A NUMERICAL AND EXPERIMENTAL INVESTIGATION OF TAYLOR FLOW
INSTABILITIES IN NARROW GAPS AND THEIR RELATIONSHIP TO
TURBULENT FLOW IN BEARINGS
A Dissertation
Presented to
The Graduate Faculty of The University of Akron
In Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy
Dingfeng Deng
August, 2007
A NUMERICAL AND EXPERIMENTAL INVESTIGATION OF TAYLOR FLOW
INSTABILITIES IN NARROW GAPS AND THEIR RELATIONSHIP TO
TURBULENT FLOW IN BEARINGS
Dingfeng Deng
Dissertation
Approved: Accepted:
______Advisor Department Chair Dr. M. J. Braun Dr. C. Batur
______Committee Member Dean of the College Dr. J. Drummond Dr. G. K. Haritos
______Committee Member Dean of the Graduate School Dr. S. I. Hariharan Dr. G. R. Newkome
______Committee Member Date R. C. Hendricks
______Committee Member Dr. A. Povitsky
______Committee Member Dr. G. Young ii
ABSTRACT
The relationship between the onset of Taylor instability and appearance of what is
commonly known as “turbulence” in narrow gaps between two cylinders is investigated.
A question open to debate is whether the flow formations observed during Taylor
instability regimes are, or are related to the actual “turbulence” as it is presently modeled
in micro-scale clearance flows.
This question is approached by considering the viscous fluid flow in narrow gaps
between two cylinders with various eccentricity ratios. The computational engine is
provided by CFD-ACE+, a commercial multi-physics software. The flow patterns,
velocity profiles and torques on the outer cylinder are determined when the speed of the
inner cylinder, clearance and eccentricity ratio are changed on a parametric basis.
Calculations show that during the Taylor vortex regime velocity profiles in the radial
direction are sinusoidal with pressure variations in the axial direction even for the case of
the “long journal bearing” (L/D>2). For the concentric case ( ε = 0), both velocity and pressure profiles are axisymmetric and time-independent during the Taylor vortex regime. During the wavy vortex regime the radial velocities maintain their sinusoidal profiles, while pressure varies in both axial and circumferential directions. Both velocity and pressure profiles are non-axisymmetric and time-dependent. An order of magnitude analysis of the Navier-Stokes equation terms shows that the inertia, viscous, pressure and
iii the Reynolds stress terms are equally significant during the transition regime (Taylor to wavy vortex regimes).
Based on these findings, a new model for predicting the flow behavior in long and short journal bearing films in the transition regime is proposed. The new model indicates that the velocity profiles are sinusoidal and depend on the local Reynolds number (or
Taylor number) and the position in the axial direction. Unlike the modified turbulent viscosity of the most accepted models (Constantinescu, Ng-Pan, Hirs and Gross et al.), the viscosity used in the new model is kept at its laminar value. A comparison is made between the results of this model and the four most accepted turbulence models mentioned above.
Experimental torque measurements and flow visualization are performed for three kinds of oils with different viscosities. It is shown that in general there is a good agreement between the numerical and experimental torques except those in turbulent regime. The different flow patterns in the turbulent regime accounts for the difference between the numerical and experimental torques. Comparison between numerical and experimental flow patterns is also made and it shows that they match well in the Couette,
Taylor and Wavy regimes. There is a discrepancy between numerical and experimental flow patterns in Pre-wavy regime. The reason is that the vorticity is strong enough to destroy the organized Taylor vortices but it is not strong enough to hold the particles in the center areas of the vortices during experiments in this regime.
In general there is a good agreement between the numerical and experimental results including torque measurements and flow patterns. The new model for predicting the flow behavior in journal bearing films in the transition regime is justified. iv
ACKNOWLEDGEMENTS
I would like to express my appreciation and gratitude to my advisor, Dr. M.J. Braun, for his continuous and invaluable advisement. I am glad that I met Dr. Braun during my first semester of graduate school. Since then he has been my professional advisor, life mentor and personal friend. It is obvious that I have achieved my goals under the five years’ guidance of Dr. Braun.
I would like to thank my committee members, Dr. Drummond, Dr. Hariharan, Mr.
Hendrick, Dr. Povitsky and Dr. Young, for their comments and guidance during the
course of my dissertation.
I would also like to thank Dale Ertley for his work on building the installation for my
experiments and Bob Shardy for his work on installing and trouble-shooting the CFD
software package.
Additionally I received unlimited support and courage from my friends and family. I
would not have accomplished this without them.
v
TABLE OF CONTENTS
Page
LIST OF TABLES...... xii
LIST OF FIGURES...... xiv
NOMENCLATURE...... xxiii
CHAPTER
I. THEORETICAL INTRODUCTION AND LITERATURE REVIEW...... 1
1.1 Introduction...... 1
1.2 Theories of Fluid Instability in Concentric Cylinders and Literature Review...... 4
1.2.1 Rayleigh’s Stability Criterion...... 4
1.2.2 Linear Theory of the Instability...... 8
1.2.3 Literature Review on Linear Theory...... 12
1.2.4 Weakly Nonlinear Theory...... 15
1.2.5 Literature Review on Weakly Nonlinear Theory...... 18
1.2.6 Theory of Wavy Taylor Vortices and Literature Review...... 20
1.3 Theories of Fluid Instability in Eccentric Cylinders and Literature Review...... 25
1.3.1 DiPrima’s Local Theory and Literature Review...... 25
1.3.2 DiPrima and Stuart’s Non-local Theory and Literature Review...... 31
1.4 Theories of Turbulence in Bearings and Literature Review...... 33
vi 1.4.1 Constantinescu’s Theory...... 33
1.4.2 Ng-Pan Theory...... 37
1.4.3 Bulk Flow Theory of Hirs...... 39
1.4.4 Literature Review on Theories of Turbulence in Bearings...... 42
II. SCOPE OF WORK...... 47
2.1 Numerical Simulations and Calculations of Long and Short Bearings...... 47
2.2 The Nature of “Instability” & “Turbulence” in Small Gap Journal Bearings...... 47
2.3 New Models for Transition Flow of Thin Films in Long and Short Bearings...... 48
2.4 Experimental Verification of the Numerical Results...... 49
III. NUMERICAL ALGORITHM...... 50
3.1 General Introduction to CFD-ACE+...... 50
3.2 Numerical Methodology Adopted by CFD-ACE+...... 50
3.2.1 Discretization...... 51
3.2.2 Velocity-Pressure Coupling...... 55
3.2.3 Boundary Conditions...... 57
3.2.4 Solution Methods...... 58
3.3 Flow Module...... 61
IV. NUMERICAL RESULTS (LONG BEARING)...... 62
4.1 Introduction...... 62
4.2 Geometry and Boundary Condition Applications...... 63
4.2.1 Geometry...... 63
4.2.2 Boundary Condition Applications ...... 67 vii 4.3 Convergence Criteria and Numerical Accuracy...... 67
4.3.1 Gridding and Grid Convergence...... 67
4.3.2 Time Step and Time Step Convergence...... 69
4.4 Multiplicity and Transition of Taylor-Couette Flow...... 71
4.4.1 Introduction...... 71
4.4.2 The Influence of T b on the Number of Taylor Cells...... 73
4.4.3 The Influence of T b on the Number of Waves...... 82
4.4.4 Summary on Multiplicity and Transition...... 88
4.5 Effect of Clearance on Taylor Vortices Induced Instability...... 88
4.5.1 Concentric Cylinder Case with Clearance of 0.01 in...... 89
4.5.2 Concentric Cylinder Case with Clearance of 0.13 in...... 104
4.5.3 Effect of Clearance on Taylor Vortices Induced Instability...... 126
4.6 Effect of Eccentricity Ratio on Taylor Vortices Induced Instability...... 129
4.6.1 Effect of Eccentricity Ratio on the Torque and Magnitude of Critical Taylor Number...... 129
4.6.2 Effect of Eccentricity Ratio on the Friction Factor...... 131
4.6.3 Effect of Eccentricity Ratio on Maximum Vortex Intensity...... 132
4.6.4 Effect of Eccentricity Ratio on Recirculation...... 134
4.7 A New Model for Transition Flow of Thin Films in Long Journal Bearings...... 135
4.7.1 The Geometry and Coordinates ...... 136
4.7.2 The Case with ε = 0 ...... 136
4.7.3 The Case with ε = 0.2 ...... 144
4.7.4 Proposed Transition Reynolds Equation Model ...... 146 viii 4.7.5 Coefficients Used in Transition Reynolds Equation Model ...... 148
4.7.6 Comparison between Our Model and Turbulence Models...... 158
V. NUMERICAL RESULTS (SHORT BEARING)...... 165
5.1 Introduction...... 165
5.2 Geometry and Coordinates...... 165
5.3 The Case with ε = 0...... 167
5.4 The Case with ε = 0.2...... 174
5.5 Proposed Transition Reynolds Equation Model for Short Bearings...... 182
5.6 Comparison between our Model and Turbulence Models for Short Bearing...183
VI. DESCRIPTION OF EXPERIMENTAL INSTALLATION ...... 186
6.1 Test Loop...... 186
6.2 Drive System...... 187
6.3 Test Section (Apparatus)...... 188
6.4 Torque Measuring System...... 191
6.5 Visualization System ...... 192
6.5.1 Creation of the Laser Sheet...... 192
6.5.2 Visualization of the Test Section ...... 192
6.5.3 The Tracer Particels ...... 195
6.6 Working Fluid...... 195
VII. CALIBRATIONS...... 197
7.1 Calibration of the Strain Gage ...... 197
7.2 Torque Measurements for the Case of Air between the Cylinders...... 203
VIII. EXPERIMENTAL PROCEDURE ...... 205 ix 8.1 Experimental Start-up ...... 205
8.1.1 Preparation of the Experimental Apparatus...... 205
8.1.2 Preparation of the Experiments...... 206
8.2 Experimental Procedure...... 207
8.2.1 Quick Pass...... 207
8.2.2 Individual Regime Study ...... 208
IX. EXPERIMENTAL RESULTS ...... 209
9.1 Results for the Case of Oil I [ = 0.006 kg/(m-s), ρ = 778 kg/m 3]...... 209
9.1.1 Torque Measurements for the Case of Oil I...... 209
9.1.2 Visualization of Flow Patterns for the Case of Oil I...... 213
9.2 Results for the Case of Oil II [ = 0.038 kg/(m-s), ρ = 961 kg/m 3]...... 231
9.2.1 Torque Measurements for the Case of Oil II ...... 231
9.2.2 Visualization of Flow Patterns for the Case of Oil II ...... 234
9.3 Results for the Case of Oil III [ = 0.099 kg/(m-s), ρ = 1048 kg/m 3] ...... 247
9.3.1 Torque Measurements for the Case of Oil III...... 248
9.3.2 Visualization of Flow Patterns for the Case of Oil III...... 251
X. COMPARISON BETWEEN NUMERICAL AND EXPERIMENTAL RESULTS..265
10.1 Comparison of Torque-speed Graphs ...... 265
10.2 Comparison of Flow Patterns for the Case of Oil I ...... 269
XI. CONCLUSIONS ...... 283
11.1 Conclusion to the Numerical Investigation...... 283
11.2 Conclusion to the Experimental Verification ...... 285
11.3 Future Work...... 286 x BIBLIOGRAPHY...... 288
APPENDICES...... 295
APPENDIX A: THE RESULTS OF CALIBRATION OF STRAIN GAGE...... 296
APPENDIX B: VOLTAGE MEASUREMENTS FOR THE CASE OF AIR...... 298
APPENDIX C: VOLTAGE MEASUREMENTS FOR THE CASE OF OIL I...... 300
APPENDIX D: VOLTAGE MEASUREMENTS FOR THE CASE OF OIL II....302
APPENDIX E: VOLTAGE MEASUREMENTS FOR THE CASE OF OIL III ...304
xi
LIST OF TABLES
Table Page
1.1 Critical Parameters for the Onset of Taylor Vortices for Various Values of η ...... 14
1.2 Comparison of Coefficients for the Calculations of Taylor Vortex Torques...... 19
1.3 Constants for Calculating k x and k z at Different Reynolds Numbers...... 39
4.1 Grid Convergence Experiments (C = 0.13 in., ε = 0.0 and ω = 1,550 rpm)...... 68
4.2 Time Step Convergence Experiments (C = 0.13 in., ε = 0.0 and ω = 4,000 rpm)...... 70
4.3 Torque and Flow Type vs. Speed for the Case of C = 0.01 in., ε = 0.0...... 89
4.4 Torque and Flow Type vs. Speed for the Case of C = 0.13 in., ε = 0.0...... 104
4.5 Positions of Points on a Typical Trajectory in Taylor Vortex Regime (Axial Direction)...... 121
4.6 Positions of Points on a Typical Trajectory in Taylor Vortex Regime (Radial Direction)...... 121
4.7 Positions of Points on a Typical Trajectory in Wavy Vortex Regime (Axial Direction)...... 122
4.8 Positions of Points on a Typical Trajectory in Wavy Vortex Regime (Radial Direction)...... 122
4.9 Recirculation Region with Different Speed when ε = 0.6...... 135
4.10 Order of Magnitude Analysis of the Navier-Stokes Equation in x-direction (C = 0.01 in., ε = 0.2)...... 162
xii
4.11 Order of Magnitude Analysis of the Navier-Stokes Equation in z-direction (C = 0.01 in., ε = 0.2)...... 163
6.1 Properties of the Working Fluid (Oil I)...... 195
6.2 Properties of the Working Fluid (Oil II)...... 196
6.3 Properties of the Working Fluid (Oil III)...... 196
7.1 Correlation Coefficients of the Curve Fit for the Calibration of the Strain Gage...... 200
9.1 Correlation Coefficients of the Curve Fit for Couette Flow (Oil I)...... 210
9.2 Correlation Coefficients of the Curve Fit for Pre-wavy Flow (Oil I)...... 211
9.3 Correlation Coefficients of the Curve Fit for Wavy Flow (Oil I)...... 211
9.4 Correlation Coefficients of the Curve Fit for Turbulent Flow (Oil I)...... 211
9.5 Correlation Coefficients of the Curve Fit for Couette Flow (Oil II)...... 233
9.6 Correlation Coefficients of the Curve Fit for Pre-wavy Flow (Oil II)...... 233
9.7 Correlation Coefficients of the Curve Fit for Wavy Flow (Oil II)...... 234
9.8 Correlation Coefficients of the Curve Fit for Couette Flow (Oil III)...... 250
9.9 Correlation Coefficients of the Curve Fit for Taylor Flow (Oil III)...... 250
9.10 Correlation Coefficients of the Curve Fit for Pre-wavy Flow (Oil III)...... 251
9.11 Correlation Coefficients of the Curve Fit for Wavy Flow (Oil III)...... 251
xiii
LIST OF FIGURES
Figure Page
1.1 Velocity Components in Cylindrical Coordinates...... 5
1.2 Rayleigh’s Stability Diagram [After Koschmieder (1993)]...... 7
1.3 Coordinates of Concentric Cylinders...... 9
1.4 The Relationship among Reynolds Number, Taylor Number, Curvature Effect and Flow Pattern...... 26
1.5 A Plot of Values of Rr C / R1 & RP C / R1 for Stable & Unstable Flow...... 27
1/2 1.6 Critical Value of Rr(C/R1) as a Function of ε for Various Values of θ...... 29
1.7 Critical Speed as a Function of ε for Various Values of Clearance Ratio...... 30
1.8 Lower and Upper Bounds for Instability in Bearing-like Clearance Ratios...... 32
1.9 Turbulence Coefficients G x and G z vs. Reynolds Number...... 43
1.10 Friction Factor vs. Reynolds Number...... 44
1.11 Operating Ranges for Typical Oil and Air Bearings...... 45
3.1 A Three-dimensional Computational Cell (Control Volume)...... 51
3.2 A Two-dimensional Cell (Control Volumes) ...... 52
3.3 Computational Boundary Cell ...... 57
3.4 Solution Flowchart...... 59
xiv 4.1 Geometric Description of the Cylinders (Not at scale)...... 63
4.2 Torque vs. Taylor Number for Various λs when C = 0.01 in. and ε = 0.0 ...... 64
4.3 Torque vs. Taylor Number for Different Aspect Ratio (L/C = 4 and 6)...... 64
4.4 Wavelength of Taylor Vortices for Different Aspect Ratios (L/C=2, 4 and 6) ...65
4.5 Torque vs. Time at Different CFLs (C = 0.13 in., ε = 0.0, ω = 4000 rpm, Re = 323.55, Ta = 120.65) ...... 70
4.6 Geometry for Studying the Influence of T b on the Number of Taylor Cells (Not at Scale) ...... 73
4.7 Velocity of Inner Cylinder vs. Time (T b is the Time Needed to Increase the Speed of Inner Cylinder from Rest to 263.33 m/s)...... 74
4.8 Number of Taylor Cells vs. Time...... 75
4.9 w - Velocity in Axial Direction vs. Time ...... 75
4.10 Maximum Vorticity vs. Time ...... 78
4.11 P – Generated Pressure vs. Time...... 78
4.12 The Process of Formation of Taylor Cells for the Case of Impulsive Start...... 79
4.13 The Process of Mergence of Taylor Cells for the Case of Impulsive Start ...... 80
4.14 Velocity of Inner Cylinder vs. Time...... 82
4.15 Wave Number in Circumferential Direction vs. Time...... 82
4.16 P – Generated Pressure vs. Time...... 83
4.17 The Development of Waves for the Case of T b = 0.002 s ...... 84
4.18 Torque vs. Ta and Re for the Case of C = 0.01 in., ε = 0.0 ...... 90
4.19 Velocity and Pressure Profiles (Re=283.22, Ta =28.46, C = 0.01 in.)...... 92
4.20 Trajectory of a Typical Particle in Couette Regime (Re=283.22, Ta =28.46, C = 0.01 in.) (Not at scale)...... 93
xv 4.21 Velocity and Pressure Profiles (Re=424.82, Ta =42.70, C = 0.01 in.)...... 94
4.22 Trajectory of a Typical Particle in Taylor Regime (Re=424.82, Ta =42.70, C = 0.01 in.) (Not at scale)...... 95
4.23 Velocity and Pressure Profiles (Re=708.04, Ta =71.16, C = 0.01 in.)...... 96
4.24 Velocity and Pressure Profiles (Re=708.04, Ta =71.16, C = 0.01 in.) (Not at scale)...... 97
4.25 Trajectory of a Typical Particle in Wavy Regime (Re=708.04, Ta =71.16, C = 0.01 in.) (Not at scale)...... 98
4.26 Velocity and Pressure Profiles of One Wave when Ta =71.16 [Not at scale in b) and c)]...... 103
4.27 Torque vs. Ta and Re for the Case of C = 0.13 in., ε = 0.0 ...... 105
4.28 Velocity and Pressure Profiles (C=0.13 in., Ta =30.16, A in Figure 4.25) ...... 108
4.29 Velocity and Pressure Profiles (C=0.13 in., Ta =46.75, B in Figure 4.25) ...... 109
4.30 Velocity and Pressure Profiles (C=0.13 in., Ta =66.35, C in Figure 4.25) ...... 112
4.31 w-velocity and Pressure Contours at Different z for the Wavy Vortex Regime (C=0.13 in., ω=4000 rpm, Re=323.55, Ta =120.65), D in Figure 4.25...... 113
4.32 w-velocity and Pressure Contours at Different r for the Wavy Vortex Regime (C=0.13 in., ω=4000 rpm, Re=323.55, Ta =120.65), D in Figure 4.25...... 114
4.33 w-velocity and Pressure Contours with Superimposed Velocity Vectors at Different θ for the Wavy Vortex Regime, D in Figure 4.25...... 115
4.34 Comparison of Velocity and Pressure Contours for Taylor Vortex Flow and Wavy Vortex Flow (C = 0.13 in.) ...... 117
4.35 Comparison of Iso-curves and Iso-surfaces of w and P for Taylor Vortex Flow and Wavy Vortex Flow (C = 0.13 in.) ...... 118
4.36 Trajectory of a Typical Particle in Taylor Vortex Regime (Re=125.38, Ta =46.75, C = 0.13 in.)...... 123 xvi 4.37 Trajectory of a Typical Particle in Wavy Vortex Regime (Re=323.55, Ta =120.65, C = 0.13 in.)...... 124
4.38 Comparison of Our Calculations and Experiments by Koschmieder ...... 125
4.39 Comparison of Velocity Profiles of Different Clearances...... 126
4.40 Torque vs. Taylor No. and Reynolds No. for Various Clearances ...... 127
4.41 First and Second Critical Taylor Numbers for Various Clearances ...... 128
4.42 Torque - Ta and Ta cr - Eccentricity Ratio Curves...... 130
4.43 Friction Factor vs. Reynolds Number when C = 0.01 in., ε = 0.0 – 0.8 ...... 131
4.44 Flow Pattern at Different Positions when ε = 0.2, ω = 65,000 rpm...... 132
4.45 Flow Pattern at Different Positions when ε = 0.6, ω = 110,000 rpm...... 133
4.46 Geometry and Coordinates of a Journal Bearing Film ...... 136
4.47 Torque vs. Ta When C = 0.01 in., ε = 0.0 ...... 137
4.48 Velocity and Pressure Profiles as Functions of y* during Couette Region ...... 138
4.49 Velocity and Pressure Profiles as Functions of y* during Taylor Region...... 139
4.50 Velocity and Pressure Profiles as Functions of y* during Wavy Vortex Region at a Given Time t = T (0) ...... 140
4.51 Pressure Contour in the x-z Plane during Wavy Vortex Regime ...... 144
4.52 Pressure Profiles as Functions of z* during Wavy Vortex Region at a Given Time t = T (0) ...... 146
4.53 Relationship between Velocity Gradient on Outer Cylinder and U*...... 149
4.54 Coefficients a 1 as Function of Reynolds Number for C/R = 0.01 ...... 150
4.55 Coefficients a 2 as Function of Reynolds Number for C/R = 0.01 ...... 150
4.56 Coefficients a 1 as Function of Reynolds Number for C/R = 0.004 ...... 151
4.57 Coefficients a 2 as Function of Reynolds Number for C/R = 0.004 ...... 151 xvii 4.58 Coefficients a 1 as Function of Reynolds Number for C/R = 0.001 ...... 152
4.59 Coefficients a 2 as Function of Reynolds Number for C/R = 0.001 ...... 152
4.60 Coefficients A 1 as Function of Reynolds Number for Different C/R...... 153
4.61 Coefficients A 2 as Function of Reynolds Number for Different C/R...... 153
4.62 Coefficients A 1 as Function of Ta for Different C/R...... 155
4.63 Coefficients A 2 as Function of Ta for Different C/R...... 155
4.64 Coefficients A 1 as Function of Inverse of Ta for Different C/R...... 156
4.65 Coefficients A 2 as Function of Inverse of Ta for Different C/R...... 156
4.66 Comparison of Torque on Bearing between Our Model and Turbulence Models as Functions of Reynolds Number...... 158
4.67 Comparison of Torque on Bearing between Our Model and Turbulence Models as Functions of Taylor Number ...... 158
5.1 Geometric Description of the Cylinders (Not at scale)...... 166
5.2 Geometry and Coordinates of a Journal Bearing Film ...... 166
5.3 Torque vs. Ta for Short Bearing When C = 0.01 in., ε = 0.0 ...... 167
5.4 Velocity and Pressure Contours during Couette Regime (Point A in Figure 5.3)...... 168
5.5 Velocity and Pressure Profiles as Functions of y* and z* during Couette Regime (Point A in Figure 5.3)...... 169
5.6 Velocity and Pressure Contours during Taylor Vortex Regime (Point B in Figure 5.3) ...... 171
5.7 Velocity and Pressure Profiles as Functions of y* and z* during Taylor Vortex Regime (Point B in Figure 5.3)...... 172
5.8 Torque vs. Ta for Short Bearing When C = 0.01 in., ε = 0.2 ...... 175
xviii 5.9 Velocity and Pressure Contours during Couette Regime (Point C in Figure 5.8) ...... 176
5.10 Velocity and Pressure Profiles as Functions of y* and z* during Couette Regime (Point C in Figure 5.8)...... 177
5.11 Velocity and Pressure Contours during Taylor Vortex Regime (Point D in Figure 5.8)...... 178
5.12 Velocity and Pressure Profiles as Functions of y* and z* during Taylor Vortex Regime (Point D in Figure 5.8) ...... 179
5.13 Comparison of Torque on Bearing between Our Model and Turbulence Models for Short Bearing...... 184
6.1 Schematic Drawing of Test Loop ...... 186
6.2 Driving System and Test Section...... 187
6.3 Schematic Drawing of Test Section...... 189
6.4 Torque Measuring System...... 190
6.5 Visualization of the Fluid in Longitudinal Cross Section View...... 193
6.6 Visualization of the Fluid in Front View...... 194
7.1 The Principle of the Torque Measurement ...... 197
7.2 Calibration of the Strain Gage ...... 198
7.3 Calibration Curve of Force-Voltage ...... 201
7.4 Curve-fit of the Calibration of the Strain Gage ...... 203
7.5 Torque Measurements for the Case of Air between the Cylinders...... 204
9.1 Voltage Measurements for the Case of Oil I ...... 212
9.2 Voltage due to Fluid Friction vs. Speed for the Case of Oil I ...... 212
9.3 Flow Pattern in Longitudinal Cross Section View in Couette Regime (Oil I) ....213
9.4 Flow Pattern in Longitudinal Cross Section View in Taylor Regime (Oil I) ...... 214
xix 9.5 The Collapse of Taylor Vortices for the Case of ω = 200 rpm (Oil I)...... 215
9.6 Flow Pattern in Longitudinal Cross Section View in Pre-wavy Regime (Oil I)...... 216
9.7 Flow Pattern in Longitudinal Cross Section View in Wavy Regime (Oil I) ...... 218
9.8 Flow Pattern in Longitudinal Cross Section View in Turbulent Regime (Oil I) ...... 219
9.9 Flow Pattern in Front View in Couette Regime (Oil I) ...... 219
9.10 Flow Pattern in Front View in Taylor Regime (Oil I) ...... 220
9.11 Flow Pattern in Front View in Pre-wavy Regime (Oil I)...... 222
9.12 Flow Pattern in Front View in Wavy Regime (Oil I) ...... 223
9.13 The Collapse of Waves for the Case of ω = 1,350 rpm (Oil I)...... 227
9.14 Flow Pattern in Front View in Turbulent Regime (Oil I) ...... 228
9.15 Voltage Measurements for the Case of Oil II...... 231
9.16 Voltage due to Fluid Friction vs. Speed for the Case of Oil II ...... 232
9.17 Flow Pattern in Longitudinal Cross Section View in Couette Regime (Oil II)...234
9.18 Flow Pattern in Longitudinal Cross Section View in Taylor Regime (Oil II).....236
9.19 The Collapse of Taylor Vortices for the Case of ω = 650 rpm (Oil II) ...... 237
9.20 Flow Pattern in Longitudinal Cross Section View in Pre-wavy Regime (Oil II) ...... 239
9.21 Flow Pattern in Longitudinal Cross Section View in Wavy Regime (Oil II)...... 240
9.22 Flow Pattern in Front View in Couette Regime (Oil II) ...... 241
9.23 Flow Pattern in Front View in Taylor Regime (Oil II)...... 242
9.24 Flow Pattern in Front View in Pre-wavy Regime (Oil II) ...... 243
9.25 Flow Pattern in Front View in Wavy Regime (Oil II)...... 244
xx 9.26 Flow Pattern in Front View in Turbulent Regime (Oil II)...... 247
9.27 Voltage Measurements for the Case of Oil III...... 248
9.28 Voltage due to Fluid Friction vs. Speed for the Case of Oil III...... 249
9.29 Flow Pattern in Longitudinal Cross Section View in Couette Regime (Oil III)..252
9.30 Flow Pattern in Longitudinal Cross Section View in Taylor Regime (Oil III) ...253
9.31 The Collapse of Taylor Vortices for the Case of ω = 1,200 rpm (Oil III) ...... 254
9.32 Flow Pattern in Longitudinal Cross Section View in Pre-wavy Regime (Oil III)...... 255
9.33 Flow Pattern in Longitudinal Cross Section View in Wavy Regime (Oil III).....256
9.34 Flow Pattern in Front View in Couette Regime (Oil III)...... 257
9.35 Flow Pattern in Front View in Taylor Regime (Oil III)...... 258
9.36 Flow Pattern in Front View in Pre-wavy Regime (Oil III)...... 261
9.37 Flow Pattern in Front View in Wavy Regime (Oil III)...... 262
10.1 Comparison of Torque-speed Graphs for the Case of Oil I...... 266
10.2 Comparison of Torque-speed Graphs for the Case of Oil II...... 267
10.3 Comparison of Torque-speed Graphs for the Case of Oil III ...... 268
10.4 Comparison of Flow Pattern in Longitudinal Cross Section View in Couette Regime (ω = 100 rpm) (Oil I)...... 270
10.5 Comparison of Flow Pattern in Front View in Couette Regime (ω = 100 rpm) (Oil I)...... 271
10.6 Comparison of Flow Pattern in Longitudinal Cross Section View in Taylor Regime (ω = 200 rpm) (Oil I) ...... 272
10.7 Comparison of Flow Pattern in Front View in Taylor Regime (ω = 200 rpm) (Oil I) ...... 273
10.8 Comparison of Flow Pattern in Longitudinal Cross Section View in Pre-wavy Regime (ω = 500 rpm) (Oil I)...... 275 xxi
10.9 Vorticity Comparison between Pre-wavy and Wavy Regimes (Oil I) ...... 276
10.10 Comparison of Flow Pattern in Front View in Pre-wavy Regime (ω = 500 rpm) (Oil I)...... 279
10.11 Comparison of Flow Pattern in Longitudinal Cross Section View in Wavy Regime (ω = 900 rpm) (Oil I) ...... 280
10.12 Comparison of Flow Pattern in Front View in Wavy Regime (ω = 900 rpm) (Oil I) ...... 281
10.13 Comparison of Flow Pattern in Longitudinal Cross Section View in Turbulent Regime (ω = 2,100 rpm) (Oil I) ...... 282
xxii
NOMENCLATURE
C = mean radial clearance, in. e = eccentric displacement of journal, in. h = film thickness, in.
L = length of cylinders, in.
R = radius of inner cylinder, in.
R+C = radius of outer cylinder, in.
RωC Re = Reynolds number = ν
t = time, sec
t* = dimensionless time = t/T
T = period that the waves travel in the circumferential direction, sec
2C Ta = Taylor number = Re 2 2R + C C Ta = Taylor number = Re2 for small value of clearance ratios R u = velocity in the circumferential direction, in./sec
v = velocity in the radial direction, in./sec
w = velocity in the axial direction, in./sec
U = average velocity in the circumferential direction, in./sec
V = average velocity in the radial direction, in./sec xxiii W = average velocity in the axial direction, in./sec x = circumferential coordinate, in. y = radial coordinate, in. z = axial coordinate, in. y* = dimensionless radial coordinate = y/h z* = dimensionless axial coordinate = z/L
ε = eccentricity ratio = e/C
θ = circumferential coordinate, in.
λ = dimensionless axial wavelength = L/C
ν = kinematic viscosity, in. 2/sec
ρ = fluid density, lb-sec 2/in. 4
ω = rotational speed, rad/sec
xxiv
CHAPTER I
THEORETICAL INTRODUCTION AND LITERATURE REVIEW
1.1 Introduction
People have been investigating the hydrodynamic stability of rotating fluids between two cylinders, both experimentally and analytically, for more than one hundred years.
Why did such an apparently “simple problem” become so popular? In fact, investigators were amazed to see the wealthy variety of flow patterns that occur, for example, when the speed of the inner cylinder is increased and the outer cylinder is at rest. One attractive feature of these patterns is their degree of symmetries, both spatially and temporally. For example, the Couette flow and Taylor vortex flow are axisymmetric along the axis of the cylinders and are time-independent when the speed of the inner cylinder is low.
However, as the speed of inner cylinder increases, the flow pattern becomes more and more complicated, breaking more and more symmetries both spatially and temporally.
With the increase of the inner cylinder speed, the Taylor vortex flow is replaced by wavy
Taylor vortex flow, which is non-axisymmetric and time-dependent. Eventually the flow becomes “turbulent”, containing large-scale structures with many degrees of symmetry.
This model problem then appears as an ideal example of a system that progressively approaches “turbulence”, which is still one of the challenging problems in the field of fluids. 1 The very first experimental investigator was Couette (1890), who designed an apparatus consisting of two concentric cylinders, the inner one was stationary and the outer one rotating, and the space between them was filled with water. The purpose of his experiment was to determine the viscosity of water by measuring the torque exerted by water on the inner cylinder. He found that when the angular velocity was not too large, water flow was laminar and the torque was proportional to the angular velocity.
However, when the angular velocity was increased to a large value, water flow became unstable and the slope of torque-speed diagram increased.
A few years later, Mallock (1896) designed a similar apparatus but allowed the inner cylinder to rotate. He was the first one who purposely did experiments to study the hydrodynamic stability problem. He stated that the object of the experiments was chiefly to examine the limits between which the motion of the fluid in the annulus was stable, and the manner in which the stability broke down. He noticed that instability occurred easily when the inner cylinder rotated and the outer one was stationary; and instability also occurred when the outer cylinder rotated at a large angular velocity and the inner one was at rest.
However, neither Couette nor Mallock established a definite criterion for the
hydrodynamic stability of the fluids. Rayleigh (1916) was the first one who found a
criterion for the hydrodynamic stability of inviscid fluid between two concentric rotating
cylinders. Taylor (1923) in his celebrated paper derived the linear theory of instability
for viscous fluids. His theory showed that there exists a critical Reynolds number, below
which all initially infinitesimal disturbances are damped and decay to zero with
increasing time and above which the disturbances will grow with time. Moreover, 2 Taylor’s experiments showed that this instability of Couette flow leads to a new steady secondary axisymmetric flow in the form of regularly spaced vortices in the axial direction, which are commonly called Taylor vortices. Since then, theories predicting the growth of the Taylor vortices and the onset of wavy Taylor vortices with further increase of speed were proposed by a lot of investigators, such as weakly nonlinear theory by
Stuart (1958) and Davey (1962), and theory of wavy vortices by DiPrima (1961) and
Davey et al. (1968).
Stability problem of fluid between two eccentric cylinders or a journal bearing is receiving more and more attention. The reason is that there is a trend in modern technology toward increasingly higher operational speed of journal bearings and toward the use of low viscosity fluids for lubrication, which results in situations where the fluid in a bearing film becomes unstable. Since the load capacity and frictional power loss of a bearing can increase significantly when the flow undergoes a transition from laminar to vortex flow and turbulence flow, it is very important to know when and how this transition occurs.
Wilcock (1950) and Smith and Fuller (1956) were the pioneers to recognize that flow transition occurs between eccentric cylinders in journal bearings. However, DiPrima
(1963) was the first one to establish a stability criterion of fluid between two eccentric cylinders, which is also called local theory of hydrodynamic stability. To account for the disagreement between the local theory and some experimental data, DiPrima and Stuart
(1972) established their nonlocal theory.
DiPrima’s local theory and DiPrima and Stuart’s nonlocal theory define the threshold for the onset of fluid instability, which usually appears in the form of Taylor vortices in 3 eccentric cylinders. With the further increase of inner cylinder speed, the flow becomes wavy vortices and eventually becomes fully developed “turbulence”. Constantinescu
(1959), Ng and Pan (1965), and Hirs (1973) established their own turbulence theories, which govern the fully developed “turbulence” in journal bearings. All the three theories are based on well-established empiricisms relating turbulent shear stress to mean velocity gradient, such as Prandtl mixing length concept or the eddy diffusivity concept.
During the next sections in this chapter, a) theories of fluid instability in concentric cylinders, i.e. Rayleigh’s stability criterion, linear theory, weakly nonlinear theory and theory of wavy Taylor vortices, b) theories of fluid instability in eccentric cylinders, i.e. local theory and nonlocal theory, c) theories of turbulence in bearings, i.e.
Constantinescu, Ng and Pan, and Hirs’ theories will be introduced in detail.
1.2 Theories of Fluid Instability in Concentric Cylinders and Literature Review
In this section, theories of fluid instability in concentric cylinders, i.e. Rayleigh’s stability criterion, linear theory, weakly nonlinear theory and theory of wavy Taylor vortices, will be introduced.
1.2.1 Rayleigh’s Stability Criterion
The first hydrodynamic stability criterion for inviscid fluids was derived by Rayleigh
(1916). His conclusion is that the flow is stable if the square of circulation increases in the radial direction for the case of inviscid axisymmetric flow; otherwise it is unstable.
Rayleigh reached that conclusion by starting from the Euler’s equations, Equation
(1.1), in cylindrical coordinates shown in Figure 1.1. 4
Figure 1.1 Velocity Components in Cylindrical Coordinates
∂v ∂v v ∂v v 2 ∂v ∂P r + v r + θ r − θ + v r = − (1.1a) ∂t r ∂r r ∂θ r z ∂z ∂r
∂v ∂v v ∂v v v ∂v 1 ∂P θ + v θ + θ θ + r θ + v θ = − (1.1b) ∂t r ∂r r ∂θ r z ∂z r ∂θ
∂v ∂v v ∂v ∂v ∂P z + v z + θ z + v z = − (1.1c) ∂t r ∂r r ∂θ z ∂z ∂z
where
dp P = −V (1.2) ∫ ρ
where V is the potential of extraneous forces.
With the assumption that vr, v θ, v z, and P are symmetric with respect to the axis of z, i.e., independent of θ, Equations (1.1a), (1.1b), and (1.1c) become:
5 ∂v ∂v v 2 ∂v ∂P r + v r − θ + v r = − (1.3a) ∂t r ∂r r z ∂z ∂r
∂v ∂v v v ∂v θ + v θ + r θ + v θ = 0 (1.3b) ∂t r ∂r r z ∂z
∂v ∂v ∂v ∂P z + v z + v z = − (1.3c) ∂t r ∂r z ∂z ∂z
Equation (1.3b) is equivalent to:
∂ ∂ ∂ ( + v + v )( rv θ ) = 0 (1.4) ∂t r ∂r z ∂z
Therefore, the angular momentum or circulation Γ(= rv θ ) of a fluid element, per unit mass, remains constant as we follow it with its motion. Then the force acting in the radial direction can be expressed as:
v 2 Γ 2 θ = (1.5) r r 3
The kinetic energy associated to a ring of fluid at a position r from the axis can be
expressed as:
πρ Γ 2 2 2 E = πρ vθ rdr = d (r ) (1.6) k ∫ ∫ 2r 2
Suppose now that we interchange the fluid contained in two elementary rings, of equal
2 2 areas d r1 or d r2 , at r = R1 and r = R2 (where R2 > R1). The corresponding increment in kinetic energy is represented by:
πρ Γ 2 Γ 2 Γ 2 Γ 2 ∆ = − = 2 2 + 1 − 1 − 2 Ek Ek 2 Ek1 d(r )( 2 2 2 2 ) 2 R1 R2 R1 R2
πρ = d(r 2 )( Γ 2 − Γ 2 )( R −2 − R −2 ) (1.7) 2 2 1 1 2
6 2 2 Remembering that R2 > R 1, we observed that Ek is positive if Γ2 > Γ1 , i.e., the circulation always increasing outwards makes kinetic energy a minimum and thus ensures stability. This can be expressed as the stability criterion:
> 2Ω > 2Ω (R2vθ 2 ) (R1vθ1 ) or (R2 2 ) (R1 1 ) (1.8)
If we denote that = 2/ 1 and η = R1/R 2, this relation can be illustrated with the
stability diagram in Figure 1.2.
Rayleigh’s stability criterion can only be applied to the case of inviscid flow or when
viscosity does not affect the occurrence of instability. However, viscosity is a very
important property for flows satisfying the Navier-Stokes equations. When the viscosity
of a fluid is greater than a critical value, all solutions tend monotonically to a basic flow,
Couette flow for the case of fluids between two concentric cylinders. When the viscosity
is less than the critical value, disturbances of the basic flow will grow. Therefore, next
section we will discuss linear theory, which deals with viscous fluid flow.
Figure 1.2 Rayleigh’s Stability Diagram [After Koschmieder (1993)]
7 1.2.2 Linear Theory of the Instability
The first linear theory on stability of viscous fluids was proposed by Taylor (1923).
He considered a viscous flow between two concentric rotating cylinders, which were assumed to be infinitely long and rotated about their common axis with constant angular velocities 1 at r = R 1, and 2 at r = R 2 (R2 > R 1). The governing equations are Navier-
Stokes equations in cylindrical coordinates (constant density and viscosity are assumed):
∂v ∂v v ∂v v 2 ∂v r + v r + θ r − θ + v r = ∂t r ∂r r ∂θ r z ∂z
2 2 2 1 ∂p ∂ v 1 ∂v v 1 ∂ v 2 ∂vθ ∂ v − +ν r + r − r + r − + r (1.9a) g r ρ ∂r ∂r 2 r ∂r r 2 r 2 ∂θ 2 r 2 ∂θ ∂z 2
∂v ∂v v ∂v v v ∂v θ + v θ + θ θ + r θ + v θ = ∂t r ∂r r ∂θ r z ∂z
∂ ∂ 2 v ∂v v ∂ 2v ∂v ∂ 2v − 1 p +ν θ + 1 θ − θ + 1 θ + 2 r + θ gθ (1.9b) ρr ∂θ ∂r 2 r ∂r r 2 r 2 ∂θ 2 r 2 ∂θ ∂z 2
∂v ∂v v ∂v ∂v z + v z + θ z + v z = ∂t r ∂r r ∂θ z ∂z
∂ ∂ 2v ∂v ∂ 2v ∂ 2v − 1 p +ν z + 1 z + 1 z + z g z (1.9c) ρ ∂z ∂r 2 r ∂r r 2 ∂θ 2 ∂z 2
The continuity equation in cylindrical coordinates for constant density is:
1 ∂(rv ) 1 ∂v ∂v r + θ + z = 0 (1.10) r ∂r r ∂θ ∂z
If the radial velocity component vr = 0, the vertical velocity component vz = 0, the
azimuthal velocity component vθ = vθ(r), the pressure p = p(r), and the gravity is negligible, then Equation (1.9b) reduces in the steady state to:
8 2 ∂ vθ 1 ∂vθ vθ + − = 0 (1.11) ∂r 2 r ∂r r 2
The solution of Equation (1.11) is:
vθ = Ar + B / r (1.12)
where
Ω η 2 1( − µ /η 2 ) A = − 1 (1.13) 1−η 2
Ω R 2 1( − µ) B = 1 1 (1.14) 1−η 2 where = 2/ 1 and η = R1/R 2.
Let u, v, and w be the radial, azimuthal, and axial velocity components of the
disturbance respectively as shown in Figure 1.3.
Figure 1.3 Coordinates of Concentric Cylinders
9
Assume that u, v, and w are small in comparison with vθ, and that the disturbance is symmetrical, so that they are functions of r, z and t only. Neglecting terms containing products of squares of u, v, and w, the equations of disturbance may be written as:
∂ 2 ∂ ∂ 2 ∂ ∂ 2 1 p − vθ = − u + + B +ν u + 1 u + u − u (2 A )v (1.15a) ρ ∂r r ∂t r 2 ∂r 2 r ∂r ∂z 2 r 2
∂ ∂ 2 ∂ ∂ 2 = − v − +ν v + 1 v + v − v (1.15b) 0 2Au 2 2 2 ∂t ∂r r ∂r ∂z r
∂ ∂ ∂ 2 ∂ ∂ 2 1 p = − w +ν w + 1 w + w (1.15c) ρ ∂z ∂t ∂r 2 r ∂r ∂z 2
The continuity equation is:
∂u u ∂w + + = 0 (1.16) ∂r r ∂z
The six boundary conditions are:
u = v = w = 0 at r = R 1 and r = R 2 (1.17)
A solution of the disturbance is assumed to be:
= λ σt u u1 (r) cos( z)e (1.18a)
= λ σt v v1 (r) cos( z)e (1.18b)
= λ σt w w1 (r)sin( z)e (1.18c)
If the pressure terms in Equations (1.15a) and (1.15c) are eliminated, then Equations
(1.15a), (1.15b), (1.15c), and (1.16) reduce to:
∂u u 1 + 1 + λw = 0 (1.19) ∂r r 1
10 ∂ 2 1 ∂ 1 σ ν + − − λ2 − = (1.20) 2 2 v1 2Au 1 ∂r r ∂r r ν
ν ∂ ∂ 2 1 ∂ σ B ∂ 2 1 ∂ 1 σ + − λ2 − = − + −ν + − − λ2 − (1.21) 2 w1 2 A 2 v1 2 2 u1 λ ∂r ∂r r ∂r ν r ∂r r ∂r r ν
Now the six boundary conditions become:
= = = u1 v1 w1 0 at r = R 1 and r = R 2 (1.22)
There are no terms containing z in these equations, which means that the normal modes of disturbance are simple harmonic with respect to z, the wave-length being 2π/λ. Note that σ is a quantity which determines the rate of increase in a normal disturbance. If σ is positive the disturbance increases and the motion is unstable. If σ is negative the disturbance decreases and the motion is stable. If σ is zero the motion is neutral.
The solution of Equations (1.19), (1.20) and (1.21) can be obtained by means of a type
of Bessel functions. Taylor assumed a Bessel series solution for u1:
∞ = κ (1.23) u1 ∑ am B1 ( m r) m=1
κ where B1 ( m r) is the Bessel function of order 1. Then he substituted this solution into
Equations (1.19), (1.20) and (1.21) and the solutions for v1 and w1 were obtained. Finally
after he used all the boundary conditions, Taylor obtained a system of linear homogenous
equations containing a1, a 2, , a m. The number of the unknowns is the same as that of the equations, so all the unknowns can be eliminated from the equations. The resulting equation takes the form of an infinite determinant equating to zero. The eigenvalue problem with σ = 0 can be solved by the Galerkin method, and the critical Taylor number
can be expressed by the following formula:
11 π 4 (1+ C 2/ R ) Ta = 1 (1.24) c ()()− + − −1 .0 0571 1 .0 652 C / R1 .0 00056 1 .0 652 C / R1
1.2.3 Literature Review on Linear Theory
After Taylor’s first success in the calculation of the critical Taylor number by using a linear stability theory, a lot of investigators improved the linear theory on Taylor instability by both analytical and numerical methods.
The first modification of the solution of Equations (1.19) through (1.22) was introduced by Jeffreys (1928). He reduced these equations to a single sixth-order differential equation, (1.25), for the case of narrow gap and marginal instability ( σ = 0).
3 ∂ 2 4A(λc) 2 c 4 B − ()λ 2 = + (1.25) 2 c v1 2 A 2 v1 ∂r ν r
Then by reasoning the analogy between the conditions in a layer of liquid heated below and in a liquid between two concentric cylinders rotating at different rates, Jeffreys showed that for the case of = 2/ 1 → 1, the critical Taylor number is equal to:
4Ac 4 Ta = Ω = 1709 (1.26) c ν 2 1 which is in good agreement with 1706, the value obtained by Taylor’s formula, Equation
(1.24).
Chandrasekhar (1961b) solved the nondimensional version of Equations (1.19) through
(1.22) for the case of narrow gap when the marginal state is stationary ( σ = 0). He expanded v1 in a sine series of the form:
∞ = πζ (1.27) v1 ∑Cm sin( m ) m=1 12 where ζ = (r-R1)/C . Then by manipulating the secular equation, the Taylor number in the
first approximation is obtained by the formula:
2 [π 2 +(λC)2 ]3 Ta = (1.28) c 1+µ (λC)2 {1−16 (λC)2 cosh 2 (λC [](/)2/ π 2 +(λC)2 )(sinh(λC)+λC)}
Equation (1.28) is valid for any , but it is too complicated. Therefore, Chandrasekhar gave two simpler formulas depending on the value of :
2 3416 − 1− µ Ta = 1− 61.7 ×10 3 (µ → ,1 η → )1 (1.29) c + µ + µ 1 1
= − µ 4 µ → −∞ η → Ta c Ta c0 1( ) ( , )1 (1.30) where Ta c0 ( = 1182) is comparable to the Taylor number when = 0.
Harris & Reid (1964) found the same asymptotic behavior of Ta c when → - ∞ as shown in Equation (1.30), and their value of Ta c0 was 1178.6. DiPrima & Swinney
(1981) verified that Equation (1.29) is quite satisfactory for -0.25 < < 1, and Equation
(1.30) holds with quite reasonable accuracy for < -1.
Chandrasekhar (1954) and DiPrima (1961) derived an expression of the critical Taylor number as a function of the clearance C and the inner radius R1 when C is much smaller than R1:
= ( + ) Ta c 1695 1 C 2/ R1 ( C << R 1) (1.31)
which is a simplified form of Taylor’s expression, Equation (1.24).
The stability problem in a wide gap was studied numerically by Roberts (1965) for the
case of = 0, and analytically by Sparrow et al. (1964) for the case in which both cylinders rotate. Their results were in good agreement with each other, and some results
13 of Robert are listed in Table 1.1. It is shown that both the critical Taylor number and wavelength increase with the decrease of radius ratio η.
Table 1.1 Critical Parameters for the Onset of Taylor vortices for Various Values of η
η Ta c λc
0.975 1723.89 3.1268
0.950 1754.76 3.1276
0.925 1787.93 3.1282
0.900 1823.37 3.1288
0.875 1861.48 3.1295
0.850 1902.40 3.1302
0.750 2102.17 3.1355
0.650 2383.96 3.1425
DiPrima and Hall (1984) studied the eigenvalue problem for the linear stability of
Couette flow between rotating concentric cylinders to axisymmetric disturbances. It was shown by numerical calculations and by formal perturbation methods that there exist complex eigenvalues corresponding to oscillatory damped disturbances when the outer cylinder is at rest. DiPrima and Hall (1984) were the first investigators to demonstrate the existence of complex eigenvalues for the boundary value problem for the stability of
Couette flow between rotating cylinders. Their results contradict an argument by Yih
(1972) that all the eigenvalues are real when the cylinders rotate in the same direction and the circulation of the basic flow decreases in the outward radial direction. However, 14 DiPrima and Hall (1984) also showed that all of the complex eigenvalues they found correspond to damped disturbances. Thus their results do not contradict the conjectured and widely believed principle of exchange of stabilities, i.e., for a fixed axial wave number, the first mode to become unstable as the speed of the inner cylinder is increased is non-oscillatory as the stability boundary is crossed.
1.2.4 Weakly Nonlinear Theory
In linear theory only the infinitesimal disturbances have been considered, so only the initial growth of the disturbance has been determined. In other words, the linear theory of hydrodynamic stability can predict correctly the critical Taylor number, but it cannot predict the establishment of a new equilibrium flow, Taylor vortex flow, above the critical Taylor number. It is obvious that the exponential growth of the disturbances considered in linear theory is not realistic. Therefore, it is necessary to solve the nonlinear equations, which means that the higher order terms cannot be neglected.
Stuart (1958) was the pioneer of nonlinear theory when he extended the linear theory to larger amplitude and studied the mechanics of disturbance growth taking the nonlinearity of the hydrodynamic system into account. He assumed that the flow had rotational symmetry and was therefore independent of θ, and then the Navier-Stokes and
continuity equations became:
∂v ∂v v 2 ∂v ∂ ∂ 2v ∂v v ∂ 2v r + r − θ + r = − 1 p +ν r + 1 r − r + r vr vz (1.32a) ∂t ∂r r ∂z ρ ∂r ∂r 2 r ∂r r 2 ∂z 2
∂v ∂v v v ∂v ∂ 2v ∂v v ∂ 2v θ + θ + r θ + θ = θ + 1 θ − θ + θ vr vz v (1.32b) ∂t ∂r r ∂z ∂r 2 r ∂r r 2 ∂z 2
15 ∂v ∂v ∂v ∂ ∂ 2v ∂v ∂ 2v z + z + z = − 1 p +ν z + 1 z + z vr vz (1.32c) ∂t ∂r ∂z ρ ∂z ∂r 2 r ∂r ∂z 2
1 ∂(rv ) ∂v r + z = 0 (1.33) r ∂r ∂z
Stuart developed the disturbances u, v and w in Fourier series, and then the series were
substituted into Equations (1.32a) through (1.32c). After some mathematic
manipulations and the application of the boundary conditions, then he obtained the
following disturbance energy equation:
∂v v ()− ρ θ − θ = µ ()ξ 2 +η 2 + ζ 2 ∫∫uv rdrdz ∫∫ rdrdz (1.34) ∂r r where the bars denote mean values and the vorticity components ξ, η and ζ are given by
∂v ∂u ∂w 1 ∂ ξ = − η = − ζ = (rv ) (1.35) ∂z ∂z ∂r r ∂r
Equation (1.34) means that the rate of transfer of kinetic energy from the mean flow of the disturbance is balanced by the rate of viscous dissipation of kinetic energy. If assumption of small gap is adopted and a few transformations are applied, Equation
(1.34) yields the formula of equilibrium amplitude:
.5 425 ×10 4 Ta A 2 = 1( − c ) (1.36) e Re 2 Ta where Ae is the amplitude of the disturbance, Re ( = 1R1C/ν) is the Reynolds number, Ta
2 3 2 (= 1 R1C /ν ) is the Taylor number, and Ta c (=1708) is the critical Taylor number.
−1 − Equation (1.36) states that the amplitude Ae is proportional to Re 1 Ta c /Ta , which is in agreement with the results of Landau (1944) for values of Ta close to Ta c, i.e. in the weakly nonlinear regime. 16 Stuart’s energy-balance method included the effect of distortion of the mean motion, but ignored the generation of harmonics of the fundamental mode and the distortion of the velocity associated with the fundamental mode. Davey (1962) found that those were not valid mathematical approximations, so he continued the investigation of weakly nonlinear Taylor vortex flow by a rigorous perturbation expansion. Davey assumed the velocity components of the disturbances in Fourier series in harmonics of the wave number λ:
∞ = λ u ∑un (r,t)cos( n z) (1.37a) n=1
∞ = λ v ∑vn (r,t)cos( n z) (1.37b) n=1
∞ = λ w ∑ wn (r,t)sin( n z) (1.37c) n=1
Davey eliminated p and vz from Equations (1.32) and (1.33) and tried a solution of the remaining two differential equations with the following forms:
∞ = n + 2m un (r,t) A un (r) ∑ A unm (r) (1.38a) m=1
∞ = n + 2m vn (r,t) A vn (r) ∑ A vnm (r) (1.38b) m=1
Then the time-dependent amplitudes are determined from the equation:
∞ 1 dA = 2m = σ ∑ am A a0 (1.39) A at m=0
Following the assumption that the amplitude was small, the amplitude equation
became:
17 = σ + 3 + 5 dA / dt A a1 A (0 A ) (1.40) and the solution of Equation (1.40) is:
2 = σ 2σt − 2σt A K e /( 1 a1 Ke ) (1.41) where K is a constant, which depends on the origin of time t. By using numerical
methods, Davey determined the following expression of the equilibrium amplitude for the
case of small gap:
Ta A 2 = .0 3257 1( − c ) (m = )0 (1.42) e Ta
1.2.5 Literature Review on Weakly Nonlinear Theory
The validity of the amplitude Equation (1.40), which is the major contribution by
Davey (1962), was supported by the torque measurements of some investigators. When the amplitude of the motion is known, the torque can be calculated by Davey’s theory,
Equation (1.43) for the case of small gap.
= − Ω −1 + Ω T0 906 6. 1 51 64. 1 (1.43)
Davey found a good agreement of the torque predicted by the theory and the one
measured by Donnelly (1958) over the range of the Taylor number above the critical
value for which the perturbation theory is expected to be valid. Later on an empirical
formula for calculating torque was given by Donnelly and Simon (1960), Equation
(1.44), which is very similar to Equation (1.43).
= Ω −1 + Ω 36.1 T0 a 1 b 1 (1.44)
where a and b are constants determined from the experimental data.
18 Direct verification of the amplitude of Taylor vortices as a function of the Taylor number was also done by some experimenters. The first such measurements were made by Donnelly and Schwarz (1965). By measuring the amplitude of the radial component
2 of the disturbance, they found that Ae , the square of the amplitude of disturbance, varies linearly with ( Ta-Ta c), the difference between the Taylor number and the critical Taylor number when Ta is close to Ta c. Their experimental data directly verified Davey’s law of
equilibrium amplitude, Equation (1.42).
Table 1.2 Comparison of Coefficients for the Calculation of Taylor Vortex Torques
Authors Η λ A B
DiPrima and Eagles (1977) 1 3.13 9.105 × 10 -4 (-8.3 ± 0.3) × 10 -7
Reynolds and Potter (1967) 1 3.13 9.047 × 10 -4 -8.44 × 10 -7
DiPrima and Eagles (1977) 0.95 3.0 7.884 × 10 -4 (-7.1 ± 0.3) × 10 -7
Kirchgassner and Sorger (1969) 0.95 3.0 7.882 × 10 -4 -6.7 × 10 -7
DiPrima and Eagles (1977) 0.5 3.0 1.421 × 10 -4 (-3.5 ± 0.3) × 10 -8
Kirchgassner and Sorger (1969) 0.5 3.0 1.422 × 10 -4 -4.12 × 10 -8
The quintic term in the amplitude Equation (1.40) was calculated by DiPrima and
Eagles (1977). Kirchgassner and Sorger (1969) also carried the calculations to the same order. According to the results of these two papers, the mean torque on the inner cylinder with the outer one at rest is given by:
π LρV 2 1( +η)ηR T = (G + G ) (1.45) 0 1(2 −η) 2 l t
19 2 where L is the height of the cylinders, Gl = -8/(1+ η) is the dimensionless torque due to
Couette flow, and Gt is the dimensionless Taylor vortex torque. The ratio of Gt to Gl can
be expressed in the form:
= η λ [ − η λ ] + η λ [ − η λ ]2 Gt / Gl A( , ) Ta Ta c ( , ) 2/ B( , ) Ta Ta c ( , ) 4/ (1.46) where A(η, λ) and B(η, λ) are coefficients. Values of A(η, λ) and B(η, λ) obtained by
DiPrima and Eagles (1977), by Kirchgassner and Sorger (1969), and by Reynolds and
Potter (1967) are given in Table 1.2.
1.2.6 Theory of Wavy Taylor Vortices and Literature Review
According to the experimental observations of Coles (1965), the axisymmetric time- independent Taylor vortex flow becomes unstable with the further increase of Taylor number, and the instability leads to non-axisymmetric time-dependent wavy Taylor vortex flow. DiPrima (1961) was the first one who analytically studied the linear problem for the stability to non-axisymmetric disturbances. He assumed the wavy azimuthal disturbances of the form:
u(r,θ, z,t) = u(r)cos( λz)ei(σt+mθ ) (1.47a)
v(r,θ, z,t) = v(r)cos( λz)ei(σt+mθ ) (1.47b)
w(r,θ, z,t) = w(r)sin( λz)ei(σt+mθ ) (1.47c) where m is the integer number of the azimuthal waves. Equations (1.47a) through (1.47c) are introduced into the linearized Navier-Stokes equations, and only the real part of the disturbances is considered. By using the narrow gap approximation and neglecting some small terms, eventually DiPrima derived the following equations:
20 1 [DLD − (λC) 2 L]u = − (λC) 2 Ta 1( + µ)v (1.48) 2
Lv = u (1.49)
= = = = = u v Du 0 at r R1 and r R2 (1.50)
where
L = [D 2 − (λC) 2 ] − i Re[ β + mf (r)] (1.51)
D = d / dr (1.52) and β is the real part of rotation speed and f(r) is a function of position r.
The eigenvalue problem defined by equations (1.48) – (1.52) was solved with a
variational method, which yields that the critical Taylor number for the onset of wavy
vortices increases slightly as the number of waves in the azimuthal direction increases
(from 0 to 3).
Roberts (1965) used a sixth-order system to solve the non-axisymmetric linear problem
for Couette flow, and his calculation supports DiPrima’s result, i.e. the critical Taylor
number is an increasing function of the wave number.
The stability of Couette flow with respect to non-axisymmetric disturbances was
studied by Krueger et al. (1966) in the narrow gap case when both of the cylinders
rotated. They solved the eigenvalue problem by a direct numerical procedure, and it is
found that there is a critical value of 2/ 1 of approximately -0.78 (counter-rotating
cylinders), above which the critical disturbance is axisymmetric and below which it is
non-axisymmetric. The result of Krueger et al. was confirmed by experimental
observations of Andereck et al. (1986).
21 Davey et al. (1968) attacked the problem of instability of the Taylor vortex flow against perturbations, which are periodic both in the axial and azimuthal directions, and moreover travel with some phase velocity in the latter. They made a number of assumptions, of which the most important one was the narrow gap assumption. Davey et al. assumed the azimuthal velocity component is periodic with period 2 π/λ in the
nondimensional vertical direction ζ and periodic with period 2 π/k in the azimuthal
direction φ:
∞ ∞ ζ ϕ τ = τ + τ λζ + τ λζ iqk ϕ v(x, , , ) ∑v0q (x, ) ∑vcnq (x, )cos( n ) vsnq (x, )sin( n )e (1.53) q=−∞ n=1 where x is the nondimensional radial coordinate, τ the nondimensional time, and k =
2 m 0C /ν.
Formulas similar to equation (1.53) were also applied to the u and w velocity
components. The velocities were then expanded in powers and products of the amplitude
Ac(τ), As(τ), Bc(τ), Bs(τ), the amplitudes B belonging to the non-axisymmetric motions.
These expansions should be consistent with the differential equations for u, v and w, and then the amplitude equations can be derived.
Davey et al. found that the Taylor vortex flow is stable against perturbations with the same axial wavelength and phase, but unstable against perturbations differing in phase by
π/2. The critical Taylor number for the onset of wavy vortex flow is found to be about
8% above the critical Taylor number for the onset of axisymmetric Taylor vortex flow.
The azimuthal wave number is found to be one by a slight preference.
The results of Davey et al. are in qualitative agreement with some experimenters’ data.
Cole (1976) observed that the second critical Taylor number (onset of wavy vortices) was
22 8.2% above the first critical Taylor number (onset of Taylor vortices) when the aspect ratio was large ( L/C = 107) and clearance ratio was small ( C/R1 = 0.0478). Schwarz et al.
(1964) observed that the second critical Taylor number was 5% above the first one, and
the azimuthal wave number was one.
However, the preference of the wave number to be one was disproved by some
investigators. Eagles (1974) calculated the torque corresponding wavy vortex flows with m = 1, 2, 3 and 4 for the case of η = 0.95. His numerical results and the experimental results of Donnelly (1958) and Debler et al. (1969) for the torque agree quite well with the theory of Davey et al. (1968) of wavy vortex flow with m = 4 (four azimuthal waves).
This suggests that the m = 4 nonaxisymmetric mode does indeed dominate the other
nonaxisymmetric modes to produce a wavy flow with four waves in the azimuthal
direction.
Experimental data show that the second critical Taylor number (onset of wavy
vortices) depends strongly on the radius ratio η. It is 5% - 10% above the first critical
Taylor number (onset of Taylor vortices) for η = 0.95, and is very much larger, ten times of the first Taylor number or greater, for η = 0.5 according to Snyder (1969) and Debler
et al. (1969).
Numerically, Eagles (1971) extended the application by dropping the narrow gap
approximation. He calculated the stability of Taylor vortex flow by using fifth-order
terms in amplitude, and he found that the instability to the Taylor vortex flow to non-
axisymmetric disturbances at about 10% above the first critical Taylor number when η =
0.95.
23 Dando et al. (2000) used a multiple scales method to derive an equation governing the stability of Taylor vortices in a high Reynolds number flow through a curved channel of small gap. They showed that this stability of Taylor vortices is governed by a modified
Burgers equation:
+ 2 = + λ U t U U x U U xx (1.54) where x is the circumferential direction coordinate and λ is the wave length in the axial
direction. Numerical solutions of this equation show that a family of periodic traveling
waves of period 2 π/N, where N is an integer, can be calculated from a primary 2 π- periodic traveling wave by a simple transformation. It was found that the primary traveling waves are the only stable ones emerging at large times for a given initial value problem. They also showed that the final stable steady state traveling waves become increasingly steeper as the length of the system is increased.
Riechelmann and Nanbu (1997) assumed that the smallest scale of turbulence, i.e. the size of the smallest vortices could be considered being still some orders larger than the molecular mean free path λ. Then they applied the direct simulation Monte Carlo
(DSMC) method, which solves flow problems on molecular level, thus avoiding turbulence models. Riechelmann and Nanbu (1997) found that the wavy vortices were seen to move in the azimuthal direction and the frequency of the waves was in agreement with the experimental observations. They were the first investigators to simulate the wavy Taylor flow by means of a molecular approach.
The flow between two cylinders usually experiences Couette, Taylor and wavy vortices as the speed of the inner cylinder increases. This is observed by most of the investigators both numerically and experimentally. However, Lim et al. (1998) reported 24 that there exists a new flow regime when the acceleration of the inner cylinder ( dRe/ dt ) is
higher than a critical value of about 2.2 s -1. In this regime the flow pattern shows
remarkable resemblance to regular Taylor vortex flow but is of shorter axial wavelength.
However, when the acceleration is lower than 2.2 s -1, a wavy flow is found to occur for
the same Reynolds number range as usually observed. This is the first time that such a
phenomenon has been observed.
1.3 Theories of Fluid Instability in Eccentric Cylinders and Literature Review
In this section, theories of fluid instability in eccentric cylinders, i.e. local and nonlocal
theories, and the corresponding literature review will be introduced.
1.3.1 DiPrima’s Local Theory and Literature Review
DiPrima (1963) investigated analytically both the concentric and eccentric cases for
1 ≤ C ≤ 1 small clearances that correspond to the range of 3 2 . He assumed that for the 10 R1 10 concentric case with the inner cylinder rotating and the outer one stationary, the Couette flow will make a transition to Taylor vortex flow first, if the Taylor number Ta
R Ω C R Ω C (= 1 1 C / R ) reaches 41.3 before the Reynolds number Re (= 1 1 ) attains ν 1 ν
2,000. However if Re = 2,000 is reached before Ta reaches its critical value of 41.3, then the flow will make a direct transition to turbulence. The dividing point would be a value of C/R1 of roughly 1/2500 (or 0.000426); for smaller values turbulence will develop directly and for larger values Taylor vortices will occur before the appearance of
25 turbulence as the velocity is increased. The relationship among Reynolds number, Taylor number, curvature effect and flow pattern is shown in Figure 1.4. For most journal bearings C/R1 will generally be greater than 1/2500 by a factor of 2 to 5, so the instability
will be in the form of Taylor vortices if the flow is unstable.
Laminar Flow:
Transition Flow:
Turbulent Flow:
0 131 400 1000 1306 2000 2920 3000 Re
Ta C/R=0.1 0 41.3 316.2 632.5 948.7 CurvatureEffect Ta C/R=0.01 0 41.3 100 200 300
Ta C/R=0.001
0 31.6 41.3 63.2 94.9
Ta C/R=0.000426 0 20.6 41.3 61.9
Ta C/R=0.0002 0 14.1 28.3 41.3 42.4
Figure 1.4 The Relationship among Reynolds Number, Taylor Number, Curvature Effect and Flow Pattern
26 For the eccentric case, DiPrima superimposed a circumferential pressure gradient on the basic Couette flow, and then the laminar velocity distribution is given by:
y 1 ∂P V (y,θ ) = V 1( − ) + (y 2 − yh ) (1.55) 1 µ ∂θ h 2 R1
where P/ θ denotes the circumferential pressure gradient, V1 is the velocity of inner cylinder, y is the radial coordinate measured from inner cylinder, and h = C (1 + ε cos θ) is the film thickness.
200
Rr C / R1 160
120
80 Unstable 40 Stable RP C / R1 0 -120 -80 -40 0 40 80 120 -40 Unstable -80
-120
-160
-200
Figure 1.5 A Plot of Values of Rr C / R1 & RP C / R1 for Stable & Unstable Flow
27 DiPrima (1963) shows that the stability of the flow to disturbances of the Taylor vortex
type is governed by the values of the two parameters, Rr C / R1 and RP C / R1 , where
Rr (= V1C/ν) is a Reynolds number based on the velocity of the inner cylinder, and
− C 2 ∂P C V C R = = P (1.56) P µ ∂θ ν ν 12 R1
is a Reynolds number based on the average velocity due to the circumferential pressure
gradient. The stability diagram is shown in Figure 1.5. Notice that for RP = 0, Taylor’s
criterion of Ta = 41.3 is observed.
By using the Sommerfeld solution of the Reynolds equation for the infinitely long journal bearing according to Pinkus and Sternlicht (1961), the pressure gradient can be expressed by:
∂P 2C 1( − ε 2 ) = 6µV R (h − h /) h3 h = (1.57) ∂θ 1 1 0 0 2 + ε 2
where h0 is the film thickness at the cross section where P/ θ = 0. Then DiPrima (1963)
calculated the local Reynolds numbers corresponding to the velocity and pressure
gradient in terms of ε and θ:
h = + ε θ /3 2 Rrh R 1( cos ) (1.58a) R1
h 1(2 − ε 2 ) 1( + ε cos θ ) 2/1 R = −R 1( + ε cos θ ) − (1.58b) Ph + ε 2 R1 2 2 where
V C C R = 1 (1.59) ν R1
28 Combining Equations (1.58) and Figure 1.5, the critical value of R as a function of ε
and θ can be derived, Figure 1.6, which is also called DiPrima’s local theory. In Figure
1.6, the variation of Rθ(ε) with ε is shown for several values of θ. It shows that the flow
to be the least stable at the position of maximum film thickness θ = 0˚. The critical
Taylor number Ta first decreases, as the eccentricity ratio ε increases from 0, and remains below its concentric value in the range of 0 < ε < 0.6. For ε < 0.6, Ta
increases rapidly as ε increases.
C Rr R1 100
90 θ = 0˚ θ = 60˚ 80 θ = 90˚ θ = 180˚ 70
60
50
40
30 0.0 0.2 0.4 0.6 0.8ε 1.0
Figure 1.6 Critical Value of R (C/R )1/2 as a Function of ε for Various Values of θ r 1
However, DiPrima’s local theory does not match with some experimental results. Both
Vohr’s (1968) and Cole’s (1957, 1965) experimental data show that the critical speed
increases monotonically with eccentricity, Figure 1.7. Based on both of the author’s
experimental results, Vohr also found that the critical speed for onset of vortices in flow 29 between eccentric cylinders does not depend significantly on clearance ratio when C/R1 is
greater than 0.1; whereas it depends significantly on clearance ratio when 0 ≤ C/R1 ≤ 0.1,
Figure 1.7. Concerning the application of the results to cylinders having clearance ratio less than 0.01, which is usually the case for bearings, Vohr suggested that the experimental curve for C/R1 = 0.0104 be used as an upper bound curve for flow stability
while DiPrima’s theoretical curve be used as a lower bound.
Vohr also found that the maximum intensity of the vortex motion is not at the position
θ = 0˚, but at a position θ = 50˚ downstream from the maximum clearance.
C R r R 140 1
130 DiPrima,C/R1 →0 Vohr, C/R1=0.0104 120 Vohr, C/R1=0.099 Cole, C/R1=0.27 110 Cole, C/R1=0.48
100
90
80 Upper bound 70 60
50
40 Lower bound 30 0.0 0.2 0.4 0.6 0.8ε 1.0
Figure 1.7 Critical Speed as a Function of ε for Various Values of Clearance Ratio
30 1.3.2 DiPrima and Stuart’s Non-local Theory and Literature Review
DiPrima’s local theory is based on the assumption that the effect of azimuthal variation of the tangential velocity is neglected. By considering the tangential velocity as a function of the radius ‘ r’ and θ, DiPrima and Stuart (1972) derived the linearized instability equations, which are partial differential equations rather than ordinary differential equations. By letting the two small parameters, the clearance ratio C/R1 and
1/2 the eccentricity ratio ε, tend to zero in such a way that ( C/R1) is proportional to ε, they
obtained a global solution to the stability problem. Their result, which is also called
linear nonlocal theory, shows that the critical speed increases monotonically with the
eccentricity ratio ε according to the relationship:
= + C ()+ ε 2 + C ε 2 C 2 ε 4 Ta c 1695 1 .1 162 1 .1 125 0 (, ) , (1.60) R1 R1 R1
The agreement between equation (1.60) and Vohr’s (1968) experimental data for the case of C/R1 = 0.0104 is excellent for values of ε up to about 0.5, which indicates that the term of order ε4 is needed in equation (1.60) for calculating the critical Taylor number
when ε is larger than 0.5. However, agreement between (1.60) and experimental data for
the case of large clearance ratio, Vohr (1968) for the case of C/R1 = 0.099 and Kamal
(1966) for the case of C/R1 = 0.0904, is good for values of ε up to about 0.2 only. This
indicates that the term of order ε4 in equation (1.60) is more significant for the case of
large clearance ratio.
DiPrima and Stuart’s linear nonlocal theory also indicates that the maximum intensity
of vortex motion is at θ = 90˚ downstream from the maximum clearance.
31 Later on, DiPrima and Stuart’s (1975) nonlinear nonlocal theory found that the angular position of maximum vortex intensity can be any value between θ = 0˚ and θ = 90˚,
depending on the value of the supercritical Taylor number T1, which is defined as T1 =
(Ta – Ta c)/ ε. Also they found good agreement with Vohr’s observation of θ = 50˚ for maximum vortex intensity for the case of C/R1 = 0.099 and ε = 0.475.
C Rr R1 70 DiPrima Vohr, C/R1=0.0104 60 DiPrima and Stuart Upper bound
50
40 Lower bound
30 0.0 0.2 0.4 0.6 0.8ε 1.0
Figure 1.8 Lower and Upper Bounds for Instability in Bearing-like Clearance Ratios
Many experimental studies for Taylor vortex transition of eccentric cylinders have been performed, but there are no conclusive results on the verification of the two theories, local and nonlocal. Measurements made with large clearance ratios, Vohr’s (1968) and
Cole’s (1957, 1965), appear to agree better with nonlocal theory. However, measurements made with bearing-like clearance ratios, Frene and Godet (1971, 1974)
(C/R1 = 0.005), appear to support local theory. From a fundamental point of view, the
problem remains unresolved. However, from a practical point of view, Gross et al. 32 (1980) concluded that the two theories of DiPrima (1963) and of DiPrima and Stuart
(1972) can safely be said to bracket the transition zone applied to the bearing-like clearance ratios. The conclusion of Gross et al. is very close to Vohr’s suggestion that the experimental curve for C/R1 = 0.0104 be used as an upper bound while DiPrima’s
theoretical curve be used as a lower bound, Figure 1.8.
1.4 Theories of Turbulence in Bearings and Literature Review
In this section, theories of turbulence in bearings, i.e. Constantinescu, Ng and Pan, and
Hirs’ theories and the corresponding literature review will be introduced.
1.4.1 Constantinescu’s Theory
Constantinescu’s (1959) turbulent lubrication theory starts from the simplified
turbulence version of Navier-Stokes equations:
∂ ∂ ∂ P = µ U − ρ u v'' (1.61a) ∂x ∂y ∂y
∂ ∂ ∂ P = µ W − ρ v'w' (1.61b) ∂z ∂y ∂y
Then Prandtl’s mixing length hypothesis is introduced by Equation (1.62), i.e. the
Reynolds stress component − ρu v'' is proportional to the square of the mixing length l, and Equation (1.63), i.e. the mixing length vanishes at the walls and to vary linearly with the distance from the nearest wall.
∂U ∂U − ρu v'' = ρl 2 (1.62) ∂y ∂y
33 l = ky (0 ≤ y ≤ h 2/ ) (1.63a)
l = ky ' (0 ≤ y'≤ h 2/ ) (1.63b) where y' = h – y. Introducing Equation (1.63) into Equation (1.62) and introducing
Equation (1.62) into Equation (1.61a), the governing equation for a long bearing
( ∂P = ,0 W = 0 ) is obtained. For 0 ≤ y ≤ h/2 this equation has the non-dimensional ∂z
form:
2 ∂ 2 ∂u ∂u ∂u h ∂P 2 + − = (1.64) k y Re h 0 ∂ y ∂ y ∂ y ∂ y µU ∂x where
= ρ µ Re h Uh / (1.65)
y = y / h (1.66)
u = U /U (1.67)
U = U2 is velocity of the rotating surface in the x direction relative to the stationary
bearing surface ( U1 = 0).
By integrating Equation (1.64) one has the following equation:
2 ∂u ∂u ∂u Ay + + B y − C = 0 (1.68) ∂ y ∂ y ∂ y where
= 2 A k Re h (1.69)
h 2 ∂P B = − (1.70) µU ∂x
34 The integration constant C in Equation (1.68) represents the dimensionless wall stress
(∂ ∂ ) = τ µ → [Equation (1.68) reduces to C = u / y y=0 h w / U as y 0 ]. Constantinescu chose
to follow Prandtl and divided the flow regime 0 ≤ y ≤ 2/1 into two layers. He assumed
≤ ≤ that the effect of the Reynolds stress is negligible in the viscous sublayer 0 y y L , and
then Equation (1.68) reduces to
∂u + B y − C = 0 (1.71) ∂ y
≤ ≤ whereas in the turbulent outer layer y L y 2/1 the effect of molecular viscosity is negligible and then Equation (1.68) becomes
2 ∂u ∂u Ay + B y − C = 0 (1.72) ∂ y ∂ y
Equations (1.71) and (1.72) need to be solved simultaneously, and for example, at the position of maximum film pressure ( B = 0) the velocity is:
u = 1− C y' 0 ≤ y'≤ (CA ) − 2/1 (1.73a)
C − u = 1− [1+ ln ()y' CA ] (CA ) 2/1 ≤ y'≤ 5.0 (1.73b) A
C − u = [1+ ln ()y CA ] (CA ) 2/1 ≤ y ≤ 5.0 (1.73c) A
u = C y' 0 ≤ y ≤ (CA ) − 2/1 (1.73d)
where C is given by the transcendental equation
C CA 1− 2 + ln = 0 (1.74) A 4
35 A similar analysis is performed to the axial flow, which is decoupled from the circumferential flow by Constantinescu. Finally the velocities in both the axial and circumferential directions are introduced to the continuity equation, and one gets the differential equation that governs the pressure distribution in a turbulent lubricant film,
Constantinescu (1967):
∂ h3 ∂P ∂ h3 ∂P U ∂h + = (1.75) ∂ µ ∂ ∂ µ ∂ ∂ x k x x z k z z 2 x where
= 1 = + .0 8265 k x 12 .0 0260 (Re h ) (1.76a) Gx
= 1 = + .0 741 k z 12 .0 0198 (Re h ) (1.76b) Gz
Another important parameter affected by turbulence is the shear stress τs acting on the rotating inner cylinder. The expression for τs under a laminar flow condition is
h ∂p τ = τ + (1.77) s c 2 ∂x where
µU τ = (1.78) c h is the Couette shear stress. Onset of turbulence increases both τc and p/ x. The increase of p/ x is calculated through the solution of the turbulence Reynolds equation (1.75),
and the increase of τc is calculated by the following expression:
µU τ = C (1.79) c f h
36 where Cf is a turbulence Couette shear stress factor. The expression of Cf is given by
Constantinescu and Galetuse (1965):
= + .0 855 C f 1 .0 0023 (Re h ) (1.80)
1.4.2 Ng-Pan Theory
Ng (1964) and Ng and Pan (1965) also started from the simplified turbulence version of Navier-Stokes Equations (1.61a) and (1.61b). Substituting Equations (1.81a) and
(1.81b) into Equations (1.61a) and (1.61b) respectively, and integrating twice with respect to y, then the velocity distributions in circumferential and axial directions are
obtained, Equations (1.82a) and (1.82b).
∂U ε ∂U τ = µ − ρu v''= 1+ m (1.81a) xy ∂y ν ∂y
∂W ε ∂W τ = µ − ρv'w' = 1+ m (1.81b) zy ∂y ν ∂y
1 h y dy ' ∂P y y'−h 2/ U = τ + dy ' (1.82a) µ xy ∫0 + ε ν ∂ ∫0 + ε ν 2 1 m / x 1 m /
1 h y dy ' ∂P y y'−h 2/ W = τ + dy ' (1.82b) µ zy ∫0 + ε ν ∂ ∫0 + ε ν 2 1 m / z 1 m /
where τxy (h/2) and τzy(h/2) are integration constants to be determined from the boundary
conditions at y = h.
Then Ng and Pan introduced Reichardt’s empirical formula of eddy viscosity profiles:
+ ε + + y m = k y − δ tanh (1.83) ν l δ + l
37 where
y + = y (τ / ρ) /ν (1.84)
δ + k is mixing length constant and l is a constant related to the thickness of the laminar
sublayer. Introducing Equation (1.83) into Equations (1.82a) and (1.82b), and then
substituting the velocity distributions into the continuity equation, the turbulent
lubrication equation is obtained. This equation is identical to Constantinescu’s Equation
(1.75), but the coefficients kx and kz are different:
1 1 y 2/1 −η g (η) = G = d y 1− c dη (1.85a) x ∫0 ∫ 0 ()η ()η k x f c f c
−η 1 = = 1 y 2/1 η Gz d y d (1.85b) ∫0 ∫ 0 ()η k z f c where
h + f ()y = 1+ k hy + − δ + tanh y c (1.86) c c l δ + l
+ 1 + h g ()y = k hy tanh 2 y c (1.87) c c δ + 2 l
y = y / h (1.88)
+ = (τ ρ ) ν hc h c / / (1.89)
The expressions of coefficients kx and kz are too complicated because of the double
integration in Equation (1.85). However, by least-squares fitting of polynomials to
Equation (1.85), Taylor (1970) obtained a much simpler expression:
1 = = + nx k x 12 K x (Re h ) (1.90a) Gx 38 1 = = + nz k z 12 K z (Re h ) (1.90b) Gz where the constants in Table 1.3 can be used over the appropriate range of Reynolds number.
Elrod and Ng (1967) also derived an expression for τs under turbulence flow condition:
µU h ∂p τ = C + (1.91) s f h 2 ∂x and the turbulence Couette shear stress factor Cf is curve-fitted by Taylor (1970) to give
the following expression:
= + 86.0 > C f 1 .0 00232 (Re h ) Re 10 , 000 (1.92a)
= + 96.0 ≤ C f 1 .0 00099 (Re h ) Re 10 , 000 (1.92b)
Table 1.3 Constants for Calculating k x and k z at Different Reynolds Numbers
Reynolds Number Kx nx Kz nz
50,000 < Re 0.0388 0.80 0.0213 0.80
10,000 ≤ Re < 50,000 0.0250 0.84 0.0136 0.84
5,000 ≤ Re < 10,000 0.0250 0.84 0.0088 0.88
Re < 5,000 0.0039 1.06 0.0021 1.06
1.4.3 Bulk Flow Theory of Hirs
The bulk flow theory of Hirs (1973) is primarily based on the empirical finding that the relationship between wall-shear stress and average velocity can be expressed by a simple formula for pressure flow, for drag flow, and for combinations of these two basic types of
39 flow. Thus two basic formulas, Equations (1.93a) and (1.93b), can be derived based on that finding.
τ u h m0 0 = n m (1.93a) ρ 2 0 ν 2/1 um
τ u h m1 1 = n m (1.93b) ρ 2 1 ν 2/1 um
where τ0 is shear stress at a surface due to flow under the influence of a pressure gradient,
τ1 is shear stress at a surface due to the sliding of a surface, um is the mean velocity, and n0, m0, n1 and m1 are empirical constants to be fitted to the available experimental results.
Experimental results by Hirs (1974) show that m0 = m1 = -0.25 and a = n0/n1 =1.2.
Then after he introduced a fictitious pressure gradient (d P1/d x) to account for the occurrence of the drag flow component, two formulas, one for the stationary surface,
Equation (1.94a), and another for the sliding surface, Equation (1.94b), can be obtained.
d − h ( p + p ) m0 1 u h dx = n m (1.94a) ρ 2 0 ν um
d − h ( p − p ) m0 1 (u −U )h dx = n m (1.94b) ρ − 2 0 ν (um U )
On eliminating the fictitious pressure gradient d P1/d x between Equations (1.94a) and
(1.94b), one can obtain the actual pressure gradient in terms of the average velocity um
and sliding speed U:
m m d p n ρu 2 u h 0 ρ(u −U ) 2 (u −U )h 0 = − 0 m m + m m (1.95) ν ν dx 2 h h
40 Equation (1.95) is valid for the flow in the direction of the representative pressure gradient, but this direction need not coincide with the direction of the sliding speed U [ x
direction in Equation (1.95)], in which case there will be two component equations, one
in the x direction and the other in the z direction. Similar analysis can be done to get the
actual pressure gradient in both x and z directions. Finally the dimensionless pressure
flow coefficients Gx and Gz can be obtained from Equations (1.96a) and (1.96b) by substitution.
2/1 − u /U G = mx (1.96a) x (h 2 / µU )( ∂P / ∂x)
− u /U G = mz (1.96b) z (h 2 / µU)( ∂P / ∂z)
where umx and umz are the mean velocities in the x and z directions respectively. If the
pressure flow component is much smaller than the drag flow component, which is the
case of self-acting bearings operating at moderate eccentricities, Equation (1.96) then
reduces to
1( +m ) 2 0 − + G = Re 1( m0 ) (1.97a) x + h n0 2( m0 )
1( +m ) 2 0 − + = 1( m0 ) Gz Re h (1.97b) n0
where n0 = 0.066 and m0 = -0.25 for smooth surfaces at Reynolds number smaller than
105. With these values Equation (1.97) gives the approximate formulas:
= 1 = 75.0 k x .0 0687 Re h (1.98a) Gx
41 = 1 = 75.0 k z .0 0392 Re h (1.98b) Gz
The values of kx and kz must not be allowed to fall below 12, which is the value appropriate to laminar flow. The limiting Reynolds numbers for this situation are:
= 1 = = k x 12 Re 977 (1.99a) Gx
= 1 = = k z 12 Re ,2 060 (1.99b) Gz
The turbulence Couette shear stress factor Cf is given by the following expression:
= + 75.0 C f 1 .0 00818 (Re h ) (1.100)
and Cf = 1, the value appropriate to laminar flow, when Re ≤ 607.
1.4.4 Literature Review on Theories of Turbulence in Bearings
Constantinescu’s (1959) theory is the first turbulent lubrication theory, but there are
some apparent limitations in his approach. The most obvious one is that the buffer zone
between the laminar sublayer and the fully turbulent region receives no attention. This
should cause a discontinuity in the shear stress, which is not shown in his theory.
Ng and Pan’s (1965) theory predicts the same form of turbulent Reynolds equation as
Constantinescu’s theory. However, the work of the former allows consideration of
Couette and Poiseuille flows and combinations of the two and is generally accepted to be
more accurate than that of Constantinescu. The theory obtained by Hirs (1973) is
completely different and is based solely on experimental observations. It does not give a
42 turbulent Reynolds equation directly but its results can be cast into the form used by
Constantinescu (1959) and Ng and Pan (1965).
As shown in Figures 1.9 and 1.10, the law of the wall approach by Ng and Pan is in excellent agreement with the predictions of the bulk flow theory by Hirs when the
Reynolds number is greater than 10 4. There is a discrepancy between the two approaches
in the transition region, which is after the onset of Taylor vortices and before the
Reynolds number reaches 10 4. Hirs claims that in this region the law of wall used by Ng and Pan is not in good agreement with experimental data. Since Hirs’ bulk flow theory is directly based on experimental evidence it is claimed that it is more applicable in this region and in general the most reliable turbulent lubrication theory.
0.1
Gx = 1/k x Gz
Gz = 1/k z G x
0.01
Constantinescu Ng-Pan Hirs Constantinescu Ng-Pan Hirs
0.001 100 1000 10000Re 100000
Figure 1.9 Turbulence Coefficients G x and G z vs. Reynolds Number
43
0.1
8Cf /Re Constantinescu Ng-Pan Hirs
0.01
0.001 100 1000 10000Re 100000
Figure 1.10 Friction Factor vs. Reynolds Number
In summary, the three turbulence models are based on well-established experimental results. None of these models, however, accounts for the inertia effects as well as the effect of the onset of the Taylor and wavy vortices. They use the Prandtl mixing length and the corresponding eddy viscosity concepts that imply the existence of a turbulent regime, which causes increased viscosity effects. The turbulent lubrication factors, Gx
and Gz, [used in Constantinescu (1959), Ng and Pan (1965) and Hirs (1973)] account for the apparent increased viscosity due to the eddy effects. This approach accurately predicts the flow behavior and the three models are relatively close to each other when
“turbulence” is truly fully developed, i.e. Reynolds number is larger than 2,000.
However, in the transition regime, i.e. after the onset of Taylor vortices and before the full development of turbulence, the discrepancy amongst these most accepted models is significant. Note that the transition regime is very important for both oil and air bearings, 44 because a majority of them operate in this regime, often mistaken for the turbulent one.
Figure 1.11 shows the relationship between the transition regime and the operating ranges for typical oil and air bearings. Thus, for oil bearings (R = 1 to 3 in., C = 0.001 to 0.003
in. and ν = 1×10 -5m2/s), operating in ranges from 3,000 rpm to 30,000 rpm, the Reynolds number varies from 20.3 to 1,824.1 and the Ta number varies from 0.6 to 57.5. For air
bearings ( R = 1 to 3 in., C = 0.001 to 0.003 in. and ν = 1.66×10 -5m2/s), operating in ranges from 30,000 rpm to 100,000 rpm, number Re varies from 122.1 to 3,662.9 while the Ta number varies from 3.9 to 115.8.
20.3 122.1 1824.1 2000 3662.9 Re
air bearings transition regime
oil bearings
0.6 3.9 41.3 57.5 115.8 Ta
Figure 1.11 Operating Ranges for Typical Oil and Air Bearings
Neither of these turbulence models [Constantinescu (1959), Ng and Pan (1965) and
Hirs (1973)] agrees completely with the calculations of Ho and Vohr (1974), who used
the Kolmogoroff-Prandtl energy model of turbulence, but Ng-Pan’s model (1965) comes
closest. Owing to the uncertainty of the flow behavior predicted by these turbulence
models in the transition regime, Gross et al. (1980) recommended the following
45 procedure: i) use Reynolds equations when Re < 41.3 R /C (before the onset of Taylor
vortices); ii) use Ng-Pan model when Re > 2,000 (fully developed turbulent flow); iii) in
between these critical values of Re, interpolate linearly using the Re number as the
interpolating variable. The lower limit for Gx and Gz is that Gx = Gz = 1/12 (laminar values) while the upper one is a function of Re, as Re reaches 2,000. Gross et al. didn’t justify their recommendation but rather forwarded it as a “common sense” approach.
Black et al. (1975) realized the significance of the presence of the Taylor (and wavy) vortices and they developed a theory covering both the vortex and turbulent regimes.
According to this model, the Reynolds equation is expressed in terms of effective viscosities, where the latter have different expressions in the vortex and turbulent regimes, respectively. Application of the theory of Black et al. (1975) to the calculation of load capacity, bearing friction, torque, and lubricant flow rate is compared with
experimental results for a Taylor number range between 2 Ta cr and 10 Ta cr , corresponding to Reynolds numbers between 1,620 and 8,102. For this range there is good agreement between theory and experiments. Constantinescu (1975) and
Constantinescu and Galetuse (1974) realized the importance of the inertia forces, and added an extra term I* to the turbulence model to account for their effect on the momentum equations. According to Constantinescu (1975) and Constantinescu and
Galetuse (1974), I* is a function of film thickness, average velocities, and turbulent lubrication factors and itself includes several correction coefficients. Constantinescu
(1975) concluded that “fluid films operating at large Reynolds numbers are subjected not only to transition and turbulence but also to inertia forces effects”.
46
CHAPTER II
SCOPE OF WORK
2.1 Numerical Simulations and Calculations of Long and Short Bearings
The numerical simulations and calculations of long and short bearings will be
performed by using a commercial code, CFD-ACE+ (2003), produced by ESI Group.
The work will focus on the structure of the fluid flow patterns in a section of fluid film
between two cylinders (finite axial length, 0˚ ≤ θ ≤ 360˚) where the inner one is rotating, and the outer one is stationary. To simulate the infinitely long bearings, periodic boundary conditions are applied to the two ends of the section of fluid film. To simulate the short bearings, zero pressure boundary conditions are applied to the two ends of the finite section of fluid film. The numerical simulations will concentrate on the flows in a micro-scale clearance (0.13 in. to 0.005 in.) with various eccentricity ratios (0.0 to 0.8) between the two cylinders. The work will concentrate on the flow pattern change, velocity profile change and torque generation under conditions when speed, clearance and eccentricity are changed on a parametric basis.
2.2 The Nature of “Instability” and “Turbulence” in Small Gap Journal Bearings
It will be shown from the results that there are two critical points for the onset of: (i) the first Taylor instability (appearance of Taylor vortices) and (ii) the second Taylor 47 instability (appearance of wavy Taylor vortices), as they are indicated by both the flow pattern changes and the inflections of the torque-speed curve. The inflection points in the torque-speed graphs coincide with the flow pattern changes. The Reynolds and Taylor numbers will be calculated for these inflection points and the onset of flow instabilities
(“turbulent” flow called by some literature) will be discussed versus the critical values of these dimensionless numbers. It will be shown that the slope change in the torque-speed graph is due to the change of the average velocity gradient on the outer cylinder wall.
This finding and not an increase in apparent viscosity is the cause for the inflection points on the torque-speed graph. Eventually, the question of “Are the flow formations observed in Taylor instability regimes the actual ‘turbulence’ as it is presently modeled in micro-scale clearance flow?” will be answered.
2.3 New Models for Transition Flow of Thin Films in Long and Short Bearings
The velocity profiles, pressure profiles across the radial direction and the average velocity gradients on the outer cylinder wall will be presented in a quantitative manner for both Taylor and wavy vortex regimes or the transition regime. Order of magnitude analyses of the Navier-Stokes equation component terms will be performed for the transition regime. Two new models (one for long bearing and the other for short bearings), which are fundamentally different from the three most accepted turbulence models [Constantinescu (1959), Ng and Pan (1965) and Hirs (1973)], for predicting flow behavior of thin films in transition regime of long and short journal bearings will be proposed and justified. A comparison will be made between the results of our models and the three turbulence models. 48 2.4 Experimental Verification of the Numerical Results
Experiments will be performed to verify the numerical results. The work will concentrate the flow visualizations and the torque measurements, which will be done simultaneously. It will be shown that numerical calculations and the experimental results appear to be in good agreement with each other. Applications of the new model of long bearings to the calculation of torque will be compared with experimental results in the range of Taylor and wavy vortex regimes.
49
CHAPTER III
NUMERICAL ALGORITHM
3.1 General Introduction to CFD-ACE+
CFD-ACE+ (2003) is a set of computer programs for multi-physics computational analysis. It introduces several modules, such as flow module, heat transfer module, and turbulence module etc, which are governed by different partial differential equations or
PDEs.
The numerical method to solve these PDEs consists of the discretization of the PDEs on a computational grid, the formation of a set of algebraic equations, and the solution of the algebraic equations. The numerical method yields a discrete solution of the field, which is comprised of the values of the variables at the cell centers. In this chapter, the numerical methodology adopted in CFD-ACE+ and flow module, the only module used in present work will be presented.
3.2 Numerical Methodology Adopted by CFD-ACE+
In this section, the numerical methodology adopted by CFD-ACE+, i.e. the discretization of the differential equations, the method of the velocity-pressure coupling, the application of the boundary conditions, and the solution method will be introduced in detail. 50 3.2.1 Discretization
To start the numerical solution process, discretization of the differential equations is introduced to produce a set of algebraic equations. In CFD-ACE+, the finite-volume approach is adopted due to its attractive capability of conserving solution quantities. The solution domain is divided into a number of cells known as control volumes. In the finite volume approach of CFD-ACE+, the governing equations are numerically integrated over each of these computational cells or control volumes. An example of one such control volume is shown in Figure 3.1. The geometric center of the control volume, which is denoted by P, is also often referred to as the cell center. CFD-ACE+ employs a collocated cell-centered variable arrangement, i.e. all dependent variables and material properties are stored at the cell center P. In other words, the average value of any quantity within a control volume is given by its value at the cell center.
Figure 3.1 A Three-dimensional Computational Cell (Control Volume)
Most of the governing equations can be expressed in the form of a generalized transport equation: 51 ∂(ρφ ) r + ∇ • (ρVφ) = ∇ • (Γ∇φ) + Sφ (3.1) ∂t 14243 14243 { 123 convection diffusion source transient
This equation is also known as the generic conservation equation for a quantity φ .
Integrating this equation over a control-volume cell, the equation becomes:
∂(ρφ ) r dϑ + ∇ • (ρVφ)dϑ = ∇ • (Γ∇φ)dϑ + S dϑ (3.2) ∫ ∂ ∫ϑ ∫ϑ ∫ φ ϑ t ϑ
The transient term in Equation (3.2) is discretized as follows,
∂(ρφ ) ρφϑ − ρ 0φ 0ϑ 0 dϑ = (3.3) ∫ ∂ ∆ ϑ t t
where the superscript “0” denotes an older time, while no superscript denotes the current
or the new time. The cell volume, represented by ϑ , may change with time, in particular
when moving grids are used.
Figure 3.2 A Two-dimensional Cell (Control Volumes)
The convection term is discretized as follows: r r ∇ • ρ φ ϑ = ρφ • r = ρ φ n = φ ( V )d V ndA ( e eVe )Ae Ce e (3.4) ∫ϑ ∫A ∑ ∑ e e
52 where subscript “ e” denotes one of the faces of the cell in question, Ae is the area of face
n e, Ve represents the velocity component in the direction that is normal to the face, Ce is thus the mass flux across the face. The evaluation of φ at control volume faces is expressed differently in different schemes.
For ease of illustration, let us consider a two-dimensional control volume as shown in
Figure 3.2. Because the solution variable φ is available only at the cell-centers, the cell-
face values of φ need to be interpolated. Various interpolation schemes with varying
levels of numerical accuracy and stability are used today. In CFD-ACE+ (2003) the user
has a choice of several popular schemes, including first-order upwind, central difference,
second-order upwind, second-order upwind with limiter, smart scheme, or third-order
scheme. After preliminary numerical experimentation third-order scheme was chosen in
present calculations because it took the shortest CPU time to get the converged solutions.
The diffusion term is discretized as follows:
r ∂φ ∇ • (Γ∇φ)dϑ = Γ∇φ • ndA = Γ A (3.5) ∫ϑ ∫ ∑ e ∂ e A e n e
With the three unit vectors defined in Figure 3.2, ∂φ / ∂n can be expressed:
∂φ 1 ∂φ r r ∂φ = r r − e •τ (3.6) ∂n n • e ∂e ∂τ
Then the diffusion term becomes: r r Γ ∂φ τ • eΓ ∂φ ∇ • (Γ∇φ)dϑ = r e r A − r r e A (3.7) ∫ϑ ∑ • ∂ e ∑ • ∂τ e e n e n e e n e e
where
53 ∂φ φ −φ = E P (3.8) ∂ δ e e P,E
∂φ φ − φ = C 2 C1 (3.9) ∂τ δ e C ,2 C1
δ δ where P,E and C ,2 C1 represent the distance between E and P, and C2 and C1, respectively.
If the source term is a function of φ itself, it can be linearized as the following:
U P Sφ = S + S φ (3.10) such that S P is negative. The linearized source term is integrated over the control volume,
which results in:
ϑ = + φ ∫ Sφ d SU S P P (3.11) ϑ
= Pϑ = Uϑ where S P S and SU S .
The numerically integrated transient, convection, diffusion and source terms are assembled together, which result in the following finite difference equation:
− φ = φ + (aP S P ) P ∑ anb ab SU (3.12) nb where the subscripts “ nb ” denote values at neighboring cells, anb are known as the link
coefficients. This finite difference equation is the discrete equivalent of the continuous
flow transport equation that we started with, Equation (3.2).
Equation (3.12), in general, is nonlinear because the link coefficients anb themselves
are functions of φ P, φ nb , etc. When this equation is formulated for each computational
cell, it results in a set of coupled nonlinear algebraic equations. No direct matrix
54 inversion method is available to solve a set of nonlinear algebraic equations. Therefore, an iterative procedure in employed in CFD-ACE+ at every time step.
3.2.2 Velocity-Pressure Coupling
The continuity equation, which governs mass conservation, requires special attention because it can not be written in the form of the general convection-diffusion equation.
Moreover, it is used to determine the pressure field in the pressure-based method, which is employed in CFD-ACE+. The continuity equation can be written in the form:
∂ρ + ∇ • (ρV ) = 0 (3.13) ∂t
Integrating the above equation over the cell in Figure 3.2, the equation becomes:
ρϑ − ρ 0ϑ 0 + ρ V n A = 0 (3.14) ∆ ∑ e e e t e
n where Ve is the face-normal component of the velocity at face e, which is obtained by the inner product of the velocity vector ( u, v, w) and the face-normal unit vector ( nx, ny, nz),
n = + + Ve ue nx ve n y we nz (3.15)
Solutions of the three momentum equations yield the three Cartesian components of
velocity. Even though pressure is an important flow variable, no governing PDE for
pressure is presented. Pressure-based methods use the continuity equation to formulate
an equation for pressure. In CFD-ACE+, the SIMPLEC (Semi-Implicit Method for
Pressure-Linked Equations Consistent) scheme, which was originally proposed by Van
Doormal and Raithby (1984), has been adopted.
The finite difference form of the x-momentum equation can be written as:
55 = + − aPuP ∑ anb uab SU ∑ Pe Ae nxe (3.16) nb P e P
with the subscript P again indicating that the equation is written for cell center P.
The pressure field should be provided to solve the above equation for u, but it is not
known in advance. If the equation is solved with a guessed pressure P*, it will yield
velocity u*, which satisfies the following equation:
* = * + − * aPuP ∑ anb unb SU ∑ Pe Ae nxe (3.17) nb P e P
In general, u* will not satisfy the continuity equation. The strategy is to find corrections to and so that an improved solution can be obtained. Let u' and P' stand for corrections, then:
u = u * + u′ (3.18)
P = P* + P′ (3.19)
An expression for u' p can be obtained by subtracting Equation (3.17) from Equation
(3.16):
′ = ′ − ′ aPuP ∑ anb uab ∑ Pe Ae nxe (3.20) nb P e P
which gives an expression for u' p if u' nb is approximated by u' p:
−1 u′ = P′A n (3.21) P − ∑ e e xe aP ∑ anb e P
Then substitute all the corrected velocity components, Equation (3.18) into Equation
(3.14), a pressure correction equation can be derived:
′ = ′ + aP PP ∑ anb Pab Sm (3.22) nb 56 where Sm represents the mass correction or mass source in the control volume:
ρ 0ϑ 0 − ρ *ϑ S = P − ρ V n A (3.23) m ∆ ∑ e e e t e
The SIMPLEC procedure can be summarized as follows:
1) Guess a pressure field P*.
2) Obtain u*, v*, and w* by solving discretized momentum equation (3.17).
3) Obtain P' by solving Equation (3.22).
4) Calculate P from Equation (3.19).
5) Calculate u, v, and w from Equation (3.18).
6) Solve the discretized equations for other flow variables, such as enthalpy, turbulent quantities etc.
7) Treat the corrected pressure P as a new guessed P*, return to step 2 and repeat the procedure until converged solution is obtained.
3.2.3 Boundary Conditions
Figure 3.3 Computational Boundary Cell
57 A control cell adjacent to the west boundary of the calculation domain is shown in
Figure 3.3. A fictitious boundary node labeled B is shown. The finite-volume equation for node P will be constructed as:
φ = φ + φ + φ + aP P aE E aN N aS S S (3.24)
where coefficient aW is set to zero after the links to the boundary node are incorporated into the source term S in its linearized form:
= + φ S SU S P P (3.25)
φ If the boundary value is fixed as B , the source term is modified as:
= + φ SU SU aW B (3.26)
= − S P S P aW (3.27)
At zero-flux boundaries, such as adiabatic walls for heat and symmetric boundaries for
any scalar variables, the boundary link coefficients are simply set to zero without
modifying source terms.
3.2.4 Solution Methods
CFD-ACE+ uses an iterative, segregated solution method wherein the equation sets for
each variable are solved sequentially and repeatedly until a converged solution is
obtained. The overall solution procedure is shown in Figure 3.4.
Note that all the parameters, which indicate how many times a procedure is repeated,
can be specified by the user. These are the number of iterations (NITER) and the number
of time step (NSTEP) in the case of a transient simulation. The number of iterations to be
performed is dictated by the overall residual reduction desired. At each iteration the
58 program will calculate a residual for each variable, which is the sum of the absolute value of the residual for that variable at each computational cell. In general, a five order of magnitude reduction in the residual is expected before declaring that convergence has been obtained.
Figure 3.4 Solution Flowchart
Under-relaxation of the dependent and auxiliary variables is used to constrain the change in the variable from iteration to iteration in order to prevent divergence of the solution procedure. For all dependent variables, this is done by modifying Equation
(3.12) in the following way,
59 + φ = φ + + φ * aP 1( I) P ∑ anb ab SU aP I P (3.28) nb
* where φ P is the current value of φ P. At convergence, when there is no change in φ P from one iteration to the next, the equation is not modified at all by the addition of these terms. Prior to convergence, however, they provide a link between the new value of φ P
* and the current value φ P . The larger the value of I is, the stronger the under-relaxation
will be. The values of I in the range of 0.2 to 0.8 are common in CFD-ACE+.
The type of linear equation solver is crucial for a solution method because CPU time
and memory requirements strongly depend on it. Two types of linear equation solvers
are available in CFD-ACE+. They are Conjugate Gradient Squared + Preconditioning
(CGS + Pre) Solver and Algebraic MultiGrid (AMG) Solver. However, only AMG
solver is adopted in our calculations due to its two major advantages: a) CPU time only
increase in proportion to the number of unknowns in the equations; b) faster convergence.
The basic idea of AMG is to use hierarchy of girds, from fine to coarse, to solve a set
of equations, with each grid being particularly effective for removing errors of
wavelength characteristic of the mesh spacing on that grid. The solution method can be
illustrated for a two-grid system:
1) On the fine grid (original grid) obtain residual after performing a few iterations,
2) Perform iterations on the coarse grid to obtain corrections, with the fine grid
residual being imposed as a source term,
3) Interpolate the corrections to the fine grid and update the fine grid solution,
4) Repeat the entire procedure until the residual is reduced to the desired level.
60 3.3 Flow Module
CFD-ACE+ offers several modules, such as flow module, heat transfer module, and turbulence module etc, but flow module is the most important one and is also the only module used for most of our numerical calculations.
The flow module allows CFD-ACE+ to simulate almost any fluid (gas or liquid) flow problem. Both internal and external flows can be simulated to obtain velocity and pressure fields.
The governing equations for the flow module represent mathematical statements of the conservation laws of physics for flow, namely, a) The mass of a fluid is conserved; b)
The time rate of change of momentum equals the sum of the forces on the fluid
(Newton’s second law). These two laws can be used to develop a set of equations
(known as the Navier-Stokes equations) for CFD-ACE+ to solve numerically.
The flow module can also be used in conjunction with many of the other modules, such as turbulence module, heat transfer module etc, in CFD-ACE+ to perform multi-physics analyses.
61
CHAPTER IV
NUMERICAL RESULTS (LONG BEARING)
4.1 Introduction
This chapter is concerned the relationship between the onset of Taylor instability and
appearance of what is commonly known as “turbulence” in narrow gaps between two
infinitely long cylinders with various eccentricity ratios. A question that we open to
debate is whether the flow formations observed during Taylor instability regimes are, or
are related to the actual “turbulence” as it is presently modeled in microscale clearance
flows. To start answering this question the viscous flow in small gaps (0.005 in. to 0.13
in.) between two cylinders with eccentricity ratios varying from 0.0 to 0.8, is investigated
using CFD-ACE+ as a computational engine.
On the problem of hydrodynamic stability of rotating fluids between two infinitely
long cylinders with the inner one rotating and the outer one at rest, there are four
controlling parameters: (i) the speed of the inner cylinder ω, (ii) the clearance between
the two cylinders C, (iii) the dimensionless axial wavelength of the instability λ (one can
not simulate the fluid between two infinitely long cylinders, instead one can study one
section of fluid with height L = λC due to the periodic behavior of the fluid instability), and (iv) the eccentricity ratio ε. Strategically and numerically it is impracticable to vary
all the parameters at one time. We have picked λ = 2 after a preliminary investigation 62 (see “The Geometry and Boundary Condition Applications” section in this chapter); the angular velocity ω needs to be changed as the main parameter of the problem. Therefore,
either C or ε needs to be kept constant at any given time. In the following sections the
effect of C and ω while ε = 0 (Section 4.5) and the effect of ε and ω while C = 0.01 in.
(Section 4.6) on the Taylor vortices induced instability will be presented. Finally a new
model for predicting the flow behavior in long journal bearing films in the transition
regime is proposed in Section 4.7.
4.2 Geometry and Boundary Condition Applications
In this section, the geometry adopted in the investigation of the fluid stability problem
and the application of the boundary conditions will be introduced.
4.2.1 Geometry
The viscous fluid flow in the gap between two cylinders with various eccentricities
will be investigated by using CFD-ACE+ as a computational engine. The geometry is
presented in Figure 4.1. Detail shown in c)
ω z R+C θ θ r L r e z R θ
r
a) b) c)
Figure 4.1 Geometric Description of the Cylinders (Not at scale) 63
0.1325 Torque λ = 2.2 (N.m) λ = 2.1 0.1300 λ = 2.0 λ = 1.9 λ = 1.8 0.1275
0.1250
0.1225 Ta 0.5 0.1200 41.0 41.5 42.0 42.5
Figure 4.2 Torque vs. Taylor Number for Various λs when C = 0.01 in. and ε = 0.0
Torque Torque λ = 2 (L/C = 6) (N.m) λ = 2 (L/C = 4) (N.m) 0.27 0.41
0.40
0.26 0.39
0.38 0.25
0.37 0.5 Ta 0.5 Ta 0.24 0.36 41.0 41.5 42.0 42.5 41.0 41.5 42.0 42.5
a) b)
Figure 4.3 Torque vs. Taylor Number for Different Aspect Ratio (L/C = 4 and 6)
64
L
L
L λC λC λC
C C C
a) b) c)
Figure 4.4 Wavelength of Taylor Vortices for Different Aspect Ratios (L/C = 2, 4 and 6)
65 The two cylinders have a finite length ofL = λ C , where λ represents the dimensionless
axial wavelength that we have chosen to be 2. After preliminary numerical
experimentation we chose this axial wavelength because it appears to be the wave length
at which the fluid is most unstable at the lowest rotational speed. For example as shown
in Figure 4.2 for the case of C = 0.01 in. and ε = 0.0, for λ ≠ 2 the critical Taylor number
is always larger than 41.52, which is the critical Taylor number for λ = 2 when C = 0.01 in. and ε = 0.0, therefore the disturbance with λ = 2 will grow first. Once the instability sets in at λ = 2, the other flow instabilities with λ ≠ 2 can not occur.
If the aspect ratio (L/C) is an even integer, the instability with λ = 2 will always set on when the critical speed is reached. The critical Taylor numbers for the cases of L/C = 4 and 6 are 41.52, Figures 4.3a and 4.3b, which is the same critical Taylor number when
L/C = 2. The wavelengths of Taylor vortices for different aspect ratios (L/C = 2, 4 and 6) are shown in Figure 4.4. For other clearances and eccentricities, similar numerical experimentation was also performed and it shows that the Taylor type instabilities with λ
= 2 always set on first.
The radius of the outer cylinder is fixed, R+C = 1.0 in., and the average clearance, C, between the two cylinders varies from 0.13 to 0.005 in., Figure 4.1. Therefore, the radius of the inner cylinder varies from 0.87 to 0.995 in. correspondingly. The centers of the cylinders are set at a distance ‘e’ apart. The corresponding eccentricity ratio, ε = e/C,
varies from 0.0 to 0.8. The angular speed of the inner cylinder ω, the average clearance
C, and the eccentricity ratio ε are the three controlling parameters.
66 4.2.2 Boundary Condition Applications
The top and bottom faces of fluid between the two cylinders are set as periodic boundary conditions, due to the λ axial wavelength characteristic. That means that the
values of the variables and material properties at the nodes on the top and bottom faces
are the same if their coordinates are the same in circumferential and radial directions but
at a distance λ apart in the axial direction. The outer cylinder is set as a non-slip wall boundary condition. The inner cylinder wall is set as a rotating non-slip wall, and the angular velocity vector in the z-direction is prescribed, as shown in Figure 4.1.
4.3 Convergence Criteria and Numerical Accuracy
In this section, the convergence criteria adopted in CFD-ACE+, grid convergence and
time step convergence will be introduced in detail.
4.3.1 Gridding and Grid Convergence
The three-dimensional grid for the fluid is obtained by extruding the two-dimensional
annulus-like x-y grid into the z-direction, Figure 4.1. The number of nodes varies in the
radial direction (12 – 20 nodes), circumferential direction (160 – 600 nodes), and axial
direction (12 – 18 nodes) as a function of clearance and eccentricity ratio. Also the
number of nodes in the circumferential direction for each circumferential quarter (Figure
4.1a shows that the fluid annulus is divided into four quarters) varies with the different
eccentricity ratio. The total number of cells for the entire annulus thus varies from
40,500 to 72,600.
67 CFD-ACE+ uses absolute convergence criteria, which for the pressure field usually,
require convergence of the residual of the order of 1.0E-4. For the cases considered here
we have used a convergence criterion of 1.0E-4 for each primitive variable (u, v, w, and
p). To improve convergence, the under-relaxation factors (URF) are adjusted
individually for each one of these variables. Thus, for velocities and pressure the typical
URFs vary between 0.1 and 0.8.
Table 4.1 Grid Convergence Experiments (C = 0.13 in., ε = 0.0, and ω = 1,550 rpm)
a) b) c)
Grid / Variables 13×108×12 20×160×18 30×240×26
Torque (N.m) 0.1513 0.1541 0.1552
Error (%) 1.85 0.71
Pressure (N/m 2) 679.8 731.8 757.6
Error (%) 7.65 3.53
Grid convergence experiments were also performed in order to ensure that the results presented below are fully converged and are grid independent. Since one had to choose between ultra accurate grid convergence and acceptable computation time for these parametric calculations, we decided that grid convergence within 5.0% is an acceptable threshold. Table 4.1 details the maximum error in torque on the outer cylinder and the generated pressure between three consecutive grids used for calibration of the grid convergence for the case of C = 0.13 in., ε = 0.0, and ω = 1,550 rpm. The baseline grid
was 13×108×12 as can be seen in column a) of Table 4.1. Two other denser grids were
68 also used for the same geometry. Each subsequent grid was increased by a factor of 1.5 per direction. For our computations we chose the grid of 20×160×18 detailed in column b), which yields results that are different at most by 3.53% from the results obtained with the densest grid in column c). Similar grid convergence experiments were also taken for the cases of other clearances, eccentricity ratios and angular speeds.
4.3.2 Time Step and Time Step Convergence
Each time step must be small enough and converged in order to get accurate and converged results for a transient case, which is needed for calculating the wavy vortices.
The time step can be calculated from the CFL [CFL stands for Courant, Friedrichs and
u∆t v∆t w∆t Lewy, Courant et al. (1967)] number: CFL = + + , where u, v, and w are the ∆x ∆y ∆z
characteristic speeds in the circumferential, radial, and axial directions respectively, t is the time step, x, y, and z are the sizes of the control volume in the circumferential,
radial, and axial directions respectively. Generally the allowable CFL numbers for
simple viscous fluid flow with implicit time integration range from 0.1 to 1.2, Courant et
al. (1967). However, for a specific problem one needs to do time step convergence
experiments to optimize the CFL number.
Figure 4.5 details the torque, wave numbers and CPU time at different CFL numbers
for the case of C = 0.13 in., ε = 0.0, and ω = 4,000 rpm ( Ta =120.65). There is no wave
at all or the wave number is 5 when CFL is larger than 0.5. However, the wave number
is always 6 when CFL is smaller or equal to 0.5. CFL of 0.4 is picked for our
calculations by the trade-off between the CPU time and the accuracy of the torque, which
69 is listed in both Figure 4.5 and Table 4.2. Similar time step convergence experiments were also taken for the transient cases of other clearances, eccentricities, and speeds.
0.020
0.018
0.016
0.014
CFL=2.0, No w aves, CPU=1(5.5hr.) 0.012 CFL=1.0, 5 w aves, CPU=4.4 Torque (N.m) Torque CFL=0.625, 5 w aves, CPU=13.1 0.010 CFL=0.5, 6 w aves, CPU=12 CFL=0.4, 6 w aves, CPU=5 0.008 CFL=0.3, 6 w aves, CPU=6.7
0.006 0.00 0.30 0.60 0.90 1.20 Time (s)
Figure 4.5 Torque vs. Time at Different CFLs
(C = 0.13 in., ε = 0.0, ω = 4000 rpm, Re = 323.55, Ta = 120.65)
Table 4.2 Time Step Convergence Experiments (C=0.13 in., ε=0.0, and ω=4,000 rpm)
a) b) c)
CFL/ Variable 0.5 0.4 0.3
Torque (N.m) 0.8306 0.8225 0.8155
Error (%) 0.98 0.85
CPU Time 12.0 5.0 6.7
70 4.4 Multiplicity and Transition of Taylor-Couette Flow
Multiplicity and transition of Taylor-Couette flow are two of the most attractive features in studying the hydrodynamic stability of rotating fluids between two cylinders.
Both of them will be presented in detail in this section.
4.4.1 Introduction
Multiplicity of Taylor-Couette flow is one of the most attractive features in studying the hydrodynamic stability of rotating fluids between two cylinders. Experimentally,
Coles (1965) observed as many as 20 or 25 different states at a given speed for the case of the inner cylinder rotating and the outer one at rest. According to Coles, each state is defined by the number of Taylor cells in the axial direction and the number of waves in circumferential direction.
Both analytically and experimentally, Benjamin (1978a, 1978b) attacked the problem of bifurcation phenomena in steady flows of a viscous fluid. He concluded that the flow strongly depended on two parameters, one the Reynolds number Re and the other the aspect ratio Г (length to clearance, L/C). For a given value of Г, a unique mode with an
even number of Taylor cells emerges as the primary mode as the speed of the inner
cylinder is gradually increased to a given value from rest. The other mode is also existed
as the secondary mode , which appears with sudden increase of the inner cylinder’s speed
to the desired value.
Benjamin and Mullin (1982) observed 15 different stable steady flows under the same
geometrical and dynamical boundary conditions ( η = 0.6, Г = 12.61, and Re = 359). The
12-cell primary mode was easily produced by slowly raising the speed of the inner 71 cylinder to the prescribed value. The other 14 secondary modes were realized by the manipulation of the acceleration rate of the inner cylinder speed and first at smaller or larger values of Г.
Blennerhassett and Hall (1979) investigated the stability of circular Couette flow in a
finite-length cylindrical annulus by using linear theory. Their results showed that the
instability changed from a two-cell to a four-cell primary flow at Г ≈ 1.3. This analytical work is close to Benjamin’s (1978b) experimental observation that the primary flow changed from a two-cell to a four-cell mode when Г ≈ 1.85.
Transition between those stable steady solutions is another attractive feature in
studying the hydrodynamic stability of rotating fluids between two cylinders. Benjamin
and Mullin (1982) concluded that there are at least N-1 additional unstable solutions if N
of distinct steady solutions are realizable. Then one question is raised: why and how the
transition happens between those steady solutions? Coles (1965) identified two distinct
kinds of transition in Taylor-Couette flow between concentric rotating cylinders, namely
spectral and catastrophic transitions. The spectral transition is characteristic of the
motion when the inner cylinder rotating and the outer one at rest. This kind of transition
may be viewed as a cascade process, in which energy is transferred through a discrete
spectrum to progressively higher frequencies in a two-dimensional wave-number space.
Ohmura et al. (1994) concluded that vorticity played an important role on the transition
between steady solutions. For the case of η = 0.615, Г = 3.0 and end wall boundaries,
Ohmura et al. found that higher acceleration rates of inner cylinder speed produce such strong vorticity for the Taylor cells on the end walls that a pair of cells is induced toward
72 the center and four cells are formed, while lower acceleration rates promote diffusion of vorticity into the center of the annulus and two cells are formed.
The next two sections will deal with the influence of T b, the time needed to increase the speed of the inner cylinder from rest to the prescribed value, on the number of Taylor cells formed in axial direction and waves formed in the circumferential direction. Section
4.4.2 will investigate the influence of T b on the number of Taylor cells, and the fluid in
finite length annulus ( η = 0.99, Г = 3.0 and end wall boundaries) will be studied with T b as a parameter. Section 4.4.3 will investigate the influence of T b on the number of waves
in the circumferential direction, and the fluid in finite length annulus ( η = 0.99, Г = 2.0
and periodic boundaries) will be studied with T b as a parameter.
4.4.2 The Influence of T b on the Number of Taylor Cells
Detail shown in c) ω R+C z θ r θ L r R z
θ r
a) b) c)
Figure 4.6 Geometry for Studying the Influence of T on the Number of Taylor Cells b (Not at Scale)
73 The geometry for studying the influence of Tb on the number of Taylor cells is
presented in Figure 4.6. The length of the two cylinders is L = 0.762 mm (0.03 in.) and
the clearance is C = 0.254 mm (0.01 in.) corresponding to Г = L/C = 3. The radii of the inner and the outer cylinders are R = 25.146 mm (0.99 in.) and R+C = 25.4 mm (1.0 in.).
The top and bottom faces of the annulus are set as no-slip wall boundary conditions.
The outer cylinder wall is set as a no-slip wall boundary condition. The inner cylinder wall is set as a rotating wall, and the angular velocity vector in the z-direction is prescribed, Figure 4.6.
The time (T b) needed to increase the speed of inner cylinder from rest to prescribed speed (263.33 m/s) is varied in a parametric manner. Figure 4.7 shows the velocity of the inner cylinder vs. time for different T b’s.
300
250
200
150
Impulsive Start 100 Tb = 0.0002 s Tb = 0.001 s 50 Tb = 0.002 s
Velocity of Velocity (m/s) Cylinder Inner 0 0 0.001 0.002t (s) 0.003
Figure 4.7 Velocity of Inner Cylinder vs. Time
(T b is the Time Needed to Increase the Speed of Inner Cylinder from Rest to 263.33 m/s)
74
C D E 5
4
3 B F
A 2 Impulsive Start 1 Tb = 0.0002 s
Number Number of Taylor Cells Tb = 0.001 s Tb = 0.002 s 0 0 0.001 0.002t (s) 0.003
Figure 4.8 Number of Taylor Cells vs. Time
D E F 7
6 C 5 2→4 4→2 4
3 w(m/s) Impulsive Start B 2 Tb = 0.0002 s 0→2 Tb = 0.001 s 1 Tb = 0.002 s A 0 0→2 0 0.001 0.002t (s) 0.003
Figure 4.9 w - Velocity in Axial Direction vs. Time
75 Figure 4.8 presents the number of Taylor cells formed in the axial direction as a function of time for different T b’s. It is shown that cell number becomes two, and then becomes four, and finally turns back to two as the time increases for the case of impulsive start (T b = 0 s) and T b = 0.0002 s. However, the cell number becomes two only as the time increases for the case of T b = 0.001 s and T b = 0.002 s.
The formation of Taylor cells is indicated by w, the velocity in axial direction. Figure
4.9 presents the velocity w as a function of time for different T b’s. It is shown that w, which can be used to denote the disturbance amplitude A(t), exhibits an initial exponential growth when the time t is small for all the cases (notice that this exponential growth is obvious only for the cases of T b = 0.001 s and T b = 0.002 s on the current time scale). Qualitatively our calculations reach a good agreement with the numerical computational results by Neitzel (1984), and experimental observations by Donnelly and
Schwarz (1965), i.e. the disturbance amplitude grows exponentially when t is small.
Figure 4.9 also shows that w is always increasing where there is a change of cell numbers. For example for the case of impulsive start, w always increases when cell number becomes two (“0 →2” is labeled to represent this moment), when cell number turns to four from two (“2 →4” is labeled to represent this moment), and when cell
number turns back to two from four (“4 →2” is labeled to represent this moment). Also for the case of T b = 0.002 s, w increases when cell number becomes two (“0 →2” is labeled to represent this moment).
There is only one stable solution (cell number is two) for all the T b’s, Figure 4.8.
However, there exists a transition solution (cell number is four) for the cases of impulsive start (T b = 0 s) and T b = 0.0002 s. How does T b cause the cell number to change? 76 Ohmura et al. (1994) concluded that vorticity played an important role on the formation of cells: higher acceleration rates of inner cylinder speed produce such strong vorticity on the end walls that a pair of cells is induced toward the center (four cells formed), while lower acceleration rates promote diffusion of vorticity into the center of the annulus (two cells formed).
Does vorticity really play an important role on the formation of the cells? Figure 4.10 presents the maximum vorticity generated as a function of time for different T b’s. The
maximum vorticity at where cell number becomes two (“0 →2”) is not much different from the one at where cell number turns to four from two (“2 →4”) and the one at where cell number turns back to two (“4 →2”) for the case of impulsive start. Therefore, the
conclusion can be made that at least the maximum vorticity doesn’t play a big role on the
formation or the mergence of Taylor cells. Instead the maximum vorticity is only the
consequence of the velocity of inner cylinder if one observes the similarity of Figures 4.7
and 4.10.
The formation of different cell numbers is not caused by the maximum vortictiy.
Instead, the generated pressure, P, plays an important role on the formation of cells.
Figure 4.11 show the generated pressure as a function of time for different T b’s. P at
where cell number becomes two (“0 →2”) is much smaller than the one at where cell number turns to four from two (“2 →4”), and P at where cell number turns back to two
(“4 →2”) is much larger than the former two ones for the case of impulsive start. The conclusion that P is the direct cause of w, the indication of the formation of cells can be
made if one observes the similarity of Figures 4.9 and 4.11.
77
2→4 4→2 A B C D E F
0→2 80 70 60
50
40
30 Impulsive Start Tb = 0.0005 s 20 Tb = 0.001 s Tb = 0.002 s 0→2
Maximum Vortictiy (E+5 1/s) 10
0 0 0.001 0.002t (s) 0.003
Figure 4.10 Maximum Vorticity vs. Time
F 5 E 4→2 D 2→4 4 C 3
2
P (E+5N/m^2)