of cells

Coarse 107 × 12 107 × 2 5 × 12 21 × 12 95 × 12 2950
Standard 160 × 18 160 × 3 7 × 18 32 × 18 142 × 18 6618
Fine 240 × 27 240 × 5 11 × 27 48 × 27 231 × 27 15510

121
 1e+00

1e-01
Ux_0
Uy_0
1e-02 p_0
Residual

1e-03

1e-04

1e-05

1e-06
0 100 200 300 400
Number of Iterations
Figure 4.4: Residual using Newtonian (solid curves) and non-Newtonian
(dashed curves) fluids for the refined mesh for the nozzle.

Coarse Mesh
0.25 Standard Mesh
Fine Mesh
0.2
Velocity [m/s]

0.15

0.1

0.05

0
-0.02 -0.01 0 0.01 0.02
x-coordinate [m]
Figure 4.5: Velocity along the centerline for the single phase calculations
using Newtonian (solid curves) and non-Newtonian (dashed curves) for the
nozzle. The vertical dashed lines at x = 0 and x = 5.5 mm indicate the
contracting part of the domain.

122
 6
Coarse Mesh
5 Standard Mesh
Kinematic pressure [m /s ]
2

Fine Mesh
2

0
-0.03 -0.02 -0.01 0 0.01 0.02
x-coordinate [m]
Figure 4.6: Pressure along the centerline for the single phase calculations
using Newtonian (solid curves) and non-Newtonian (dashed curves) for the
nozzle. The vertical dashed lines at x = 0 and x = 5.5 mm indicate the
contracting part of the domain.

123
4.3 Drop Tracking Along Streamlines

In this section, the break up investigations are performed for droplets moving along

different streamlines, see Figure 4.7. Each drop started in the fully developed flow in

the upstream channel, and the streamlines were labeled by the initial y-coordinate of the

drop’s center. The streamlines ranged from y = 0 (centerline) to y = 2.7 mm (close to the

upstream channel wall, located at y = 3 mm). Figure 4.8 shows the shear rates, γ̇ = |γ̇|

where γ̇ = ∇v + (∇v)T is the rate-of-strain tensor, along various streamlines for both

the Newtonian fluid (solid curves) and non-Newtonian fluid (dashed curves) as a function

of particle transit time along the streamline. For each streamline, t = 0 corresponds to

the beginning of the contraction section of the domain (at x = 1 mm). Characteristic of

shear-thinning fluids, their shear rates are smaller than those for the Newtonian fluid close

to the centerline and larger than those for the Newtonian fluid close to the wall.

Dynamic meshing is used in the simulations to have accurate resolution around the

Table 4.4
DynamicMeshDict parameters for the nozzle.

refineInterval field lowerRefInterval upperRefInterval

1 alpha (α ) 0.1 0.9
unrefineLevel nBufferLayersR nBufferLayers maxRefinement maxCells
10 1−2 1 3−6 400000

interface. The parameters used in dynamicMeshDict are as follows: refine interval

equal to one, field alpha α , lower refine interval equal to 0.1, upper refine interval equal

124
to 0.9, number of buffer layers for unrefinement equal to 1, number of buffer layers for

the refinement ranging between 1 − 2 where one is used for large drops and two is used

for small drops, and maximum refinement ranging between 3 − 6 to establish the mesh

independence of drop breakup. Table 4.4 summaries these parameters. Figure 4.9 shows

the mesh refinement around a drop in two regions of the domain: in the upstream portion

of the domain, where the shear rates are relatively low and the drop remains nearly circular,

and in the downstream portion of the domain, where the shear rates are high and the

drop elongates dramatically before breaking up. Recall that the refinements are shown

as diagonals, although the cells are actually partitioned into rectangles.

Before the 2D planar simulations were performed, axisymmetric simulations were

accomplished, keeping in mind that the axisymmetric simulation are valid only for droplets

along the center line (see Figure 4.10). From the simulation, break up did not occur along

the center line and the same results are observed in the 2D planar simulation. Moreover,

no breakup occurred on streamlines y < 0.75 in the 2D simulations since the maximum

drop radius that can be placed in the nozzle is 3 mm and larger drops are needed to get the

breakup along these streamlines.

The critical drop size is defined to be either the largest drop size for which a drop does

not break up in the domain, or the smallest drop size for which breakup occurs. Figure 4.11

shows a drop deformation and breakup along the streamline y = 1.5 for the non-Newtonian

continuous phase fluid. Figure 4.12 is presented to show the critical breakup radius at

different streamlines (top) and shear rate in the downstream channel (bottom) for both

125
 Figure 4.7: Nozzle streamlines at y equal to 0.75, 1, 1.5, 2, and 2.5
millimeter.

continuous phase fluids. Two curves are shown for each of these fluids. In each case,

the lower and upper curves correspond to no breakup and breakup, respectively, so that

the critical drop size lies somewhere between these two curves. For both fluids, the

critical breakup radius is smaller when the streamline is farther from the centerline until

about streamline y = 1.5 or y = 2.0 mm. Moreover, there is a rapid decreasing in the

critical breakup radius for streamlines less than or equal to y = 1.5, corresponding to the

downstream shear rate of γ̇ ≈ 500 1/s, while it slows down after that remaining almost the

same. Note that the critical radius agrees between the two fluids after streamline y = 1.5.

Therefore, the fluid rheology has an effect on the critical breakup radius along streamlines

close to the center line, i.e y < 1.5. This effect makes the critical breakup radius larger in

the case of Newtonian fluid where the opposite was expected since the shear rate is larger.

A possible theory that could explain the results is because the non-Newtonian fluid has

different viscosity ratios, affecting the critical breakup radius as will be discussed later. For

126
 1000

800

Shear Rate [1/s] 600

y=2.7 mm
y=2.5 mm
400 y=2.25 mm
y=2.0 mm
y=1.5 mm
200 y=1.0 mm
y=0.75 mm

0
-0.4 -0.2 0 0.2 0.4 0.6 0.8
Streamline Transit Time [s]

Figure 4.8: Shear rates as a function of transit time along a set of

streamlines for the Newtonian (solid curves) and non-Newtonian (dashed
curves) continuous phase fluid. Along each streamline, t = 0 corresponds
to beginning of the contraction at x = 1 mm.

μd
example, for the non-Newtonian fluid, the viscosity ratio λ = μc ranges from λ = 0.62

along y = 0.75 mm to λ = 0.65 along y = 1.0 mm, while for the Newtonian continuous

phase, the viscosity ratio is somewhat smaller, remaining at λ = 0.56.

The capillary number represents the ratio of viscous forces to interfacial forces and is
rγ̇ηc
calculated using the formula Ca = σ , where r is the radius of the droplet, γ̇ is the shear

rate of the continuous phase, ηc is the continuous phase viscosity, and σ is the interfacial

tension. The critical capillary number is a number for which the droplet breaks up if the

capillary number is greater than the critical one (Ca > Cacrit) and does not break up if

127
 Figure 4.9: Mesh around the drop interface in the low-shear-rate upstream
(top) and high-shear-rate downstream (bottom) portions of the domain for
the nozzle.

the capillary number is less than the critical one (Ca < Cacrit). The general behavior seen

for the critical radius is also observed for the critical capillary number Cacrit as shown

in Figure 4.13, where the critical capillary number Cacrit is calculated using the constant

shear rate and viscosity values in the downstream channel. The critical capillary numbers

decrease with distance from the centerline before becoming approximately constant. Also,

the the critical capillary numbers for the Newtonian fluid are larger than those for the

128
 Figure 4.10: Drop deformation at t = 0.01, 0.08, and 0.13 s for the nozzle.

non-Newtonian fluid for streamlines close to the centerline.

The shear rate and drop breakup location along different streamlines for both fluids

is shown in Figure 4.14. The location is given by the x-coordinate of the center of

the elongated drop and is indicated by a circle along the streamline. The upper graph

129
 Figure 4.11: Drop deformation and breakup for streamline y = 1.5 at t =
0.02, 0.42, and 0.5 s for the non-Newtonian continuous phase for the nozzle.

corresponds to the Newtonian continuous phase and the lower graph corresponds to the

non-Newtonian continuous phase. The vertical dashed lines at x = 0 and x = 5.5 mm

indicate the contracting part of the domain (see Figure 4.1), while the vertical line at

130
 0.4
Newtonian

Critical Drop Radius [mm]

Non-Newtonian
0.3

0.2

0.1

0
0.5 1 1.5 2 2.5 3
Streamline Label Y [mm]

0.4
Newtonian
Critical Drop Radius [mm]

Non-Newtonian
0.3

0.2

0.1

0
0 200 400 600 800 1000 1200
Streamline Shear Rate in Die [1/s]

Figure 4.12: Critical drop sizes as a function of the streamline position

(top) and the downstream shear rate (bottom) for the Newtonian and
non-Newtonian continuous phase for the nozzle.

x = 7 mm indicates the end of the nozzle used in experiments. The breakup occurs near

the beginning of the downstream channel for streamlines far from the centerline y  2.25

mm, but it occurs closer to the end of the downstream channel for streamlines near the

centerline. The breakup position gives an idea on how long the nozzle should be to produce

break up along different streamlines. The simulations indicate that within the length of the

131
 2.5

Critical Capillary Number Cacrit

Newtonian
2
Non-Newtonian

1.5

0.5

0
0.5 1 1.5 2 2.5 3
Streamline Label Y [mm]
Figure 4.13: Critical Capillary number as a function of the streamline
position for the Newtonian and non-Newtonian continuous phase for the
nozzle.

actual nozzle used in experiments breakup would occur only along streamlines close to the

wall y ≥ 2.25. Table 4.5 gives the breakup locations for super-critical drop sizes along

streamlines y = 1.5 and y = 2.0 for the non-Newtonian continuous phase. There appears to

be little effect of super-critical drop size on breakup location, for drop sizes relevant to this

geometry.

Grace [57] has constructed a plot of the critical capillary number as a function of the
μd
viscosity ratio λ = μc for Newtonian/Newtonian fluid systems in unbounded simple shear

flow as shown in Figure 4.15. In the graph, the droplet does not breakup if the capillary

number is under the curve and the breakup occurs when the capillary number is above the

132
 1200

Y=2.7
1000 Y=2.5
Y=2.25
Y=2.0

Shear Rate [1/s]

800
Y=1.5
Y=1.0
600 Y=0.75

400

200

0
-5 0 5 10 15 20 25
Axial Position x [mm]
1200

Y=2.7
1000 Y=2.5
Y=2.25
Y=2.0
Shear Rate [1/s]

800
Y=1.75
Y=1.5
600 Y=1.0
Y=0.75

400

200

0
-5 0 5 10 15 20 25
Axial Position x [mm]

Figure 4.14: Breakup position of a drop along a given streamline in the

Newtonian continuous phase (top) and non-Newtonian continuous phase
(bottom) for the nozzle.

curve. The droplet does not breakup if the viscosity ratio is greater than a number around 4.

Instead the large viscosity of the disperse phase relative to the continuous phase causes the

droplet to rotate. Moreover, the minimum critical capillary number is when the viscosity

ratio somewhere between about 0.6 and 1. This is because the viscosity of the disperse is

133
 Table 4.5
Drops breakup location for streamlines y = 1.5 and y = 2 in the
non-Newtonian continuous phase for the nozzle.

breakup breakup drop

Streamline drop radius (mm)
location (mm) center (mm)
y = 1.5 0.041 17-20 18.5
y = 1.5 0.090 14-18 16.0
y = 1.5 0.148 11-16 13.5
y = 1.5 0.196 11-18 14.5
y = 1.5 0.245 10-17 13.5
y = 2.0 0.017 08-10 09.0
y = 2.0 0.090 09-13 11.0
y = 2.0 0.157 08-13 10.5
y = 2.0 0.202 08-16 12.0
y = 2.0 0.245 07-16 11.5

less than the viscosity of the continuous phase which makes the deformed droplet horizontal

and aligned with the flow field. As a result, the viscosity ratio has an effect on the critical

capillary number which implies that it affects the droplet breakup radius.

To study the relation between the critical capillary number and viscosity ratio for the

Newtonian fluid at streamlines y = 1 and y = 2, Figure 4.16 is presented. Many droplets

are tracked along these streamlines with different viscosity ratios to get the curves. The

critical capillary number for each case is between the lower and upper curve for each case.

The shear rate used in the computation of the capillary number is the streamwise constant

shear rate in the downstream channel, namely γ̇ = 352 s−1 and 710 s−1 for streamline

y = 1.0 mm and 2.0 mm, respectively. The dashed vertical lines in the figure represent the

range of viscosity values reached for the original drop in the non-Newtonian continuous

phase fluid. The critical capillary number reaches the minimum when the viscosity ratio

134
 Figure 4.15: Critical capillary number vs viscosity ratio (Grace curve).

is between 0.56 and 1 and they show a similar behavior to the Grace plot where small

changes are observed at streamline y = 2 due in part to small critical breakup radius. The

critical capillary number decreases when the viscosity ratio increases from 0.56 to 0.75

at streamline y = 1 and remains constant at streamline y = 2. In particular, the viscosity

ratio has an effect on the critical capillary and radius numbers where low (λ ≤ 0.56) and

high (λ ≥ 1) viscosity ratio increase these numbers. The decrease in the critical capillary

number at streamline y = 1, helps explain why the critical drop radius (Figure 4.12) and

critical capillary number (Figure 4.13) are larger for the Newtonian continuous phase than

for the non-Newtonian continuous phase for streamlines closer to the centerline, such as

y = 1 mm, even though the shear rates are larger for the Newtonian case. Likewise, the

nearly constant value of Cacrit between λ = 0.56 and λ = 0.79 along streamline y = 2.0 mm

helps explain why there is little or no difference between the critical drop radius (Fig. 4.12)

135
and critical capillary number (Fig. 4.13) along streamlines farther from the centerline, such

as y = 2.0 mm.
3

y=1.0 mm

Critical Capillary Number Cacrit

2.5 y=2.0 mm

1.5

0.5

0
0 0.5 1 1.5 2
Viscosity Ratio λ

Figure 4.16: Critical capillary number as a function of viscosity ratio along

two streamlines in the Newtonian continuous phase for the nozzle. The
dashed vertical lines represent the range of viscosity ratios encountered for
the original drop viscosity (see Table 4.1).

4.4 Summary and Conclusions

In summary, the droplet breakup conditions inside a spray nozzle is analyzed for a simple

emulsion. The study is achieved by the simulation of two-phase flow where the droplet is a

Newtonian fluid and the outer phase is either Newtonian or non-Newtonian. Because a large

number of drops must be tracked in order to determine breakup conditions, the simulations

were performed in two dimensions using the modified interDyMFoam solver. A first

136
step, single phase flow calculations are performed to study the mesh independence for the

outer phase fluid. Then, the two-phase flow is solved using dynamic refinement mesh

to have an accurate resolution around the interface. The simulations on a single phase

flow revealed a mesh independence after analyzing velocity and pressure on three different

meshes. The refined mesh is used as a basic mesh for the two-phase flow.

For both continuous phase fluids, there was an initial rapid decrease in critical drop size as

distance from the centerline of the nozzle increased, i.e., as the shear rates experienced by

the drop increased. Starting at approximately half-way between the centerline and the wall,

the critical drop radius became approximately constant or decreased only slightly.

By noting the location of breakup within the nozzle, it was determined that drops near

the centerline break up only for very long dies. Within the length of the nozzle used in

experiments, only drops closer to the wall broke up for the flow rate considered.

It was also found that close to the centerline of the nozzle, critical drop sizes were larger

for the Newtonian continuous phase than for the non-Newtonian continuous phase, even

though the shear rates were larger along these streamlines for the Newtonian fluid. The

explanation for this is partly due to the viscosity ratios reached in the simulations. This

was illustrated by determining critical capillary numbers for a range of viscosity ratios

along two streamlines in the Newtonian continuous phase. The resulting Cacrit -vs-λ curves

resembled the well-known Grace curve for steady simple shear flow.

From these simulations we can get general idea about the droplet breakup conditions inside

a spray nozzle. Additional simulations and experimental validation are needed.

137
Chapter 5

Summary and Future Work

In summary, this study focused on the dynamic meshing in a two-phase flow solver. The

3D solver has been modified to allow for dynamic meshing around fluid-fluid interfaces in

two-dimensional planar and axisymmetric geometries. Moreover, the procedure is modified

to allow for computing the deformation and breakup of drops or bubbles that are very small

relative to the mesh of the flow domain. This is necessary to avoid mass loss when tracking

small drops or bubbles through flow fields.

To validate the modifications, the modified dynamic meshing code for two-dimensional

planar geometry was applied to two test problems: drop deformation and breakup in linear

shear flow, and drop formation and detachment from a micro T-channel. The modified

dynamic meshing code for axisymmetric geometry was applied to a bubble rising from

a pore into a static liquid. In these test problems, we studied computational time, mesh

139
independence, and mass accuracy. Comparisons were made with the two-phase flow solver

without dynamic meshing.

To investigate the validity of the modified code, simulation and performances have been

discussed in detail. It was found that the modified code produces accurate results with

much less CPU time in comparison to simulations without dynamic meshing. A summary

of the conclusions from the validation tests are given in Section 3.8.

The modified code was then applied to study the droplet breakup conditions inside a spray

nozzle for a single emulsion, where droplets of various sizes were tracked through the

flow field within the nozzle to determine the conditions under which they break up. The

goal was to determine the largest drop sizes for which the breakup does not occur. The

critical radius and capillary number at different streamlines with various shear rates were

found along with the breakup location for both Newtonian and non-Newtonian continuous

phase fluids. Furthermore, the effect of viscosity ratio on the critical capillary number in a

Newtonian continuous phase fluid was determined for two streamlines.

The simulations showed that the critical drop sizes decreased rapidly as distance from

the centerline increased, before becoming approximately constant. The simulations also

showed that the fluid rheology has an effect on the critical drop size at streamlines near

the centerline, however, there was no effect far from the centerline. This behavior was

attributed to the range of viscosity ratios reached in the simulations and the Grace curves

which where produced for this geometry. A summary of conclusions can be found in

Section 4.4.

140
Although the modified dynamic meshing procedure was applied to fluid-fluid interfaces in

two-phase flow problems, it can be adapted to other regions in the domain and for other

types of flow problems in 2D planar and axisymmetric geometries.

Future Work

This thesis provides a preliminary investigation into the breakup conditions of emulsion

droplets inside a spraying nozzle. The study used a nozzle with diameter 1 mm,

downstream cylinder of length 20 mm, and flow rate corresponding to velocity v =

(0.03, 0, 0) m/s. To get more detailed information about the droplet breakup conditions

inside a spray nozzle, additional simulations are needed. The effect of different parameters

on the droplet breakup can be study in the future. Some of those parameters are:

1. Geometric parameters such as the nozzle diameter and length of the contraction

region.

2. Flow parameters, such as the inlet velocity or flow rate.

3. Material parameters, such as the viscosity of the continuous phase μc , interfacial

tension σ , infinity shear rate η∞ , and n in the Bird-Carreau model.

An additional improvement to the code can be realized by replacing the volume of fluid

(VOF) method with the coupled level set-volume fluid method (CLSVOF). The coupled

level set-volume fluid method uses (1) the VOF method to calculate the volume fraction

function α since it is mass conservative and (2) the level set method to calculate the

141
curvature κ since it is smoother. Also, the breakup conditions can be further studied

when the droplet exits the nozzle. Finally, the breakup conditions can be investigated

using 3D simulations to get more realistic results but this will require significantly more

computational time, even when using dynamic meshing. However, the information

obtained in our 2D simulations will serve to guide our choice of 3D simulations.

142
References

[1] A. Aserin, Multiple emulsions: technology and applications. Hoboken, New Jersery:
Wiley & Sons, Inc, 2007.

[2] K. Pays, J. Giermanska-Kahn, B. Pouligny, J. Bibette, and F. Leal-Calderon,

“Coalescence in surfactant-stabilized double emulsions,” Langmuir, vol. 17(25),
pp. 7758–7769, 2001.

[3] J. Sander, L. Isa, P. Rühs, P. Fischer, and A. Studart, “Stabilization mechanism of

double emulsions made by microfluidics,” Soft Matter, vol. 8, pp. 11471–11477,
2012.

[4] M. Ficheux, L. Bonakdar, F. Leal-Calderon, and J. Bibette, “Some stability criteria

for double emulsions,” Langmuir, vol. 14, pp. 2702–2706, 1998.

[5] S. Frasch-Melnik, F. Spyropoulos, and I. Norton, “W1/o/w2 double emulsions

stabilised by fat crystals - formulation, stability and salt release,” Journal of Colloid
and Interface Science, vol. 350(1), pp. 178–185, 2010.

[6] N. Garti, “Progress in stabilization and transport phenomena of double emulsions in

food applications,” Lebensmittel-Wissenschaft and Technologie, vol. 30, pp. 222–235,
1997.

[7] Y. B. Li, S. G. Zangh, and J. G. Li, “perimental and theoretical approaches on

uniform droplets formation from a rationed rotating membrane system,” Chemical
Engineering Science, vol. 66, pp. 788–796, 2011.

[8] R. Lutz, A. Aserin, L. Wicker, and N. Garti, “uble emulsions stabilized by a charged
complex of modified pectin and whey protein isolate,” Colloids and Surfaces B:
Biointerfaces, vol. 72, pp. 121–127, 2009.

[9] M. Aghbashlo, H. Mobli, A. Madadlou, and S. Rafiee, “The correlation of wall

material composition with flow characteristics and encapsulation behavior of fish oil
emulsion,” Food Research International, vol. 49, pp. 379–388, 2012.

143
[10] H. C. Carneiro, R. V. Tonon, C. R. Grosso, and M. D. Hubinger, “Encapsulation
efficiency and oxidative stability of flaxseed oil microencapsulated by spray drying
using different combinations of wall materials,” Journal of Food Engineering,
vol. 115, pp. 443–451, 2013.

[11] B. N. Dubey and E. J. Windhab, “Iron encapsulated microstructured emulsion particle

formation by prilling process and its release kinetics,” Journal of Food Engineering,
vol. 115(2), pp. 198–206, 2013.

[12] C. Tang and X. Li, “Microencapsulation properties of soy protein isolate and storage
stability of the correspondingly spray-dried emulsions,” Food Research International,
vol. 52, pp. 419–428, 2013.

[13] B. N. Dubey, M. R. Duxenneuner, , and E. J. Windhab, “Synthesis of functional food

powder of simple and multiple emulsions through prilling process,” Proceedings of
11th International Congress on Engineering and Food, 2011a.

[14] J. M. Ballester, N. Fueyo, and C. Dopazo, “Combustion characteristics of heavy

oil-water emulsions,” Fuel, vol. 75(6), pp. 695–705, 1996.

[15] C. D. Bolszo, A. A. Narvaez, V. G. McDonell, D. R. Dunn, and W. A. Sirignano,

“Pressure-swirl atomization of water-in-oil emulsions,” Atomization and Sprays,
vol. 20(12), pp. 1077–1099, 2010.

[16] L. Broniarz-Press, M. Ochowiak, J. Rozanski, and S. Woziwodzki, “The atomization

of water-oil emulsion,” Experimental Thermal and Fluid Science, vol. 33,
pp. 955–962, 2009.

[17] W. Kim, T. Yu, and W. Yoon, “Atomization characteristics of emulsified fuel oil
by instant emulsification,” Journal of Mechanical Science and Technology, vol. 26,
pp. 1781–1791, 2012.

[18] J. Schroeder, A. Kleinhans, Y. Serfert, S. Drusch, H. P. Schuchmann, and V. Gaukel,

“Viscosity ratio: A key factor for control of oil drop size distribution in effervescent
atomization of oil-in-water emulsions,” Journal of Food Engineering, vol. 111(2),
pp. 265–271, 2012.

[19] A. Tratnig, G. Brenn, T. Strixner, P. Fankhauser, N. Laubacher, and M. Stranzinger,

“Characterization of spray formation from emulsions by pressureswirl atomizers for
spray drying,” Journal of Food Engineering, vol. 95(1), pp. 126–134, 2009.

[20] Y. Sun and C. Beckermann, “Sharp interface tracking using the phase-field equation,”
Journal of Computational Physics, vol. 220(1), pp. 626–653, 2007.

[21] J. Sethian and P. Smereka, “Level set methods for fluid interface,” Annual Review of
Fluid Mechanics, vol. 35(1), pp. 341–372, 2003.

144
[22] S. Osher and R. Fedkiw, Level Set Methods and Dynamic Implicit Surfaces. New
York: Springer-Verlag New York, Inc., 2003.

[23] S. S. Deshpande, L. Anumolu, and M. F. Trujillo, “Evaluating the performance of

the two-phase flow solver interfoam,” Computational science and discovery, vol. 5,
014016, 2012.

[24] W. Rider and D. Kothe, “Reconstructing volume tracking,” comput. Phys., vol. 141,
p. 112, 1998.

[25] M. J. Berger and J. Oliger, “Adaptive mesh refinement for hyperbolic partial
differential equations,” Comp. Phy., vol. 53, pp. 484–512, 1984.

[26] D. J. Mavriplis and A. D. Gosman, “Adaptive meshing techniques for viscous flow
calculations on mixed element unstructured meshes,” Int. J. Num. Meth. Fluids,
vol. 34, pp. 93–111, 2000.

[27] D. J. Mavriplis, “Accurate multigrid solution of the euler equations on unstructured

and adaptive meshes,” AIAA J., vol. 28, pp. 213–221, 1990.

[28] S. Z. Pizadeh, “An adaptive unstructured grid method by grid subdivision, local
remeshing and grid movement,” AIAA, vol. 99, p. 3255, 1999.

[29] R. W. Anderson, R. B. Pember, and N. S. Elliott, “An arbitrary lagrangian-eulerian

method with adaptive mesh refinement for the solutions of the euler equations,” Comp.
Phys., vol. 199, pp. 86–617, 2004.

[30] W. J. Coirier, “An adaptively-refined, cartesian, cell-based scheme for the euler and
navier-stokes equations,” NASA Technical Memorandum, p. 106754, 1994.

[31] J. D. Hunt, An Adaptive 3D Cartesian Approach for the Parallel Computation of

Inviscid Flow About Static and Dynamic Configurations. PhD thesis, The University
of Michigan, 2004.

[32] Q. Xue, Development of adaptive mesh refinement scheme and conjugate heat transfer
model for engine simulations. PhD thesis, Iowa State University, 2009.

[33] OpenFOAM User Guide, 2.1.0 ed., December 2011.

[34] M. D. and M. R., Drops and bubbles in interfacial research. The Netherlands:
Elsevier, 1998.

[35] B. Lafaurie, C. Nardone, R. Scardovelli, S. Zaleski, and G. Zanetti, “Modelling

merging and fragmentation in multiphase flows with surfer,” Comput. Phys., vol. 113,
pp. 134–147, 1994.

145
[36] S. Popinet and S. Zaleski, “A front-tracking algorithm for accurate representation of
surface tension,” Int. J. Numer. Methods Fluids, vol. 30, pp. 775–793, 1999.

[37] L. X.-D., F. R.P., and K. M., “A boundary condition capturing method for
poissonâĂŹs equation on irregular domains,” Comput. Phys., vol. 160, pp. 151–178,
2000.

[38] K. M., F. R.P., and L. X.-D., “A boundary condition capturing method for multiphase
incompressible flow,” Sci. Comput., vol. 15, pp. 323–360, 2000.

[39] M. Meier, G. Yadigaroglu, and B. Smith, “A novel technique for including surface
tension in plic-vof methods,” Eur. J. Mech. B Fluids, vol. 21, pp. 61–73, 2002.

[40] M. Meier, Numerical and experimental study of large steam-air bubbles injected in a
water pool. PhD thesis, Swiss Federal Institute of Technology, 1999.

[41] J. U. Brackbill, D. B. Kothe, and C. Zemach, “A continuum method for modeling

surface tension,” Computational Physics, vol. 100, p. 335, 1992.

[42] H. Jasak, Error analysis and estimation in the Finite Volume method with appli-
cations to fluid flows. PhD thesis, Imperial College, University of London, 1996.

[43] J. D. Hoffman, Numerical Methods for Engineers and Scientists. New York:
McGrawHill, 1992.

[44] P. Wesselin, Principles of Computational Fluid Dynamics. Heidelberg: Springer,

2001.

[45] C. Rhie and W. Chow, “A numerical study of the turbulent Âřow past an isolated
airfoil with trailing edge separation,” in AIAA-82-0998, AIAA/ASME 3rd Joint
Thermophysics, Fluids, Plasma and Heat Transfer Conference, (St.Louis, Missouri),
1982.

[46] S. Patankar, Numerical heat transfer and fluid flow. Taylor & Francis, 1980.

[47] M. Peric, A finite volume method for the prediction of three-dimensional fluid flow in
complex ducts. PhD thesis, Imperial College London (University of London), 1985.

[48] R. I. Issa, “Solution of the implicitly discretised fluid flow equations by

operator-splitting,” Journal of Computational Physics, vol. 62, pp. 40–65, 1986.

[49] H. Rusche, Computational Fluid Dynamics of Dispersed Two-Phase Flows at High

Phase Fractions. PhD thesis, mperial College of Science, Technology and Medicine,
2002.

[50] S. M. Damián, An Extended Mixture Model for the Simultaneous Treatment of Short
and Long Scale Interfaces. PhD thesis, Universidad Nacional del Litoral, 2013.

146
[51] T. Young, “An essay on the cohesion of fluids,” Phil. Trans. R. Soc. Lond., vol. 65,
p. 95, 1805.
[52] B. Dubey, M. Duxxenneuner, C. Küchenmeister, P. Fischer, and E. Windhab,
“Influences of rheological behavior of emulsions on the spraying process,” in 24rd
Annual Conference on Liquid Atomization and Spray Systems, (Estoril, Portugal),
September 2011.
[53] B. Dubey, M. Duxenneuner, and E. Windhab, “Prilling process: an alternative for
atomization and producing solid particles of emulsions,” 23rd Annual Conference on
Liquid Atomization and Spray Systems, Brno, Czech Republic, 2010.
[54] J. Uddin, S. Decent, and M. Simmons, “The effect of surfactants on the instability of
a rotating liquid jet,” Fluid Dynammics Research, vol. 40, pp. 827–851, 2008.
[55] L. Broniarz-Press, M. Ochowiak, J. Rozanski, and S. Woziwodzki, “The atomization
of waterâĂŞoil emulsions,” Experimental Thermal and Fluid Science, vol. 33,
pp. 955–962, 2009.
[56] F. Tanner, S. Srinivasan, T. Althaus, K. Feigl, and E. Windhab, “Modeling and
validation of the crystallization process in food sprays,” 1th Triennial International
Conference on Liquid Atomization and Spray System, Vail, Colorado, USA, 2009.
[57] H. P. GRACE, “ispersion phenomena in high viscosity immiscible fluid systems
and application of static mixers as dispersion devices in such system,” Chemical
Engineering Communications, vol. 14, pp. 225–277, 1982.
[58] R. Barrett, M. Berry, T. F. Chan, J. Demmel, J. Donato, J. Dongarra, V. Eijkhout,
R. Pozo, C. Romine, and H. Van der Vorst, Templates for the solution of linear
systems: building blocks for iterative methods. No. 43, Society for Industrial and
Applied Mathematics, 1987.
[59] F. A. Morrison, Understanding Rheology. New York: Oxford University Press, Inc.,
2001.
[60] J.-D. Yu, S. Sakai, and J. A. Sethian, “A coupled level set projection method applied
to ink jet simulation,” Interface and Free Boundaries, vol. 5, p. 459, 2003.
[61] D. Wong, M. Simmons, S. Decent, E. Parau, and A. King, “Break-up dynamics and
drop size distributions created from spiralling liquid jets,” International Journal of
Multiphase Flow, vol. 30, pp. 499–520, 2004.
[62] R. B. Bird, R. C. Armstrong, and O. Hassager, Dynamics of Polymeric Liquids. New
York: John Wiley and Son Inc, 2nd ed., 1987.
[63] W. Shyy, Computational Modeling for Fluid Flow and Interfacial Transport.
Amsterdam: Elsevier, corrected ed., 1997.

147
[64] B. N. Datta, Numerical linear algebra and applications, vol. 116. Society for
Industrial and Applied Mathematics, 2010.
[65] I. Demirdžić and M. Perić, “Space conservation law in finite volume calculations
of fluid flow,” International journal for numerical methods in fluids, vol. 8, no. 9,
pp. 1037–1050, 1988.
[66] J. H. Ferziger and M. Perić, Computational methods for fluid dynamics, vol. 3.
Springer Berlin etc, 2001.
[67] L. Hogben, Handbook of linear algebra. Chapman & Hall, 2007.
[68] K. Hutter and K. Jöhnk, Continuum methods of physical modeling: continuum
mechanics, dimensional analysis, turbulence. Springer Verlag, 2004.
[69] H. Jasak, A. Jemcov, and J. Maruszewski, “Preconditioned linear solvers for large
eddy simulation,” in CFD 2007 Conference, CFD Society of Canada, 2007.
[70] H. Jasak and H. Rusche, “Dynamic mesh handling in openfoam,” in Proceeding of the
47th Aerospace Sciences Meeting Including The New Horizons Forum and Aerospace
Exposition, Orlando, Florida, 2009.
[71] H. Jasak and Z. Tuković, “Dynamic mesh handling in openfoam applied to
fluid-structure interaction simulations,” in Proceedings of the V European Conference
on Computational Fluid Dynamics (ECCOMAS CFD 2010)(Lisbon, Portugal, 14-17
June 2010), JCF Pereira AS, Pereira JMC,(Eds.).(27), 2010.
[72] C. Kleinstreuer, Biofluid dynamics: Principles and selected applications. CRC, 2006.
[73] A. Krishnamoorthy and D. Menon, “Matrix inversion using cholesky decomposition,”
arXiv preprint arXiv:1111.4144, 2011.
[74] T. Lucchini, “Running openfoam tutorials.”
[75] E. W. Merrill, “Rheology of blood,” Physiol Rev, vol. 49, no. 4, pp. 863–88, 1969.
[76] S. V. Patankar and D. B. Spalding, “A calculation procedure for heat, mass and
momentum transfer in three-dimensional parabolic flows,” International Journal of
Heat and Mass Transfer, vol. 15, no. 10, pp. 1787–1806, 1972.
[77] Y. Saad, Iterative methods for sparse linear systems. Society for Industrial and
Applied Mathematics, 2003.
[78] J. F. Steffe, Rheological methods in food process engineering. Freeman Press, 1996.
[79] M. Stranzinger, Numerical and Experimental Investigations of Newtonian and
Non-Newtonian Flow in Annular Gaps with Scraper Blades. PhD thesis, Swiss
Federal Institute of Technology (ETH), 1999.

148
[80] Y. Takeda, “Velocity profile measurement by ultrasound doppler shift method,”
International journal of heat and fluid flow, vol. 7, no. 4, pp. 313–318, 1986.

[81] L. N. Trefethen and D. Bau III, Numerical linear algebra. No. 50, Society for
Industrial and Applied Mathematics, 1997.

[82] H. K. Versteeg and W. Malalasekera, An Introduction to Computational Fluid

Dynamics. USA: Pearson Education Limited, 2nd ed., 2007.

149
Appendix A

interFoam and interDyMFoam solvers

• The source code for the interFoam solver is

/*--------------------------------------------*\
========= |
\\ / F ield | OpenFOAM: The Open
\\ / O peration | Source CFD Toolbox
\\ / A nd | Copyright (C) 2011
\\/ M anipulation | OpenFOAM Foundation
---------------------------------------------------
License
This file is part of OpenFOAM.

OpenFOAM is free software: you can redistribute it

and/or modify it under the terms of the GNU General
Public License as published by the Free Software
Foundation, either version 3 of the License, or
(at your option) any later version.

OpenFOAM is distributed in the hope that it will be

useful, but WITHOUT ANY WARRANTY; without even the
implied warranty of MERCHANTABILITY or FITNESS FOR
A PARTICULAR PURPOSE. See the GNU General Public
License for more details.

You should have received a copy of the GNU General

Public License along with OpenFOAM. If not,
see <http://www.gnu.org/licenses/>.

151
Application
interFoam

Description
Solver for 2 incompressible, isothermal immiscible
fluids using a VOF (volume of fluid) phase-fraction
based interface capturing approach.

The momentum and other fluid properties are of the

"mixture" and a single momentum equation is solved.

Turbulence modelling is generic, i.e. laminar, RAS

or LES may be selected.

For a two-fluid approach see twoPhaseEulerFoam.

\*---------------------------------

#include "fvCFD.H"
#include "MULES.H"
#include "subCycle.H"
#include "interfaceProperties.H"
#include "twoPhaseMixture.H"
#include "turbulenceModel.H"
#include "interpolationTable.H"
#include "pimpleControl.H"
#include "ker.H"

// * * * * * * * * * * * * * * * *

int main(int argc, char *argv[])

{
#include "setRootCase.H"
#include "createTime.H"
#include "createMesh.H"

pimpleControl pimple(mesh);

#include "initContinuityErrs.H"
#include "createFields.H"
#include "readTimeControls.H"
#include "correctPhi.H"

152
 #include "CourantNo.H"
#include "setInitialDeltaT.H"

// * * * * * * * * * * * * * * *

Info<< "\nStarting time loop\n" << endl;

while (runTime.run())
{
#include "readTimeControls.H"
#include "CourantNo.H"
#include "alphaCourantNo.H"
#include "setDeltaT.H"

runTime++;

Info<< "Time = " << runTime.timeName()

<< nl << endl;

twoPhaseProperties.correct();

#include "alphaEqnSubCycle.H"

while (pimple.loop())
{
#include "UEqn.H"

// --- Pressure corrector loop

while (pimple.correct())
{
#include "pEqn.H"
}

if (pimple.turbCorr())
{
turbulence->correct();
}
}

runTime.write();

153
 Info<< "ExecutionTime = " <<
runTime.elapsedCpuTime() << " s"
<< " ClockTime = " <<
runTime.elapsedClockTime() << " s"
<< nl << endl;
}

Info<< "End\n" << endl;

return 0;
}

// ******************************************//

• The source code of interDyMFoam solver is

/*--------------------------------------------------*\
========= |
\\ / F ield | OpenFOAM: The Open
\\ / | Source CFD Toolbox
\\ / O peration |
\\ / A nd | Copyright (C) 2011
\\/ M anipulation | OpenFOAM Foundation
------------------------------------------------------
License
This file is part of OpenFOAM.

OpenFOAM is free software: you can redistribute it

and/or modify it under the terms of the GNU General
Public License as published by the Free Software
Foundation, either version 3 of the License, or
(at your option) any later version.

OpenFOAM is distributed in the hope that it will be

useful, but WITHOUT ANY WARRANTY; without even the
implied warranty of MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE. See the GNU General Public
License for more details.

You should have received a copy of the GNU General

Public License along with OpenFOAM. If not,
see <http://www.gnu.org/licenses/>.

154
Application
interDyMFoam

Description
Solver for 2 incompressible, isothermal immiscible
fluids using a VOF (volume of fluid) phase-fraction
based interface capturing approach, with optional
mesh motion and mesh topology changes including
adaptive re-meshing.

\*----------------------------------*/

#include "fvCFD.H"
#include "dynamicFvMesh.H"
#include "MULES.H"
#include "subCycle.H"
#include "interfaceProperties.H"
#include "twoPhaseMixture.H"
#include "turbulenceModel.H"
#include "pimpleControl.H"

// * * * * * * * * * * * * * * * * * *//

int main(int argc, char *argv[])

{
#include "setRootCase.H"
#include "createTime.H"
#include "createDynamicFvMesh.H"
#include "initContinuityErrs.H"
#include "createFields.H"
#include "readTimeControls.H"

pimpleControl pimple(mesh);

surfaceScalarField phiAbs("phiAbs", phi);

fvc::makeAbsolute(phiAbs, U);

#include "correctPhi.H"
#include "CourantNo.H"
#include "setInitialDeltaT.H"
// * * * * * * * * * * * * * * * * * * * //
Info<< "\nStarting time loop\n" << endl;

155
while (runTime.run())
{
#include "readControls.H"
#include "alphaCourantNo.H"
#include "CourantNo.H"

#include "setDeltaT.H"

runTime++;

Info<< "Time = " << runTime.timeName()

<< nl << endl;

scalar timeBeforeMeshUpdate =
runTime.elapsedCpuTime();

{
volVectorField Urel("Urel", U);

if (mesh.moving())
{
Urel -= fvc::reconstruct(fvc::meshPhi(U));
}

// Do any mesh changes

mesh.update();
}

if (mesh.changing())
{
Info<< "Execution time for mesh.update() = "
<< runTime.elapsedCpuTime() -
timeBeforeMeshUpdate
<< " s" << endl;

gh = g & mesh.C();
ghf = g & mesh.Cf();
}

if (mesh.changing() && correctPhi)

{
#include "correctPhi.H"

156
 }

if (mesh.changing() && checkMeshCourantNo)

{
#include "meshCourantNo.H"
}

twoPhaseProperties.correct();

#include "alphaEqnSubCycle.H"

// --- Pressure-velocity PIMPLE corrector loop

while (pimple.loop())
{
#include "UEqn.H"

// --- Pressure corrector loop

while (pimple.correct())
{
#include "pEqn.H"
}

if (pimple.turbCorr())
{
turbulence->correct();
}
}

runTime.write();

Info<< "ExecutionTime = " <<

runTime.elapsedCpuTime() << " s"
<< " ClockTime = " <<
runTime.elapsedClockTime() << " s"
<< nl << endl;
}

Info<< "End\n" << endl;

return 0;
}

157
// ******************************** //

158
Appendix B

Modifications to interDyMFoam

In this section, if a new line is added to the code, then the initial “AB” is added to the right
of the line. If multiple consecutive lines are added, then the initial “AB....” is added to the
first line and “AB” to the last line.

• Modifications to dynamicRefineFvMesh library.

1. Create a new library called dynamicRefineFvMesh2D in

src/dynamicFvMesh library by using the command
cp -r dynamicRefineFvMesh dynamicRefineFvMesh2D
2. Change the name in files from dynamicRefineFvMesh to
dynamicRefineFvMesh2D by using the command
“sed -i s/dynamicRefineFvMesh/dynamicRefineFvMesh2D/g file name”
where the file name can be dynamicRefineFvMesh2D.C and
dynamicRefineFvMesh2D.H
3. Add dynamicRefineFvMesh2D/dynamicRefineFvMesh2D.C to the
file src/dynamicFvMesh/Make/files
4. To extend the number of buffer layers for refinement, part of the following is
added to the dynamicRefineFvMesh2D.C source code;

const label nBufferLayersR =

readLabel(refineDict.lookup("nBufferLayersR")); //AB
PackedBoolList refineCell(nCells());

if (globalData().nTotalCells() < maxCells)

{

159
 selectRefineCandidates
(
lowerRefineLevel,
upperRefineLevel,
vFld,
refineCell
);
for (label i = 0; i < nBufferLayersR; i++) //AB ....
{
extendMarkedCells(refineCell)
} //AB

• Modifications to dynamicMesh library.

The modification to this library is done by modifying the
hexRef8.C source code file which is located in the directory
src/dynamicMesh/polyTopoChange/polyTopoChange.

1. Create new files, hexRef82D.C and hexRef82D.H by using the command

cp hexRef8.C hexRef82D.C
cp hexRef8.H hexRef82D.H
2. Change the name in files from hexRef8 to hexRef82D
3. Add polyTopoChange/polyTopoChange/hexRef82D.C to the file
src/dynamicMesh/Make/files
4. The lines added to the source code hexRef82D.C are given below:
(a) Give the cells that are chosen for refinement a number greater than zero (1
is used here)
labelList cellMidPoint(mesh_.nCells(), -1);

forAll(cellLabels, i)

{
label cellI = cellLabels[i];
cellMidPoint[cellI] = 1; //AB
}
(b) The empty faces and the edges on the empty faces are the only faces and
edges visible for partitioning
for (label faceI = mesh_.nInternalFaces();
faceI < mesh_.nFaces(); faceI++) //AB....
{
const label & patchID =

160
 mesh_.boundaryMesh().whichPatch(faceI);
if (isA<emptyPolyPatch>(mesh_.
boundaryMesh()[patchID]))
{
isDivisibleFace[faceI] = true;
const labelList& fEdges = mesh_.faceEdges(faceI);
forAll(fEdges, i)
{
label edgeJ = fEdges[i];

isDivisibleEdge[edgeJ] = true;

}
}
} //AB
(c) Give the visible edges a number (1234 is used here)
forAll(cellMidPoint, cellI)

if (cellMidPoint[cellI] >= 0)

{
const labelList& cEdges = mesh_.cellEdges(cellI);

forAll(cEdges, i)

{
label edgeI = cEdges[i];

const edge& e = mesh_.edges()[edgeI];

if
(
pointLevel_[e[0]] <= cellLevel_[cellI]
&& pointLevel_[e[1]] <= cellLevel_[cellI]
&& isDivisibleEdge[edgeI] //AB
)
{
edgeMidPoint[edgeI] = 12345;
}

161
 }

}
(d) Add a point in the middle of the visible faces
forAll(faceMidPoint, faceI)
{
if (faceMidPoint[faceI] >= 0
&& isDivisibleFace[faceI]) //AB
{
const face& f = mesh_.faces()[faceI];

faceMidPoint[faceI] = meshMod.setAction

(
polyAddPoint
(
(
faceI < mesh_.nInternalFaces()

? mesh_.faceCentres()[faceI]

: bFaceMids[faceI-mesh_.nInternalFaces()]

), // point
f[0], // master point
-1, // zone for point
true // supports a cell

);
// Determine the level of the corner points
// and midpoint will be one higher.

newPointLevel(faceMidPoint[faceI])
= faceAnchorLevel[faceI]+1;

}
}
(e) The corner points
labelListList cellAnchorPoints(mesh_.nCells());
{

162
labelList nAnchorPoints(mesh_.nCells(), 0);

forAll(cellMidPoint, cellI)
{
if (cellMidPoint[cellI] >= 0)
{
cellAnchorPoints[cellI].setSize(8);

}
}

forAll(cellMidPoint, cellI) //AB....

{
const cell& cFaces = mesh_.cells()[cellI];

forAll(cFaces, i)
{
label faceI = cFaces[i];

const face& f = mesh_.faces()[faceI];

forAll(f, fp)

{
label pointI = f[fp];

if
(
isDivisibleFace[faceI] //AB

&& cellMidPoint[cellI] >= 0

&& pointLevel_[pointI] <= cellLevel_[cellI]

)
{
if (nAnchorPoints[cellI] == 8)

{
dumpCell(cellI);

163
 FatalErrorIn

(
"hexRef82D::setRefinement(const labelList&"

", polyTopoChange&)"

) << "cell " << cellI

<< " of level " << cellLevel_[cellI]

<< " uses more than 8 points of equal or"

<< " lower level" << nl

<< "Points so far:" << cellAnchorPoints[cellI]

<< abort(FatalError);

cellAnchorPoints[cellI][nAnchorPoints[cellI]++]
= pointI;

}
}
(f) Split the empty faces to four new faces and the other faces to two new faces
forAll(faceMidPoint, faceI)
{
if (faceMidPoint[faceI] >= 0
&& affectedFace.get(faceI))
{
// Face needs to be split and hasn’t yet been
//done in some way (affectedFace - is impossible
//since this is first change but just for
//completeness)

const face& f = mesh_.faces()[faceI];

// Has original faceI been used (three faces

//added, original gets modified)

164
bool modifiedFace = false;
label anchorLevel = faceAnchorLevel[faceI];

if (isDivisibleFace[faceI]) //AB

{
face newFace(4);
forAll(f, fp)

{
label pointI = f[fp];

if (pointLevel_[pointI] <= anchorLevel)

{
// point is anchor. Start collecting face.

DynamicList<label> faceVerts(4);

faceVerts.append(pointI);

// Walk forward to mid point

// - if next is +2 midpoint is +1
// - if next is +1 it is midpoint
// - if next is +0 there has
// to be edgeMidPoint

walkFaceToMid
(
edgeMidPoint,
anchorLevel,
faceI,
fp,
faceVerts
);

faceVerts.append(faceMidPoint[faceI]);

walkFaceFromMid
(
edgeMidPoint,
anchorLevel,

165
 faceI,
fp,
faceVerts
);

// Convert dynamiclist to face.

newFace.transfer(faceVerts);

// Get new owner/neighbour

label own, nei;

getFaceNeighbours
(
cellAnchorPoints,
cellAddedCells,
faceI,
pointI, // Anchor point
own,
nei
);

if (debug)
{
if (mesh_.isInternalFace(faceI))
{
label oldOwn = mesh_.faceOwner()[faceI];
label oldNei = mesh_.faceNeighbour()[faceI];
checkInternalOrientation
(
meshMod,
oldOwn,
faceI,
mesh_.cellCentres()[oldOwn],
mesh_.cellCentres()[oldNei],
newFace
);

166
else
{
label oldOwn = mesh_.faceOwner()[faceI];
checkBoundaryOrientation
(
meshMod,
oldOwn,
faceI,
mesh_.cellCentres()[oldOwn],
mesh_.faceCentres()[faceI],
newFace
);
}
}

if (!modifiedFace)
{
modifiedFace = true;
modFace(meshMod, faceI, newFace, own, nei);
}

else
{
addFace(meshMod, faceI, newFace, own, nei);
}
}
}
}

else //AB....

{
face newFace(2);

forAll(f,fp)
{
label pointI = f[fp];

label nextpointI = f[f.fcIndex(fp)];

label edgeI = meshTools::findEdge

(mesh_, pointI, nextpointI);

167
if (edgeMidPoint[edgeI] >=0)

{
DynamicList<label> faceVerts(4);

label pointJ = f[f.rcIndex(fp)];

faceVerts.append(pointI);

walkFaceToMid
(
edgeMidPoint,
anchorLevel,
faceI,
fp,
faceVerts
);

walkFaceFromMid
(
edgeMidPoint,
anchorLevel,
faceI,
f.rcIndex(fp),
faceVerts
);

faceVerts.append(pointJ);

newFace.transfer(faceVerts);

label own, nei;

getFaceNeighbours
(
cellAnchorPoints,
cellAddedCells,
faceI,
pointI,
own,
nei
);

168
if (debug)
{
if (mesh_.isInternalFace(faceI))
{
label oldOwn = mesh_.faceOwner()[faceI];
label oldNei = mesh_.faceNeighbour()[faceI];

checkInternalOrientation
(
meshMod,
oldOwn,
faceI,
mesh_.cellCentres()[oldOwn],
mesh_.cellCentres()[oldNei],
newFace
);
}

else
{
label oldOwn = mesh_.faceOwner()[faceI];

checkBoundaryOrientation
(
meshMod,
oldOwn,
faceI,
mesh_.cellCentres()[oldOwn],
mesh_.faceCentres()[faceI],
newFace
);
}
}

if (!modifiedFace)

{
modifiedFace = true;

modFace(meshMod, faceI, newFace, own, nei);

169
 }

else
{
addFace(meshMod, faceI, newFace, own, nei);
}
}
}
} //AB

// Mark face as having been handled

affectedFace.unset(faceI);

}
(g) Get the anchor cell in 2D case where the corresponding vertices on the
empty faces have the same anchor cell
Foam::label Foam::hexRef82D::getAnchorCell
(
const labelListList& cellAnchorPoints,
const labelListList& cellAddedCells,
const label cellI,
const label faceI,
const label pointI
) const
{
if (cellAnchorPoints[cellI].size())
{
label index = findIndex(cellAnchorPoints
[cellI], pointI);
if (index != -1)
{
if (index >= 4) //AB....
{
if (index == 4)
{
index = 8;
}
index = 8 - index;
} //AB

170
return cellAddedCells[cellI][ index];
}
// pointI is not an anchor cell.
// Maybe we are already a refined face so
// check all the face vertices.

const face& f = mesh_.faces()[faceI];

forAll(f, fp)
{
label index = findIndex(cellAnchorPoints
[cellI], f[fp]);

if (index != -1)
{
if (index >= 4) //AB....
{
if (index == 4)
{
index = 8;
}
index = 8 - index;
} //AB
return cellAddedCells[cellI][index];
}
}

// Problem.

dumpCell(cellI);
Perr<< "cell:" << cellI << " anchorPoints:"
<< cellAnchorPoints[cellI] << endl;

FatalErrorIn("hexRef82D::getAnchorCell(..)")
<< "Could not find point " << pointI
<< " in the anchorPoints for cell "
<< cellI << end
<< "Does your original mesh obey the 2:1
constraint and"
<< " did you use consistentRefinement to
make your cells to refine"
<< " obey this constraint as well?"
<< abort(FatalError);

171
 return -1;
}
else
{
return cellI;
}
}
(h) Add internal faces
if (faceMidFnd == midPointToFaceMids.end())
{
midPointToFaceMids.insert(edgeMidPointI,
edge(faceMidPointI, -1));
}

else
{
edge& e = faceMidFnd();

if (faceMidPointI != e[0])
{
if (e[1] == -1)
{
e[1] = faceMidPointI;

changed = true;
}
}
if (e[0] != -1 && e[1] != -1)

{
haveTwoFaceMids = true;
}

// Pout << "face edge " << e << endl;

// Check if this call of storeMidPointInfo

// is the one that completed all the
// nessecary information.

if (changed && haveTwoAnchors) //AB....

172
 {
const cell& cFaces = mesh_.cells()[cellI];

label face1 = -1;

forAll(cFaces, i)
{
label faceJ = cFaces[i];

if (cellMidPoint[faceJ] != faceMidPointI

&& cellMidPoint[faceJ] >= 0

&& cellMidPoint[faceJ] != 123456789 )

{
face1 = faceJ;
}
}

const edge& anchors = midPointToAnchors

[edgeMidPointI];

label index = findIndex(cellAnchorPoints

[cellI], anchorPointJ);

if (findIndex(cellAnchorPoints[cellI],
anchorPointJ) == 0)
{
index = 4;
}

if (findIndex(cellAnchorPoints[cellI],
anchorPointJ) == 4)
{
index = 8;
}

label point1 = cellAnchorPoints

[cellI][8 - index];

label edgeMidPointJ = -1;

173
const face& f = mesh_.faces()[face1];

const labelList& fEdges =

mesh_.faceEdges(face1);

DynamicList<label> newFaceVerts(4);

if (faceOrder == (mesh_.faceOwner()[faceI]
== cellI))

{
label anch = findIndex(f, point1);
if (pointLevel_[f[f.rcIndex(anch)]]
<= cellLevel_[cellI])

{
label edgeJ = fEdges[f.rcIndex(anch)];

edgeMidPointJ = edgeMidPoint[edgeJ];

else

{
label edgeMid = findLevel(face1, f, f.rcIn
dex(anch), false, cellLevel_[cellI] +1);

edgeMidPointJ = f[edgeMid];

} //AB

newFaceVerts.append(faceMidPointI);

// Check & insert edge split if any

insertEdgeSplit

(
edgeMidPoint,
faceMidPointI, // edge between faceMid
edgeMidPointI, // and edgeMid

174
 newFaceVerts
);

newFaceVerts.append(edgeMidPointI);

insertEdgeSplit
(
edgeMidPoint,
edgeMidPointI,
edgeMidPointJ, //AB
newFaceVerts
);

newFaceVerts.append(edgeMidPointJ);
newFaceVerts.append(cellMidPoint[face1]);

else
{
label anch = findIndex(f, point1);

if (pointLevel_[f[f.fcIndex(anch)]]
<= cellLevel_[cellI])
{
label edgeJ = fEdges[anch];
edgeMidPointJ = edgeMidPoint[edgeJ];
}

else
{
label edgeMid = findLevel(face1, f, f.fcIn
dex(anch), true, cellLevel_[cellI] + 1);

edgeMidPointJ = f[edgeMid];

newFaceVerts.append(edgeMidPointJ);

insertEdgeSplit
(

175
 edgeMidPoint,
edgeMidPointJ, //AB
edgeMidPointI,
newFaceVerts
);

newFaceVerts.append(edgeMidPointI);

insertEdgeSplit
(
edgeMidPoint,
edgeMidPointI,
faceMidPointI,
newFaceVerts
);

newFaceVerts.append(faceMidPointI);

newFaceVerts.append(cellMidPoint[face1]);

face newFace;

newFace.transfer(newFaceVerts);
(i) get the points that can be unrefined
Foam::labelList Foam::hexRef82D::
getSplitPoints(
const label axis, const scalar axisVal)//AB
const
{
if (debug)

checkRefinementLevels(-1, labelList(0));

if (debug)

176
{

Pout<< "hexRef82D::getSplitPoints :"

<< " Calculating unrefineable points"

<< endl;

if (!history_.active())

FatalErrorIn("hexRef82D::
getSplitPoints()")<< "Only call if
constructed with history capability"

<< abort(FatalError);

// Master cell

// -1 undetermined

// -2 certainly not split point

// >= label of master cell

labelList splitMaster(mesh_.nPoints(), -1);

labelList splitMasterLevel(mesh_.nPoints()
, 0);

// Unmark all with not 8 cells

for (label pointI = 0; pointI <

mesh_.nPoints(); pointI++)

177
 {
const labelList& pCells =
mesh_.pointCells(pointI);

vector coord = mesh_.points()[pointI];

if (pCells.size()!=4||
coord[axis]>axisVal) //AB
{
splitMaster[pointI] = -2;

// Unmark all with different master cells

const labelList& visibleCells =

history_.visibleCells();

forAll(visibleCells, cellI)

const labelList& cPoints =

mesh_.cellPoints(cellI);

if (visibleCells[cellI] != -1 &&
history_.parentIndex(cellI) >= 0)

label parentIndex =
history_.parentIndex(cellI);

178
 // Check same master.

forAll(cPoints, i)

label pointI = cPoints[i];

label masterCellI = splitMaster[pointI];

if (masterCellI == -1)

splitMaster[pointI] = parentIndex;

splitMasterLevel[pointI] =
cellLevel_[cellI] - 1;

else if (masterCellI == -2)

else if

(masterCellI != parentIndex)

|| (splitMasterLevel[pointI] !=
cellLevel_[cellI] - 1)

179
 {

splitMaster[pointI] = -2;

else

forAll(cPoints, i)

label pointI = cPoints[i];

splitMaster[pointI] = -2;

After the modifications are done, run wmake libso to compile the dynamicMesh and
dynamicFvMesh libraries.
The modifications for axisymmetric geometry will be the same but by replacing
dynamicFvMesh2D with dynamicFvMeshAxi and hexRef82D with
hexRef8axi. The cells on the centerline should have special treatment as follows;

1. Split the empty faces to four new faces and the other faces to two new faces

forAll(faceMidPoint, faceI)

180
{
if (faceMidPoint[faceI] >= 0 &&
affectedFace.get(faceI))
{
const face& f = mesh_.faces()[faceI];

bool modifiedFace = false;

label anchorLevel = faceAnchorLevel[faceI];

if (isDivisibleFace[faceI]) //AB
{

face newFace(4);

forAll(f, fp)
{
label pointI = f[fp];

if (pointLevel_[pointI] <= anchorLevel)

{
// point is anchor. Start collecting face.

DynamicList<label> faceVerts(4);

faceVerts.append(pointI);

// Walk forward to mid point.

// - if next is +2 midpoint is +1
// - if next is +1 it is midpoint
// - if next is +0 there has to be
// edgeMidPoint

walkFaceToMid
(
edgeMidPoint,
anchorLevel,
faceI,
fp,
faceVerts
);

faceVerts.append(faceMidPoint[faceI]);

181
walkFaceFromMid
(
edgeMidPoint,
anchorLevel,
faceI,
fp,
faceVerts
);

// Convert dynamiclist to face.

newFace.transfer(faceVerts);

label own, nei;

getFaceNeighbours
(
cellAnchorPoints,
cellAddedCells,
faceI,
pointI, // Anchor point
own,
nei
);

if (debug)
{

if (mesh_.isInternalFace(faceI))
{
label oldOwn = mesh_.faceOwner()[faceI];

label oldNei = mesh_.faceNeighbour()[faceI];

checkInternalOrientation
(
meshMod,
oldOwn,
faceI,
mesh_.cellCentres()[oldOwn],
mesh_.cellCentres()[oldNei],
newFace

182
 );
}

else
{
label oldOwn = mesh_.faceOwner()[faceI];

checkBoundaryOrientation
(
meshMod,
oldOwn,
faceI,
mesh_.cellCentres()[oldOwn],
mesh_.faceCentres()[faceI],
newFace
);
}
}

if (!modifiedFace)
{
modifiedFace = true;

modFace(meshMod, faceI, newFace, own, nei);

}

else
{
addFace(meshMod, faceI, newFace, own, nei);
}
}
}
}

else //AB....
{
face newFace(2);

forAll(f,fp)
{
label pointI = f[fp];
label nextpointI = f[f.fcIndex(fp)];

183
label edgeI = meshTools::findEdge(mesh_,
pointI, nextpointI);
if (edgeMidPoint[edgeI] >=0)
{
label pointJ = f[f.rcIndex(fp)];
label prevpointJ = f[f.rcIndex
(f.rcIndex(fp))];
label edgep = meshTools::findEdge(mesh_,
pointI, pointJ);

if (edgeMidPoint[edgep] >=0)
{
DynamicList<label> faceVerts(3);
faceVerts.append(pointI);

walkFaceToMid
(
edgeMidPoint,
anchorLevel,
faceI,
fp,
faceVerts
);

walkFaceFromMid
(
edgeMidPoint,
anchorLevel,
faceI,
fp,
faceVerts
);

newFace.transfer(faceVerts);
}

else
{
DynamicList<label> faceVerts(4);

faceVerts.append(pointI);

184
walkFaceToMid
(
edgeMidPoint,
anchorLevel,
faceI,
fp,
faceVerts
);

walkFaceFromMid
(
edgeMidPoint,
anchorLevel,
faceI,
f.rcIndex(fp),
faceVerts
);

faceVerts.append(pointJ);

newFace.transfer(faceVerts);
}

label own, nei;

getFaceNeighbours
(
cellAnchorPoints,
cellAddedCells,
faceI,
pointI,
own,
nei
);

if (debug)
{
if (mesh_.isInternalFace(faceI))
{
label oldOwn = mesh_.faceOwner()[faceI];

label oldNei = mesh_.faceNeighbour()[faceI];

185
checkInternalOrientation
(
meshMod,
oldOwn,
faceI,
mesh_.cellCentres()[oldOwn],
mesh_.cellCentres()[oldNei],
newFace
);
}

else
{
label oldOwn = mesh_.faceOwner()[faceI];

checkBoundaryOrientation
(
meshMod,
oldOwn,
faceI,
mesh_.cellCentres()[oldOwn],
mesh_.faceCentres()[faceI],
newFace
);
}
}

if (!modifiedFace)
{
modifiedFace = true;

modFace(meshMod, faceI, newFace, own, nei);

}

else
{
addFace(meshMod, faceI, newFace, own, nei);
}
}
}
} //AB

186
 affectedFace.unset(faceI);
}
}

2. Add internal faces

if (changed && haveTwoAnchors) //AB....

{
const cell& cFaces = mesh_.cells()[cellI];

label face1 = -1;

forAll(cFaces, i)
{
label faceJ = cFaces[i];

if (cellMidPoint[faceJ] != faceMidPointI

&& cellMidPoint[faceJ] >= 0

&& cellMidPoint[faceJ] != 123456789 )

{
face1 = faceJ;
}
}

const edge& anchors = midPointToAnchors

[edgeMidPointI];

label index = findIndex(cellAnchorPoints

[cellI], anchorPointJ);

if (findIndex(cellAnchorPoints[cellI],
anchorPointJ) == 0)
{
index = 4;
}

if (findIndex(cellAnchorPoints[cellI],
anchorPointJ) == 4)
{
index = 8;
}

187
label point1 = cellAnchorPoints[cellI]
[8 - index];

label edgeMidPointJ = -1;

const face& f = mesh_.faces()[face1];

const labelList& fEdges = mesh_.

faceEdges(face1);

DynamicList<label> newFaceVerts(4);

if (faceOrder == (mesh_.faceOwner()[faceI]
== cellI))
{
label anch = findIndex(f, point1);

if (pointLevel_[f[f.rcIndex(anch)]]
<= cellLevel_[cellI])
{
label edgeJ = fEdges[f.rcIndex(anch)];
edgeMidPointJ = edgeMidPoint[edgeJ];
}

else
{
label edgeMid = findLevel(face1, f, f.rcIn
dex(anch), false, cellLevel_[cellI] +1);

edgeMidPointJ = f[edgeMid];
} //AB

newFaceVerts.append(faceMidPointI);

insertEdgeSplit
(
edgeMidPoint,
faceMidPointI, // edge between faceMid
edgeMidPointI, // and edgeMid
newFaceVerts
);

188
newFaceVerts.append(edgeMidPointI);

if (edgeMidPointI!=edgeMidPointJ)
{
insertEdgeSplit
(
edgeMidPoint,
edgeMidPointI,
edgeMidPointJ, //AB
newFaceVerts
);

newFaceVerts.append(edgeMidPointJ);
}

newFaceVerts.append(cellMidPoint[face1]);
}

else
{
label anch = findIndex(f, point1);

if (pointLevel_[f[f.fcIndex(anch)]]
<= cellLevel_[cellI])
{
label edgeJ = fEdges[anch];

edgeMidPointJ = edgeMidPoint[edgeJ];
}

else
{
label edgeMid = findLevel(face1, f, f.fcIn
dex(anch), true, cellLevel_[cellI] + 1);

edgeMidPointJ = f[edgeMid]; //AB

}

if (edgeMidPointI!=edgeMidPointJ)
{
newFaceVerts.append(edgeMidPointJ);

189
insertEdgeSplit
(
edgeMidPoint,
edgeMidPointJ, //AB
edgeMidPointI,
newFaceVerts
);
}

newFaceVerts.append(edgeMidPointI);

insertEdgeSplit
(
edgeMidPoint,
edgeMidPointI,
faceMidPointI,
newFaceVerts
);

newFaceVerts.append(faceMidPointI);

newFaceVerts.append(cellMidPoint[face1]);
}

face newFace;

newFace.transfer(newFaceVerts);

190
Appendix C

nozzle

The actual values in the nozzle simulations for both fluids

Table C.1
Critical breakup radius, capillary number, viscosity ratio, and the breakup
position in the nozzle for the non-Newtonian fluid

viscosity break up
y MaxRef drop radius(mm) ca
ratio position (mm)
0.75 3,4 0.2950-0.3300 1.224-1.37 0.623 15-23
1.00 3,4 0.1460-0.1810 0.808-1.00 0.649 20-26
1.50 5,6 0.0155-0.0408 0.129-0.34 0.686 17-20
1.75 5,6 0.0107-0.0313 0.104-0.31 0.700 12-14
2.00 5,6 0.0085-0.0170 0.094-0.19 0.709 08-10
2.25 5,6 0.0057-0.0159 0.071-0.20 0.718 07-09
2.50 5,6 0.0057-0.0159 0.079-0.22 0.725 05-07
2.70 5,6 0.0063-0.0159 0.094-0.24 0.729 05-07

191
 Table C.2
Critical breakup radius, capillary number, viscosity ratio, and the breakup
position in the nozzle for the Newtonian fluid

viscosity break up
y MaxRef drop radius (mm) ca
ratio position (mm)
0.75 3,4 0.3570-0.3990 1.851-2.07 0.56 24-26
1.00 3,4 0.1876-0.1960 1.300-1.36 0.56 20-25
1.50 5,6 0.0190-0.0300 0.198-0.31 0.56 09-11
2.00 5,6 0.0083-0.0146 0.116-0.21 0.56 09-11
2.25 5,6 0.0059-0.0110 0.088-0.17 0.56 06-08
2.50 5,6 0.0033-0.0100 0.060-0.17 0.56 05-07
2.70 5,6 0.0065-0.0135 0.123-0.25 0.56 05-07

Table C.3
Critical breakup radius, capillary number, viscosity ratio, and the breakup
position in the nozzle for the Newtonian fluid at stream line y = 1

viscosity break up
y MaxRef drop radius (mm) ca
ratio position (mm)
1.00 3,4 0.3543-0.4120 2.454-2.85 0.25 12-17
1.00 3,4 0.2446-0.2811 1.694-1.95 0.4 16-22
1.00 3,4 0.1876-0.1960 1.300-1.36 0.56 20-25
1.00 3,4 0.1640-0.1875 1.136-1.30 0.725 20-25
1.00 3,4 0.1640-0.1875 1.136-1.30 1 16-22
1.00 3,4 0.3345-0.3780 2.317-2.62 2 22-25.5

192
 Table C.4
Critical breakup radius, capillary number, viscosity ratio, and the breakup
position in the nozzle for the Newtonian fluid at stream line y = 2

viscosity break up
y MaxRef drop radius(mm) ca
ratio position (mm)
2.00 5,6 0.0145-0.0295 0.202-0.41 0.25 09-11
2.00 5,6 0.0083-0.0146 0.116-0.21 0.56 09-11
2.00 5,6 0.0083-0.0146 0.116-0.21 0.725 09-11
2.00 5,6 0.0083-0.0146 0.116-0.21 1 11-13
2.00 5,6 0.0141-0.0294 0.197-0.41 2 09-11

193