s4310658 Final Thesis

Development and Application of Multiphysics Simulation Tools
for Foil Thrust Bearings Operating with Carbon Dioxide
Kan Qin
Master of Engineering
A thesis submitted for the degree of Doctor of Philosophy at
The University of Queensland in 2017
School of Mechanical and Mining Engineering,

Queensland Geothermal Energy Centre of Excellence
Abstract
Foil bearings can enable different turbomachinery architecture. The use of the cycle’s working fluid
within the bearings results in an oil-free and compact turbomachinery system. Using CO2 as the
operating fluid for a foil bearing creates new operating and new modelling challenges. These in-
clude turbulent flow within the film, non-negligible inertia forces, high windage losses, non-ideal gas
behaviour and reduced rotordynamic damping. Since the flow phenomena within foil bearings are
complex, involving fluid flow, structural deformation and heat transfer, use of the conventional Reyn-
olds equation is not proven to be suitable for foil bearings with CO2 as the operating fluid. To address
these modelling issues, a multi-physics multi-timescale simulation tool including fluid, structure and
thermal solvers was developed to predict the performance of foil bearings and to create insight on
their operations with CO2 . New flow physics and operation challenges for foil thrust bearings with
CO2 were built and described in details next.
To model the fluid flow within foil bearings, the modifications of the in-house computational fluid
dynamics solver are first presented to enable laminar simulations within foil bearings. To reduce
the computational cost for turbulent simulations, a compressible wall function is implemented. The
checker-boarding effect, due to the high aspect ratio cell is eliminated by a fourth-order artificial
dissipation term, while maintaining the second order spatial accuracy. These modifications result in
a fast and stable solver for turbulent simulations of CO2 foil thrust bearings without contaminating
the flow field. For the fluid-structure simulation, the in-house computational fluid dynamics solver is
modified by implementing a moving grid capability. This capability is validated with inviscid, viscous
and turbulent flow cases. A separate bespoke finite difference code based on the Kirchhoff plate
equation for the circular thin plate is developed in Python to solve the structural deformation within
foil thrust bearings. A fluid-structure coupling strategy and the corresponding mapping algorithm are
employed for steady state and time-accurate transient simulations.
Using the developed fluid-structure simulation tool, the steady state performances of foil thrust
bearings with CO2 are investigated. The centrifugal inertia effects are found to be significant for foil
thrust bearings with CO2 . In the ramp region, they generate an additional inflow close to the rotor
inner edge, resulting in a higher peak pressure. Contrary in the flat region, the inertia force creates
a rapid mass loss through the bearing outer edge, which reduces pressure in this region. These dif-
ferent flow fields alter bearing performance compared to conventional air foil bearings. Conventional
Reynolds equation cannot account for the irregular radial velocity profiles that are driven by strong
inertial effects. In addition, the turbulence effects increase load capacity and power loss simultan-
eously. The steady state simulations indicate that both load and power loss increase linearly with the
decreasing rotor to top foil separations and the increasing rotational speeds. A slower rate is observed
for power loss. The rotational speed has a larger effect on power loss compared than the rotor to top
foil separations.
Finally, a heat conduction solver is added to the fluid-structure simulation tool. This results in a
multi-physics multi-timescale tool for the fluid-structure-thermal simulation. The coupling strategy is
then proposed and validated with different test cases. The heat transfer models of the solid structures
within foil thrust bearings are discussed. Numerical simulations of foil thrust bearings with air and
CO2 are performed at the same load condition. It is found that foil thrust bearings with CO2 signific-
antly benefit from increased convective cooling on the rear surface of the rotor, if the rotor operates in
a high pressure CO2 environment. The centrifugal pumping that naturally occurs in CO2 bearings due
to the high fluid density provides a new and effective cooling mechanism for the CO2 bearing. The
fluid-structure modelling approach is found to applicable at the rotational speed less than 30 000 rpm.
However, the thermal solver has to be included when foil bearings are operating at higher rotational
speeds. This is due to the large deflection caused by thermal stresses.
This project is the first work of its kind to use the high fidelity multi-physics multi-timescale
simulation tool to simulate foil thrust bearings with CO2 . The results reveal new flow physics, steady
state performances of foil bearings at different operating conditions.
Declaration by Author
This thesis is composed of my original work, and contains no material previously published or
written by another person except where due reference has been made in the text. I have clearly stated
the contribution by others to jointly-authored works that I have included in my thesis.
I have clearly stated the contribution of others to my thesis as a whole, including statistical as-
sistance, survey design, data analysis, significant technical procedures, professional editorial advice,
and any other original research work used or reported in my thesis. The content of my thesis is the
result of work I have carried out since the commencement of my research higher degree candidature
and does not include a substantial part of work that has been submitted to qualify for the award of
any other degree or diploma in any university or other tertiary institution. I have clearly stated which
parts of my thesis, if any, have been submitted to qualify for another award.
I acknowledge that an electronic copy of my thesis must be lodged with the University Library and,
subject to the policy and procedures of The University of Queensland, the thesis be made available
for research and study in accordance with the Copyright Act 1968 unless a period of embargo has
been approved by the Dean of the Graduate School.
I acknowledge that copyright of all material contained in my thesis resides with the copyright
holder(s) of that material. Where appropriate I have obtained copyright permission from the copyright
holder to reproduce material in this thesis
Publications during candidature
Journal Articles
1. Qin, K., Gollan, R. and Jahn, I.H., Application of a Wall Function to Simulate Turbulent
Flows in Foil Bearings at High Rotational Speeds, Tribology International 115, 2017,
546-556.
2. Qi, J., Reddell, T., Qin, K., Hooman, K. and Jahn, I.H., Supercritical CO2 Radial Tur-
bine Design Performance as a Function of Turbine Size Parameters, ASME Journal of
Turbomachinery 139(8), 2017, 081008.
3. Qin, K., Jahn, I.H. and Jacobs, P.A., Effect of Operating Conditions on the Elasohydro-
dynamic Performance of Foil Thrust Bearings for Supercritical CO2 Cycles. ASME
Journal of Engineering for Gas Turbines and Power. 139(4), 2016, 042505.
4. Qin, K., Jahn, I.H., Gollan, R.J. and Jacobs, P.A., Development of a Computational Tool
to Simulate Foil Bearings for Supercritical CO2 Cycles. ASME Journal of Engineering
for Gas Turbines and Power. 138(9), 2016, 092503.
5. Qin, K., Jahn, I.H., and Jacobs, P.A.. Development of a Fluid-Structure Model for Gas-
Lubricated Bump-Type Foil Thrust Bearings. Applied Mechanics and Materials. 846,
2016, 169-175.
Conference Papers
1. Qin, K., Jacobs, P.A., Keep, J.A. and Jahn, I.H.. A Three-dimensional Fluid-Structure-
Thermal Simulation of Bump-Type Foil Thrust Bearings. 1st Global Power and Propul-
sion Forum, Zürich, Switzerland. GPPF-2017-22, 2017.
2. Qin, K., Jahn, I.H., and Jacobs, P.A.. Prediction of Dynamic Characteristics of Foil Thrust
Bearing Using Computational Fluid Dynamics. 20th Australasian Fluid Mechanics Con-
ference, Perth, Australia. 2016.
3. Qin, K., Jahn, I.H., and Jacobs, P.A.. Effect of Operating Conditions on the Elasohydro-
dynamic Performance of Foil Thrust Bearings for Supercritical CO2 Cycles. ASME Turbo
Expo 2016, Seoul, South Korea. 2016.
4. Qi, J., Reddell, T., Qin, K., Hooman, K. and Jahn, I.H., Supercritical CO2 Radial Turbine
Design Performance as a Function of Turbine Size Parameters. ASME Turbo Expo 2016,
Seoul, South Korea. 2016.
5. Qin, K., Jahn, I.H. and Jacobs, P.A., Validation of a Three-dimensional CFD Analysis of
Foil Bearingss with Supercritical CO2 . 19th Australasian Fluid Mechanics Conference,
Melbourne, Australia. 2014.
Techanical Reports
1. Qin, K. and Jahn, I.H., Structural Deformation of a Circular Thin Plate with Combinations
of Fixed and Free Edges. School of Mechanical and Mining Engineering, The University
of Queensland. 2015/05.
2. Jahn, I.H. and Qin, K., e3prepToFoam: a Mesh Generator for OpenFOAM. School of
Mechanical and Mining Engineering, The University of Queensland. 2015/04.
Publications included in this thesis
As allowed by the University of Queensland Policy 4.60.07 (Alternate Thesis Format Options), this
thesis partly comprises of publications. These publications form the main parts of the thesis. Four of
them (Chapter 3, 4, 5 and 7) have been published in peer reviewed publications with the statements
of authorship and contribution provided, and the rest one (Chapter 6) will be submitted to a peer-
reviewed journal.
Qin, K., Gollan, R. and Jahn, I.H., Application of a Wall Function to Simulate Turbulent Flows
in Foil Bearings at High Rotational Speeds. Tribology International. 115, 2017, 546-556. –
Incorporated in Chapter 3.
Contributor Statement of contribution

Problem specification (70 %)
Numerical simulations (100 %)
Kan Qin
Result analysis (80 %)
Drafting and writing (70 %)
Ingo Jahn Result analysis (20 %)
Rowan Gollan Drafting and writing (5 %)
Qin, K., Jahn, I.H. and Jacobs, P.A., Effect of Operating Conditions on the Elasohydrodynamic Per-
formance of Foil Thrust Bearings for Supercritical CO2 Cycles. ASME Journal of Engineering
for Gas Turbines and Power. 139(4), 2016, 042505. – Incorporated in Chapter 5.

Kan Qin
Peter Jacobs Drafting and writing (5 %)
Qin, K., Jacobs, P.A., Keep, J.A. and Jahn, I.H.. A Three-dimensional Fluid-Structure-Thermal Sim-
ulation of Bump-Type Foil Thrust Bearings. 1st Global Power and Propulsion Forum, Zürich,
Switzerland. GPPF-2017-22, 2017. – Incorporated in Chapter 7.

Kan Qin Result analysis (80 %)
Joshua Keep Drafting and writing (5 %)
Ingo Jahn
Qin, K., Jahn, I.H., and Jacobs, P.A.. Prediction of Dynamic Characteristics of Foil Thrust Bearing
Using Computational Fluid Dynamics. 20th Australasian Fluid Mechanics Conference, Perth,
Australia. 2016. – Incorporated in Chapter 6.

Kan Qin
Ingo Jahn
Qin, K., Jahn, I.H., Gollan, R.J. and Jacobs, P.A., Development of a Computational Tool to Simulate
Foil Bearings for Supercritical CO2 Cycles. ASME Journal of Engineering for Gas Turbines
and Power. 138(9), 2016, 092503. – Incorporated in Chapter 4.

Kan Qin
Rowan Gollan
Peter Jacobs
Qin, K., Jahn, I.H., and Jacobs, P.A.. Development of a Fluid-Structure Model for Gas-Lubricated
Bump-Type Foil Thrust Bearings. Applied Mechanics and Materials. 846, 2016, 169-175. –
Incorporated in Chapter 4.

Kan Qin Numerical simulations (100 %)
Contribution by others to the thesis
Dr. Ingo Jahn, Dr. Rowan Gollan and Dr. Peter Jacobs are the co-authors of the publications incor-
porated in this thesis.
Statement of parts of the thesis submitted to qualify for the award of another degree
None.
Acknowledgements
I want to express my deepest and most sincere gratitude to my principle supervisor, Dr. Ingo Jahn.
Thanks for your guidance and constant encouragement throughout this project, for sharing your know-
ledge and passion for science and research, for constantly providing new ideas and insights. Finally,
thanks for always making time!
I would like to gratefully acknowledge my associate supervisor Dr. Peter Jacobs for offering the
opportunity to start my PhD candidature with this interesting project and for his technical guidance
of his CFD code: Eilmer.
I thank Dr. Rowan Gollan for giving me advice in debugging code and proposing many inspiring
suggestions for future work. Thanks also go to Dr. Paul Petrie-Repar for sharing his experience in
writting CFD code and many discussions on damping. I would like to thank Dr. Wilson Chan for
the valuable discussion on the k-ω turbulence model. Also, I would like to thank Dr. Carlos Ventura
for valuable suggestions at weekly meeting during the early stage of my candidature and Dr. Andrew
Rowlands for his help during my PhD application.
This project is not possbile without the supoort from the Australian Solar Thermal Research Initi-
ative (ASTRI), a project supported by Australian Government. I acknowledge the China Scholarship
Council, The Univeristy of Queensland and the Queensland Geothermal Energy Centre of Excellence
for the financial support.
The simulations detailed on this thesis were undertaken with the assistance of resources from the
University of Queensland High Performance Computing Facility’s Barrine and Tinaroo clusters, the
National Computational Infrastructure (NCI), which is supported by the Australian Government, and
the Pawsey Supercomputing Centre, funded by from the Australian Government and the Government
of Western Australia.
Thanks to all my other good buddies at UQ, Yuchen Dai, Peixing Dong, Xiaoxiao Li, Yu Liu,
Suoying He, Jacob Hess, Xin Kang, Joshua Keep, Mohsen Modirshanechi, Mostafa Odabaee, Yubiao
Sun, Phil Swann, Jianhui Qi, Jianyong Wang, Xurong Wang, Han Wei, Mohd Fairuz Zakariya; thanks
for all the coffee breaks, BBQ, laughs, parties and discussions. Thanks for making my time at UQ so
enjoyable.
Finally and most importantly, the biggest thanks of all goes to my grandparents, parents and
Yujiao. Thanks for your encouragement and support, for believing in me and for your love.
Keywords
Computational Fluid Dynamics, Multi-physics Simulations, Foil Thrust Bearings, Supercritical CO2
Cycles, Turbulent Flow, Centrifugal Inertia Effects
Australian and New Zealand Standard Research Classifications (ANZSRC)
091307 Numerical Modelling and Mechanical Characterisation, 50%

091309 Tribology, 50%
Fields of Research (FoR) Classification (FOR)
0913 Mechanical Engineering, 100%
xiii
Contents
List of Figures xvii
List of Tables xxvi
Nomenclature xxviii
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Aims and Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Literature Review 9
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Background on Supercritical CO2 Power Cycle . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Components for Supercritical CO2 Cycles . . . . . . . . . . . . . . . . . . . 13
2.3 Experimental Work on Foil Thrust Bearings . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Fluid Simulation for Gas Bearings . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4.1 Classical Reynolds Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4.2 Turbulent Reynolds equation . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.3 Computational Fluid Dynamics . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5 Structural model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.6 Thermal Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.7 Gap in Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3 Fluid Simulations 41
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Description of Fluid Solver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3 Modification of Fluid Solver for Laminar Simulations of Foil Bearings . . . . . . . . 42
3.3.1 Moving Wall Boundary Condition . . . . . . . . . . . . . . . . . . . . . . . 42
3.3.2 Periodic Boundary Condition . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.3 Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.4 Taylor-Couette Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.5 Hydrostatic Air Thrust Bearing . . . . . . . . . . . . . . . . . . . . . . . . 46
3.3.6 Wavy Thrust Bearing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4 Modification of Fluid Solver for Turbulent Simulations of Foil Bearings . . . . . . . 54
3.4.1 Wilcox’s 2006 k-ω model . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.4.2 Wall Modelled Turbulence Model for Foil Bearing Applications . . . . . . . 57
3.4.3 Stabilisation Method for Turbulent Simulations . . . . . . . . . . . . . . . . 70
3.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4 Fluid-Structure Coupled Simulations 79

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2 Moving Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2.1 Convective Flux Update with Moving Grid . . . . . . . . . . . . . . . . . . 80
4.2.2 Viscous Flux Update with Moving Grid . . . . . . . . . . . . . . . . . . . . 83
4.3 Fluid-Structure Simulation: Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.3.1 Random Grid Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.3.2 Constant Velocity Piston . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.3.3 Projectile in Tube . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.3.4 Pitching NACA0012 Airfoil . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.3.5 Oscillating Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.3.6 Pitching NACA64A010 Airfoil (Turbulent) . . . . . . . . . . . . . . . . . . 94
4.4 Structural Deformation Solver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.4.1 Top Foil Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.4.2 Bump Foil Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.5 Fluid-Structure Simulation: Coupling Strategy . . . . . . . . . . . . . . . . . . . . . 102
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
xv
5 Foil Bearing Steady State Performance 109
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.2 Steady State Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.3 Validation for Steady State Performance . . . . . . . . . . . . . . . . . . . . . . . . 110
5.4 Selection of Gas Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.5 Comparison Between Reynolds Equation and Eilmer . . . . . . . . . . . . . . . . . 113
5.6 Effect of Operating Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.6.1 Selection of Operating Conditions . . . . . . . . . . . . . . . . . . . . . . . 125
5.6.2 Centrifugal Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.6.3 Turbulence Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.7 Revisit of Computational Domain: Steady State Performance . . . . . . . . . . . . . 136
5.8 Steady State Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6 Foil Bearing Dynamic Performance 145

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.2 Dynamic Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.3 Verification of Dynamic Analysis Capability . . . . . . . . . . . . . . . . . . . . . . 148
6.4 Revisit of Computational Domain: Dynamic Performance . . . . . . . . . . . . . . . 149
6.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
7 Fluid-Structure-Thermal Simulations 153

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.2 Computational Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.2.1 Heat Conduction Solver . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.2.2 Coupling Method: Fluid-Thermal . . . . . . . . . . . . . . . . . . . . . . . 154
7.2.3 Conjugate Couette Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
7.2.4 Conjugate Nozzle Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
7.2.5 Coupling Method: Fluid-Structure-Thermal . . . . . . . . . . . . . . . . . . 158
7.3 Application to Foil Thrust Bearings . . . . . . . . . . . . . . . . . . . . . . . . . . 159
7.3.1 Heat Transfer Model for Rotor . . . . . . . . . . . . . . . . . . . . . . . . . 160
xvi
7.3.2 Heat Transfer Model for Stator . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.4 Comparison Between Air and CO2 Foil Thrust Bearings . . . . . . . . . . . . . . . 165
7.5 Preliminary Performance Investigation . . . . . . . . . . . . . . . . . . . . . . . . . 172
7.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
8 Conclusions and Future Work 181

8.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
8.2 Recommendations for Foil Bearing Designs . . . . . . . . . . . . . . . . . . . . . . 184
8.3 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
References 187
A Discretisation of Reynolds equation 207
B Description of Eilmer 211
C Source Code 215

C.1 Moving Wall Boundary Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
C.2 Periodic Boundary Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
C.3 Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
D Description of Wilcox’s 2006 k-ω model 219
xvii
List of Figures
1.1 Schematic of sCO2 turbo-alternator-compressor, taken from Ref. [8]. . . . . . . . . . 2

1.2 Schematic diagram of a hydrostatic bearing, taken from Ref. [10]. . . . . . . . . . . 3
1.3 Schematic diagram of a tapered-land bearing, taken from Ref. [12]. . . . . . . . . . . 4
1.4 Tilting pad bearing, taken from Ref. [15]. . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Metal mesh bearing, taken from Ref. [21]. . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Schematic of foil thrust bearing, taken from Ref. [23]. . . . . . . . . . . . . . . . . . 6
2.1 Plant layout of a sCO2 Brayton cycle, taken from Ref. [31]. . . . . . . . . . . . . . . 10
2.2 (a) operating conditions at SNL and (b) projected system operation by Dostal et al. [4],
taken from Ref. [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Direct normal irradiation in Australia, taken from Ref. [35]. . . . . . . . . . . . . . . 11
2.4 Test rig by Singh [39]; (a): System layout; (b): Control element. . . . . . . . . . . . 12
2.5 Photograph of the SNL sCO2 Brayton loop [42]. . . . . . . . . . . . . . . . . . . . 13
2.6 Sandia sCO2 centrifugal compressor rotating at 75 000 rpm, taken from Ref. [7]. . . . 14
2.7 Variation of thermophysical properties of CO2 : dynamic viscosity µ ((a), solid line);
thermal conductivity κ ((a), dash line); density ρ ((b), solid line); and specific heat
capacity Cp ((b), dash line) of CO2 as a function of temperature at a pressure of
8 MPa, taken from Ref. [49]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.8 Bulk film temperature (averaged over film thickness) of loaded bearing overlapped on
the pressure profiles for CO2 with real gas effect at 40 000 rpm, taken from Ref. [71]. 17
2.9 Prototype bearing with one top foil removed, taken from Ref. [23]. . . . . . . . . . . 18
2.10 Nomenclature for foil thrust bearings, taken from Ref. [76]. . . . . . . . . . . . . . . 19
2.11 Torque versus load at 23 000 rpm, taken from Ref. [23]. . . . . . . . . . . . . . . . . 19
2.12 Measured trailing edge temperature gradients, taken from Ref. [77]. . . . . . . . . . 20
2.13 Coordinate system for small perturbations, taken from Ref. [84]. . . . . . . . . . . . 22
2.14 Comparison between Reynolds equation and experiment, taken from Ref. [76]. . . . 23
2.15 Non-dimensional pressure distribution of foil thrust bearings obtained from Reynolds
equation, taken from Ref. [76]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.16 Load and torque of foil thrust bearings obtained from Reynolds equation, taken from
Ref. [76]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
′
2.17 A plot of the G functions for each of the Ng and Pan [87], Constantinescu [88], and
Hirs [89] turbulence models as a function of rotational Reynolds number, taken from
Ref. [24]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.18 Non-dimensional pressure distribution of foil thrust bearings obtained from turbulent
Reynolds equation, taken from Ref. [8]. . . . . . . . . . . . . . . . . . . . . . . . . 27
2.19 Load and power loss of foil thrust bearings obtained from turbulent Reynolds equa-
tion, taken from Ref. [8]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.20 Pressure distribution in a long slider bearing using Reynolds equation and CFD, taken
from Ref. [91]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.21 Nomenclature for bump strips, taken from Ref. [76]. . . . . . . . . . . . . . . . . . 31
2.22 Bump deflection for various load distributions, taken from Ref. [97]. . . . . . . . . . 33
2.23 Bump deflection for various load distributions, taken from Ref. [100]. . . . . . . . . 34
2.24 Structural stiffness and damping, taken from Ref. [75]. . . . . . . . . . . . . . . . . 35
2.25 Vertical deflection of the bump top points for friction coefficients 0.0 and 0.2 (right),
taken from Ref. [108]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.26 (a): Thermal resistance at the bump contacts to the top foil and the bearing housing,
along with the thermal resistance model of a bump arc, (b): simplified CFD model of
an inlet mixing chamber, taken from Ref. [109]. . . . . . . . . . . . . . . . . . . . . 37
2.27 Simplified thermal domains of the surrounding structures around the radial foil bear-
ing close to turbine and thermal boundary conditions, taken from Ref. [111]. . . . . . 38
3.1 Computational domain for Taylor-Couette flow, taken from Ref. [122]. . . . . . . . . 43
3.2 Comparison of temperature (a) and axially averaged velocity (b) profiles in radial
direction at low pressure condition. . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
xix
3.3 Comparison of temperature (a) and axially averaged velocity (b) profiles in radial
direction at intermediate pressure condition. . . . . . . . . . . . . . . . . . . . . . . 46
3.4 Parameters variation of Taylor-Couette Flow at high pressure condition; (a): pressure;
(b): temperature. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.5 Temperature and velocity for Taylor-vortices case. . . . . . . . . . . . . . . . . . . . 48
3.6 Schematic of hydrostatic air thrust bearing geometry, taken from Ref. [127]. . . . . . 49
3.7 Steady state pressure distribution in an inward pressurised bearing for different bear-
ing numbers: (a): wide bearing; (b): narrow bearing. . . . . . . . . . . . . . . . . . 50
3.8 Comparison of zero radial velocity line at different bearing number. . . . . . . . . . 50
3.9 Schematic of the geometry (a) and coordinate (b) of the wavy thrust bearings, taken
from Ref. [128]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.10 Schematic of symmetric wavy film thickness and comparison of pressure at center
radius between Eilmer and Reynolds equation. . . . . . . . . . . . . . . . . . . . . . 52
3.11 Schematic of a 4:1 wavy film thickness and comparison of pressure at center radius
between Eilmer and Reynolds equation. . . . . . . . . . . . . . . . . . . . . . . . . 53
3.12 (a) Experimental apparatus (taken from Ref. [134]) and (b) schematic diagram of the
simulation domain for Couette and Poiseuille flows, dimensions in mm. . . . . . . . 56
3.13 (a) Velocity distribution for turbulent plane Couette flow, (b) Comparison with the
law of the wall. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.14 Schematic diagram of the simulation domain for flat plate, in mm. . . . . . . . . . . 61
3.15 Performance comparison with different grid spacings for adiabatic walls, (a) friction
coefficient; (b) law of the wall. 380 mm from leading edge, Rex =2.39 × 106 . . . . . . 63
3.16 Performance comparison with different grid spacings for constant temperature walls,
(a) friction coefficient; (b) heat flux. . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.17 Law of the wall and temperature, 380 mm from leading edge, Rex =2.39 × 106 . . . . 66
3.18 Performance comparison with different grid spacings for various cases. . . . . . . . 68
3.19 Comparison of different turbulence models for case 5. . . . . . . . . . . . . . . . . . 70
3.20 Convergence history of norms based on turbulent kinetic energy for different mesh
sizes: a) L2 norms, b) L∞ norms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.21 (a) Norms and (b) observed order of accuracy. . . . . . . . . . . . . . . . . . . . . . 73
xx
3.22 Pressure contour of the rigid foil thrust bearings, rotational speed: 30 000 rpm, rotor
to top foil separation: 22 µm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.23 Velocity contour of the rigid foil thrust bearings, rotational speed: 30 000 rpm, rotor
to top foil separation: 22 µm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.24 Comparison of tangential and radial velocities at a circumferential angle of 44.46°. . 75
4.1 Schematic diagram for the moving interface. An , Bn , Cn , Dn are the vertices at time
n, An+1 , Bn+1 , Cn+1 , Dn+1 is the vertices at time n + 1. Aif , Bif , Cif , Dif are the
effective vertices at time n + 1, used to calculate the swept volume Vif . WA , WB ,
WC , WD grid-velocities associated with the vertices. . . . . . . . . . . . . . . . . . 81
4.2 Ghost cell configuration for a slip wall condition. . . . . . . . . . . . . . . . . . . . 82
4.3 Schematic diagram for tangential moving velocity. . . . . . . . . . . . . . . . . . . 83
4.4 (a) Initial two-dimensional mesh for random grid motion, (b) final two-dimensional
mesh. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.5 Comparison results of Eilmer and analytical solution, (a) density, (b) pressure, (c)
temperature, (d) velocity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.6 Schematic diagram of projectile in tube, dimensions in mm. . . . . . . . . . . . . . . 87
4.7 Comparison of pressure at end wall and piston face. . . . . . . . . . . . . . . . . . . 88
4.8 Comparison of results between Eilmer, L1D, and ideal solution, (a) simulation time
and piston position, (b) piston velocity and piston position. . . . . . . . . . . . . . . 88
4.9 Computational domain for NACA0012 airfoil. . . . . . . . . . . . . . . . . . . . . . 90
4.10 Grid convergence study for pitching NACA0012. . . . . . . . . . . . . . . . . . . . 90
4.11 Instantaneous pressure coefficient, (a) 0.52°, ↓, downward stroke, (b) −0.54°, ↑, up-
ward stroke, (c) −2.00°, ↑, upward stroke, (d) 2.01°, ↓, downward stroke, (e) 2.34°,
↓, downward stroke, (f) −2.41°, ↓, downward stroke. . . . . . . . . . . . . . . . . . 91
4.12 Comparison between Eilmer and experiment, (a) normal force coefficient, (b) moment
coefficient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.13 Computational domain for the oscillating plate. . . . . . . . . . . . . . . . . . . . . 93
4.14 Grid convergence study of oscillating plate: (a) velocity distribution at 1 ms, (b) Fric-
tion coefficient over the fine mesh in terms of representative cell size at 1 ms. . . . . 94
xxi
4.15 Comparison of results between Eilmer and analytical solution, (a) velocity profile at
different phase angles, (b) local skin friction coefficient. . . . . . . . . . . . . . . . . 95
4.16 Grid convergence study for the pitching NACA64A010 airfoil. . . . . . . . . . . . . 96
4.17 Comparison of results between Eilmer and experiment [161], (a) lift force coefficient
Cl vs angle of attack α, (b) moment coefficient Cm vs angle of attack α. . . . . . . . 96
4.18 Schematic diagram for forces and moments. . . . . . . . . . . . . . . . . . . . . . . 98
4.19 (a) Schematic diagram of top foil in foil bearings, (b) deformation comparison between
ANSYS and the Kirchhoff plate structural deformation code at different radii. . . . . 99
4.20 Grid convergence study: (a) & (b) local deflection, (c) & (d) maximum deflection. . . 101
4.21 Flow chart of the bump foil model, taken from [100]. . . . . . . . . . . . . . . . . . 103
4.22 Structural damping coefficient, CB . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.23 Stencil for mapping method at the fluid-structure interface, (a): from Eilmer to the
structure solver, (b): from the structure solver to Eilmer. . . . . . . . . . . . . . . . . 105
4.24 Pressure and deflection contours at the fluid-structure interface, the mapping process
is (a)→(b)→(c)→(d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.25 Comparison of deflection by the structure solver and mapped deflection for Eilmer. . 108
5.1 Side view of foil thrust bearings, not to scale. . . . . . . . . . . . . . . . . . . . . . 110
5.2 Computational domain for foil thrust bearings, not to scale. . . . . . . . . . . . . . . 111
5.3 Grid convergence study for foil thrust bearings: (a) local deflection at the medium
radius, (b) pressure at the medium radius. . . . . . . . . . . . . . . . . . . . . . . . 112
5.4 Comparison of measurements and numerical simulation at fixed rotational speed of

21 000 rpm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.5 Comparison of maximum deformation between Eilmer and numerical results from
Ref. [26]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.6 Pressure-Enthalpy diagram for CO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.7 Comparison of thermodynamic properties at the medium radius with the rotational
speed of 30000 rpm: (a) Pressure, (b) Compressibility factor, (c) Density, (d) Temper-
ature, (e) Dynamic viscosity, (f) Thermal conductivity. . . . . . . . . . . . . . . . . 115
xxii
5.8 Comparison between Eilmer-laminar and Reynolds equation, (a) Pressure (in Pa) con-
tour for low density condition, (b) Pressure (in Pa) contour for high density condition,
(c) Pressure (in Pa) contour for low density condition, (d) Pressure (in Pa) contour for
high density condition, (e) & (f) Comparison of pressure at the medium radius. . . . 117
5.9 Pressure increase contour for foil thrust bearings, (a) CO2 , rotational speed: 60 000 rpm,
(b) Air, rotational speed: 60 000 rpm. . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.10 Comparison of non-dimensional pressure distribution at the medium radius between

Eilmer and Reynolds equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.11 Ramp region: pressure increase, radial and tangential velocity at different circumfer-
ential angles. Rotational speed: 60 000 rpm, ramp ends at 15°, (Rotor at top, stator at
bottom). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.12 Flat region: pressure increase, radial and tangential velocity at different circumfer-
ential angles. Rotational speed: 60 000 rpm, flat starts at 15°, (Rotor at top, stator at
bottom). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.13 Approximate streamlines at 90% (rotor) and 10% (stator) film thickness (radial com-
ponent is exaggerated). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.14 The architecture of a supercritical CO2 turbomachinery system. . . . . . . . . . . . . 125
5.15 Density (solid line, in kg/m3 ) and dynamic viscosity (dashed line, in ×106 kg/ms) at
different operating conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.16 Rotational Reynolds number at different operating conditions (based on 5 µm initial

gap and 50 000 rpm). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.17 Performance comparison at different operating conditions from laminar fluid solver,
solid line: load capacity in N, dashed line: torque in Nmm, Eilmer-laminar. Rota-
tional speed: 50 000 rpm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.18 Pressure increase (p − pa ) at the different operating pressures. Operating temperature

fixed at 400 K, Eilmer-laminar. Rotational speed: 50 000 rpm. . . . . . . . . . . . . . 128
5.19 Radial velocity profile in the ramp region (circumferential angle: 5°) for different
operating conditions, bold line: Vr = 0, Eilmer-laminar. . . . . . . . . . . . . . . . . 129
xxiii
5.20 Radial velocity profile in the ramp region (circumferential angle: 25°) for the different
operating conditions, bold line: Vr = 0. Note, the inner radius velocity profile are so
small that they are difficult to see, Eilmer-laminar. . . . . . . . . . . . . . . . . . . . 130
5.21 Streamlines and radial velocity close to stator and rotor. Operating condition: 0.1 MPa
and 400 K, ρ=1.33 kg m−3 . the rotational direction is anti-clockwise, Eilmer-laminar. 132
5.22 Streamlines and radial velocity close to stator and rotor. Operating condition: 1.4 MPa
and 400 K, ρ=19.01 kg m−3 , the rotational direction is anti-clockwise, Eilmer-laminar. 133
5.23 Local deflection (in µm) for two operating conditions, rotational direction: anti-
clockwise, Eilmer-laminar. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.24 Performance comparison for different flow regimes, the operating temperature is fixed
as 400 K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.25 Local deflection (in µm) at an operating pressure of 1.4 MPa and operating temperat-
ure of 400 K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.26 Comparison of tangential velocity profile at laminar and turbulent flow at the medium
radius with the circumferential angle of 24.20°. . . . . . . . . . . . . . . . . . . . . 137
5.27 Pressure and radial velocity close to stator and rotor. Operating condition: 4.0 MPa
and 400 K, the rotational direction in anti-clockwise, Eilmer-turbulent. . . . . . . . . 138
5.28 Schematic diagram of different computational domains for foil thrust bearings. . . . 139
5.29 Comparison of pressure distribution at different computational domains, rotational
speed is 40 000 rpm and load is 200 N, notable differences in pressure at start of ramp
and end of flat regions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.30 Gauge Pressure (unit in kPa) and deflection (unit in µm) contours at different rotor to
top foil separation, rotational speed of 30 000 rpm. . . . . . . . . . . . . . . . . . . . 142
5.31 Steady state performance of foil thrust bearings under different rotational speed and
initial film thickness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.1 Schematic diagram of rotor perturbation motion. . . . . . . . . . . . . . . . . . . . . 146

6.2 History plot of reaction force and rotor motion. . . . . . . . . . . . . . . . . . . . . 148
6.3 Comparison of the calculated stiffness and damping coefficient at different cycles. . . 149
6.4 Comparison of the calculated stiffness and damping coefficient at different parameters. 150
xxiv
6.5 Comparison of CFD and numerical results from San Andrés [26], rotational speed:
21 000 rpm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.6 Comparison of foil thrust bearings performance at different computational domain,
(a): synchronous stiffness; (b): synchronous damping. . . . . . . . . . . . . . . . . . 151
7.1 Schematic diagram for conjugate Couette flow. . . . . . . . . . . . . . . . . . . . . 156

7.2 Comparison of temperature distribution between fluid-thermal simulation and analyt-
ical solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
7.3 Boundary conditions for conjugate nozzle flow, coloured by temperature. . . . . . . 157
7.4 (a) Temperature distribution along outer wall of nozzle [178], used as the nonuniform
temperature boundary condition for numerical simulation, (b) comparison of tem-
perature distribution along inner wall of nozzle between numerical simulation and
experiment [178]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
7.5 Schematic diagram for fluid-structure-thermal simulations. . . . . . . . . . . . . . . 159
7.6 Schematic diagram for layout of foil thrust bearings. . . . . . . . . . . . . . . . . . 160
7.7 Computational domain for the rotor. . . . . . . . . . . . . . . . . . . . . . . . . . . 160
7.8 Computational domain for the stator, (a): Stator in three dimensional, (b): Stator with
top foil removed in three dimensional. . . . . . . . . . . . . . . . . . . . . . . . . . 162
7.9 Schematic diagram of heat transfer from top foil to thrust plate. . . . . . . . . . . . . 163
7.10 Schematic diagram of the additional geometry to account for the contact resistance. . 164
7.11 Schematic diagram of the deflection of the bump foil. . . . . . . . . . . . . . . . . . 164
7.12 Schematic diagram of computational mesh of stator. . . . . . . . . . . . . . . . . . . 165
7.13 Computational domain of a foil thrust bearing. . . . . . . . . . . . . . . . . . . . . . 166
7.14 Air foil thrust bearings interface temperatures, (a): rotor; (b): stator; (c): heat flow
chart for complete bearing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
7.15 CO2 foil thrust bearings interface temperatures, (a): rotor; (b): stator; (c): heat flow
chart for complete bearing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
7.16 Comparison of streamline between air and CO2 thrust bearings in the groove, (a) air;
(b) CO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
7.17 Comparison of (a) heat flow and maximum top foil temperature (b) for different con-
vection coefficients within the bump channels. . . . . . . . . . . . . . . . . . . . . . 172
xxv
7.18 Comparison of rotor temperature (in K) at power loss, (a) 318 N and 417 W, (b) 100 N
and 352 W, rotational speed is 30 000 rpm. . . . . . . . . . . . . . . . . . . . . . . . 173
7.19 Comparison of rotor deflection (in m) at different power loss, (a) 318 N and 417 W,
(b) 100 N and 352 W, rotational speed is 30 000 rpm. . . . . . . . . . . . . . . . . . 174
7.20 Temperature and deflection at the rotor, load of 318 N, power loss of 417 W, rotational
speed is 30 000 rpm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.21 Comparison of stator temperature (in K) at different load and power loss, (a) 318 N
and 417 W, (b) 100 N and 352 W, rotational speed is 30 000 rpm. . . . . . . . . . . . 175
7.22 Comparison of stator deflection (in m) at different power loss, (a) 318 N and 417 W,
(b) 100 N and 352 W, rotational speed is 30 000 rpm. . . . . . . . . . . . . . . . . . 175
7.23 Comparison of rotor temperature (in K) at different rotational speed, (a) 40 000 rpm,
load is 275 N, power loss is 815 W, (b) 50 000 rpm, load is 318 N, power loss is 1273 W.176
7.24 Comparison of rotor deflection (in m) due to thermal stress at different rotational
speed, (a) 40 000 rpm, load is 275 N, power loss is 815 W, (b) 50 000 rpm, load is
318 N, power loss is 1273 W. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
7.25 Comparison of rotor temperature (in K) at different rotational speed, (a) 40 000 rpm,
load is 275 N, power loss is 815 W, (b) 50 000 rpm, load is 318 N, power loss is 1273 W.177
7.26 Comparison of stator deflection (in m) due to thermal stress at different rotational
speed, (a) 40 000 rpm, load is 275 N, power loss is 815 W, (b) 50 000 rpm, load is
318 N, power loss is 1273 W. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
7.27 Comparison of bearing performance for different modelling approaches. . . . . . . . 178
xxvi
List of Tables
2.1 Geometry and physical parameters for thrust foil bearings in Ref. [26, 73]. . . . . . . 19
2.2 The Ng and Pan turbulent coefficients for various critical Reynolds numbers. Applic-
able to Equations 2.8 and 2.9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1 Parameters for simulations, used match the experimental conditions as reported in
Ref. [122]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2 Rotational speed (in rpm) for different bearing speed parameter for perfect air and
dense air. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3 Wavy thrust bearing physical and operational characteristics. . . . . . . . . . . . . . 51
3.4 Comparison results of load (in N) and friction torque (in Nmm) obtained from Reyn-
olds equation and Eilmer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.5 Operating condition of foil thrust bearings from SNL’s test loop [8]. . . . . . . . . . 54
3.6 Details of El Telbany and Reynolds tests. . . . . . . . . . . . . . . . . . . . . . . . 67
3.7 Comparison of frictional velocity (in m/s) at different grid spacings. The low stress
wall is the moving wall in cases 1,2,3 and 5. . . . . . . . . . . . . . . . . . . . . . . 69
3.8 Comparison of time step used by different modeling approaches. . . . . . . . . . . . 74
3.9 Matrix of rotational speeds and minimum rotor to top foil separation to show trends
in k4 required to stabilise simulations. . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.1 Operating condition of foil thrust bearings from SNL’s test loop [8]. . . . . . . . . . 100
4.2 Comparison of pressure force between Eilmer and the structure solver. . . . . . . . . 106
5.1 Performance comparison of foil thrust bearings with different computational meshes. 111
5.2 Performance comparison of the fixed geometry thrust bearings between Eilmer-laminar
and Reynolds equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.3 Operating conditions for CO2 and air. . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.1 Geometry and parameters for the foil thrust bearing. . . . . . . . . . . . . . . . . . . 166
xxviii
Nomenclature
Acronyms
2D Two dimensional
3D Three dimensional
AGARD Advisory Group for Aerospace Research and Development
AUSM Advection Upstream Splitting Method
CFD Computational Fluid Dynamics
CFL Courant-Friedrichs-Lewy
CST Concentrated Solar Thermal
CTDNS Compressible Temporal Direct Numerical Simualtion
DNS Direct Numerical Simulation
DSMC Direct Simulation Monte Carlo
EFM Equilibrium Flux Method
FDM Finite Difference Method
LES Large Eddy Simualtion
NACA National Advisory Committee for Aeronautics
NASA National Aeronautics and Space Administration
RANS Reynolds-averaged Navier–Stokes
Re Reynolds number.
REFPROP Reference Fluid Thermodynamic and Transport Properties Database
sCO2 Supercritical CO2
SNL Sandia National Laboratories
Greek Symbols
α Angle of attack.
α0 Pitching range.
αm Mean angle of attack.
β Under-relaxation factor.
κ Reduced frequency.
µ Dynamic viscosity.
ν Poisson’s ratio.
ω Rotational speed.
φ Scalar quantity.
ρ Density.
τwall Wall shear stress.
θ Angle.
Roman Symbols
∆U Change of conserved flow quantities U.
∆t Time step.
n̂ Outward-facing unit normal of the control surface.
t̂ Tangential unit-vector associated with a boundary interface.
Fi Convective flux.
Fv Viscous flux.
L Momentum flux.
Q An array of source terms.
U An array of conserved flow quantities per unit volume.
u Flow velocity vector.
wa Interface grid-velocity obtained by averaging the velocity of four vertices.
wif Effective interface moving velocity.
A Area of the control face.
Aif Interface area.
C Equivalent damping.
c Chord length.
CB Structural damping.
xxx
Cb Fluid film damping.
Cf Skin friction coefficient.
Cl Lift force coefficient.
Cm Momentum coefficient.
D Stiffness of thin plate.
DT Diffusivity of solid materials.
E Total energy.
F Lift force.
G Mass flux.
H Energy flux.
h Thickness.
K Equivalent stiffness.
KB Structural stiffness.
Kb Fluid film stiffness.
M Momentum.
p Pressure acting on the thin plate.
pa Ambient pressure.
r Radius.
S Bounding surface.
T Friction torque.
u+ Non-dimensional velocity.
V Cell volume.
w Deflection.
y+ Non-dimensional distance.
Subscripts
f Fluid.
i Interface.
s Solid.
xxxi
Chapter 1
Introduction
1.1 Motivation
The sCO2 power cycle is considered a promising next generation power cycle. It has the potential to
offer better overall economics due to a higher thermal conversion efficiency and lower capital cost.
Feher [1] designed the first sCO2 power cycles at last 60s. At the same time, Angelino [2, 3] proposed
a liquid phase compression gas turbine. The system efficiency was larger than that of regenerative
Brayton cycles and the approximately same with that of regenerative Rankine cycles.
CO2 is an abundant, non-toxic, stable and relatively inert working fluid with a critical temperature
close to ambient temperature in many locations (31 ◦C). Therefore, sCO2 cycles do not require low
temperature cooling fluid, which is not available at many places, especially at places with abundant
solar energy. Supercritical CO2 Brayton cycle can offer higher efficiency compared to steam cycles at
temperatures larger than 550 ◦C [4]. This cycle also has a compact architecture compared to Rankine
cycles due to the higher density fluid. This also reduces the cost of maintenance, installation, and
operation [5].
sCO2 turbomachinery system is a critical component operating at high speeds with the support
from bearings. Figure 1.1 shows the schematic diagram of the sCO2 turbomachinery commissioned
at SNL. Two journal bearings and one thrust bearing are used to support axial and radial loads, re-
spectively. In sCO2 turbomachinery, thrust loads arise from the unequal axial pressure distribution
on rotors, while radial loads are from the radial force on rotors and the shaft weight. The thrust
load from a 100 kW sCO2 turbine designed at QGECE is approximately 1000 N due to high absolute
pressure and pressure difference across turbine, and the shaft rotational speed can be 160 000 rpm to
ensure a higher total-to-static efficiency [6]. This presents critical challenges to the bearing operating
2 Chapter 1 Introduction
Figure 1.1: Schematic of sCO2 turbo-alternator-compressor, taken from Ref. [8].
conditions. Many early tests of sCO2 turbomachinery at SNL [7] used ball bearings. However ball
bearings are limited to a lifetime between 20 hours and 2000 hours [7]. This is too short for practical
applications with targeted lifes greater than 20 000 hours.
Other alternatives to rolling-element bearings are fluid bearings. Long operation life is not a prob-
lem for fluid bearings compared to rolling-element bearings Also, the need for additional oil supply
systems is removed, since the cycle fluid (CO2 for sCO2 power cycles) is used as the working fluid
for the bearings. Fluid bearings are typically used in high load and high rotational speed conditions,
where conventional rolling-element bearings would have short life (as encountered at the small sCO2
cycle test rig at SNL [7]) [9]. Fluid bearings are bearings in which the load is supported by a thin
layer of rapidly moving pressurised liquid or gas between the rotor and stator. Because of no contact
between moving objectives, the sliding friction is minimised, allowing fluid bearings to have lower
friction, wear and vibration than many other types of bearings. Fluid bearings can be further classified
into two types depending on pressurisation types, hydrostatic and hydrodynamic bearings.
Hydrostatic bearings refer to a configuration where the lubricant with high pressure is fed into the
bearings by an external pump. Therefore, the formed load capacity is used to compensate loads from
turbines and compressors. The schematic diagram of a hydrostatic bearing is depicted in Figure 1.2.
The orifice in the recess connects the high pressure fluid from the external pump with the bearing
Motivation Section 1.1 3
Figure 1.2: Schematic diagram of a hydrostatic bearing, taken from Ref. [10].
chamber. In hydrostatic bearings, mechanical contact and wear can be avoided, even at the very low
rotational speed and lightly load. The second advantage is good stiffness. High stiffness is maintained
under different loads and rotational speeds [9]. However, the drawback is that auxiliary components
are needed to enable its operation. The cost and system complexity increase accordingly.
On the contrary to hydrostatic bearings, hydrodynamic bearings rely on a high speed rotor to
pressurise the fluid. The pressure in the fluid film is increased due to a wedge shape formed by the
rotor and stator. The increased pressure provides forces to compensate the load from turbines or
compressors. Compared to hydrostatic bearings, hydrodynamic bearings do not require the external
pressurisation and experience weight reduction due to the elimination of the additional oil supply
system, and substantial savings in maintenance costs [11]. Depending on the support structure at the
stator, fluid bearings can be further divided into fixed-geometry bearings (e.g. tapered-land bearings)
and flexible supporting bearings (e.g. tilting pad, bump-type, and metal mesh bearings).
Tapered-land bearings are simple and compact fixed geometry bearings, which are able to operate
reliably for long periods [13]. Their schematic diagram is shown in Figure 1.3. The geometry forms
the ramp and flat regions between the rotor and stator. The pressure is increased due to the convergent
shape in the ramp region. Since the tapered-land bearings are sensitive to load, speed and working
fluid, a custom design is commonly required for the specific operating conditions [14].
One challenge of the aforementioned fixed-geometry bearing are manufacturing uncertainties [16].
This results in a large performance uncertainty during operation. Therefore, flexible supporting bear-
Figure 1.3: Schematic diagram of a tapered-land bearing, taken from Ref. [12].
Figure 1.4: Tilting pad bearing, taken from Ref. [15].
ings are introduced, which are able to increase tolerance with respect to deformations of the rotor and
stator and to manufacturing variation. Flexible bearings consist of tilting pad bearing [15], bump-type
foil bearings [11], metal mesh foil bearings [17] or the hybrid bump and metal mesh bearings that is
recently proposed by Feng et al. [18]. Tilting pad bearings usually have a multiple number of pads.
A four pad bearing is shown in Figure 1.4. Each pad is able to rotate about a pivot freely. For tilting
pad bearings, destabilising forces are minimal. Thus, the bearings are not the effect of rotordynamic
instability [15]. Tilting pad bearings are utilised as the fluid-film bearing for high-speed applica-
tions [15]. Metal mesh bearings use the ring shaped metal mesh to support the top foil as depicted
in Figure 1.5. This bearing was employed as vibration isolaters in gas turbine engines initially [19].
Metal mesh bearings have received increased attention in recent years at Texas A&M University led
Motivation Section 1.1 5
Figure 1.5: Metal mesh bearing, taken from Ref. [21].
by San Andrés [17, 20, 21, 22]. However, this metal mesh bearing is still under development and is
not widely used in industry so far.
Bump-type foil bearings have been widely investigated in literature for air cycle machines [11].
It was successfully utilised to support the long duration tests for the sCO2 cycle test rig at SNL [7].
Therefore, this bump-type foil bearings was selected to be investigated in this project. Bump-type gas
foil bearings are composed of a smooth top foil and a corrugated bump foil. Each is made from a
thin sheet of compliant metal, usually a nickel-based alloy. In a typical configuration, the top foil is
affixed to the bearing housing on the upstream side, and on the downstream side it sits at the height
of its bump understructure, forming a partially ramped profile as indicated in Figure 1.6. During
operation, gas is drawn into the thin region between the rotor collar and top foils, and a self-generated
hydrodynamic pressure field reacts to the thrust and radial load acting on the rotor. The bearing
geometry and operating conditions, bump foil’s mechanical properties, the top foil’s material and
coating determine the overall static and dynamic performance of the foil bearing.
Foil bearings have various advantages compared to the conventional rolling-element bearings in
turbomachinery, including improved reliability, elimination of lubrication systems, and better toler-
ance to misalignment [24]. The corrugated structure providing stiffness and damping (relative motion)
to the system makes the foil bearing unique [25]. A converging wedge is formed between the rotor
and the top foil. The application of foil bearings was identified as a revolutionary concept in gas
turbine engines. This approach provided a significant performance improvement in efficiency, speed,
Figure 1.6: Schematic of foil thrust bearing, taken from Ref. [23].
and reliability. The typically applications of foil bearings are in ranging from the magnetic disk drive
read/write heads in personal computers to the air cycle machines (aeroengine and turbochargers) [26].
Even though foil bearings have been successfully used in small sCO2 cycle test rigs to allow
a longer duration testing [7], the flow behaviour and structural behaviour, including rotordynamic
performance, of foil bearings and how it is affected by nonlinear properties and highly dense CO2
is not well understood. To date, investigations of CO2 foil bearings have been limited [27, 28].
Reynolds equation is widely used to model the pressure distribution within foil bearings. However it
is uncertain if Reynolds equation is able to capture the new flow physics within foil bearings when
operating with CO2 . Additionally, turbulence flow regimes induced by highly dense CO2 would result
in the significant viscous heating effects. Their thermal performances has to be accurately investigated
and properly managed to avoid in operation failures.
1.2 Aims and Objectives

Because of the interest in applying foil bearings to future sCO2 cycle machines and the lack of analysis
capability other than Reynolds equation based approaches, this thesis sets out to develop a higher
fidelity simulation tool. The tool developed in this thesis has the capability to perform transient fluid-
Thesis Overview Section 1.3 7
structure-thermal predictions of foil bearings. The simulation tool is presently developed for foil
thrust bearings but the methodology is transferable to foil journal bearings. The objectives of this
PhD project are:
1. To develop a high fidelity multi-physics multi-timescale simulation tool to analyse the perform-
ance of foil bearings for a range of working fluids, including high density CO2 ;
2. To provide the new insight to the operation of foil thrust bearings with CO2 .
1.3 Thesis Overview

In this thesis, the aims and objectives outlined in Section 1.2 are achieved using a systematic approach.
The thesis starts with discussion of relevant literature to highlight the theoretical basis of the work
and experimental studies of foil bearings. This is followed by the development of a multi-physics and
multi-timescale computational tool, build on the in-house CFD code Eilmer. This tool is developed
and validated progressively until the full fluid-structure-thermal operation of a CO2 foil thrust bearing
can be completed. The gradual increase in complexity with each chapter allows the reader to progress
with a logical flow of ideas and conclusions. A brief breakdown of each chapter is given here:
Chapter 2 Literature Review
An overview of the sCO2 power cycle is given in this chapter to active the need to design foil thrust
bearings. Theoretical and experimental approaches for foil journal and thrust bearings are then dis-
cussed to highlight the inadequacy of these approaches for CO2 foil bearings.
Chapter 3 Fluid Simulations
Chapter 3 describes the main modifications of the in-house CFD solver Eilmer to allow the simulation
of the fluid flow within foil bearings. The modifications allow Eilmer to simulate laminar and turbu-
lent flows in foil bearings. The corresponding verification and validation cases are then presented to
test the suitability of the modified CFD solver.
Chapter 4 Fluid-Structure Coupled Simulations
Chapter 4 documents the inclusion of the moving grid framework into the in-house CFD solver and
the development of the structural deformation solver to simulate the deformation of the top and bump
foils, followed by the mapping algorithm and coupling strategy. The coupling strategy is then em-
ployed to demonstrate the calculation process of the fluid-structure solver to obtain the steady state
and dynamic performances.
Chapter 5 Foil Bearing Steady State Performance

Chapter 5 documents the validation of the developed fluid-structure simulation tool for air foil thrust
bearings. The gas model is selected for real gas properties of CO2 . Simulation results for rigid
and flexible bearings are compared to highlight the significance of centrifugal inertia and turbulence
effects introduced by the high density CO2 . It is found that centrifugal inertia effects increase the load
capacity in the ramp region, while decrease the load capacity at the flat region. Also, turbulence effects
largely increase the load and power loss. The computational domain used for foil thrust bearings is
then investigated by comparing the steady state performances. Finally, the steady state performance
under different rotational speeds and rotor to top foil separations is investigated.
Chapter 6 Foil Bearing Dynamic Performance
Chapter 6 details the procedure to calculate dynamic stiffness and damping coefficients using the
developed fluid-structure solver. This is followed by verification cases for air foil thrust bearings. The
computational domain is again revisited by comparing the dynamic performances.
Chapter 7 Fluid-Structure-Thermal Simulations
Chapter 7 details the computational framework for the fluid-structure-thermal simulation of foil thrust
bearings. Under the same load condition (110 N), the fluid-structure-thermal simulation are performed
for foil bearings with air and CO2 , respectively. It is found that the groove is beneficial to remove
viscous heat generation for CO2 foil bearings due to centrifugal inertia effect. The bearing per-
formance is compared using fluid-structure and fluid-structure-thermal modelling approaches. The
fluid-structure simulation is applicable at the low rotational speed. However, fluid-structure-thermal
modelling approach is needed at the rotational speed larger than 40 000 rpm.
Chapter 8 Conclusions and Future Work
Chapter 8 summarises the findings of the work and discusses avenues for future research.
Chapter 2
Literature Review
2.1 Introduction
In the previous chapter, we highlighted the operating requirement and experiment failure from SNL’s
tests using conventional rolling-element bearings. One solution to enable this promising cycle is to use
foil bearings. In this chapter, the background literature of sCO2 power cycles and major components
is described. Experimental studies for foil thrust bearings are then discussed. Reynolds equation, as
an effective model for foil bearings, is then reviewed, followed by the modified Reynolds equation
for turbulent flows. Next, the CFD approach applied to air bearings is reviewed. This confirms
that Reynolds equation is an accurate numerical method to investigate air bearings. The second part
focuses on analytical models for structural deformation within foil bearings. This highlights the
complexity in predicting bump foil stiffness. In addition, the influence of temperature within foil
bearings is reviewed and the effect of deflection due to thermal stresses is highlighted.
2.2 Background on Supercritical CO2 Power Cycle

In more recent years, sCO2 power cycles have drawn attention in many applications, including, nuc-
lear power [4], coal power [29], and renewable energy [30]. Dostal [4] showed that the sCO2 cycle has
a higher efficiency compared to steam cycles at temperatures above 550 ◦C. Turchi [30] developed and
benchmarked different sCO2 cycle configurations for CST applications, including, simple Brayton
cycle, recompression Brayton cycle with reheat, partial cooling cycle with recompression and reheat,
and recompression with main-compression intercooling and reheat. It was concluded that the partial
cooling cycles and recompression with main compression intercooling together with reheat are able
of achieving greater than 50 % efficiency. These intercooled cycles expanded the temperature dif-
10 Chapter 2 Literature Review
Figure 2.1: Plant layout of a sCO2 Brayton cycle, taken from Ref. [31].
ferential in heat exchanger, which is favourable for concentrating solar power systems with variable
thermal energy storage. The flowchart of the sCO2 Brayton power cycle commissioned at SNL [7] is
shown in Figure 2.1. This is a split-flow recompression system, where the additional sCO2 turboma-
chinery systems are employed for recompression. The typical operating condition at SNL proposed
by Dostal et al. [4] are depicted in Figure 2.2. For the sCO2 cycle at SNL, a high-side pressure of
10.1 MPa, a low-side pressure of 7.6 MPa, a hot-side resource fluid temperature of over 450 ◦C and
a cold-side temperature of 30 ◦C were considered. The operating pressure can be even higher. A
hot-side pressure of 20 MPa and temperature of over 700 ◦C were proposed by Dostal et al. [4].
Besarati et al. [32] added an organic Rankine cycle system to different sCO2 cycle configurations
as a bottoming cycle to utilise the waste heat. Different working fluids were examined for the bot-
toming cycles. He stated that by adding an appropriate bottoming cycle he was able to increase the
cycle efficiency by 3 to 7 % under specific conditions. Kim et al. [33] discussed sCO2 power cycles
for waste heat recovery from gas turbine with three different cycle configurations: a simple cycle, a
cascade cycle and a split cycle. The split cycle was found to produce the highest power at investigated
operating conditions. A hybrid transcritical or supercritical CO2 cycle using both low and high tem-
perature heat sources was investigated by Kim et al. [34]. The proposed cycles were able to maximise
the power output of the CO2 power cycle and offer the advantage of load leveling in conjunction with
the temperature thermal energy storage.
Background on Supercritical CO2 Power Cycle Section 2.2 11
Figure 2.2: (a) operating conditions at SNL and (b) projected system operation by Dostal et al. [4], taken from
Ref. [31].
Figure 2.3: Direct normal irradiation in Australia, taken from Ref. [35].
In Australia, due to the abundant solar energy resources (see Figure 2.3), there is a growing interest
in deploying sCO2 power cycles for solar power applications. Singh et al. [36, 37, 38] performed
(a) (b)
Figure 2.4: Test rig by Singh [39]; (a): System layout; (b): Control element.
considerable studies on the dynamic characteristics of direct-heated sCO2 Brayton cycle in a solar
thermal power plant. The dynamic characteristics was investigated. His work highlighted the poten-
tial for mass-flow rate control in summer and control of compressor inlet conditions in winter [36].
The effect of the relative hot-to-cold side volume-ratios was also investigated [37]. Furthermore, an
extremum-seeking-control and inventory-control method for closed Brayton cycles was presented by
Singh et al. [38]. Simulations indicated that this control method can be achieved with a fixed-CO2 in-
ventory in both summer and winter. A small test-rig was constructed at the University of Queensland
to test the dynamic performance of supercritical CO2 cycles [39, 40].
Besides these theoretical studies about the potential application of sCO2 power cycles, SNL con-
ducted pioneering work on the experimental operation of sCO2 cycles [5, 7, 41, 42]. Figure 2.5 depicts
the SNL’s Megawatt-scale sCO2 split-flow recompression Brayton cycle. This provided preliminary
experimental results for sCO2 cycles and the performance of critical components using sCO2 as the
working fluid. These included recuperation and waste heat rejection heat exchangers, turbines, bear-
ings, compressors, seals and heat exchangers. Iverson [41] pointed out that one challenge for solar
thermal applications was the transient nature of the solar resource. A dataset for stable sCO2 Brayton
cycle operation that was used as validation for numerical simulations was also presented.
Besides, Conboy et al. [42] presented experimental results regarding the control of sCO2 cycles,
especially during the startup and shutdown processes. A protected shutdown in a motor was employed
Figure 2.5: Photograph of the SNL sCO2 Brayton loop [42].
to prevent damage of foil bearings at speeds of less than 25 000 rpm, the lift-off speed for their foil
bearings.
2.2.1 Components for Supercritical CO2 Cycles

Before realising sCO2 cycles, a number of challenges need to be overcome, making this an active
area of research. The key components of this cycles, including turbine, compressor, heat exchanger,
bearing and seal, need to be re-studied due to the new working fluid and especially the non-linear
behaviour and high density exhibited by sCO2 .
Turbine
Both axial and radial turbines are considered as the candidate for sCO2 power cycles [6, 43, 44]. The
high dense CO2 results in the challenge for the aerodynamic design of turbines and compressor. Low
volume flow rate leads to the design point at very low specific speeds and small sizes of turbine and
compressor as shown in Figure 2.6. A partial admission axial turbine was considered by Kang et
al. [43]. Qi et al. [6] used an automatic preliminary radial inflow design code developed by Ventura
et al. [45] to investigate sCO2 turbines at different power and rotational speeds. It was found that in-
creasing power increased the number of feasible designs. The number of feasible design was limited
when the speed is reduced. In Ref. [6], an total-to-static efficiency of 85% was attained under the se-
lected operating conditions. Since the operating point of the sCO2 turbine is far away from the critical
point, Redlich-Kwong-Aungier cubic equation is accurate enough to simulate the sCO2 turbine [43].
Apart from the aerodynamic design, the forced response of sCO2 turbines was also of interest [46].
Figure 2.6: Sandia sCO2 centrifugal compressor rotating at 75 000 rpm, taken from Ref. [7].
Compressor
For CO2 compressor design, Kus et al. [47] developed a one-dimensional code for preliminary design
and performance prediction of oil-free CO2 compressors. The high speed centrifugal compressor in a
hermetic configuration was supported on foil gas bearings, and a wide range of loss mechanisms was
considered. Lee et al. [48] developed a mean-line code called KAIST TMD for designing supercrit-
ical CO2 compressor and turbine, which was compared to the existing experimental data to verify its
capability.
The operating condition of sCO2 compressors is slightly above the critical point, which introduces
the significant non-ideal gas effect [50]. An abrupt property change is observed close to the critical
point as shown in Figure 2.7. Additionally, it is possible that the working fluid within the compressors
drops into the two phase region. These effects create challenges when designing sCO2 compressors.
Therefore, the standard assumptions might not be applicable when using one dimensional design
tools [51]. High-fidelity computational tools were also used to predict the performance of sCO2
compressors. Furusawa et al. [52] proposed a numerical method to simulate sCO2 flows with the
preconditioning method, in which thermophysical properties of fluids are generated from an in-house
software. A nonequilibrium condensation model was also applied. It was found that condensation
occurred in regions near the trailing edges. Pecnik et al. [53, 54] presented a three-dimensional
computational fluid dynamics analysis of a centrifugal compressor operating with sCO2 in the region
above the critical point. Experimental data from SNL sCO2 compression loop facility was selected as
the validation case. The ideal head coefficient computed from CFD results was higher than the one
Figure 2.7: Variation of thermophysical properties of CO2 : dynamic viscosity µ ((a), solid line); thermal
conductivity κ ((a), dash line); density ρ ((b), solid line); and specific heat capacity Cp ((b), dash line) of
CO2 as a function of temperature at a pressure of 8 MPa, taken from Ref. [49].
obtained from measurements. Differences are attributed to the simplified geometry adopted in their
work. Kim et al. [55] provided CFD investigation of a centrifugal compressor with supercritical CO2
as working fluid, numerical simulations using the k − ω SST model were found to return satisfactory
results. Their operating conditions are far from the critical point, however as the sCO2 compressor
operating condition approaches the critical point, deviations from the reference data start to become
apparent. In the more challenging case, the disagreement with experimental data might be partially
due to the modelling limitations within the two-phase regions. Moreover, RANS models might have
the limitation on the investigation of turbulence close to the critical point, where a large local property
variation was observed. Alternatively, DNS has been applied to the performance simulation for CO2
compressors and varying properties [56].
Heat Exchanger
The development of the sCO2 heat exchanger including recuperator [57], hot-side [58] and cold-
side heat exchangers [59] are of particular interest. Heat exchangers are important to sCO2 power
cycles as they are the main components for heat addition and rejection. Since CO2 exhibits different
behaviours compared to conventional heat transfer fluids [49], the work on the heat exchanger using
sCO2 is primarily focusing on the development of heat transfer correlations theoretical [60, 61] and
experimental [62, 63, 64, 65]. From numerical simulations, it was found by Dang [60] that the low
Reynolds number k −ǫ model by Jones and Launder showed the best agreement with the experimental
data compared to three other models (the mixing length model by Bellmore and Reid, and the low
Reynolds number k−ǫ models by Launder and Sharma and by Myong and Kasagi). They also showed
that the the heat transfer coefficient is not significantly influenced by the turbulent Prandtl number. A
more detailed comparison of turbulence models was conducted by He et al. [61]. DNS results were
used as a benchmark dataset for model assessment. Among the selected turbulence models (classical
k − ǫ, k − ǫ Chien, k − ω model by Wilcox, Myong and Kasagi, Yang and Shihh and V2F), the
V2F model was found to perform best. Given the numerical results, which show a certain amount of
deviation with between different turbulence models and experimental data, there is a growing interest
in employing DNS to investigate the heat transfer behaviour for sCO2 [49, 66].
Seal
Sealing is very important for sCO2 cycles due to the high operating pressures. Lost gases requires a
significant amount of energy for re-injection into the closed loop. For the first sCO2 cycle at SNL,
labyrinth seals were used [7]. Recently, dry gas seals are being considered for sCO2 power cycles
due to their low leakage performance [67, 68, 69, 70]. Zakariya and Jahn [67] investigated the effect
of highly dense CO2 and operation close to the critical point. The centrifugal effect introduced by
the highly dense CO2 was found to be beneficial for seal performance, reducing seal leakage at the
expense of opening force. It was implied that the seal operating with CO2 close to the critical point
is not detrimental to seal performance. A further parametric study was performed by Zakariya and
Jahn [68]. It was found that dry gas seals with a wide dam are preferred, since the leakage is reduced
while the change in the lift force is minimal. Moreover, Bidkar et al. [70] performed a fluid, struc-
tural, and thermal analysis for a hydrodynamic face seals. Reynolds equation was found to be not
applicable for the modelling of turbulent flow within the CO2 seal. Only a one-way coupling strategy
is implemented in his analysis. Thatte and Dheeradhada [69] performed a coupled analysis of dry gas
seal using sCO2 , including thermal deformations.
Bearing
In the small sCO2 cycle test rig at SNL, a seal is positioned upstream of the foil bearing. A scav-
enge pump at the downstream of the foil bearing maintains constant pressure in the bearing cham-
ber [8]. The resulting operating conditions (see Figure 1.1) for the foil bearing are around 1.4 MPa
Experimental Work on Foil Thrust Bearings Section 2.3 17
Figure 2.8: Bulk film temperature (averaged over film thickness) of loaded bearing overlapped on the pressure
profiles for CO2 with real gas effect at 40 000 rpm, taken from Ref. [71].
and 300 K [7, 8]. It is believed to be a compromise between ensuring a low pressure downstream of
the seal, thereby low windage losses and a condition that allows foil bearing design using conven-
tional methods [8]. At this operating condition, CO2 density is up to 26 kg/m3 , resulting in Reynolds
number larger than 2000 [8], which is different from air bearings (laminar flow). The highly dense
gas also results in a larger centrifugal inertia forces compared to air bearing (density is 1.2 kg/m3).
Conboy [8] used the modified turbulent Reynolds equation to include real gas effects in foil thrust
bearings. They found that turbulent effects can increase the load capacity and friction torque sim-
ultaneously. Kim [72] developed a modified three dimensional Reynolds equation for foil journal
bearings. Real gas effects, turbulent flow and viscous heating effects are all included in this model,
and the performance of foil journal bearings using CO2 at different rotational speed and conditions
was investigated. A more recent work from Xu and Kim [71] investigated the three dimensional
performance of a hybrid (hydrostatic and hydrodynamic) thrust foil bearings. Very high temperature
increase across the thin film (see Figure 2.8) is obtained when CO2 is used as the working fluid. This
highlighted the importance of accurate thermal prediction.
2.3 Experimental Work on Foil Thrust Bearings

Foil bearing are key components for sCO2 turbomachinery as introduced in Chapter 1. Experiments
have been conducted to investigate the steady state and dynamic performances of foil journal and
Figure 2.9: Prototype bearing with one top foil removed, taken from Ref. [23].
Table 2.1: Geometry and physical parameters for thrust foil bearings in Ref. [26, 73].
Parameters Magnitude Parameters Magnitude

Number of pads 6 Bump foil material Inconel X-750
Outer diameter 101.6 mm Bump foil thickness 0.102 mm
Inner diameter 51.0 mm Bump foil pitch 5.00 mm (averaged)
Pad arc extent 45° Bump half length 1.60 mm
Pad taper extent 15° Bump height 0.50 mm
Top foil material Inconel X-750 Top foil thickness 0.150 mm
thrust bearings. The design, fabrication and performance testing of foil thrust bearings is represented
in Ref. [23]. Their prototype foil thrust bearings is shown in Figure 2.9, which consists of six thrust
pads. Five non-uniform bump foil strips in the radial direction are mounted underneath the top foil
for each pad. The geometry and physical parameters of this prototype are provided in Table 2.1. The
side view of one thrust pad is depicted in Figure 2.10. The shape of the runner and the top foil shape
form a convergent shape (ramp region), followed by a flat region. The experimentally measured load
capacity in terms of friction torque is shown in Figure 2.11. This provides an effective starting point
for the development of foil thrust bearings. The prototype is widely used in literature to investigate
the performance of foil thrust bearings [73, 74, 75].
Dickman [73] also investigated the steady state performance of foil thrust bearings at different load
and rotational speeds. However only repeatability of experiments was assessed, and the uncertainty
analysis is missing in his work. The thermal performance of foil thrust bearings was investigated by
Experimental Work on Foil Thrust Bearings Section 2.3 19
Figure 2.10: Nomenclature for foil thrust bearings, taken from Ref. [76].
Figure 2.11: Torque versus load at 23 000 rpm, taken from Ref. [23].
Figure 2.12: Measured trailing edge temperature gradients, taken from Ref. [77].
Dykas and Radil [77, 78]. Three thermocouples were mounted in a test bearing near the trailing edge
at different radial location to calculate the temperature gradient. The results for different rotational
speed and heat loss are shown in Figure 2.12. For the experimental work conducted by Dykas et
al. [23] and Dickman [73], only the integral values, load capacity and friction torque, were measured.
The corresponding bearing clearance, pressure, and temperature distributions are not given. Radil and
Zeszotek [79] performed an experimental investigation into the temperature profile of a foil journal
bearings. It was found that both journal rotational speed and radial load are factors determining heat
generation. The rotational speed played a more significant role on the temperature increase.
Balducchi et al. [74] investigated the start-up and shut-down torque, which varies linearly with
load and does not depend on the maximum speed. The other important finding from their work is
that the thermal regime is significant during the start-up tests. Balducchi et al. [75] also studied the
dynamic performance of foil thrust bearings by employing an electromagnetic shaker, which was ver-
tically mounted to the upper part of the shaft. The dynamic characteristic of structure and foil bearings
were investigated under three different static loads (30 N, 60 N, and 90 N) and excitation frequencies
(from 150 Hz to 750 Hz). It was observed that the stiffness increases with the excitation frequency and
Fluid Simulation for Gas Bearings Section 2.4 21
static load, while the equivalent viscous damping decreases with the excitation frequency and static
load. Kim et al. [80] identified the dynamic characteristic of foil thrust bearings using base excita-
tion. A single degree-of-freedom model was utilised to investigate stiffness, damping, and loss factor
(defined by damping×excitation frequency/stiffness) of foil thrust bearings at the different dynamic
load and excitation frequency. The same trend of stiffness and damping coefficients with excitation
frequency was reported in Ref. [75].
2.4 Fluid Simulation for Gas Bearings

Different approaches have been used to simulate gas film. Reynolds equation is widely used to model
the fluid flow within foil bearings. This approach has been proven as a suitable tool. In this section,
various versions of Reynolds equation and the CFD approach are introduced.
2.4.1 Classical Reynolds Equation

Osborne Reynolds was the first to propose a two-dimensional differential equation to solve the pres-
sure distribution within fluid bearings [81]. This equation is a simplified form of the Navier-Stokes
equations with several fundamental assumptions:
1. All inertial forces are negligible. This indicates that Reynolds equation is only valid at small
Reynolds numbers;
2. The pressure gradient perpendicular to the rotor and stator is small, compared to the axial and
dP
circumferential pressure gradients, dz
= 0;
3. The velocity gradient across the film thickness is largest. Therefore, the velocity gradients in
du du
the axial and circumferential direction are neglected, dr
= dθ
= 0;
4. The body forces are negligible;
5. The flow is laminar;
6. The fluid density is constant; This assumption holds when oil is the lubricant. However, for gas
bearings, the compressible Reynolds equation can be used.
7. The fluid is Newtonian;

Figure 2.13: Coordinate system for small perturbations, taken from Ref. [84].
8. Non-slip boundaries exist at the rotor and stator.
Hence, the three dimensional Navier-Stokes equations are reduced to the two dimensional partial
differential Reynolds equation [82],
1 ∂ 1 ∂p 1 ∂ 1 ∂p ∂(ρ h)
(r ρ h3 )+ 2 (ρ h3 )=Λ . (2.1)
r ∂r µ ∂r r ∂θ µ ∂θ ∂θ
The detailed derivation process can be found in Bruckner’s PhD thesis [82], including equations
in Cartesian and Cylindrical coordinates, respectively.
To solve this partial differential equation, the finite difference method is usually utilised. Detailed
discretisation procedures can be found in Appendix A. For the iteration procedure, the Gauss-Seidel
method, the preconditioned conjugate gradient method, successive over-relaxation method, and the
multigrid method can be used. These iteration methods were reviewed and compared by Wang et
al. [83]. Once steady state performance is achieved, the load capacity F and friction torque T from
Reynolds equation are calculated as [84, 85],
Z Z
F = (p − pa )dr dθ , (2.2)
r θ
h ∂p µ ω r 2
Z Z
T = [ + ]r dr dθ . (2.3)
r θ 2 ∂θ h
Figure 2.14: Comparison between Reynolds equation and experiment, taken from Ref. [76].
For some applications, the dynamic performance of foil bearings including rotordynamic stiff-
ness and damping coefficients is also of interest. The small perturbation approach has been used
to obtain dynamic force coefficients [84]. Once the steady state results of pressure distribution and
film thickens are obtained, the steady state position of the rotor is perturbed by small translational
displacement (△z), small rotational displacement (△φ and △ψ), small translational velocity (△ż),
and small rotational velocities (△φ̇ and △ψ̇) as shown in Figure 2.13. The pressure perturbations are
then solved. Consequently, the rotordynamic stiffness and damping coefficients are calculated. The
detailed procedure can be found in Refs. [84, 86]. In summary, Reynolds equation (Equation 2.1)
can be used to calculate the steady state and dynamic performance of foil thrust bearings as long as
the flow remains laminar. The comparison of results from Reynolds equation (Equation 2.4.1) and
experiment is depicted in Figure 2.14. Good agreement is achieved.
Atmospheric pressure 0.1 MPa and temperature 298 K are usually taken as the operating condi-
tions for air foil thrust bearings [26, 76, 84, 85]. The air density at the operating condition is only
1.2 kg/m3. Reynolds number around 100 is obtained using the bearing geometry in Table 2.1. The
non-dimensional pressure distribution (normalised by 0.1 MPa), obtained from Reynolds equation
(Equation 2.4.1), is shown in Figure 2.15. The pressure increases due to the convergent shape between
the rotor and top foil. The peak pressure occurs at the end of the ramp region. The steady state per-
formance of the foil thrust bearings is shown in Figure 2.16. Under the investigated film thickness and
rotational speeds, a maximum load of 200 N is obtained, while the corresponding torque is approxim-
ately 28 Nmm (power loss is 147 W). Load and torque have a roughly linear relationship with initial
Figure 2.15: Non-dimensional pressure distribution of foil thrust bearings obtained from Reynolds equation,
taken from Ref. [76].
minimum film thickness as shown in Figure 2.16(a). Both load and friction torque increase with the
decreasing film thickness.
2.4.2 Turbulent Reynolds equation
The assumptions for Reynolds equation listed in the previous section might not be appropriate for
some cases. For example, examination of CO2 fluid film at SNL reveals Reynolds number above
1600, suggesting turbulent flows. The classical Reynolds equation was modified to include turbulence
effect by including the correction factors Gr and Gθ ,
1 ∂ 1 ∂p 1 ∂ 1 ∂p ∂(ρ h)
(r Gr ρ h3 )+ 2 (Gθ ρ h3 )=Λ . (2.4)
Turbulent lubrication is developed from the work of Ng and Pan [87], Constantinescu [88] and
Hirs [89]. Constantinescu [88] used the Prandtl mixing length to calculate Reynolds stress tensor. He
experimentally showed that at a Reynolds number below 2000, the flow is in a laminar flow regime.
For Reynolds numbers (based on film thickness) greater than 2000, turbulence begins to affect the
correction factors in an exponential manner, as shown in Figure 2.17. Using the Prandtl mixing length
concept, and assuming strong Couette flow, Constantinescu determined the following expressions,
40.5405
Gr = , (2.5)
40.5405 + (k 2 Ret )0.65
Figure 2.16: Load and torque of foil thrust bearings obtained from Reynolds equation, taken from Ref. [76].
′
Figure 2.17: A plot of the G functions for each of the Ng and Pan [87], Constantinescu [88], and Hirs [89]
turbulence models as a function of rotational Reynolds number, taken from Ref. [24].
Table 2.2: The Ng and Pan turbulent coefficients for various critical Reynolds numbers. Applicable to Equa-
tions 2.8 and 2.9.
Reynolds Number Kϕ nϕ Kr nr
50 000< Ret 0.0388 0.8 0.213 0.8
10 000< Ret <50 000 0.025 0.84 0.0136 0.84
5000< Ret <10 000 0.025 0.84 0.0088 0.88
Ret <5000 0.0039 1.06 0.0021 1.06
22.6415
Gφ = , (2.6)
22.6415 + (k 2 Ret )0.725
where:
k = 0.125 Re0.007
t . (2.7)
Ng and Pan [87] utilised the concept of the law of the wall for turbulent shear flows to calculate
turbulent stresses as the function the mean velocity gradient. To obtain the eddy diffusivity, they used
Reichardt’s formula and experimentally obtained the necessary constants for various critical Reynolds
numbers. Ng and Pan’s turbulence model is expressed by,
12
Gr = , (2.8)
12 + Kr Rent r
12
Gφ = n . (2.9)
12 + Kϕ Ret φ
The constants K and n are defined as a function of Reynolds numbers, as detailed in Table 2.2.
A further model by Hirs [89] adopts a bulk flow approach that requires no physical representation
of the turbulent transport mechanism. One advantage to such an approach is that the unknown coef-
ficients can be determined from bulk-flow measurements without determining the velocity profiles.
The Hirs model is expressed below for three different Reynolds number ranges for smooth surfaces,
306.122
Gr = if Ret < 105 , (2.10)
Re0.76
t
Gr = 1.0 if Ret < 2060 . (2.11)
174.672
Gφ = 0.75
if Ret < 105 , (2.12)
Ret
Gφ = 1.0 if Ret < 977 . (2.13)
Among these turbulence correction factors, shown in Figure 2.17, it was experimentally confirmed
that the Hirs model is the most accurate [90].
Figure 2.18: Non-dimensional pressure distribution of foil thrust bearings obtained from turbulent Reynolds
equation, taken from Ref. [8].
The pressure of 1.4 MPa and temperature of 300 K are used as the ambient condition for foil
thrust bearings operating with CO2 at SNL. The non-dimensional pressure distribution (normalised
by 1.4 MPa), using turbulent Reynolds equation, is illustrated in Figure 2.18. A similar pressure
distribution is obtained compared to Figure 2.15. Again, the peak pressure occurs at the end of the
ramp region. The steady state performance is also plotted in Figure 2.19, including load and power
loss. Here, the initial minimum film thickness is set as 5 µm. It is found that the maximum load can
be over 1000 N, but a very high power loss, around 1800 W, is obtained.
2.4.3 Computational Fluid Dynamics

Due to fluid property of CO2 , some assumptions for Reynolds equation outlined in Section 2.4.1 might
not be appropriate. For example, constant density, negligible inertia forces. It is recommended that
a more comprehensive approach like CFD can be utilised to investigate CO2 foil bearings. Chen and
(a) Load.
(b) Power loss.
Figure 2.19: Load and power loss of foil thrust bearings obtained from turbulent Reynolds equation, taken
from Ref. [8].
Structural model Section 2.5 29
Hahn [91] investigated the suitability of CFD for solving steady state performance of hydrodynamic
bearings. Various geometries were investigated including step pad bearings, slider bearings, journal
bearings and squeeze-film dampers. The suitability of neglected inertia in the derivation of Reynolds
number were investigated, which is found to be appropriate. The results from Ref. [91] for slider and
step bearings are shown in Figure 2.20. Good agreement is achieved at the very low Reynolds number
of 1, however differences are observed at the high Reynolds number (1000). Brajdic-Mitidieri [92]
validated CFD as a suitable tool for predicting the fluid flow in simple converging bearings. Guo
et al. [93] applied CFD to investigate the steady state and dynamic performances of hydrodynamic,
hydrostatic and hybrid bearings. Reasonable agreement was obtained between CFD and Reynolds
equation.
For some cases, the bearing geometry is either too complex or cavitation is presented. Papado-
poulos et al. [94] studied the flow pattern and performance of a dimpled parallel thrust bearing by
CFD. Zhang et al. [95] designed a hydrodynamic water-lubricated step thrust bearings using CFD in
conjunction with the Zwart-Gerber-Belamri cavitation model. Parametric studies were undertaken
with different pad dimensions, step heights, step positions, water film thicknesses and rotational
speeds.
In literature, good agreement is obtained between CFD and Reynolds equation [91, 92] at low
Reynolds number (see Figure 2.20). The results also confirmed the assumptions of Reynolds equa-
tion and confirmed that Reynolds equation is an adequate approach to investigate gas bearings per-
formance, when the studied bearing geometry is simple. However, the difference between CFD and
Reynolds equation is observed at large Reynolds number.
2.5 Structural model
In a typical configuration of foil bearings, the top foil is affixed to the bearing housing on one side,
and on the opposing side it sits at the height of its bump understructure. A partially ramped profile
is formed. During the hydrodynamic action, the initial top foil shape is deflected due to pressure
increase within the foil bearing. This deformation has to be modelled since it is a significant fraction
compared to the clearance between top foil and stator.
(a) Slider bearings.
(b) Step bearings.
Figure 2.20: Pressure distribution in a long slider bearing using Reynolds equation and CFD, taken from
Ref. [91].
Figure 2.21: Nomenclature for bump strips, taken from Ref. [76].
The deforming parts are top and bump foils. A variety of structural deformation models exist in
literature for journal and thrust foil bearings. The bump foil is taken as a spring-like structure with a
certain stiffness to support the top foil. The nomenclature for a bump strip is depicted in Figure 2.21.
Heshmat [11] was among the first researcher to analyse foil bearings, and to propose a simplified
method (see Equation 2.14) to calculate the equivalent stiffness of the bump foils. In his model, the
entire top foil was regarded as a single element,
2 p a s l0 3
α= ( ) (1 − ν 2 ) . (2.14)
hb E tb
Iordanoff [96] proposed a more advanced structural model that divided the bump foil into two
categories: fixed-free and free-free. For the welded bump, a fixed-free arrangement is assumed, while
a free-free end is applied for the free bumps. The local compliance for these two bump types are
defined as,
12 l03 i J (1 − ν 2 )
Fixed-free end: s = , (2.15)
E e3 sin3 (α/2)
6 l03 i I (1 − ν 2 )
Free-free end: s = . (2.16)
E e3 sin3 (α/2)
The definition of each parameter can be found in Ref. [96]. This approach is widely used in liter-
ature [26]. However, it was found that these models tend to underestimate the magnitude of the bump
foil stiffness, since the interactions between bumps are not considered. Le Lez et al. [97] proposed
a model that described the foil bump structure as a multi-degree of freedom system of interacting
bumps. Good agreement exists when compared with finite element simulations. However, a large dif-
ference exists between this model and the previous analytical models presented by Heshmat [11] and
Iordanoff [96] (see Figure 2.22), especially at high load condition. For all test cases in Figure 2.22,
the total equivalent force corresponds to a uniform pressure of 200 000 Pa. Hryniewicz et al. [98] pro-
posed a new two-dimensional bump foil model. The model was able to consider the detailed geometry
of bump foil structure and the interaction between bumps. Feng et al. [99] developed a link-spring
model to calculate the stiffness. Using the link-spring model, each bump was simplified to two rigid
links and a horizontally spaced spring. The stiffness was then determined from Castigliano’s theorem.
Gad et al. [100] also developed a structural stiffness model for Generation II foil thrust bearings. In his
study, the lateral deflection of the flat segment between bumps was investigated. His model considers
the interaction between bumps and the friction between the bump foil and the surrounding structure
was considered. A comparison of the various bump foil models is shown in Figure 2.23 [100]. The
analytical model from Gad et al. [100] provides the close agreement with finite element analyses.
Experiments were also conducted to test the structural stiffness of foil bearings. Rubio and San
Andrés [101] tested the structural stiffness of a commercial foil bearing. It was found that the stiffness
of a single bump was most sensitive to the dry-friction coefficient, the bump length, and the bump
end conditions. Furthermore, the structural stiffness increases as the bump foil deflection increases,
which was taken as the hardening effect. These phenomena correspond to the analytical model from
Le Lez et al. [97], Hryniewicz et al. [98], Feng et al. [99], and Gad et al. [100]. In addition, bump and
top foils provide coulomb damping due to the relative sliding, which is essential for stability of the
machine [102]. However, little work has been done to characterise the damping coefficients of bump
foils. A theoretical model was firstly developed by Ku and Heshmat [103] to calculate equivalent
viscous damping coefficients of a bump foil strip. This coefficient was determined based on the area
of a closed hysteresis loop of the journal center motion. The same approach was also applied by Le
Lez et al. [104] to investigate the dynamic stiffness and damping of a single bump of the foil structure.
Experiments undertaken by Balducchi et al. [75] provided the frequency dependent structure stiffness
and damping coefficients at different load conditions as shown in Figure 2.24.
Figure 2.22: Bump deflection for various load distributions, taken from Ref. [97].
Figure 2.23: Bump deflection for various load distributions, taken from Ref. [100].
Thermal Performance Section 2.6 35
(a) Stiffness. (b) Damping.
Figure 2.24: Structural stiffness and damping, taken from Ref. [75].
For the top foil, two-dimensional shell or one-dimensional beam-like models were developed
by San Andrés and Kim [105]. These models allowed sagging between bumps. Advanced two-
dimensional plate models were applied by Bruckner [82] and Lee and Kim [106] in simulating de-
formation of thrust bearings. In addition, coupling to the commercial structural-analysis software
(ANSYS) was also considered to improve the accuracy of the deformation, but with an associated
increase in computational cost [94, 107]. More recently, Lehn et al. [108] proposed a new approach
that models top and bump foils as a two-dimensional thick shell. The shear, membrane, and bending
effects are all considered. This model provides an accurate prediction for the structural deform-
ation within foil bearings and shows good agreement with finite element simulations as shown in
Figure 2.25.
2.6 Thermal Performance

While there are many inherent advantages to foil bearings, there are also several limitations. The most
important problems during experiments at high speed are high local temperatures and temperature
gradients, which causes thermal runaway of foil bearings [7]. This is due to the weak conduction
rate of the thin foil structure and low heat capacity of air. The small contact areas between top foil
and bump foils and low heat capacity prevent conductive or convective heat removal from the fluid
film. Hence, the temperature distribution within foil bearings has to be carefully considered in design,
particularly for bearings operating with high rotational speed or high load.
Figure 2.25: Vertical deflection of the bump top points for friction coefficients 0.0 and 0.2 (right), taken from
Ref. [108].
Salehi et al. [110] performed the first study to model and characterise thermal properties of gas
foil journal bearings. The Couette flow approximation to the energy equation was implemented with
the compressible Reynolds equation. This simplified method provides a reasonable agreement with
experimental data, with a maximum over-prediction of 19 %. Sim and Kim [109] presented a thermo-
hydrodynamic model that accounted for the thermal contact resistance between the top foil, bump
foil and bearings housing as shown in Figure 2.26(a). Additional thermal resistances were included
between top and bump foils, and bump foil and bearing housing. The mixing effect between the
leading and trailing edges of top foils was also investigated. The suction flow mixing ratio at the
groove region was obtained and then applied to their model. This suction flow mixing ratio λs was
defined as the ratio of the cold suction flow rate to the mixed inlet flow rate,
ṁsuc
λ= . (2.17)
ṁinlet
It was noted that a large mixing ratio provides an effective cooling mechanism in the mixing
chamber. The mixing chamber is the groove region between the leading and trailing edges of the top
foil. This mixing model was also applied in Ref. [111]. In this paper, the computational domain was
expanded to the surrounding structures including two plenums, bearing sleeve, housing, and rotor
Thermal Performance Section 2.6 37
(a)
(b)
Figure 2.26: (a): Thermal resistance at the bump contacts to the top foil and the bearing housing, along with
the thermal resistance model of a bump arc, (b): simplified CFD model of an inlet mixing chamber, taken
from Ref. [109].
Figure 2.27: Simplified thermal domains of the surrounding structures around the radial foil bearing close to
turbine and thermal boundary conditions, taken from Ref. [111].
exposed to the plenums shown in Figure 2.27. In order to maintain temperature at the rotor and top
foil below 300 ◦C, cooling air must be provided. The feasibility and effectiveness of radial injection
cooling were experimentally tested by Shrestha et al. [112]. They concluded that radial injection
cooling can provide a more uniform sleeve temperature in the circumferential direction and a smaller
axial gradient of shaft temperature compared to an axial cooling strategy.
The aforementioned work is for the thermo-hydrodynamic modelling of foil journal bearings.
Little work had been undertaken to model the temperature field for foil thrust bearings. The Couette
flow approximation of the energy equation from Salehi et al. [110] was used by Gad and Kaneko [85]
to predict the temperature distribution for air foil thrust bearings. Lee and Kim [113] conducted a
three-dimensional thermo-hydrodynamic analysis of Raleigh step air foil thrust bearings with a forced
air cooling and the optimum cooling air pressure was found for the reference simulation condition.
DellaCorte et al. [114] reported a reduction of 30 % in load capacity for a tested gas foil bearing
when the operating temperature increases from 25 ◦C to 650 ◦C. This reduction is attributed to the
structural deformation due to thermal stress and centrifugal growth. In this way, this deflection has to
Gap in Literature Section 2.7 39
be accounted in accurate performance prediction of foil bearings. Typically, linear thermal expansion
theory is applied to calculate the thermal deflection [113, 115, 116].
A numerical thermo-elastohydrodynamic analysis of a novel air radial foil thrust bearing was
reported by Lee and Kim [116]. At a rotational speed of 50 000 rpm and a load of 200 N, the maximum
temperature rise was approximately 70 K for a case where cooling flow is prescribed. However, for
typical foil journal or thrust bearings, there is no pressure gradient between the inner and outer radii,
which results in no forced cooling flow through the bump foils. Therefore, the temperature increase
in typical foil bearings, recirculating coolant by natural pumping, can be much higher than the value
reported in Ref. [113]. The turbulent thermo-elastohydrodynamic analyses of hybrid thrust bearings
and journal bearings with CO2 was conducted by Xu et al. [71] and Kim [28]. A high temperature
increase was observed in their simulations. In this case, the thermal effect on the performance of foil
thrust bearings has to be considered.
2.7 Gap in Literature

At the start of this Chapter we discussed foil bearings as the critical components for sCO2 turboma-
chinery systems. We then presented experimental studies and different modelling approaches, in-
cluding Reynolds equation and its medication for turbulent flow, and CFD for foil bearings. We also
highlighted the importance of thermal modelling for the operation of foil bearing. Several research
questions remain:
• Is Reynolds equation still applicable for foil bearings with the highly dense CO2 ?
• How does the new working fluid CO2 influence foil bearing operation, including steady state
(load and power loss) and dynamic (stiffness and damping) performances?
• Are temperature distributions affected and what are good cooling strategies?
To close this gap in literature, we now begin investigating the performances of foil thrust bearings
with CO2 using the high-fidelity multiphysics simulation tools. This approach allows us to explore
that are not accounted for by Reynolds equation.
Chapter 3
Fluid Simulations
3.1 Introduction
Despite the strong interest in utilising foil bearings for sCO2 cycles, little research has been done on
developing bearings for this application. The bearings, at the core of the turbomachinery system, play
an essential role for the design of turbomachinery components, and should be carefully studied and
designed. When used as the working fluid, high pressure supercritical CO2 is far more dense than
air, less viscous, and can be a highly non-ideal gas. These factors present challenges when predict-
ing the performance of foil bearings. There is the potential for turbulent flow, highly compressible
flow, non-linear thermodynamic properties and significant fluid inertia due to high density and high
speed operation. The conventional Reynolds equation, or its modifications, might not be adequate to
account for all these effects simultaneously. The simulation of foil bearings is complex, and consists
of three aspects: fluid flow, structural deformation and heat conduction. In this Chapter, the devel-
opment of the in-house fluid solver Eilmer for simulations of flow within foil bearings is described.
Section 3.2 describes the basic structure of the in-house CFD code Eilmer [117]. The modifications of
Eilmer for laminar and turbulent simulations and the corresponding validation cases are documented
in Sections 3.3 and 3.4, respectively.
3.2 Description of Fluid Solver

In order to accurately predict the fluid performance within foil bearings for sCO2 cycle applications,
three dimensional flow simulation is selected over conventional approaches using the two-dimensional
Reynolds equation. For the simulation of the fluid domain in foil bearings, the compressible flow
solver, Eilmer (written in C++), developed at The University of Queensland [117, 118] is used. For
42 Chapter 3 Fluid Simulations
much of the code’s history, it has been used to simulate hypersonic flows and, more recently, it has
been extended to turbomachinery flow fields [119]. Eilmer is an integrated collection of programs
that solves the compressible Navier-Stokes equations on multi-block structured grids to provide time-
accurate flow simulations. The governing equations are expressed in integral form over cell-centred
finite-volume cells, with the rate of change of conserved quantities in each cell specified as a sum-
mation of the mass, momentum and energy fluxes through the cell interfaces. Detailed description of
Eilmer can be found in Appendix B.
The governing equations are closed using equation of state. Eilmer has models for gases with con-
stant specific heat, thermally perfect gases and real gases. For real gas properties of sCO2 , a real gas
equation of state is implemented directly in the computation. Alternatively, a look-up table approach
is also implemented. For the look-up table implementation, a thermodynamic mesh based on temper-
ature T and density ρ is generated prior to the simulation with the NIST database REFPROP [120].
During simulation, a bilinear interpolation method is used to calculate the thermodynamic proper-
ties from the tabulated data. This reduces the computational cost significantly and was found to be
15 times faster than executing REFPROP equation of state functions directly, with no reduction in
solution accuracy.
3.3 Modification of Fluid Solver for Laminar Simulations of Foil

Bearings
Eilmer was developed for the high-speed compressible flow within the hypersonic regime [117]. To
simulate fluid flow with foil bearings using Eilmer, several new features were added into the solver.
This section discuss the additions for the laminar simulations of foil bearings.
3.3.1 Moving Wall Boundary Condition
The first part of the work is to implement a moving wall boundary condition for the rotor in foil
bearings. This implementation is straightforward, the translational or/and rotational velcoties are
set at the boundary interfaces directly, while other parameters are the same as for conventional wall
boundaries.
Modification of Fluid Solver for Laminar Simulations of Foil Bearings Section 3.3 43
3.3.2 Periodic Boundary Condition

Foil thrust bearings typically consist of several thrust pads, only one thrust pad is simulated to save
on computational cost. In this case, a periodic boundary condition is required to exchange flow
conditions. For the scalar variables (pressure and temperature), the parameters are swapped directly
between the connected periodic boundary faces. However, for the vector parameters, a helper function
is implemented to rotate the vectors.
3.3.3 Reconstruction
A non-zero mass flux through wall boundaries is obtained when prescribing the wall boundaries to
curved faces. This is identified as a code bug during the reconstruction process. The reconstruction is
to obtain the inviscid fluxes at cell interfaces, which is done by evaluating the left and right values at
a cell interface in the one-dimensional direction [121]. The non-zero mass flux is attributed to the use
of the cell centre data in the global frame of reference. Part of source codes for these modifications
can be found in Appendix The bug is fixed by reconstructing in the local frame of reference.
3.3.4 Taylor-Couette Flow
Figure 3.1: Computational domain for Taylor-Couette flow, taken from Ref. [122].
Taylor-Couette flow is used as the first test case to verify the moving wall as well as periodic
boundary conditions. Several examples of compressible Taylor-Couette flow from Ref. [122], an
annulus with inner radius 215.5 mm and gap width 3.1 mm are selected. The axial extent of the
Table 3.1: Parameters for simulations, used match the experimental conditions as reported in Ref. [122].
Case Low Intermediate High

pressure pressure pressure
Pressure, Pa 10 100 1000
Rotor Temperature, K 344 348 351
Stator Temperature, K 344 350 366
Taylor Number 1.1 12 105
annulus is 5 times the gap width. The computational domain is shown in Figure 3.1. The outer
cylindrical surface (the housing) of the annulus is fixed and the inner surface (the rotor) is moving
with a rotational speed of 27 600 rpm. Three different pressures are simulated to cover a range of
cases, with and without Taylor vortices. Other parameters used for the simulations are summarised in
Table 3.1.
Taylor number is commonly used to define different flow states, for incompressible flows in rotat-
ing environments, a common form of the Taylor number is,
r
Ui d d
Ta = , (3.1)
υ Ri
where d denotes with width of the gap, Ri the inner radius, and Ui the peripheral velocity of the inner
cylinder.
For the low pressure case with Taylor number less than 41.3 (critical Taylor number), the expec-
ted velocity profile is approximately linear across the narrow gap, and the temperature profile has a
parabolic shape with maximum temperature near the centre of the gap. Figure 3.2 shows the results
of velocity and temperature for different methods. The apparent difference is caused by the slip-wall
boundary condition modelled in DSMC method from Ref. [122]. A no-slip boundary condition is
used by Eilmer.
For the intermediate pressure case, shown in Figure 3.3, similar results of temperature and tan-
gential velocity exists for Eilmer and other codes [122]. The Taylor number in this case is 12 and no
vortices are expected. The profile of tangential velocities is approximately linear and the temperature
profile is parabolic. The agreement between numerical schemes is good. In the low-pressure cases,
where the Knudsen number is about 0.1, the velocity profiles exhibit increased gradients near the wall
400 1
Eilmer Eilmer
DSMC DSMC
Non-dimensional velocity
390 0.8
Temperature, K
380
0.6
370
0.4
360
0.2
350
340 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Non-dimensional radial position Non-dimensional radial position
(a) (b)
Figure 3.2: Comparison of temperature (a) and axially averaged velocity (b) profiles in radial direction at low
pressure condition.
surfaces and velocity slip at the 10% level. However, as the Knudsen number decreases to about 0.01
for the intermediate pressure cases, the DSMC simulations show that the velocity profile is nearly
linear, and the velocity slip is an order of magnitude less than at the low pressures. This verifies that
Eilmer can simulate Taylor-Couette flow with the modifications.
For the high pressure case, the Taylor number has exceeded a critical value, making the flow
fully three dimensional. Schlichting [123] suggested that in incompressible flow for Taylor number
in ranges between 41.3 and 400, laminar flow with Taylor vortices exists. In this range, the flow can
be further characterised as axisymmetric Taylor vortices, wavy Taylor vortices and others. However,
the Taylor number for the transition from axisymmetric Taylor vortices to wavy vortices is not firmly
T
established. For instance, the transition is theoretically predicted to occur at Tcrit
= 1.1 for aspect
ratio η = 0.85 for infinitely long cylinders [124, 125], whereas experiments indicate a range of higher
values between 1.14 and 1.31 for η=0.80-0.90, depending on experimental conditions [126].
For the high pressure case, the ratio of Taylor number to critical Taylor number is 2.22, and the
flow will eventually evolve into unstable wavy vortices as shown in Figure 3.4. The simulation in
Ref. [122] is performed with 2D axisymmetry, therefore a stable axisymmetric vortices is obtained.
Figure 3.4 recorded temperature and pressure variations at the circumferential angle of 0° with sim-
ulation time in Eilmer (three dimensional), where the temperature and pressure will oscillate with a
certain frequency, eventually.
400 1
Eilmer Eilmer
DSMC DSMC
390 CTDNS 0.8
Temperature, K
380
0.6
370
0.4
360
0.2
350
340 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(a) (b)
Figure 3.3: Comparison of temperature (a) and axially averaged velocity (b) profiles in radial direction at
intermediate pressure condition.
If the rotational speed of the rotor is reduced to 18 400 rpm, while the other parameters remain
the same with the high pressure case in Ref. [122], the axisymmetric vortices can be simulated. The
velocity profile (averaged over the axial direction) has changed to an S-shapes curve as shown in
Figure 3.5. This velocity profile characterises a flow with a higher gradient at the walls, due to
enhanced radial transport of fluid induced by the vortices. The axially averaged temperature profile
is much flatter than the parabolic profile of the lower Taylor number cases. This averaged shape also
exhibits steeper gradient at the walls, which induce a high heat flux. Again, these changes are due
to the presence of vortices and the associated increase in radial transport of momentum and energy
across the gap.
3.3.5 Hydrostatic Air Thrust Bearing
With high density CO2 as the operating fluid, high centrifugal inertia effects cannot be ignored. A
hydrostatic air thrust bearing is used to verify the suitability of Eilmer in simulating thrust bearing as
well as modelling the centrifugal inertia force. Garratt et al. [127] considered the centrifugal inertia
effects in high-speed hydrostatic air thrust bearings. They used a Reynolds equation modified for
compressible flow to model the dynamics of pressurised air bearings in a simplified axisymmetric
geometry. The basic air-flow characteristics were analysed for various non-dimensional speed para-
meters λ under steady-state condition when the bearing faces are fixed at a constant distance. Their
1350 400
1300
380
1250
Temperature, K
Pressure, Pa
1200 360
1300
1150 1290 340 372
1280 368
1100 1270 364
1260 320 360
1050 15.6 15.8 16 15.6 15.8 16
1000 300
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Time, ms Time, ms
(a) (b)
Figure 3.4: Parameters variation of Taylor-Couette Flow at high pressure condition; (a): pressure; (b): temper-
ature.
modified Reynolds equation is,
∂ ∂p ∂
(p r ) − λ (r 2 p2 ) = 0 . (3.2)
∂r ∂r ∂r
The speed parameter λ, which relates to different rotational speed, is defined as,
3ρ̂a Ω2 rO
2
λ= . (3.3)
10 p̂a Ks
and characterises the magnitude of the inertia effects during high speed operation. ρ̂a is air density at
atmospheric pressure, and Ks is the dimensionless parameter relating the air density and pressure in
the perfect gas law,
p = Ks ρ . (3.4)
A schematic of their hydrostatic air thrust bearing is shown Figure 3.6. The outer radius is defined
as 0.05 m, the size of inner radius will be different based on narrow bearings or wide bearings, and
the gap height of the bearing is 10 µm.
The working fluid is dense air with the constant density of 10 kg/m3 . Using this dense gas can
largely reduce the rotational speed at the same bearing number λ, see Table 3.2. The viscous heating
effect can be minimused compared to that of using perfect air.
Pressure profiles obtained using Eilmer and the modified Reynolds equation (Equation 3.2) by [127]
T[0] vel.vector Magnitude

376 405 400
370 300
200
360
100
352 3.81
(a) Temperature contour. (b) Velocity contour.

380 1
Eilmer Eilmer
375 0.8
Temperature, K
370
0.6
365
0.4
360
0.2
355
350 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(c) Averaged temperature profiles in radial direction. (d) Axially averaged velocity profiles in radial direction.
Figure 3.5: Temperature and velocity for Taylor-vortices case.
Table 3.2: Rotational speed (in rpm) for different bearing speed parameter for perfect air and dense air.
Case λ=0 λ=2 λ=4 λ=6

Perfect air 0 142 994.89 202 225.32 247 674.42
Dense air 0 49 637.97 70 198.70 85 975.49
Figure 3.6: Schematic of hydrostatic air thrust bearing geometry, taken from Ref. [127].
for an inward pressurised bearing are shown in Figure 3.7, where (a) and (b) correspond to a wide and
narrow bearing, respectively. For the wide bearing, at low rotation speeds, the inward pressurisation
gives a continuously decreasing pressure from the outer to inner radii. The pressure becomes sub-
ambient in the middle of bearing at high speeds. The reason is attributed to the mass loss close to the
rotor. For the narrow bearing, the pressure profile remains almost linear for all speed number λ, as
shown in Figure 3.7. The pressure distribution from Eilmer shows good agreement against analytical
results from the modified Reynolds Equation. Eilmer correctly captures the inertial effects, which are
the cause of the dip in pressure observed for the wide bearings.
For inward pressurised bearings, the positive pressure gradient results in a purely inward flow
without of high-speed rotation. However, with rotation, the flow near the rotor becomes outwards,
and this region of outward flow increases in size for higher rotational speeds. The boundary between
the pressure driven inward flow and the outside flow due to centrifugal effects can be monitored by
a line of zero radial velocity. Figure 3.8 compares the position of the radial velocity line between
Eilmer and the modified Reynolds equation. The results confirm that Eilmer accurately captures the
inward and outward flows.
3.3.6 Wavy Thrust Bearing
This test case is used to study the suitability of Eilmer to simulate the pressure distribution in bearings
with complex pad geometries. The modeling of a wavy thrust bearing has been studied by Zhao et
4 1.3
λ=2 Eilmer λ=2 Eilmer
3.5
Non-dimensional pressure
λ=2 Reynolds Equation λ=2 Reynolds Equation
3 λ=4 Reynolds Equation 1.2
λ=4 Reynolds Equation
λ=6 Reynolds Equation λ=6 Reynolds Equation
2.5
2
1.1
1.5
0.5 1
0.1 0.4 0.7 1 0.9 0.92 0.94 0.96 0.98 1
(a) (b)
Figure 3.7: Steady state pressure distribution in an inward pressurised bearing for different bearing numbers:
(a): wide bearing; (b): narrow bearing.
1.2
λ=2 Analytical Result
λ=λc Analytical Result
1 λ=4 Analytical Result
λ=2 Eilmer
Non-dimensional gap
λ=λc Eilmer
0.8 λ=4 Eilmer
0.6
0.4
0.2
0
0.9 0.92 0.94 0.96 0.98 1
Non-dimensional radial position
Figure 3.8: Comparison of zero radial velocity line at different bearing number.
(a) (b)
Figure 3.9: Schematic of the geometry (a) and coordinate (b) of the wavy thrust bearings, taken from
Ref. [128].
al. [128] using Reynolds equation. Their numerical simulation of a self-acting thrust bearing was
developed using a system of two circular discs moving relative to each other as shown in Figure 3.9.
The Reynolds equation for this system, written in polar coordinate, is,
∂ p 1 ∂p 1 ∂ p ∂ρ h
(ρ h3 ) + ρ h3 + 2 (ρ h3 ) = 6 η ω . (3.5)
∂r r r ∂r r ∂θ θ ∂θ
The discretization of Reynolds equation 3.5 is based on Ref. [129] and provided in Appendix A.
The solution to the equation is evaluated using the Gauss-Seidel successive under-relaxation method.
The computational domain is a single thrust pad. To solve this equation, density and dynamic viscos-
ity of working fluid are calculated from thermo-physical software REFPROP [120].
Table 3.3: Wavy thrust bearing physical and operational characteristics.
Outer radius of discs 50.8 mm

Inner radius of discs 25.4 mm
Nominal average clearance, hc 0.010 mm
Wave amplitude, g 6.35 µm
Rotational speed 25 000 rpm
Pad angular extent 45°
Working fluid CO2
1.5e-05 2
Eilmer
1.8 Reynolds Equation
Dimensional pressure
1e-05
1.6
Film thickness, m
5e-06 hc 1.4
g
1.2
0
1
-5e-06 h=gsin(8θ-π/2) 0.8
0.6
-1e-05
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
Angle Angle
(a) (b)
Figure 3.10: Schematic of symmetric wavy film thickness and comparison of pressure at center radius between
Eilmer and Reynolds equation.
As a first comparison case a wavy bearing with a sinusoidal structure was studied. The geometry is
shown in Figure 3.10(a). The remaining geometry and operating parameters are defined in Table 3.3.
High pressure CO2 is used as the working fluid for this case, and the ambient conditions are 1.4 MPa
and 300 K, corresponding to SNL’s test conditions [8]. The inner and outer surfaces are regarded as
fixed pressure boundary condition, while the rotor and pad are modelled as moving and stationary
fixed temperature walls, respectively. The other two surfaces are connected with a periodic boundary
condition. A grid independence study was conducted for Eilmer as well as the finite difference code
for Reynolds equation. Figure 3.10(b) compares the results of pressure distribution at mean radius
between Eilmer and Reynolds equation. Good agreement between Reynolds equation and Eilmer
demonstrates the ability of Eilmer to correctly simulate wavy bearings.
To study a more foil bearing like geometry, which is characterised by a long compression ramp
and followed by an almost instantaneous expansion, case 2 explores the flow in wavy bearing with
a skewed sin-wave geometry. The ratio of compression to expansion part is now changed to 4:1 as
shown in Figure 3.11(a). The resulting pressure at the bearing mean radius for Reynolds equation and
Eilmer are shown in Figure 3.11(b). Good agreement is also achieved.
The results for load and friction torque integrated from Reynolds equation and Eilmer are shown
in Table 3.4. The relative error between Eilmer and Reynolds equation for both cases is almost the
same, 70 N for load and 14 Nmm for friction torque. The cause of this difference has to be further
1.5e-05 2
Eilmer
1.8 Reynolds Equation
1e-05
1.6
Film thickness, m
5e-06 hc 1.4
g
1.2
0
1
-5e-06 h=gsin(5θ-π/2) 0<θ<π/5 0.8

h=gsin(20θ-3π/2) π/5<θ<π/4
0.6
-1e-05
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
Angle Angle
(a) (b)
Figure 3.11: Schematic of a 4:1 wavy film thickness and comparison of pressure at center radius between
Eilmer and Reynolds equation.
Table 3.4: Comparison results of load (in N) and friction torque (in Nmm) obtained from Reynolds equation
and Eilmer.
Case Load Load Friction torque Friction torque

Eilmer Reynolds equation Eilmer Reynolds equation
1 374.20 308.51 64.41 50.45
2 723.60 650.46 64.35 50.49
Table 3.5: Operating condition of foil thrust bearings from SNL’s test loop [8].
Parameters Value
Working fluid CO2
Ambient pressure 1.4 MPa
Ambient temperature 300 K
Inner diameter 50.8 mm
Outer diameter 106 mm
Designed rotational speed 75 000 rpm
Designed minimum film thickness 5 µm
investigated.
3.4 Modification of Fluid Solver for Turbulent Simulations of

Foil Bearings
We now turn our attention to the case of turbulent flow within the bearing. High Reynolds numbers
are likely, due to the high density and low dynamic viscosity of high pressure CO2 . For foil bearings
running without a pressure gradient between inner and outer radii, the fluid flow is pure Couette
flow [130], and the transition to turbulence is governed by the rotational Reynolds number,
ρωrh
Re = , (3.6)
µ
when Re < Ret , the bearing flow remains laminar and when Re > Ret turbulence appears. Although
no experimental study has been undertaken to determine the limits, Souchet [131] recommended the
following values:
Ret = 1600 . (3.7)
The operating condition for foil bearings tested in the SNL are listed in Table 3.5. These conditions
result in a rotational Reynolds number of 2656, which is substantially larger than the critical value
recommended by Souchet [131]. Therefore, the flow regime within foil bearings is expected to be
turbulent at design condition.
Foil thrust bearings operating with CO2 exhibit turbulent flow due to the high density and low
Modification of Fluid Solver for Turbulent Simulations of Foil Bearings Section 3.4 55
viscosity of the working fluid. To close the momentum equations this requires a model for the eval-
uation of the turbulent shear stress. The Wilcox’s 2006 low-Reynolds k − ω turbulence model [132]
was previously implemented in Eilmer [133]. Section 3.4.1 details the implemented turbulence model
and its validation. However, this model has a high computational cost due to the requirement for fine
meshes close to wall boundaries and simulation instabilities due to very high aspect ratio cells within
the foil thrust bearing geometry. To reduce the computational cost and stabilise turbulent simulations,
a compressible wall function and a fourth-order artificial dissipation term are implemented in Eilmer
as discussed in Section 3.4.2 and 3.4.3.
3.4.1 Wilcox’s 2006 k-ω model
The Favre-averaged mass, momentum and energy conservation equations and the equations defining
the Wilcox’s 2006 k-ω model are described in Appendix D.
The most important differences between Wilcox’s 2006 k-ω model and earlier versions are the
addition of a “cross diffusion” term and a built-in “stress-limiter” modification. The addition of
“cross diffusion” (see σd in the ω equation) was suggested as a remedy for the original k-ω model’s
sensitivity to the freestream value of ω. The “stress-limiter” modification makes the eddy viscosity
a function of k, ω and, effectively, the ratio of turbulence-energy production to turbulence-energy
dissipation.
Turbulent Couette Flow Using Wall-Resolved Turbulence Model
The k-ω turbulence model is already implemented in Eilmer and validated for hypersonic flowfield
simulation [133], however, it is yet to be validated for turbulent flow in foil bearings. To validate
the turbulence model for the flow in foil bearings, an experiment of turbulent plane Couette flow
conducted by El Telbany et al. [134, 135] is selected as the test case. Their test rig [134, 135] was a
2440 mm long parallel walled channel, with the main measurement station at the centre of the channel
and 1980 mm from the blowing end. The belt forming the moving floor is 1200 mm wide. To eliminate
possible flapping, the belt is supported from underneath by an aluminimum-surfaced plate. For their
test case, the height between two parallel plates was 66 mm, while the moving speed of the belt was
12.84 m/s. Since the measured region is located at the centre of the whole test rig, the side walls had
little influence on the flow regime and a two-dimensional domain can be used for the simulation.
The corresponding experimental apparatus and computational domain is shown in Figure 3.12.
(a)
Moving wall, low stress
(Blowing End)
Fixed Pressure
66
Fixed Pressure
Stationary wall, high stress Pout
Pin
2440
(b)
Figure 3.12: (a) Experimental apparatus (taken from Ref. [134]) and (b) schematic diagram of the simulation
domain for Couette and Poiseuille flows, dimensions in mm.
The top boundary is a stationary fixed temperature wall, the bottom is a fixed temperature wall with
fixed translational velocity, and the west and east boundaries of the simulation domain are defined as
mixed boundaries. At these mixed boundaries, a fixed pressure is used, while zero gradient is applied
for velocities. The turbulence intensity Iturb and turbulent-to-laminar viscosity ratio µturb /µlam are
set to inflow values if the direction of mass flux is inward, otherwise their values are obtained from the
flow cells within the domain and adjacent to the boundary. The values of turbulence kinetic energy k
and specific rate of dissipation ω for turbulence model are then computed as,
3
k = (Iturb u)2 , (3.8)
2
k µlam
ω=ρ ( ) , (3.9)
µlam µturb
where u is the velocity of the cell adjacent to the boundary. El Telbany et al. [135] measured the
turbulence intensity in the experiment to be approximately 0.1. We used this value and a turbulent-to-
laminar viscosity ratio of 1.0 as the inflow values in the simulation.
Three different computational meshes were used. A coarse mesh (240×80 cells), a medium mesh
(480×160 cells) and a fine mesh (960×320 cells). For these meshes, cells were clustered to the walls
to ensure that the first y + value was lower than 1.0 and there were at least 15 cells in the wall-normal
direction within the boundary layer. The velocities at points between the two parallel plates were
extracted and compared for the different meshes as shown in Figure 3.13(a). The maximum relative
error between the coarse and fine mesh was 2.04 %, whereas it was 0.49 % between the medium and
fine mesh. The medium mesh was then selected for subsequent analysis.
Figure 3.13(a) compares the velocity profile across the channel between Eilmer and the experi-
ment. The figure shows good agreement, which indicates that the velocity profile between the parallel
plates is adequately captured. In addition, it is instructive to replot the data of Figure 3.13(a) in the
typical semi-logarithmic manner. This is done in Figure 3.13(b). In the viscous layer [136], y + < 10,
the non-dimensional velocity u+ and distance y + have the relationship,
u+ = y + . (3.10)
and in the overlap layer [136], y + > 10, the logarithmic velocity profile is expected,
1
u+ = ln y + + 5.0 . (3.11)
0.41
Figure 3.13(b) compares the law of the wall to the simulation data. The y + value for the first cell
is less than 0.3, which is sufficient for turbulent flow simulation using Eilmer as suggested by Chan
et al. [133]. The frictional velocity calculated from Eilmer is 0.292 m/s, and El Telbany et al. [134]
experimentally determined the friction velocity to be 0.282 m/s. The relative error is less than 0.3 %,
which is also acceptable. Good agreement is achieved between the law of the wall and Eilmer with
Wilcox’s 2006 k − ω turbulence model.
3.4.2 Wall Modelled Turbulence Model for Foil Bearing Applications
For turbulent simulations, the Wilcox’s 2006 k − ω turbulence model [132] is implemented in Eilmer.
As shown by Chan et al. [133], the non-dimensionalised normal distance of the first cell from the
wall, y + , has a non-trivial effect on the accuracy of surface skin friction and heat flux prediction. For
turbulence models that integrate through the viscous sublayer, at least one cell has to be within the
viscous sublayer to ensure accurate solutions. This results in very fine cells close to wall boundaries.
1
Coarse mesh Viscous sublayer
Medium mesh 25 Log-law region
0.8 Fine mesh Eilmer

Experiment 20
0.6
15
+
u
0.4 10
0.2 5
0
0 0.1 1 10 100 1000 10000
0 0.5 1 1.5 2 +
y
Non-dimensional height
(a) (b)
Figure 3.13: (a) Velocity distribution for turbulent plane Couette flow, (b) Comparison with the law of the wall.
For explicit time-stepping solvers such as Eilmer [117], these fine cells result in very small time steps
due to CFL stability criterion that is imposed.
An alternative is to employ a wall function to model the lower section of the boundary layer. This
section describes the implementation of a wall model to avoid having to resolve the viscous sublayer
in Eilmer. Since the high-gradient region of the fluid nearest to the wall is modeled with empirical
relationships, the first cell center may be placed further away from the wall [137]. This approach
reduces the number of cells required to discretise the flowfield and increases the maximum allowable
time step.
Wall functions are typically utilised as an efficient approach to reduce computational cost. Several
wall functions exist in the literature and are implemented by various CFD codes. Typically a piecewise
wall function is utilised [138], and different velocity profiles are activated depends on the local y +
value at the first cell from the wall. In OpenFOAM [139], an automatic near-wall treatment is used in
conjunction with SST k − ω turbulence model, and the formulation for the velocity profile near the
wall is,
U
uvis
τ = (3.12a)
y+
U
ulog
τ = 1 (3.12b)
κ
ln(y + ) +C
uτ = [(uvis 4 log 4 0.25
τ ) + (uτ ) ] (3.12c)
However, this does not consider compressibility effects, which become significant especially for
high speed flows. The wall function developed by Nichols et al. [137] is able to include these factors,
and has been implemented by Gao et al. into their CFD code [140]. This wall function has been
demonstrated to provide a virtually grid independent solution when used with the Spalart-Allmaras
one-equation model and the k − ω model for both attached and separated flows [132, 137]. The
assumptions of this wall function are defined by Nichols and Nelson [137]. The wall function uses
the unified law of the wall,
y + = u+ + ywhite
+
− e−κ B [1 + κ u+ + (κ u+ )2 /2 + (κ u+ )3 /6] , (3.13)
+
where y + is the non-dimensional wall distance, u+ denotes the non-dimensional velocity, ywhilte is the
non-dimensional wall distance for incompressible adiabatic flow, κ and B are constants, taken as 0.4
and 5.5, respectively. Equation 3.13 ensures that the y + value at the first cell can be within different
ranges (viscous sublayer, buffer layer and logarithm regions). The effects of compressibility and flow
+
are addressed by the non-dimensional parameter, ywhite defined as,
+
√
ywhite = exp((κ/ Γ)(sin−1 [(2 Γ u+ − β)/Q] − φ))e−κ B . (3.14)
with variables defined as,
γ u2τ
Γ = , (3.15)
2 cp Tw
qw µw
β = , (3.16)
ρw Tw kw uτ
1
Q = (β 2 + 4 Γ) 2 . (3.17)
The non-dimensional parameter Γ models compressibility effects, and the parameter β models heat
transfer effects. The temperature distribution within the boundary layer is given by the Crocco-
Busemann equation,
T = Tw [1 + β u+ − Γ (u+ )2 ] , (3.18)
where T is the temperature at the first cell and Tw is the wall temperature. For adiabatic walls, the
heat flux qw = 0, and Equation 3.18 can be reduced to,
γ u2
T = Tw − , (3.19)
2 cp
where u is the cell velocity, cp is the specific heat at constant pressure and γ is a constant equal to
P r 1/3 . Hence, the boundary-layer profile is defined in terms of the shear stress τw and the wall heat
transfer qw . Effectively this means that for a given velocity and temperature in the first cell the heat
flux and shear stress can be defined and vice versa.
The implementation of this compressible wall function in Eilmer for adiabatic and constant tem-
perature walls follows the process suggested by Nichols et al. [137]
1. Set the wall velocity to zero for non-moving body problems or to the grid velocity for moving
wall cases.
2. (a) For adiabatic walls: use the velocity and temperature at the center of the first cell, to
solve Equation 3.19 for the wall temperature Tw and extrapolate the pressure from the cell
centre to the wall to obtain pressure at the wall. The wall density ρw is obtained using the
equation of state.
(b) For constant temperature walls: extrapolate the pressure from the first cell centre to the
wall and solve for the wall density using the equation of state with the given wall temper-
ature Tw .
3. (a) For adiabatic walls: iteratively solve Equation 3.13 to obtain the wall shear stress τw in
the local coordinates using the distance of the first cell centre from the wall.
(b) For constant temperature walls: iteratively solve Equation 3.13 and 3.18 for the wall shear
stress τw and heat transfer qw in the local coordinate using the distance of the cell centre
from the wall.
4. Rotate the stress tensor into the Cartesian coordinates and update the viscous flux at the wall.
Once the wall shear stress and heat transfer (for constant temperature wall) have been calculated,
the turbulence transport variables can be calculated at the centre of the first cell. The eddy viscosity
µt is given by,
µt ∂y + + (κ u+ )2 µw+1
= 1 + white
+
− κ e−κB
(1 + κ u + )− , (3.20)
µw ∂y 2 µw
where µw and µw+1 are the molecular viscosity at the wall and first cell centre from the wall,
+
∂yW
respectively. hite
∂y +
is given by,
+
√
∂ywhite + κ Γ (2 Γ u+ − β)2 1
= 2 y white [1 − ]2 . (3.21)
∂y + Q Q2
The values of the transport model turbulence variables in the first cell must also be defined. The
turbulent kinetic energy k and specific turbulent dissipation ω at the first cell from the wall for the
k − ω turbulence model are given by,
6 µw
ωi = , (3.22)
0.075 ρw y 2
6 uτ
ωo = p , (3.23)
Cµ κ y
q
ω = ωi2 + ωo2 , (3.24)
ω µt
k = . (3.25)
ρ
This two level model for ω was suggested by Veiser [141] and implemented as part of the com-
pressible wall function proposed by Nichols et al. [137], is implemented in Eilmer.
Flow Over Flat Plate Using Wall-Modeled Turbulence Model
The wall function described previously was incorporated into the in-house computational fluid dy-
namics solver Eilmer. The intention of this implementation is to reduce the computational cost of tur-
bulent simulations for foil bearings, hence representative test cases (flat plate, Couette and Poiseuille
flows) are studied to validate the wall function.
The first validation case is air flow over a flat plate. The turbulent friction correlations by van
Driest are used for comparison with numerical simulations [136]. Because of its simplicity, this test
case is used as a fundamental case to validate the initial implementation of wall functions.
The schematic diagram of the simulation is shown in Figure 3.14. The left and north boundary
surfaces are taken as inflow conditions, where static pressure, temperature and velocity of the air flow
are prescribed. In this case, the inflow condition is set to 0.1 MPa, 300 K and 100.0 m/s. The right
boundary surface is modeled as outflow with a zero gradient condition. The south is defined as a wall
boundary. For this test case, two wall boundary conditions, adiabatic and constant temperature, are
y Inflow Outflow
400
300
Wall Boundary
x
400
Figure 3.14: Schematic diagram of the simulation domain for flat plate, in mm.
used. The velocity boundary profiles (including the calculation of y + value) extracted at a position
380 mm from the leading edge are shown in Figures. 3.15 and 3.16.
For the adiabatic wall case, the predicted values of the friction coefficient along the lower wall with
varying initial grid wall spacing using the k − ω turbulence model are shown in Figure 3.15(a). Also
shown are theoretical values from van Driest [136, 142], with a 10 % uncertainty range. The y + = 1
results were created without using a wall function and are included for reference. The wall function
predictions using y + =10, 20, 30 and 50 are in close agreement with the fully resolved simulation (y + =
1). Reasonable agreement also exists between the predictions and the van Driest correlation with the
simulation consistently under predicting. For y + =100 and 200, some deviation exists at Reynolds
number below 0.5 × 106 , but good agreement exists for higher Reynolds numbers. Velocity profiles
and the law of the wall are shown in Figure 3.15(b). Again the y + = 1 results obtained without using
the wall functions are included for reference. Close agreement exists between the simulated profiles,
far from the wall.
The corresponding results for a constant temperature wall (320 K) are shown in Figure 3.16.
Again, the values from van Driest [136, 142] and the y + = 1 results, created without using a wall-
function are included. The wall function predictions using y + =10, 20, 30 and 50 are in close agree-
ment with the fully resolved simulation (y + =1). Also reasonable agreement exists between the pre-
dictions and the van Driest correlation. For y + = 100, again a deviation of friction coefficient exists
around Reynolds number of 0.5 × 106 and good agreement is achieved at higher Reynolds numbers.
The predicted heat transfer is shown in Figure 3.16(b). The differences between the different grid
spacings are larger than for the friction coefficient predictions.
It is found that the absolute error between wall function and experimental results (var Driest) are
5.17 % for the adiabatic cases and 12.29 % for the constant wall temperature cases, similar to the
values cited by Nichols et al. [137]. Velocity and temperature profiles at 380 mm from the leading
0.01 +
y =1
+
0.009 y+=10
y+=20
Friction coefficient, cf
0.008 y+=30
y =50
+
0.007 y+=100
y =200
0.006 van Driest II with 10% uncertainty
0.005
0.004
0.003
0.002
0 0.5 1 1.5 2 2.5
6
Rex, 10
(a)
25
20
15 +
y =1
+
y+=10
+
u
y+=20
10 y+=30
y =50
+
y+=100
5 y =200
Viscous sublayer
Log-law region
0
1 10 100 1000 10000
+
y
(b)
Figure 3.15: Performance comparison with different grid spacings for adiabatic walls, (a) friction coefficient;
(b) law of the wall. 380 mm from leading edge, Rex =2.39 × 106 .
edge are shown in Figure 3.17. These are in reasonable agreement with each other and with theory
for all of the wall spacings tested.
Based on the verification studies, it has been shown that a y + value less than 100 is required to
ensure the accurate prediction of the shear stress and heat transfer. With these test cases for adiabatic
and constant temperature walls, the implementation of wall functions is validated for these simple
cases.
Turbulent Couette and Poisullie Flows Using Wall-Modeled Turbulence Model
The next test cases (Couette and Poisullie flows) are representative of flows in foil bearings. They can
be used to validate the suitability of this wall function for the bearing flowfield. Five test cases from
the experimental results given by El Telbany and Reynolds [134, 135] are considered. These include
pure Couette flow, pure Poisullie flow and hybrid Couette and Poiseuille flow, which are the typical
flows within foil bearings. The test rig [134, 135] described in literature uses 2440 mm long parallel
walls, with the main measurement station at the centre of the channel and 1980 mm from the blowing
end. For test cases used in this section, the height between two parallel walls was 66 mm, and the belt
forming the moving wall has a translational velocity of 12.84 m/s. The pressure difference between
the inlet and outlet of the moving belt is attained by a blower. The exact value of pressure difference
across the moving belt is not indicated in literature, but it can be obtained from the shear stress listed
for the experiments [134, 135].
A two-dimensional computational domain is used as the measured region is located in the centre
of the belt. The schematic diagram is shown in Figure 3.12. The left and right boundary patches
are modeled as fixed pressure. By varying the pressure difference between inlet and outlet, different
forms of hybrid Couette and Poisullie flows are created. The north boundary is taken as a fixed
temperature wall with a prescribed translational velocity, while the south is regarded as a stationary
fixed temperature wall.
The experimental results used for comparison are listed in Table 3.6. The case index is in accord-
ance with the literature [134, 135]. uτ 1 indicates the high stress wall, while uτ 2 is for the low stress
wall. Case 1 is the pure Couette flow, and there is no pressure difference across the moving belt,
while case 15 is the pure Poiseuille flow, as indicated by a non-moving belt. With a certain amount
of pressure difference across the moving belt, case 2, 3 and 5 are regarded as hybrid Couette and
Poiseuille flows, but Couette flows are still dominant. For these hybrid flows, the belt velocity is set
0.01 +
y+=10
0.009 y+=20
y+=30
Friction coefficient, cf
0.008 y =50
+
+
y =100
0.007 y =1, no wall model
van Driest II with 10% uncertainty
0.006
0.005
0.004
0.003
0.002
0 0.5 1 1.5 2 2.5
6
Rex, 10
(a)
0.01 +
y+=10
0.009 y+=20
2
y+=30
Wall heat flux, MW/m
0.008 y =50
+
+
y =100
0.007 y =1, no wall model
van Driest II with 10% uncertainty
0.006
0.005
0.004
0.003
0.002
0 0.5 1 1.5 2 2.5
6
Rex, 10
(b)
Figure 3.16: Performance comparison with different grid spacings for constant temperature walls, (a) friction
coefficient; (b) heat flux.
25
20
15
+
y+=10
+
u
y+=20
10 y+=30
y =50
+
+
y =100
5 y =1, no wall model
Viscous sublayer
Log-law region
0
1 10 100 1000 10000
+
y
(a)
6 +
y+=10
y+=20
5 y+=30
y =50
+
4 +
y =100
y =1, no wall model
y, mm
0
295 300 305 310 315 320 325
Temperature, K
(b)
Figure 3.17: Law of the wall and temperature, 380 mm from leading edge, Rex =2.39 × 106 .
Table 3.6: Details of El Telbany and Reynolds tests.
Case Ub , m/s uτ 1 , m/s uτ 2 , m/s Pressure difference, Pa

1 12.84 0.282 0.282 0
2 12.84 0.328 0.233 2.3771
3 12.84 0.362 0.1809 4.3782
5 12.84 0.383 0.1305 5.7765
15 0 0.659 0.659 38.6913
as 12.84 m/s. All the cases were generated by running the belt in the same direction as the blown flow
of air. The pressure differences in Table 3.6 are not provided in literature, but are calculated from the
friction velocity of the high and low stress walls,
(ρu2τ 1 − ρu2τ 2 )Lb

Couette flow: case 2, 3 and 5 ∆p = , (3.26)
Hb
(ρu2τ 1 + ρu2τ 2 )Lb
Poisullie flow: case 15 ∆p = . (3.27)
Hb
The resulting velocity profiles and the law of the wall for different test cases with various grid
spacings are plotted in Figure 3.18. The boundary layer velocity profile is normalised with the belt
moving velocity of 12.84 m/s and the y distance is normalised by the half height of the parallel chan-
nel, 33 mm. The results at y + = 1 are created without a wall function and plotted as the reference.
In all cases, good agreement is attained between the different numerical models. Good agreement is
also achieved in comparison to the experimental data for case 2, 3 and 15, both at the high-stress and
low-stress walls. The solution accuracy still maintains at different grid spacings is good for hybrid
Couette and Poiseuille flow cases. The highest y + values for the different test cases shown in Fig-
ure 3.18 are obtained by using 5 uniform cells across the channel. This is the coarsest grid, which still
presents good agreement to the experimental data. Case 5 shows some differences between experi-
mental data and numerical simulations, but the results from numerical simulations are self-consistent
at different grid spacings. For the plot of the law of the wall in Figure 3.18(d), (f) and (h), even if the
results from the wall function is self-consistent, a difference is spotted between wall-modelled and
wall-resolved approaches, and the magnitude of this difference is gradually increasing from cases 2 to
5. This deviation is caused by the over-predicted wall shear stress from the wall function as reported
in Table 3.7.
1 30
0.8 25
20
0.6 + y+=10
y =10 +
u+
+ 15 y+=20
0.4 y+=20 y =50
y =50 10 +
+ +
y =100
0.2 y =100 y =1, no wall model
+ 5
y =1, no wall model Viscous sublayer
Experiment: case 1 Log-law region
0 0
0 0.5 1 1.5 2 1 10 100 1000 10000
+
Non-dimensional height y
(a): Case 1, boundary layer. (b): Case 1, law of the wall.

1 30
0.8 25
20
0.6 + y+=10
y =10 +
u+
+ 15 y+=20
0.4 y+=20 y =50
y =50 10 +
+ +
y =100
0.2 y =100 y =1, no wall model
+ 5
y =1, no wall model Viscous sublayer
0 0
0 0.5 1 1.5 2 1 10 100 1000 10000
+

1 30
0.8 25
20 +
0.6 + y =10
y+=10 y++=20
u+
15
0.4 y =20 y =50
+ +
y =50
+
10 y =100
y =100 +
0.2 + y =1, no wall model
y =1, no wall mdoel 5 Viscous sublayer
0 0
0 0.5 1 1.5 2 1 10 100 1000 10000
+

1 30
0.8 25
20 +
0.6 + y =10
y+=10 y++=20
u+
15
0.4 y =20 y =50
+ +
y =50
+
10 y =100
y =100 +
y =1, no wall model 5 Viscous sublayer
0 0
0 0.5 1 1.5 2 1 10 100 1000 10000
+

1 30
0.8 25
20 +
0.6 + y =10
y =10 y+=20
u+
+ 15
0.4 y =20
+ y+=50
y
+
=50 10 y+=100
y =100 +
y =1, no wall model 5 Viscous sublayer
0 0
0 0.5 1 1.5 2 1 10 100 1000 10000
+
Figure 3.18: Performance comparison with different grid spacings for various cases.
Table 3.7: Comparison of frictional velocity (in m/s) at different grid spacings. The low stress wall is the
moving wall in cases 1,2,3 and 5.
High stress wall

Case index Exp. y+ = 1 y + = 10 y + = 20
1 0.282 0.292 0.293 0.292
2 0.328 0.328 0.339 0.335
3 0.362 0.369 0.373 0.372
5 0.383 0.389 0.398 0.397
15 0.659 0.637 0.656 0.655
Low stress wall
Case index Exp. y+ = 1 y + = 10 y + = 20
1 0.282 0.292 0.293 0.292
2 0.233 0.245 0.251 0.247
3 0.1809 0.208 0.212 0.206
5 0.1305 0.179 0.182 0.178
15 0.659 0.637 0.656 0.655
To further validate the implemented wall functions, the friction velocities from experimental res-
ults and numerical simulations are listed in Table 3.7. Good agreement exists between experiment
and simulations. Despite differences in velocity profile far from the wall (center of channel) the wall
shear stress is predicted correctly. It is found that a y + value less than 100 can achieve accurate
result for turbulent Couette and Poiseuille flows and that at least 5 cells are required to correctly sim-
ulate the velocity profile. The wall function method for Couette and Poiseuille type flows has been
throughly validated here. This indicates that the wall function can be applied for the bearing flowfield
simulations.
Suitability of Wilcox’s 2006 k − ω Turbulence Model for Foil Bearings
To test the suitability of the implemented Wilcox’s 2006 k−ω for the fluid flow within foil bearings the
experimental results of case 5 from the previous section are used for a turbulence model comparison.
Figure 3.19 depicts the results for different turbulence models. The result of Wilcox’s 2006 k − ω
model is from Eilmer, while the results for other turbulence models (SST k − ω, Spalart Allmaras,
standard k − ǫ and Nonlinear k − ǫ Shih) are all obtained using the simpleFoam solver [139]. It can be
seen from Figure 3.19 that both standard k − ǫ and Nonlinear k − ǫ Shih show a large difference with
experimental data, while good agreement between other turbulence models (Wilcox’s 2006 k − ω,
SST k − ω, Spalart Allmaras) is attained. However a certain deviation to experimental data exists
0.8
0.6
0.4
Wilcox k-ω
SST k-ω
Spalart Allmaras
0.2 Standard k-ε
Nonlinear k-ε Shih
Experiment: case 5
0
0 0.5 1 1.5 2
Figure 3.19: Comparison of different turbulence models for case 5.
for all turbulence models. The implemented Wilcox’s 2006 k − ω is perhaps best for this test case.
The reason for the remaining deviation could be that the current turbulence models cannot handle this
flow or errors in the experimental data. Unfortunately uncertainty bands for experimental data are not
given by Telbany et al. [134]. It is recommended that a more accurate simulations such as large eddy
simulation or direct numerical simulation are conducted to further investigate this deviation.
3.4.3 Stabilisation Method for Turbulent Simulations
The film thickness within foil bearings is typically of the order of microns, where as the foil bearing
dimensions are of the order 10s of mm. For example, the NASA foil thrust bearing has inner and outer
radii of 25.4 mm and 50.8 mm [73], respectively. This particular geometry results in cells with a high
aspect ratio (over 1000). The resulting ill-conditioned meshes can cause checker-boarding (numerical
oscillation) during turbulent simulations with Eilmer. The checker-boarding phenomenon is not new
to the CFD community. For the CFD code developed by Dawes [143], an adaptive artificial viscosity
term is added to the governing equations to control odd-even point solution decoupling and to suppress
oscillations in regions with strong pressure gradients. Using the diffusion and anti-diffusion equations
as a filter numerical diffusion is also reduced [144]. To stabilise turbulent simulations in Eilmer, the
artificial dissipation terms proposed by Jameson, Schmidt and Turkel [145] were selected to dampen
the high-frequency oscillations. For this method, an artificial dissipative flux is added into governing
equations, written as,
(2) (4) (2) (4)

dj+ 1 = dj+ 1 − dj+ 1 = ǫj+ 1 ∆wj+ 1 − ǫj+ 1 (∆wj+ 3 − 2∆wj+ 1 + ∆wj− 1 ) , (3.28)
2 2 2 2 2 2 2 2 2
where,  
ρj+1 − ρj
 
(2)
∆wj+ 1 = (ρ u)j+1 − (ρ u)j  . (3.29)
 
2  
(ρ u H)j+1 − (ρ u H)j
The spectral radius of the Jacobian matrix in cell j is rj = |u| + c, where c is the local speed of
(2) (4)
sound. The dissipative coefficient ǫj+ 1 and ǫj+ 1 are switched on and off by a pressure sensor,
2 2
pj+1 − 2pj + pj−1

sj = | | . (3.30)
pj+1 + 2pj + pj−1
Interface values of the spectral radius and sensor are defined as,
rj+ 1 = max(rj+1 , rj ) , sj+ 1 = max(sj+1 , sj ) . (3.31)

2 2
Then,
(2) (4) (2)
ǫj+ 1 = k2 sj+ 1 rj+ 1 , ǫj+ 1 = max(0, k4 rj+ 1 − c4 ǫj+ 1 ) . (3.32)
2 2 2 2 2 2
The typical values of the constants k2 , k4 and c4 are,
1 1
k2 = 0.2 ∼ 1 , k4 = ∼ , c4 = 2 . (3.33)
256 16
(2)
In this method, the second-order dissipation term dj+ 1 is used to suppress overshooting caused
2
(4)
by shocks and the fourth-order dissipation term dj+ 1 is used to damp the spurious oscillations in
2
smooth regions [145]. Therefore, for the current simulations the second term can be neglected and
only the fourth-order artificial dissipation term is added to the system. This term provides additional
dissipation to suppress spurious numerical oscillations when the grid size is not small enough to
render the physical viscosity and will not contaminate the physical solutions [146]. However, for
cells close to wall boundary conditions, these artificial dissipation terms are turned off to eliminate
over-dissipation as suggested by Goncalves et al. [147]. For the implementation within Eilmer, this
10000 10000
Eilmer, 8x8 cells Eilmer, 8x8 cells
1000 Eilmer, 32x32 cells 1000 Eilmer, 32x32 cells
Artificial dissipation, 8x8 cells 100 Artificial dissipation, 8x8 cells
100 Artificial dissipation, 16x16 cells
Artificial dissipation, 16x16 cells
L∞ Norm
L2 Norm
Artificial dissipation, 32x32 cells Artificial dissipation, 32x32 cells

10 Artificial dissipation, 64x64 cells 10 Artificial dissipation, 64x64 cells
1 1
0.1 0.1
0.01 0.01
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
Time, ms Time, ms
(a) (b)
Figure 3.20: Convergence history of norms based on turbulent kinetic energy for different mesh sizes: a) L2
norms, b) L∞ norms.
artificial dissipation is turned off in the first four cells from the wall.
Verification of Eilmer with Artificial Dissipation Using the Method of Manufactured Solution
After the implementation of the fourth-order artificial dissipation in Eilmer, the verification is per-
formed to test the influence of this addition on the fluid solver. Here the method of manufactured
solutions is utilised. The detailed introduction of this verification method applied to Eilmer including
inviscid, viscous and RANS solvers can be found in Ref. [117, 148]. In this section, the verification
is applied to the RANS solver within Eilmer, and Wilcox’s 2006 k − ω turbulence model is selected.
The computational setup is the same as outlined in Ref. [148], where exact Dirichlet values obtained
from the manufactured solution are prescribed at boundaries. A squared computational domain with
a structured grids is investigated. Three different computational meshes (8×8, 16×16, 32×32 and
64×64) are employed to investigate the order of accuracy. The discretisation error is evaluated by L2
and L∞ norms, as defined in Ref. [117].
The time-dependent data of the L2 and L∞ norms of the turbulent kinetic energy from different
meshes are plotted in Figure 3.20. Other parameters, like density, dissipation rate can also be utilised
in Figure 3.20. The results are obtained from Eilmer with and without artificial dissipation (k4 =0.02),
respectively. Good agreement is achieved between these two approaches in terms of the steady state
and dynamic performances. Also, the difference between two approaches exists in the initial state,
but reduces at the final steady state.
10 3
2.5
Order of accuracy, P
1 2
Norm
1.5
0.1 1
Eilmer, L2
Eilmer, L∞ Eilmer, L2
2nd Order Slope 0.5 Eilmer, L∞
Artificial dissipation, L2 Artificial dissipation, L2
Artificial dissipation, L∞ Artificial dissipation, L∞
0.01 0
0.01 0.1 0.01 0.1
∆x, m ∆x, m
(a) (b)
Figure 3.21: (a) Norms and (b) observed order of accuracy.
Figure 3.21(a) shows the steady state error norms in terms of cell sizes. The order of accuracy can
be determined from the norm values [117] and the results are depicted in Figure 3.21(b). The results
from Eilmer and Eilmer with artificial dissipation both indicates an order of two is achieved. The
variation of order of accuracy between these two approaches is gradually reduced as the mesh density
increases. It is noted that a second order of accuracy is also achieved by investigating L2 and L∞
norms of the specific dissipation and density, which are not shown here. These result proves that the
implemented fourth-order artificial dissipation has no influence on the accuracy of the RANS solver
in Eilmer, and the second order accuracy is still ensured.
Influence of Artificial Dissipation on Checker-Boarding
For turbulent simulations of foil thrust bearings, the time step determined from a CFL number of 0.5
in conjunction with the predictor-corrector time stepping method results in a large checker-boarding
effect and divergence of the simulation. Alternatively, it is found that this checker-boarding effect
can be minimized by selecting a smaller time step. However, the time step has to be extremely
small to suppress the checker-boarding effect. Numerical experiments were carried out to the test the
suitability of the fourth-order artificial dissipation in providing solution stability without affecting the
solution due to excessive dissipation. The test case is the fluid flow within a rigid foil thrust bearings
with rotational speeds of 20 000 rpm, 30 000 rpm and 40 000 rpm (geometry is defined in Table 2.1).
Ambient pressure and temperature are selected as 1.4 MPa and 300 K, respectively. The rotor to top
foil separation is chosen as 22 µm. The artificial dissipation constant k4 is set as 0.02 in this case. A
Checker-boarding
44.46°
Figure 3.22: Pressure contour of the rigid foil thrust bearings, rotational speed: 30 000 rpm, rotor to top foil
separation: 22 µm.
mesh of 48×96×15 was selected. Grid independence was previously established.
Table 3.8: Comparison of time step used by different modeling approaches.
Rotational Eilmer k4 =0 Eilmer k4 =0.02 Speed-up

speed ∆t Wall clock ∆t Wall clock
20 000 rpm 1.231 × 10−9 s 20 912 s 2.481 × 10−9 s 10 741 s 1.95
30 000 rpm 7.185 × 10−10 s 35 479 s 2.338 × 10−9 s 11 343 s 3.13
40 000 rpm 3.123 × 10−10 s 81 313 s 2.171 × 10−9 s 12 383 s 6.57
The time steps used for simulations with and without artificial dissipation (k4 =0.02) are compared
in Table 3.8. It is noted that the time step used for Eilmer without dissipation in Table 3.8 cannot
entirely remove the checker-boarding effect, but is able to provide a stable computation. It is found
that high rotational speed presents a critical requirement for the time step in Eilmer. The additions of
artificial dissipation ensure a comparatively large time step can be maintained. The simulations shown
in Table 3.8 were performed on two 12 core Intel Xeon E5-2680V3 2.5 GHz CPU workstations,
and the wall clock time was recorded for comparison. The speed-up capability with the artificial
dissipation increases with the rotational speed, and the highest value is 6.57 times at the rotational
speed of 40 000 rpm.
The pressure and velocity contours at the rotational speed of 30 000 rpm are shown in Figure 3.22
Checker-boarding
44.46°
Figure 3.23: Velocity contour of the rigid foil thrust bearings, rotational speed: 30 000 rpm, rotor to top foil
separation: 22 µm.
160 160
Rotor, k4=0 Stator, k4=0
Rotor, k4=0.005 Stator, k4=0.005
Tangential velocity, m/s
120 120 40
35
30
140 25
80 135 80 20
130 15
125 10
120 45 46 47 48
40 115 40
110
45 46 47 48
0 0
30 35 40 45 50 30 35 40 45 50
Radius, mm Radius, mm
(a) Tangential velocity, close to rotor. (b) Tangential velocity, close to stator.
10 10
Rotor, k4=0 Stator, k4=0
8 8
2.6 2
6 2.4 6 1.8
2.2 1.6
4 2 4 1.4
45 46 47 48 45 46 47 48
2 2
0 0
30 35 40 45 50 30 35 40 45 50
Radius, mm Radius, mm
(c) Radial velocity, close to rotor. (d) Radial velocity, close to stator.
Figure 3.24: Comparison of tangential and radial velocities at a circumferential angle of 44.46°.
and 3.23. These results are obtained from simulations with and without artificial dissipation using the
time steps listed in Table 3.8. The study of foil thrust bearings at this rotational speed was also per-
formed with the k4 of 0.005. This additional case is used to investigate the impact of different k4 on
the simulations. It is shown in Figure 3.22 that the pressure contour is similar for these two modeling
approaches and selected k4 , even if there is a small check-boarding at the top right corner for the case
without artificial dissipation. The checker-boarding effect is more obvious for velocity contours as
highlighted with by the circle in Figure 3.23. This effect can be fully eliminated by the addition of
the fourth-order artificial dissipation. To further investigate the influence of the artificial dissipation,
tangential and radial velocities at the first cell from the rotor and stator at an angle of 44.46° are com-
pared in Figure 3.24. Due to the checker-boarding effect, the tangential and radial velocities exhibit
a zigzag distribution in the radial direction. The fluctuation amplitude of the tangential and radial
velocity are as high as 15 m/s and 0.2 m/s, respectively. This non-physical behaviour is successfully
smoothed with the addition of the artificial dissipation, of either k4 =0.005 or k4 =0.02. The relative
difference between different k4 (0.005 and 0.02) is less than 0.3 %. This shows that this artificial dis-
sipation can be used to stabilise the simulations without creating new non-physical effects or affecting
the solution accuracy.
Selection of Appropriate Damping Coefficients
This section investigates the minimum k4 to stabilise the turbulent simulations for Couette type flow.
It is noted that the checker-boarding effect is not observed in two dimensional simulations for the
hybrid pressure and shear driven flows. The three dimensional turbulent simulations are performed for
CO2 foil thrust bearings, with the geometry in accordance to Table. 2.1. The pressure and temperature
boundary conditions are set as 1.4 MPa and 300 K. The results are obtained by evaluating the turbulent
simulations of foil thrust bearings at three rotational speeds (20 000 rpm, 30 000 rpm and 40 000 rpm)
and three minimum rotor stator separation (14 µm, 18 µm and 22 µm). A computational mesh of
48×96×15 is set for all cases, resulting in different cell aspect ratios.
The minimum requirement for k4 to stabilise the turbulent simulations for foil thrust bearings are
listed in Table 3.9. For the selected operating conditions, k4 less than 0.01 is typically sufficient. It
is found that the minimum k4 varies with rotational speed and cell aspect ratio. For fixed rotational
speeds, the higher cell aspect ratios require large k4 to stabilise simulations. A similar trend is found
for increasing rotational speed at fixed cell aspect ratio, which requires a larger k4 to stabilise the
Conclusions Section 3.5 77
Table 3.9: Matrix of rotational speeds and minimum rotor to top foil separation to show trends in k4 required
to stabilise simulations.
14 µm 18 µm 22 µm
20 000 rpm 0.006 0.004 0.003
30 000 rpm 0.008 0.006 0.005
40 000 rpm 0.01 0.009 0.007
turbulent simulations of foil thrust bearings. As discussed previously, k4 between 0.005 and 0.02
have a negligible influence on the flow field. Therefore, it is recommended that a k4 larger than 0.01
is selected to ensure stable foil bearing simulations.
3.5 Conclusions
This chapter describes the development and validation of the CFD code Eilmer for foil bearing sim-
ulations. Three different test cases are simulated to verify the predictions from Eilmer. The moving
wall and periodic boundary conditions are verified with Taylor Couette flow, while the suitability of
Eilmer in simulating centrifugal and inertial effects is verified by simulations of high speed air thrust
bearings. For wavy thrust bearings with more complex geometry, CFD results show good agreement
compared to Reynolds equation for symmetric wavy as well as a high compression geometry. For
turbulent simulations, two new features: wall functions and a fourth-order artificial dissipation, are
added to Eilmer. The implemented compressible wall function from Nichols et al. [137] is able to
correctly simulate hybrid pressure and shear driven flows, typical for foil bearings. A y + value at
the first cell from the wall of less than 20 is required to accurately calculate the wall shear stress
for the studied geometries. Wilcox’s 2006 k − ω model is found to be one of most accurate turbu-
lence models for hybrid pressure and shear driven flows compared to other models including: SST
k − ω, Spalart Allmaras, standard k − ǫ and Nonlinear k − ǫ Shih. It is also demonstrated that the
inclusion of a fourth-oder artificial dissipation has no influence on the spatial accuracy of Eilmer, and
that second order accuracy is still maintained. The fourth-order artificial dissipation can speed up
turbulent simulations within foil thrust bearings, while ensuring a stable and accurate computation
without contaminating the fluid flow. For the test case used in this Chapter, a speed-up of 6.57 times
is achieved at a rotational speed of 40 000 rpm. The minimum k4 value for stabilising turbulent sim-
ulations increases in terms of the rotational speed and cell aspect ratio. It is recommended that a k4
dissipation constant between 0.01 and 0.02 is selected.

Chapter 4
Fluid-Structure Coupled Simulations
4.1 Introduction
To model the influence of structural deformation of the foil elements on the fluid flow, a moving grid
capability is added to Eilmer. In this chapter, the fundamental theory of moving grids and structural
deformation is addressed, and a small number of test cases are presented to test the suitability of the
developed solvers, both Eilmer with moving grid and the structural deformation solver for foil bear-
ings. Section 4.2 details the theory behind the moving grid. Section 4.3 investigates the suitability of
Eilmer to simulate a range of test cases with moving grid. Section 4.4 documents the structural de-
formation model for the top and bump foils and the associated verification case. Finally, the coupling
strategy between fluid and structural deformation solver is introduced and verified in Section 4.5.
4.2 Moving Grid
Due to the formulation of Eilmer, which uses boundary conforming grids, to effectively couple fluid
and structure simulation, a moving grid has to be implemented. The governing equations for the
compressible flow solver are still the same as Equation B.1, but the terms within the convective and
viscous flux expressions need to be altered to account for grid movement.
80 Chapter 4 Fluid-Structure Coupled Simulations
4.2.1 Convective Flux Update with Moving Grid
For the convective flux with moving grid, the interface moving velocity wif is incorporated into the
convective flux Fi [149, 150],
   
G ρ (u − wif ) · n̂
   
Fi =  L  =  ρ (u − wif )u · n̂ + p n̂  . (4.1)
   
   
H ρ E (u − wif ) · n̂ + p u · n̂
′ ′ ′
so that the convective flux Fi relative to the interface (mass flux G , momentum flux L and energy
′
flux H ) can be determined from the left and right interface flow states as,
   
′
G ρ (u − wif ) · n̂
   
′  ′  
Fi =  L  =  ρ (u − wif )(u − wif ) · n̂ + p n̂  . (4.2)

   
′
H ρ E (u − wif ) · n̂ + p (u − wif ) · n̂
These can be calculated with flux scheme like Liou and Steffen’s AUSM scheme [151], Wada and
Liou’s AUSMDV scheme [152] and Macrossan’s EFM scheme [153]. The convective flux relative to
the interface can then be transformed to the global reference frame using the following transforma-
tion [153],
′
G = G , (4.3)
′ ′
L = L + G wif , (4.4)
′ 1 ′ ′
H = H + G |wif | + L · wif . (4.5)
2
The key parameter is the effective interface moving velocity wif as reported by Petrie-Repar [149]
and Johnston [150]. Correct selection ensures that the Geometric Conservation Law is satisfied, which
implies that any extra volume accumulated by one cell is accounted for by loss of the same volume
from other cells or movement of the domain boundaries. Thus, for a single cell, the effective swept
volumes Vif , defined by two faces [ABCD]n and [ABCD]if , must be equal to volume change due to
the discretised grid motion defined by the volume between ([ABCD]n and [ABCD]n+1 ), as depicted
Moving Grid Section 4.2 81
Bn+1
WB
Bn Bif
WA An+1
An Aif
Cn Cif
WC Cn+1
Dn Dif
WD
Dn+1
Figure 4.1: Schematic diagram for the moving interface. An , Bn , Cn , Dn are the vertices at time n, An+1 ,
Bn+1 , Cn+1 , Dn+1 is the vertices at time n + 1. Aif , Bif , Cif , Dif are the effective vertices at time n + 1,
used to calculate the swept volume Vif . WA , WB , WC , WD grid-velocities associated with the vertices.
in Figure 4.1. The effective interface moving velocity wif can then be defined as,
Vif
wif · n̂ = . (4.6)
Aif ∆t
Ambrosi et al. [154] stated a method to calculate the effective interface velocity for two-dimensional
meshes. This can be found in Petrie-Repar’s PhD thesis [149]. The motion of boundary edges in two
dimensions is composed of two triangles and the effective moving area in two dimensions is then
calculated with the sum of these two triangles. For three-dimensional meshes, Grandy [155] presen-
ted a way to calculate the effective volume. The motion of the boundary interface is divided into six
pyramids for three dimensional meshes and then the effective volume is the sum of these individual
pyramid.
The time-evolution equation for the conserved quantities (Equation B.5) can now be integrated
in time. The predictor-corrector scheme is selected to integrate the governing equation and the cell-
averaged conserved quantities in Equation B.5 are advanced from time n to time n + 1, which is
Wall
Boundary
Uf2 Uf1 Ug1 Ug2

QO
wif
Figure 4.2: Ghost cell configuration for a slip wall condition.
shown as,
X
∆U1 = (Fi · S)n ∆t , (4.7)
V n Un + ∆U1
U1 = n+1
, (4.8)
X V
∆U2 = (Fi · S)1 ∆t , (4.9)
∆U2 − ∆U1
Un+1 = U1 + . (4.10)
2V n+1
The superscripts 1 and 2 indicate intermediate results used in the predictor-corrector time-step.
The convective component of boundary conditions is implemented by filling in the ghost-cell data
and then applying the normal reconstruction and flux calculation without further discrimination of
the boundary cells. This approach works for solid walls, inflow, and outflow boundaries, as shown
in Figure 4.2. For computing gas transport due to convective fluxes, the interface moving velocity at
the wall boundary should be considered when reflecting the velocities in ghost cells to ensure a zero
mass flux through wall boundaries [150]. This can be stated as,
(uf 1 − wif ) · n̂ = −(ug1 − wif ) · n̂ , (4.11)
(uf 2 − wif ) · n̂ = −(ug2 − wif ) · n̂ . (4.12)

Fluid-Structure Simulation: Test Cases Section 4.3 83
4.2.2 Viscous Flux Update with Moving Grid

Viscous fluxes, Fv in Eilmer are evaluated separately to the convective fluxes, Fi and after the grid
has moved. Internal to the mesh, the moving grid does not have to be included in the evaluation of
viscous fluxes, as viscous fluxes are updated after the grid has moved. However, the moving grid
must be considered for the wall boundary conditions. Viscous boundary conditions in Eilmer use data
specified at cell faces that lie along the boundary surface. At those locations, the tangential component
of the interface moving velocity must be included in the boundary velocity to correctly compute the
spatial derivatives used in the evaluation of the viscous fluxes. In the current implementation, the
tangential velocity at cell interfaces is approximated by averaging the vertex velocities,
wA + wB + wC + wD
wa = , (4.13)
4
wif = (wa · t̂) t̂ . (4.14)
wA wB
wa
wif
wD wC
Figure 4.3: Schematic diagram for tangential moving velocity.
The schematic diagram of the tangential moving velocity wif obtained from the vertices wA , wB ,
wC and wD are shown in Figure 4.3.
4.3 Fluid-Structure Simulation: Test Cases

A number of test cases were simulated to verify and validate the moving grid implementation in
Eilmer. In this section, a randomly moving grid, a constant velocity piston in a tube, and an accelerat-
ing projectile in a tube, the inviscid flow about a pitching airfoil, the viscous flow over an oscillating
plate, and turbulent flow about a pitching airfoil are discussed. Together, these examples provide a
thorough demonstration of the moving-grid capability that has been implemented in Eilmer.
4.3.1 Random Grid Motion

The first test case is random grid motion, an approach that has been used by several authors [149, 150,
156, 157] developing CFD code with moving grids. It verifies that mass, momentum, and energy are
conserved in Eilmer.
The initial 10×10 two-dimensional mesh for this test case is shown in Figure 4.4(a). The ini-
tial density, pressure and temperature in the simulation domain are 1.168 kg/m3 , 100 kPa and 300 K,
respectively, and the velocity is set to zero. The internal vertices of the mesh, are assigned random
velocities with the x and y components varying between −0.4 m/s and 0.4 m/s, and the velocity of the
external vertices are set to zero. If the scheme is correct, density, pressure and temperature should
remain constant and the flow velocity zero.
The simulation was run for 607 steps to a simulation time of 5.0 ms. The final mesh is shown
in Figure 4.4(b). Density, pressure and temperature remained constant and the maximum change in
velocity for all cells within the simulation domain is 9.521 × 10−14 m/s. This is close to machine
precision and verifies that the code satisfies the Geometric Conservation Law. A three-dimensional
mesh with random grid motion was also simulated, the initial states were the same as used for the
two-dimensional mesh, and the maximum change in velocity was less than 1.0 × 10−13 m/s, while
the other parameters remained the same. This confirms Geometric Conservation Law is also satisfied
for 3D meshes.
These results verify the calculation of the effective interface moving velocity for 2D and 3D
meshes. At the same time, the correct implementation of predictor-corrector time-stepping scheme in
Eilmer is also tested.
4.3.2 Constant Velocity Piston

The problem of a piston moving with a constant velocity in an inviscid shock tube has an analytical
solution [158] and is used to verify that the code can solve the Euler equations accurately in a moving
reference frame.
This test case is performed on a grid of 320×20 cells. The 2D simulation domain is 0.8 m long
and 0.04 m wide. It is filled initially with quiescent perfect air at ρ=1 kg/m3 and p=100 kPa. The
0.1
0.08
0.06
Y, m
0.04
0.02
0
0 0.02 0.04 0.06 0.08 0.1
X, m
(a)
0.1
0.08
0.06
Y, m
0.04
0.02
0
0 0.02 0.04 0.06 0.08 0.1
X, m
(b)
Figure 4.4: (a) Initial two-dimensional mesh for random grid motion, (b) final two-dimensional mesh.
4
Eilmer
Analytical Solution 100000
3.5
3 80000
Density, kg/m3
Pressure, Pa
Piston
2.5
60000
2
Piston
40000
1.5
20000
1
Eilmer
Analytical Solution
0.5 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
X, m X, m
(a) (b)
200
Eilmer
Analytical Solution
300
150
250
Temperature, K
Velocity, m/s
200 100
150 Piston
Piston
50
100
50 Eilmer 0
Analytical Solution
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
X, m X, m
(c) (d)
Figure 4.5: Comparison results of Eilmer and analytical solution, (a) density, (b) pressure, (c) temperature, (d)
velocity.
piston moves at a constant velocity of u=154 m/s, v=0 m/s. Profiles of pressure, temperature, density
and x-velocity are extracted from the centerline after 800 µs and compared with the analytical solution
derived from wave theory. The numerical results are in excellent agreement with analytical values as
shown in Figure 4.5. Note that there is a deviation in the density and temperature profile near the
piston location. This is also presented by authors of [156, 157, 159]. This case demonstrates the
correct implementation of wall boundaries that have a velocity component normal to the wall.
4.3.3 Projectile in Tube

The motion of a projectile in a tube is examined to verify the ability of the code to solve coupled fluid-
structure interaction. This case has been analysed by L1D [160], a quasi-one-dimensional Lagrangian
code for modeling of free-piston-driven shock tunnels. Moreover, an analytical solution for this flow
problem, is reported in by Jacobs [160].
The simulation geometry is a tube with a constant diameter of 10 mm. The reservoir gas extends
from 0 to 4 m. There is no gas in front of the piston. The length of the piston is 0.01 m, with mass
0.001 kg. The schematic diagram is illustrated in Figure 4.6. The initial flow condition are the same as
the test case in Ref. [160]. The density, pressure and temperature are 1.0 kg/m3 , 100 kPa and 348.4 K,
respectively.
Piston
Reservior Gas
4000
Figure 4.6: Schematic diagram of projectile in tube, dimensions in mm.
The simulation starts at t = 0 with the release of the projectile. The projectile accelerates along
the tube and allows the driver gas to expand behind it. A grid of 96×10 cells is used to simulate the
tube behind the projectile. The pressure at the end wall and piston face are recorded and compared
with the result from L1D [160] in Figure 4.7. The change of pressure at end wall at t ≈10 ms is
due to the expansion propagating to the left into the quiescent driver gas reflecting off the end of the
tube. The decreases of pressure on piston face at t ≈23 ms is caused by the reflected expansion wave
reaching the piston [160]. The position and velocity of the piston at each time step are also recorded.
The comparison with L1D, and ideal solution, is shown in Figure 4.8. There is a deviation between
piston position and velocity in results from Eilmer after x = 3 and the ideal solution. The same was
also noted in L1D. The ideal solution assumes an infinitely long driver and an vacuum in front of
the projectile. The deviation is caused by the expansion that reaches the projectile. This test case
indicates Eilmer is capable of simulating coupled inviscid fluid-structure problems.
4.3.4 Pitching NACA0012 Airfoil

The pitching NACA0012 airfoil is to test moving grids used for inviscid simulations. Experimental
results from AGARD report [161] document the unsteady transonic flow over a NACA0012 airfoil.
The airfoil undergoes pitching oscillations around a point on the chord, one quarter chord length from
120000
End wall-Eilmer
End wall-L1D
Piston face-Eilmer
100000 Piston face-L1D
80000
Pressure, Pa
60000
end wall
40000
piston face
20000
0
0 0.01 0.02 0.03 0.04 0.05
Time, s
Figure 4.7: Comparison of pressure at end wall and piston face.
0.06 250
Eilmer Eilmer
L1D L1D
Ideal Ideal
0.05 200
0.04
Velocity, m/s
150
Time, s
0.03
100
0.02
50
0.01
0 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Position, m Positon, m
(a) (b)
Figure 4.8: Comparison of results between Eilmer, L1D, and ideal solution, (a) simulation time and piston
position, (b) piston velocity and piston position.
the leading edge. The selected case is AGARD test case CT5. The free stream velocity, U∞ , is
parallel to the x axis and the Mach number is set to M=0.755, corresponding to a Reynolds number
of 5.5 × 106 . The oscillation cycle is defined as,
α = αm + α0 sin(ω t) , (4.15)
where the mean angle of attack is αm =0.016°, and the amplitude of the pitching oscillation is
α0 =2.51°. ω is defined as the circular pitch frequency, which is calculated from the reduced frequency,
κ=0.0814, defined as,
ωc
κ= . (4.16)
2 U∞
In many computations of this problem, the mesh is fixed and the free stream velocity U∞ is rotated,
however, the free-stream air velocity is set as constant and the entire mesh is rotated, including the
airfoil. The grid velocity is calculated according to the angular velocity of the pitching airfoil and the
distance between the grid point and the center of rotation.
Figure 4.9 shows the computational domain for the inviscid simulation of the pitching NACA0012
airfoil. The left boundary is regarded as a velocity inlet, the right is set to an extrapolation outflow
boundary condition, and the airfoil surface is modeled as a fixed temperature wall. The outer bound-
aries are 10 chord lengths away from the airfoil. Three different sets of computational meshes are
used for the grid independence study. These are a coarse mesh of 500×160 cells, a medium mesh of
750×240 cells and a fine mesh of 1000×320 cells.
The comparison of normal force coefficient is shown in Figure 4.10. The maximum relative er-
ror between the fine and coarse mesh is 4.64 %, whereas it is 1.32 % between the fine and medium
mesh. The medium computational mesh of 750×240 cells is selected for examination of the pressure
distribution and force coefficients.
Figure 4.11 compares instantaneous surface pressure between Eilmer and experiment [161]. The
pressure distribution at different angles of attack is well captured by the simulation. Since the pressure
is strongly influenced by the shock/boundary interaction, and the present simulation is inviscid, there
is a difference in the indicated shock location.
The experiment also calculated the normal force and moment coefficients during one pitching
loop. The experimental data, along with a the corresponding results from the numerical simulation,
Figure 4.9: Computational domain for NACA0012 airfoil.
0.6
Coarse mesh
Medium mesh
0.4 Fine mesh
Normal force coefficient
0.2
-0.2
-0.4
-0.6
-3 -2 -1 0 1 2 3
Angle of attack, deg
Figure 4.10: Grid convergence study for pitching NACA0012.

1.5 1.5
Eilmer upper Eilmer upper
Eilmer lower Eilmer lower
1 Experiment upper 1 Experiment upper
Pressure coefficieent
Experiment lower
Experiment lower
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-1.5 -1.5
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1
X, m X, m
(a) (b)
1.5 1.5
Experiment lower
Experiment lower
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-1.5 -1.5
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1
X, m X, m
(c) (d)
1.5 1.5
Experiment lower
Experiment lower
0.5 0.5
0 0
-0.5 -0.5
-1 -1
-1.5 -1.5
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1
X, m X, m
(e) (f)
Figure 4.11: Instantaneous pressure coefficient, (a) 0.52°, ↓, downward stroke, (b) −0.54°, ↑, upward stroke,
(c) −2.00°, ↑, upward stroke, (d) 2.01°, ↓, downward stroke, (e) 2.34°, ↓, downward stroke, (f) −2.41°, ↓,
downward stroke.
(a) (b)
0.6 0.04
Eilmer Eilmer
Experiment 0.03 Experiment
Pitching moment coefficient

0.4
Normal force coefficient
0.02
0.2 0.01
0 0
-0.01
-0.2
-0.02
-0.4
-0.03
-0.6 -0.04
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
Angle of attack, deg Angle of attack, deg
Figure 4.12: Comparison between Eilmer and experiment, (a) normal force coefficient, (b) moment coefficient.
are shown in Figure 4.12. The agreement is similar to what has been observed for other solvers, for
example as shown in Ref. [162].
It is concluded that this test case validates the components of the code to simulate inviscid flow
through a moving grid. Turbulent flow through the moving grid will be discussed separately, in
Section 4.3.6.
4.3.5 Oscillating Plate

In order to verify the implementation of the viscous fluxes in the presence of a moving grid, the flow
known as Stokes’s second flow [123] is examined. This flow is the result of an infinite flat plate
undergoing a time-periodic oscillatory translational motion with an stagnant fluid. The time-periodic
plate velocity u(t)plate is given by,
u(t)plate = umax cos(ω t) , (4.17)
where umax is the maximum plate velocity, ω is the frequency of oscillation, and t is time. The
resulting flow is time-periodic finally. Stokes derived an analytical solution for this flow [123], given
by,
√ω r
ω
−y
u(y, t) = umax e 2ν cos(ω t − y ) , (4.18)
2ν
where ν is the kinematic viscosity of the working fluid. This equation can be solved to obtain the
fixed pressure
extrapolate extrapolate
boundary boundary
condition condition
y
moving wall with the oscillating velocity
Figure 4.13: Computational domain for the oscillating plate.
instantaneous velocity profile. The wall shear stress τwall can be derived from Equation 4.18 and is
written as, r
du(y, t) ων
τwall = = ρ uwall (sin(ω t) − cos(ω t)) . (4.19)
dy
y=0 2
The corresponding skin friction coefficient Cf is,
τwall
Cf = . (4.20)
0.5 ρ u2max
To simulate this flow, the computational domain of 40 mm×10 mm, shown in Figure 4.13, was
used. The bottom boundary was set as a wall with an oscillating translational velocity. The maximum
velocity of the plate was umax = 100 m/s, and the frequency of oscillation was 2000 rad/s. The left
and right boundaries were taken as an extrapolation boundary conditions, possible due to the infinite
plate width, while the top was set to fixed pressure. The initial condition of the working fluid (air,
in this case) was 0.1 MPa and 300 K. Three different computational meshes were used for a grid
independence study. These were a coarse mesh of 24×30 cells, a medium mesh of 48×60 cells, and
a fine mesh of 96×120 cells. For each of these meshes, cells were clustered toward the oscillating
wall to ensure the boundary layer is correctly captured. The boundary layer profile and the associated
friction coefficient at a simulation time of 1 ms were extracted from these meshes and are shown in
Figure 4.14. The relative differences in friction coefficient between fine and coarse meshes was less
than 1.19 %, and it was 0.09 % between fine and medium meshes. It appears that the medium mesh
has adequately captured the interesting features of this flow.
The instantaneous velocity profile at different phase angles is compared with Equation 4.18 in
Figure 4.15. The simulations results show very good agreement with the analytical solution. The
relative error is less than 0.02 %, as shown in Figure 4.15(a). Figure 4.15(b) shows the time history of
0.001
Coarse mesh
Medium mesh
0.0008 Fine mesh
0.0006
Y, m
0.0004 1.04
Cf/Cf,fine
1.02
0.0002 1
0.98
0.96
0
-40 -20 0 20 40 60 80 100 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
Velocity, m/s Representative cell size, ∆
(a) (b)
Figure 4.14: Grid convergence study of oscillating plate: (a) velocity distribution at 1 ms, (b) Friction coeffi-
cient over the fine mesh in terms of representative cell size at 1 ms.
the local skin friction coefficient compared with the analytic result computed by Equation 4.20. Again,
very good agreement is evident. It is concluded that this test case verifies the correct implementation
of the viscous fluxes for moving grid in Eilmer as well as the correct application of the tangential
interface velocity at wall boundary interface when computing the spatial derivatives of flow quantities.
4.3.6 Pitching NACA64A010 Airfoil (Turbulent)

Finally, the simulation of viscous unsteady flow at transonic conditions is performed for the NACA
64A010 airfoil for which experimental data was reported by Davis [161]. For this test case, the flow
is fully attached to the airfoil during pitching and allows us to use a two-dimensional simulation. The
high Reynolds number associated with the flow provides an opportunity to test the code components
for computing turbulent flow in the presence of a moving grid and pressure gradients. The pitching
motion of the airfoil is defined as,
α(t) = αm + α0 sin(ω t) , (4.21)
where α(t) is the instantaneous angle of attack, α0 is the pitching range, and αm is the mean angle
of attack. The angular frequency ω is related to the reduced frequency κ, which is defined in Equa-
tion 4.16. The freestream Mach number is M∞ =0.796, α0 =1.01°, αm =0°, and κ=0.202. The Reynolds
number based on chord length is 1.3 × 107 , and the airfoil rotates about the quarter-chord point.
0.001 0.01
Eilmer--0 deg Eilmer
Analytical--0 deg Analytical solution
Local skin friction coefficient

Eilmer--90 deg
0.0008 Analytical--90 deg
0.005
Eilmer--180 deg
Analytical--180 deg
0.0006 Eilmer--270 deg
Y, m
Analytical--270 deg 0
0.0004
-0.005
0.0002
0 -0.01
-100 -50 0 50 100 0 0.5 1 1.5 2 2.5 3 3.5 4
Velocity, m/s Time, ms
(a) (b)
Figure 4.15: Comparison of results between Eilmer and analytical solution, (a) velocity profile at different
phase angles, (b) local skin friction coefficient.
The computational domain for the turbulent simulation of the pitching of this airfoil was essen-
tially the same as the pitching NACA 0012 airfoil, shown in Figure 4.9. The left boundary was
regarded as velocity inlet, while the right was set as a extrapolation boundary condition. The airfoil
surface was modeled as fixed temperature wall. These outer boundaries were again 10 chord lengths
away from the airfoil. Three different computational meshes were used for a grid independence study.
These were a coarse mesh of 360×120 cells, a medium mesh of 720×240 cells and a fine mesh of
1080×360 cells. The cell spacing close to the wall is set to achieve y + < 1 along the airfoil and there
were at least 15 cells in the wall-normal direction within the boundary layer. The k-ω two equation
turbulence model was used in this simulation without a wall function.
The comparison of result for lift force is shown in Figure 4.16. The relative differences between
coarse and medium mesh are less than 6.14 %, while they are only 1.7 % between medium and fine
mesh. As the medium mesh calculation has a reasonable computational cost, it was selected for the
following analysis.
The results for the lift and moment coefficients Cl and Cm for a pitching loop using Eilmer are
shown in Figure 4.17. The lift force coefficient Cl in Figure 4.17(a) is close to experiment, however,
the computation of the momentum coefficient Cm misses two pocket-like features in the experimental
data between the maximum and minimum angles of attack. This difference was also evident in sim-
ulation results by Barakos and Drikakis [163]. Since aerodynamic moment is very sensitive to shock
0.15
Coarse mesh
Medium mesh
0.1 Fine mesh
Lift force coefficient
0.05
-0.05
-0.1
-0.15
-1.5 -1 -0.5 0 0.5 1 1.5
Angle of attack, deg
Figure 4.16: Grid convergence study for the pitching NACA64A010 airfoil.
0.15 0.015
Eilmer Eilmer
Experiment Experiment
Pitching moment coefficient
0.1 0.01
Lift force coefficient
0.05 0.005
0 0
-0.05 -0.005
-0.1 -0.01
-0.15 -0.015
-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5
Angle of attack, deg Angle of attack, deg
(a) (b)
Figure 4.17: Comparison of results between Eilmer and experiment [161], (a) lift force coefficient Cl vs angle
of attack α, (b) moment coefficient Cm vs angle of attack α.
Structural Deformation Solver Section 4.4 97
locations, the details of shock/boundary-layer interaction may affect the moment significantly. Des-
pite these small differences, it is concluded that Eilmer is capable of accurately simulating a turbulent
flow through a moving grid.
4.4 Structural Deformation Solver

Foil bearings consist of top and bump foils as shown in Figure 2.9. The corrugated bump foil is
utilised to support the plate-like top foil. During operation, the top and bump foils are deformed, due
to the pressure force. The deformed parts have to to considered as the deflections are of similar size
as the film thickness.
4.4.1 Top Foil Model

In the current study, the top foil is approximated as a circular thin plate structure. The polar coordin-
ates r and θ are used when solving the bending problems for the circular plate, and the governing
equation, from Ref. [164] is,
∂2 1 ∂ 1 ∂ ∂ 2 w 1 ∂w 1 ∂2w q ∂2w
( 2+ + )( + + 2 )= + ρt ht 2 , (4.22)
∂r r ∂r r 2 ∂θ2 ∂r 2 r ∂r r ∂θ2 D ∂ t
where w is the local deflection, r is the local radius, θ is the local angle, ρt and ht are the density and
thickness of the top foil, D is the stiffness of the thin pad (top foil) and q is the force acting on the top
foil, defined as,
q = p − KB w − CB ẇ . (4.23)
Here KB and CB are the stiffness and damping coefficients of the bump foils that provide support to
the top foil. The stiffness of the thin plate D is defined as,
E h3
D= . (4.24)
12 (1 − µ3 )
The above governing equation is a special case of the Kirchhoff plate equation. It is simplified
considerably for isotropic and homogeneous plates for which the in-plane deformations can be neg-
lected. The theory assumes that a mid-surface plane can be used to represent a three-dimensional
plate in two-dimensional form. The assumptions are [165].
• Straight lines normal to the mid-surface remain straight after deformation;

'r
P ( r ,T )
M rT VT
Vr
MTr
Mr
MT
'T
Figure 4.18: Schematic diagram for forces and moments.
• Straight lines normal to the mid-surface remain normal to the mid-surface after deformation;
• The thickness of the plate does not change during a deformation;
• This equation does not include in-plane tension and is only applicable for plates that carry load
in bending.
This governing equation is only accurate for small deflections and for thin plates subjected to perpen-
dicular forces and bending moments. This is applicable for the structural deformation of the top foil,
as it has three free and one fixed edge and only carries load in bending.
The bending or twisting moment in the polar system are defined as Mr , Mθ and Mrθ , and the shear
forces are Vr and Vθ , shown in Figure 4.18. The bending or twisting moments Mr , Mθ , Mrθ and Mθr
in the polar system are defined as [164],
∂2w 1 ∂w 1 ∂2w
Mr = −D [ + µ ( + )] , (4.25)
∂r 2 r ∂r r 2 ∂θ2
1 ∂w 1 ∂2w ∂2w
Mθ = −D [ + 2 + µ ] , (4.26)
r ∂r r ∂θ2 ∂r 2
1 ∂2w 1 ∂w
Mrθ = Mθr = −D (1 − µ) ( + 2 ) . (4.27)
r ∂r∂θ r ∂θ
Free edge
0.0003
FDM-inner radius
ANSYS-inner radius
0.00025 FDM-medium radius
ANSYS-medium radius
FDM-outer radius
Defomration, m
0.0002
ANSYS-outer radius
Free edge
0.00015 outer radius
medium radius
0.0001 inner radius
Free edge
5e-05
y
0
0 5 10 15 20 25 30 35 40 45
x Angle, deg
Fixed end
(a) (b)
Figure 4.19: (a) Schematic diagram of top foil in foil bearings, (b) deformation comparison between ANSYS
and the Kirchhoff plate structural deformation code at different radii.
The edge forces in radial direction Vr and in tangential direction Vθ are defined as [164],
∂ ∂2w 1 ∂w 1 ∂2w 1 − µ ∂ 1 ∂2w 1 ∂w

Vr = −D [ ( + + 2 ) + ( − 2 )] , (4.28)
∂r ∂r 2 r ∂r r ∂θ 2 r ∂θ r ∂r∂θ r ∂θ
1 ∂ ∂2w 1 ∂w 1 ∂2w ∂ 1 ∂2w 1 ∂w
Vθ = −D [ ( + + 2 ) + (1 − µ) ( − 2 )] . (4.29)
r ∂θ ∂r 2 r ∂r r ∂θ 2 ∂r r ∂r∂θ r ∂θ
For the structural deformation of the top foil, the boundary conditions are shown in Figure 4.19(a):
the attachment edge is a fixed end. The other boundaries are set as free edges, with constant r and θ.
In addition, a nonuniform pressure, p(r, θ) defined by Equation 4.23, is prescribed on the top foil, as
shown in Figure 4.18. The corresponding boundary conditions, constraints are:
• Fixed end at constant angle: θ = 0

∂w
w = 0|θ=0 = 0 , (4.30)
∂θ θ=0
• Free edge at constant radius: r = r0
Mr = 0|r=r0 Vr = 0|r=r0 , (4.31)

Table 4.1: Operating condition of foil thrust bearings from SNL’s test loop [8].
Parameters Value
Working fluid CO2
Ambient pressure 1.4 MPa
Ambient temperature 300 K
Inner diameter 0.0508 m
Outer diameter 0.1060 m
Designed rotational speed 75 000 rpm
Designed minimum film thickness 5 µm
• Free edge at constant angle: θ = α
Mθ = 0|θ=α Vθ = 0|θ=α . (4.32)
The finite difference method (FDM) is used to solve the bending Equation 4.22. The expanded
form of Equation 4.22, neglecting the time term is,
∂4w 2 ∂3w 1 ∂2w 1 ∂w 2 ∂4w 2 ∂3w 4 ∂2w 1 ∂4w q

4
+ 3
− 2 2
+ 3
+ 2 2 2
− 3 2
+ 4 2
+ 4 4
= . (4.33)
∂r r ∂r r ∂r r ∂r r ∂r ∂θ r ∂θ ∂r r ∂θ r ∂θ D
To verify the implementation, a test case of a 45° segment of a circular thin plate is selected. The
thickness of this plate is chosen as 150 µm, which is a typical thickness for the top foil [23], and a
uniform pressure of 70 Pa is prescribed on the top. The inner and outer radii are the same as for the
bearing used by SNL listed in Table 4.1. The material selected is stainless steel with a modulus of
elasticity of 200 GPa and a Poisson’s ratio of 0.3. The deflections are compared with the results from
the static structural solver in ANSYS [166], which is a well-verified solver for this type of deformation
problem.
A grid independence study was performed for both codes, and results are shown in Figure 4.20.
The deflection at the inner, medium and outer radii of the circular thin plate from the developed
structural deformation solver is compared between three different meshes: coarse mesh (100×120
cells), medium mesh (200×240 cells) and fine mesh (400×480 cells). The comparison between these
different meshes is good, with the maximum relative error being less than 0.23 %. The maximum
deflection in terms of representative cell size, ∆, is plotted in Figure 4.20(c) and (d). For these 2D
0.0002 0.0002
Coarse mesh Coarse mesh
Medium mesh Medium mesh
Fine mesh Fine mesh
0.00015 Outer radius 0.00015 Outer radius
Deformation, m
Deformation, m
0.0001 Medium radius 0.0001 Medium radius
5e-05 Inner radius 5e-05 Inner radius
0 0
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
Angle, deg Angle, deg
(a) Kirchhoff plate solver. (b) ANSYS.

1.04 1.04
1.02 1.02
d/dfine
d/dfine
1 1
0.98 0.98
0.96 0.96
0 0.002 0.004 0.006 0.008 0.01 0 0.002 0.004 0.006 0.008 0.01
Representative cell size, ∆ Representative cell size, ∆
(c) Kirchhoff plate solver. (d) ANSYS.
Figure 4.20: Grid convergence study: (a) & (b) local deflection, (c) & (d) maximum deflection.
p
calculations, the representative cell size, ∆, is computed as 1/Ncells, where Ncells is the total number
of cells for each mesh. The medium mesh shows the converged result for the maximum deflection and
is selected for the structural deformation solver in the subsequent analysis. A grid independence study
for the finite element solver in ANSYS was also undertaken for different meshes: coarse mesh (41 299
nodes), medium mesh (67 383 nodes) and fine mesh (113 851 nodes). The computed deformations are
shown in Figure 4.20(b). The maximum relative error between different meshes is less than 0.047 %,
so the coarse mesh is used in the following analysis.
The comparison result between these two codes is shown in Figure 4.19(b). The maximum relative
error between the structural deformation code based on Kirchhoff plate equation and ANSYS is less
than 3.5 %. It is concluded that this structural deformation solver is suitable for studying foil bearings.
4.4.2 Bump Foil Model
An advanced bump foil model introduced by Gad et al. [100] is selected as introduced in Section 2.5.
This bump foil stiffness model considered the interaction between bumps and the friction between
bump foil and the surrounding structure, and is considered to be the most accurate bump foil model
available in literature [100]. The assumptions of his model are:
• The bump foil displacement is from the fixed end to the free end;
• The deflection of the top foil follows the bump foil deflection and no relative deflection exists;
• The deflection is elastic;
• Bump ends are assumed to be a rigid support.
The calculation procedure is sequential. The calculation process is detailed in Figure 4.21.
Structural damping is required for dynamic analysis, however studies on the bump foil damping
coefficient are limited. Balducchi et al. [75] experimentally determined the structural damping coeffi-
cients at various excitation frequency and load conditions. They found that the variation of structural
damping with respect to frequency is small, allowing CB to be treated as being a function of load only.
Using the data from Balducchi et al. [75], a polynomial can be created, relating structural damping,
CB to load, FZ (see Figure 4.22). This polynomial is used in the following analysis to obtain the
structure damping coefficients.
4.5 Fluid-Structure Simulation: Coupling Strategy

Two coupling strategies are possible, weak and strong. Weak coupling between the gas-dynamic
and structural-deformation solvers is when one of the solvers is held, while the other advances for
a number of iterations. Weak coupling can provide a relatively fast way to reach steady-state or
quasi-steady state solution, however the interface fluxes are only balanced in the final steady state.
Conversely, strong coupling maintains the conservation laws at each step, and is therefore essential
for time-evolving solutions. Strong coupling also means that the grid must be adapted at each flow
time step and this was done for the transient validation cases discussed in this section. For the steady-
state performance estimation of foil thrust bearings discussed later, the grid motion is executed at
a comparatively larger but fixed time step ∆ts to save computational cost. The calculation update
sequence is as follows:
1. The fixed-geometry foil thrust bearings are simulated with Eilmer;
2. The simulation with Eilmer is paused at t + ∆ts ;

Fluid-Structure Simulation: Coupling Strategy Section 4.5 103
Figure 4.21: Flow chart of the bump foil model, taken from [100].
1200
Structural damping, Ns/m

1000
800
600
400
200
Experimental data
0.75
30.5x -206
0
20 40 60 80 100 120 140 160
Load, N
Figure 4.22: Structural damping coefficient, CB .
3. Pressure load is passed from Eilmer to the structural deformation code;
4. The structural solution is updated;
5. Deflection is passed from the structural deformation code to Eilmer;
6. Grid in Eilmer is updated and fluid domain simulation continues;
7. Repeat steps 2-6 until the steady-state solution is achieved.
Since the fluid solver uses a cell-centred finite-volume method, and the structural deformation
code used a node-based finite-difference formulation, and as the solvers require different meshes for
grid independence, the data is stored in a different mesh for each solver. A mapping algorithm is used
when passing pressure values from Eilmer to the structural deformation code and deflection from the
structural deformation code back to to Eilmer.
Figure 4.23 shows the stencils for mapping pressure and deflection at the fluid-structure interface.
When passing pressure from Eilmer to the structure solver (pressure is stored at the cell centre in
Eilmer and vertices for the structure solver), the mapping method firstly searches for neighbouring
cell centre from Eilmer for each vertex in the structure solver. A bilinear interpolation method is
then used to calculate pressure for the vertex of the structure solver as shown in Figure 4.23(a).
This is only applied when four neighbouring cells are found; otherwise the mapping method will
search for the nearest cell centre instead. The same mapping method is also applied when passing the
deflection from the structure solver to Eilmer (from structure vertex to fluid mesh vertex), as shown in
Fluid-Structure Simulation: Coupling Strategy Section 4.5 105
Fluid,
cell centre
Structure,
vertex
Circumferential
Radial
(a)
Fluid, vertex
Structure, vertex
Circumferential
Radial
(b)
Figure 4.23: Stencil for mapping method at the fluid-structure interface, (a): from Eilmer to the structure
solver, (b): from the structure solver to Eilmer.
Table 4.2: Comparison of pressure force between Eilmer and the structure solver.
Radius range Angle range Force, structure Force, Eilmer Relative error
25.4 mm to 50.8 mm 0° to 45° 10.73 N 10.80 N 0.69 %
25.4 mm to 50.8 mm 11.48° to 12.85° 0.74 N 0.74 N 0%
25.4 mm to 50.8 mm 15.63° to 15.93° 0.1259 N 0.1262 N 0.22 %
25.4 mm to 50.8 mm 18.29° to 19.31° 0.364 N 0.363 N 0.23 %
25.4 mm to 50.8 mm 25.71° to 27.81° 0.475 N 0.477 N 0.37 %
25.4 mm to 50.8 mm 36.24° to 37.96° 0.21 N 0.22 N 0.39 %
Figure 4.23(b). This coupling approach usually requires fine meshes at the fluid-structure interfaces
to ensure physical conservation [167]. The following analysis has shown that the respective fluid
and structural meshes at grid independence state from the two solvers are adequate. For example, a
48×96×15 mesh for Eilmer and a 200×240 mesh for the structure solver.
Figure 4.24 shows the pressure and deflection contours at the fluid-structure interface for these
two solvers indicating the order of the mapping process ((a)→(b)→(c)→(d)). These contours indic-
ate the same pattern (pressure and mapped pressure, deflection and mapped deflection), however, a
more detailed comparison is undertaken to verify the accuracy. To verify the mapped pressure from
Eilmer to the structure solver, the pressure forces acting on specific regions of the top foil for the two
solvers are compared. As shown in Table 4.2, the relative error for pressure forces between these
two solvers is less than 0.7 %. This is sufficiently accurate for mapping pressure at the fluid-structure
interface. To verify the mapped deflection field, the deflection solved by the structure solver and the
mapped deflection for Eilmer are shown in Figure 4.25. This indicates a relative error of 1 %, which
is sufficiently accurate.
4.6 Conclusions
In this section, a computational tool to simulate the three-dimensional elastohydrodynamic operation
of foil thrust bearings is developed. This computational problem consists of two parts, fluid simulation
and structural deformation. The CFD code Eilmer was modified to include a moving grid capability.
Different cases were provided to test the suitability of Eilmer to simulate inviscid, viscous and turbu-
lent flows. Additionally, a new finite difference solver was developed for the structural deformation
within the foil bearing. This is verified with the commercial software ANSYS, and the relative differ-
0.035 48000 0.035 48000
42000 42000
0.030 0.030
36000 36000
0.025 0.025
30000 30000
0.020 0.020
Y, m
Y, m
24000 24000
0.015 0.015
18000 18000
0.010 0.010
12000 12000
0.005 6000 0.005 6000
0.020 0.025 0.030 0.035 0.040 0.045 0.050 0 0.000 0.020 0.025 0.030 0.035 0.040 0.045 0.050 0
X, m X, m
(a) Eilmer, pressure in Pa. (b) Structure solver, mapped pressure in Pa.
0.035 5.6 0.035 5.6
4.8 4.8
0.030 0.030
4.0 4.0
0.025 0.025
3.2 3.2
0.020 0.020
Y, m
Y, m
2.4 2.4
0.015 0.015
1.6 1.6
0.010 0.8 0.010 0.8
0.005 0.0 0.005 0.0
0.000 0.020 0.025 0.030 0.035 0.040 0.045 0.050 −0.8 0.000 0.020 0.025 0.030 0.035 0.040 0.045 0.050 −0.8
X, m X, m
(c) Structure solver, deflection in µm. (d) Eilmer, mapped deflection in µm.
Figure 4.24: Pressure and deflection contours at the fluid-structure interface, the mapping process is
(a)→(b)→(c)→(d).
10
Sturcture solver-inner radius
Eilmer-inner radius
8 Sturcture solver-medium radius
Eilmer-medium radius
Local deflection, µm
6 Sturcture solver-outer radius
Eilmer-outer radius
-2
0 5 10 15 20 25 30 35 40 45
Angle, deg
Figure 4.25: Comparison of deflection by the structure solver and mapped deflection for Eilmer.
ence is less than 3.5 %. A coupling strategies between fluid and structural deformation solvers wass
developed to obtain the steady state and dynamic performances of foil thrust bearings. Within the
coupling strategy, a mapping algorithm is used to exchange pressure and deflection between the two
solvers. A difference of less than 1 % is found. So far, the work has been verified through comparison
to numerical results from literature and good agreement has been observed.
Chapter 5
Foil Bearing Steady State Performance
5.1 Introduction
Steady state performance of foil thrust bearings, using the fluid-structure simulation tool developed
in Chapter 4 is studied in this Chapter. First, Section 5.2 outlines the parameters used to evaluate
bearing performance. Section 5.3 then briefly details the validation of the developed fluid-structure
simulation tool by comparing to experimental data of air foil thrust bearings. Section 5.4 selects the
gas model for real gas properties of CO2 . Section 5.5 compares results from Reynolds equation and
Eilmer to highlight the new flow physics introduced by high density CO2 . Section 5.6 compares the
steady state performance at different operating conditions. Section 5.7 revisits the selection of the
computational domain for foil thrust bearings by comparing the steady state performances. Finally,
Section 5.8 presents the steady state performance at different rotational speeds and rotor-to-stator
separations.
5.2 Steady State Performance Evaluation

The fluid-structure simulation starts with a stationary flow and a initial rotor to top foil separation
between the rotor and stator defined by the relaxed bump foil height and top foil as shown in Fig-
ure 5.1. Once the steady state pressure distribution is obtained, the load Fz and power loss Pw are the
two key parameters used to evaluate steady state performances. They are defined as,
Z
Fz = (p − pa ) dA , (5.1)

dUθ
Z
Pw = wr rµ dA , (5.2)
dz z=zrotor
110 Chapter 5 Foil Bearing Steady State Performance
Circumferential
Rotor
Rotor to stator Rotor to top foil
separation separation
Foil Block
Top Foil Bump Foil Housing Groove Ramp Region Flat Region
Figure 5.1: Side view of foil thrust bearings, not to scale.
where p is the pressure in the film, pa is the ambient pressure, A is area, wr is the rotational speed, µ
dUθ
denotes the viscosity, and dz
is the gradient of the tangential velocity close to the rotor surface.
5.3 Validation for Steady State Performance

Dickman [73] presented experimental results for three identical thrust foil bearings operating with
shaft speeds as high as 40 000 rpm under increasing loads. Since air was used as operating fluid in
these experiments, viscous heating effects will be low, and the results can be used as a test case for
fluid-structure coupled simulation of foil thrust bearings. Table 2.1 lists the geometry and material
properties of the tested thrust foil bearing. The bearing configuration, materials, and the experimental
results, are regarded as a benchmarking test model for foil thrust bearings [26]. Dickman [73] did not
release all information about the tested foil thrust bearings so some geometric items in Table 2.1 have
been obtained from San Andrés et al. [26].
The corresponding computational domain of the foil thrust bearings is shown in Figure 5.2. It con-
sists of two parts, namely, the ramp and the flat region. The top surface of the computational domain
is the rotor while the bottom surface is the top foil. The rotational direction is shown in Figure 5.2.
Boundary conditions at the surrounding surfaces are modeled as fixed pressure and temperature, the
top wall is regarded as a moving wall with fixed temperature, and the bottom wall (top foil) is set as a
fixed temperature wall and coupled with the structural deformation solver. Three different computa-
tional meshes are considered for the grid independence study. These are a coarse mesh of 24×48×10
cells, a medium mesh of 48×96×15 cells and fine mesh of 72×144×20 cells.
Validation for Steady State Performance Section 5.3 111
Flat Region
Ramp Region
z
y
x
Figure 5.2: Computational domain for foil thrust bearings, not to scale.
Table 5.1: Performance comparison of foil thrust bearings with different computational meshes.
Parameters Coarse mesh Medium mesh Fine mesh

Lift force 59.89 N 60.82 N 61.03 N
Friction torque 0.0161 Nm 0.0161 Nm 0.0162 Nm
During operation, the fluid is drawn from the ramp to the flat region. The non-dimensional pres-
sure and local deflection at the medium radius of the computational domain of the foil thrust bearings
are extracted and compared, as shown in Figure 5.3 and summarised in Table 5.1. For the pressure
field, the relative difference between the coarse and medium meshes is 0.24 %, and between the me-
dium and fine meshes it is less than 0.1 %. For the local deflection, the relative difference between
coarse and medium mesh is 0.87 % and 0.27 % between the medium and fine mesh. The medium
mesh is considered in the subsequent discussion.
Figure 5.4 compares measurement from Dickman [73] and numerical simulation at a rotational
speed of 21 000 rpm. There is good agreement between predicted and measured data for small to
moderate loads. This is also reported by San Andrés et al. [26] where it is also noted that the rapid
increase in drag at the highest applied load in the tests is caused by rubbing contact between the top
foil and the shaft collar. Simulations for these highest load values are not performed.
Figure 5.5 shows the numerical result of the maximum deformation of the top foil under the same
minimum film thickness. There is a small difference between Eilmer results and the results from
San Andrés et al. [26] this is attributted to the use of different fluid and structural models. Ref. [26]
6 1.5
Coarse mesh Coarse mesh
Medium mesh Medium mesh
5 Fine mesh 1.4 Fine mesh
4
1.3
3
1.2
2
1.1
1
0 1
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
Circumferential angle, deg Circumferential angle, deg
(a) (b)
Figure 5.3: Grid convergence study for foil thrust bearings: (a) local deflection at the medium radius, (b)
pressure at the medium radius.
0.06
Test Data, Dickman
Drag torque, Nm
0.05 Eilmer
0.04
0.03
0.02
0.01
0
0 50 100 150 200
Load, N
Figure 5.4: Comparison of measurements and numerical simulation at fixed rotational speed of 21 000 rpm.
used the Reynolds equation for the flow modelling and a finite element plate model to determine
the elastic deformation of a top foil and its supporting bump strip layers. Despite these differences,
the overall numerical performance of the present simulations is acceptable, and the suitability of this
fluid-structure simulation tool for foil bearings is demonstrated.
5.4 Selection of Gas Models

Conboy [8] reported an operating condition of 1.4 MPa and 300 K for the foil thrust bearings used
during the preliminary tests at SNL. The position of this operating condition on the Pressure-Enthalpy
diagram for CO2 is shown in Figure 5.6.
In this section, the simulation results obtained using an ideal gas model (perfect equation of state
Comparison Between Reynolds Equation and Eilmer Section 5.5 113
Maximum deformation, µm
12
San Andres et al
10 Eilmer
8
6
4
2
0
0 5 10 15 20
Minimum film thickness, µm
Figure 5.5: Comparison of maximum deformation between Eilmer and numerical results from Ref. [26].
and Sutherland’s law for transport properties) and results obtained using a look-up table generated
by REFPROP [120] are compared. The different thermodynamic and transport properties. Pressure,
compressibility factor, density, temperature, dynamic viscosity and thermal conductivity at the me-
dium radius of the foil thrust bearings are plotted in Figure 5.7. The difference in results obtained
using the ideal gas model and look-up table are between 1 % and 10 %. The largest deviation is in
density. This is due to difference in equation of state, ideal gas (perfect-gas equation of state, Suth-
erland law for viscosity and conductivity) and REFPROP (Span and Wagner [168] for equation of
state). The compressibility factor from the look-up table is around 0.93 as shown in Figure 5.7(b),
which indicates slight non-ideal gas behaviour at this operating point. In addition, the transport prop-
erties of CO2 do not follow the conventional relationship that is described by Sutherland’s law, as
indicated in Figures 5.7(e) and (f). Finally, it is noted that the optimum operating condition of foil
thrust bearings for sCO2 cycles is unknown, and may be close to the critical point. To ensure the foil
thrust bearings operation at any operating point can be evaluated reliably, the look-up table approach
is used for the subsequent analysis in this thesis.
5.5 Comparison Between Reynolds Equation and Eilmer

This section compares foil bearing predictions attained from using Reynolds equation and the CFD
method (Eilmer). For laminar flow between a rotating and stationary disc, the simplified Reynolds
equation is given as (see Section 2.1),
1 ∂ 1 ∂p 1 ∂ 1 ∂p ∂(ρ h)
(r ρ h3 )+ 2 (ρ h3 )=Λ . (5.3)
130 kg/m3
605 K
280 K 305 K 330 K 355 K 380 K 405 K 110 kg/m3
90 kg/m3
10 70 kg/m3
Pressure, MPa
50 kg/m3
30 kg/m3
Operating point at SNL 10 kg/m3

1.4 MPa, 300 K
1
0 200 400 600 800 1000

Enthalpy, kJ/kg
Figure 5.6: Pressure-Enthalpy diagram for CO2 .
When required, additional correction factors can be included to account for turbulent effects [8].
However, the current comparison is restricted to laminar formulation shown in Equation 2.4, and the
laminar simulation for Eilmer. The load capacity F and friction torque T for the Reynolds equation
solution are calculated as [84, 85],
Z Z
F = (p − pa )dr dθ , (5.4)
r θ
h ∂p µ ω r 2
Z Z
T = [ + ] r dr dθ . (5.5)
r θ 2 ∂θ h
The accuracy of Reynolds equation to model air foil thrust bearings is well established [26, 76,
84]. This section explores the differences that exist between the two modelling approaches for two
operating conditions:
Low density: working fluid, CO2 , pressure, 0.1 MPa, temperature, 300 K, density, 1.77 kg/m3;
High density: working fluid, CO2 , pressure, 1.4 MPa, temperature, 300 K, density, 26.61 kg/m3.
The studied bearing geometry is based on Table. 2.1, however as the aim is to compare fluid
1.05 Ideal gas model

Ideal gas model 1.04
Look-up table Look-up table
1.04 1.02
Compressibility factor
1.03 Error relative to: 1
pressure increase < 4.5%
0.98
1.02
0.96 Ramp region Flat region
1.01
0.94
1 Ramp region Flat region 0.92
0.99 0.9
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
(a) (b)
28 305
ideal gas model Ideal gas model
Look-up table Look-up table
27.5
304 Ramp region Flat region Error: < 1%
27
Temperature, K
Density, kg/m3
26.5 303
26 Ramp region Flat region Error: about 5%
302
25.5
25
301
24.5
24 300
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
(c) (d)
1.54e-05 0.018
Ideal gas model Ideal gas model
Thermal conductivity, W/(m K)
Look-up table 0.0178 Look-up table

1.535e-05
Dynamic viscosity, kg/(m s)
1.53e-05 0.0176
1.525e-05 0.0174
1.52e-05 Ramp region Flat region 0.0172 Ramp region Flat region Error: about 10%
1.515e-05 Error: about 1% 0.017
1.51e-05 0.0168
1.505e-05 0.0166
1.5e-05 0.0164
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
(e) (f)
Figure 5.7: Comparison of thermodynamic properties at the medium radius with the rotational speed of
30000 rpm: (a) Pressure, (b) Compressibility factor, (c) Density, (d) Temperature, (e) Dynamic viscosity,
(f) Thermal conductivity.
Table 5.2: Performance comparison of the fixed geometry thrust bearings between Eilmer-laminar and Reyn-
olds equation.
Operating Methodology Load Torque

conditions capacity Nmm
N
0.1 MPa, 300 K Reynolds equation 621.38 93
0.1 MPa, 300 K Eilmer-laminar 629.56 96
1.4 MPa, 300 K Reynolds equation 624.46 96
1.4 MPa, 300 K Eilmer-laminar 1137.46 97
solvers, structural deformations are not considered. The rotational speed is set to 50 000 rpm and the
initial rotor-to-stator separation is set to 5 µm. The illustration is depicted in Figure 5.1, which is the
clearance between rotor and top foil.
The results from Eilmer and Reynolds equation are shown in Figure 5.8. At the low density
condition (0.1 MPa, 300 K), the solutions from Reynolds equation and Eilmer are almost identical.
On the contrary, at the high density condition (1.4 MPa, 300 K), a higher peak pressure is predicted in
the ramp region by Eilmer. This difference in predicted pressure distribution manifests as a notable
performance difference as summarised in Table. 5.2. For the low density conditions close agreement
exists, however for the high density condition the variation in load capacity is 45 %. The cause of this
difference is one of the foci of this Chapter.
For high pressure CO2 (dense gas) new physical effects not included in the Reynolds equation in-
fluence performance. As CFD methods perform a more comprehensive analysis of the flow behaviour,
Eilmer is used to create new insight how dense gas operation affects foil bearing performance.
This following parts are devoted to explain the cause of different load coupling between Reynolds
equation and Eilmer. The geometry of the foil thrust bearing is the same as listed in Table 2.1 and the
minimum film thickness (rotor to top foil separation) is set to 16 µm. A rigid bearing is used, as this
allows a more direct comparison between the cases.
The difference is revealed by comparing the simulation to Reynolds equation. To compare the
operation with a low and high density gas, the same bearing was simulated, operating with air and
CO2 and with the operating conditions listed in Table 5.3.
Maps of pressure increase relative to the operating pressure are shown in Figure 5.9. The flow
inside the foil thrust bearings is a typical shear driven flow and the pressure increase is mainly created
400000 35 2200000
35
30 30
2000000
25 25
300000
20 20
Y, mm
Y, mm
1800000
15 15
200000
10 10 1600000
5 5
20 25 30 35 40 45 50 100000 20 25 30 35 40 45 50 1400000
X, mm X, mm
(a) Eilmer-laminar. (b) Reynolds equation.
400000 35 2200000
35
30 30
2000000
25 25
300000
20 20
Y, mm
Y, mm
1800000
15 15
200000
10 10 1600000
5 5
0 20 25 30 35 40 45 50 100000 0 20 25 30 35 40 45 50 1400000
X, mm X, mm
(c) Eilmer-laminar. (d) Reynolds equation.
400000 2.2e+06
Reynolds Equation 2.1e+06 Reynolds Equation
350000
Pressure, Pa
Pressure, Pa
Eilmer 2e+06 Eilmer

300000 1.9e+06
Ramp region Flat region
250000 1.8e+06 Ramp region Flat region
200000 1.7e+06
1.6e+06
150000 1.5e+06
100000 1.4e+06
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
(e) Low density condition. (f) High density condition.
Figure 5.8: Comparison between Eilmer-laminar and Reynolds equation, (a) Pressure (in Pa) contour for low
density condition, (b) Pressure (in Pa) contour for high density condition, (c) Pressure (in Pa) contour for low
density condition, (d) Pressure (in Pa) contour for high density condition, (e) & (f) Comparison of pressure
at the medium radius.
Table 5.3: Operating conditions for CO2 and air.
Parameters CO2 Air

Pressure 1.4 MPa 0.1 MPa
Temperature 300 K 300 K
Density 26.93 kg/m3 1.16 kg/m3
Viscosity 15.17 × 10−6 kg/ms 18.57 × 10−6 kg/ms
Flat region
Pressure increase
1.6e+05
1.2e+5
8e+4
4e+4
y
0
-1.2e+04
x
Ramp region
(a) CO2
Flat region
Pressure increase
1.6e+05
1.2e+5
8e+4
4e+4
y
0
-1.2e+04
x
Ramp region
(b) Air
Figure 5.9: Pressure increase contour for foil thrust bearings, (a) CO2 , rotational speed: 60 000 rpm, (b) Air,
rotational speed: 60 000 rpm.
by viscous forces and the convergent geometry. At both conditions, the maximum pressure is at the
end of the ramp and at the start of the flat region. The two conditions (Air and CO2 ) display quite
different pressure distributions with CO2 showing a much increased peak magnitude. To compare the
results, the non-dimensional pressure parameter [8] is used,
6 ω µo R2 2
Λ= ( ) , (5.6)
po h2
where subscript o indicates the operating condition, h2 is the minimum film thickness and R2 is the
outer radius of the foil thrust bearing. Hence, the pressure increase relative to the operating pressure
is proportional to,
∆p ∝ pa (Λ − 1) . (5.7)
where pa is the ambient pressure. The non-dimensional pressure distribution at two different ro-
tational speeds (30 000 rpm and 60 000 rpm) is compared between Eilmer and Reynolds equation
(Equation 5.8) in Figure 5.10. As this version of Reynolds equation doesn’t include inertia forces, the
comparison highlights effects of the fluid inertia. As indicated in Figures 5.10(a) and (c), the pressure
distribution between Eilmer and Reynolds equation are in good agreement if air is the operating fluid,
however, different results are shown for CO2 . The results from Eilmer show a higher pressure in the
ramp region compared to Reynolds equation, and a lower pressure in the flat region (Figures 5.10(b)
and (d)).
To further explore this difference, Figure 5.11 and Figure 5.12 show flow properties (pressure
increase, radial and tangential velocity) for a number of radial slices on the ramp and the flat region
obtained using Eilmer. These figures highlight how fluid is transported by the combination of shear,
pressure and inertia effects. Figure 5.13 shows sketches of streamlines close to the rotor and stator (at
90% and 10% film thickness) created to highlight the flow pattern. These figures show a significant
difference in gas flow, internal to the bearing.
In the air case, fluid is drawn into the bearing from the ramp start, and pressurised as the viscous
forces push the fluid up the ramp into a convergent gap. The highest pressure is generated along
1
the bearing centre line (rcenter ≈ 2
(rinner + router )) and fluid flows radially inwards and radially
outwards towards the fixed pressure boundary conditions. The same process continues after the ramp,
where fluid continues to flow from the centre line towards the inner and outer edge. Effectively a
2 1.12
Reynolds Equation Reynolds Equation
Eilmer 1.1 Eilmer
1.8 1.08
1.06
Ramp region Flat region 1.04
1.4 1.02
1
1.2
0.98
Sub-ambient pressure
1 0.96
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
(a) Air, 30 000 rpm (b) CO2 , 30 000 rpm

2 1.12
Reynolds Equation Reynolds Equation
Eilmer 1.1 Eilmer
1.8 Ramp region Flat region 1.08

1.06
1.6
1.04
1.4 1.02
1
1.2
Sub-ambient pressure
1 0.96
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
(c) Air, 60 000 rpm (d) CO2 , 60 000 rpm
Figure 5.10: Comparison of non-dimensional pressure distribution at the medium radius between Eilmer and
Reynolds equation.
fixed amount of air is drawn up the ramp into the bearing and this depletes as air leaks to the inner
and outer edges, as illustrated by the streamline sketches in Figures 5.13(a) and (b). This resembles a
fluid flow that is dominated by the pressure and viscous forces.
The flow field for CO2 is very different. Again, fluid is drawn into the bearing from the ramp start,
however, the fluid close to the rotor now experiences a significant radial force due to inertial effects.
Hence, while the flow field close to the stator is similar to the flow experienced with air, predominantly
outwards flow exists close to the rotor. See Figure 5.11(a) and (b) for radial velocity profiles, where the
bold line corresponds to the location of vradial = 0. Only towards the end of the ramp (θ > 12.21 deg),
dP
when the pressure inside the bearing is much higher than the surrounding pressure (large dr
) does
the flow direction at the bearing inner radius reverse, as shown in Figure 5.11(c). This illustrates that
for CO2 bearings, inertia effects have a significant impact on the flow field. The overall effects are
illustrated in the streamline sketches shown in Figures 5.13(c) and (d). In addition to changing the
flow, the strong inertia effects also create a mechanism that locally enhances pressure increase and
lift generation on the ramp. The strong inertia force acting on the fluid close to the bearing inner edge
balances the pressure gradient. This manifests as a radial velocity, vradial ≈ 0 in the third of the film
closest to the rotor and inner edge. Consequently, the amount of gas that is drawn into the bearing
from the start of the ramp depletes less quickly and also receives some additional in-flow close to the
rotor.
After the ramp, the CO2 flow continues to be different to that of air. Here, the inertial effects
continue to dominate and the flow has a strong outwards component. This is especially clear at high
speeds and shown in Figure 5.12. The result is a more rapid mass loss through the outer edge of the
bearing. As a consequence, a region with sub-ambient pressure can be generated in the flat region.
This is because the available flow area in the radial direction increases with radius, thus creating
a restriction (location of smallest flow area) at the bearing inner edge. The same effect has also
been identified by Garratt et al. [127] and Pinkus et al. [169], who studied inertial effects in film
riding bearings. Consequently, rather than enhancing the lift generation, the flat region can now be
detrimental to total bearing lift. Furthermore, in an actual bearing, this subambient pressure may lead
to separation of the top foil from bump foil, which will cause further issues.
Looking back at Figure 5.10, which compares the Eilmer results with the solution to Reynolds
equation, the large differences seen for CO2 confirm the importance of using a full 3D CFD in order
160000 160000 160000

CO2 CO2 CO2
140000 Air 140000 Air 140000 Air
120000 120000 120000
100000 100000 100000
∆p, Pa
∆p, Pa
∆p, Pa
80000 80000 80000
60000 60000 60000
40000 40000 40000
20000 20000 20000
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Non-dimensional radius Non-dimensional radius Non-dimensional radius
Pressure increase ∆p
Normalised height 0 100 m/s 0 100 m/s

Normalised height 0 100 m/s
Normalised height
0 0.33 0.67 1 0 0.33 0.67 1 0 0.33 0.67 1
CO2 : Radial velocity profile
0 500 m/s 0 500 m/s 0 500 m/s

Normalised height Normalised height Normalised height
0 0.33 0.67 1 0 0.33 0.67 1 0 0.33 0.67 1
CO2 : Tangential velocity profile
Normalised height 0 100 m/s

Normalised height 0 100 m/s Normalised height 0 100 m/s
0 0.33 0.67 1 0 0.33 0.67 1 0 0.33 0.67 1
Air: Radial velocity profile
0 500 m/s 0 500 m/s 0 500 m/s

0 0.33 0.67 1 0 0.33 0.67 1 0 0.33 0.67 1
Air: Tangential velocity profile

(a) 3.55° (b) 9.05° (c) 12.21°
Figure 5.11: Ramp region: pressure increase, radial and tangential velocity at different circumferential angles.
Rotational speed: 60 000 rpm, ramp ends at 15°, (Rotor at top, stator at bottom).
160000 160000 160000

CO2 CO2 CO2
140000 Air 140000 Air 140000 Air
120000 120000 120000
100000 100000 100000
∆p, Pa
∆p, Pa
∆p, Pa
80000 80000 80000
60000 60000 60000
40000 40000 40000
20000 20000 20000
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Non-dimensional radius Non-dimensional radius Non-dimensional radius
pressure increase ∆p
Normalised height 0 100 m/s Normalised height 0 100 m/s 0 100 m/s
Normalised height
0 0.33 0.67 1 0 0.33 0.67 1 0 0.33 0.67 1
CO2 : Radial velocity profile
0 500 m/s Normalised height 0 500 m/s 0 500 m/s

Normalised height Normalised height
0 0.33 0.67 1 0 0.33 0.67 1 0 0.33 0.67 1
CO2 : Tangential velocity profile
Normalised height 0 100 m/s Normalised height 0 100 m/s Normalised height 0 100 m/s
0 0.33 0.67 1 0 0.33 0.67 1 0 0.33 0.67 1
Air: Radial velocity profile
0 500 m/s 0 500 m/s 0 500 m/s

0 0.33 0.67 1 0 0.33 0.67 1 0 0.33 0.67 1
Air: Tangential velocity profile

(a) 24.20° (b) 33.66° (c) 41.02°
Figure 5.12: Flat region: pressure increase, radial and tangential velocity at different circumferential angles.
Rotational speed: 60 000 rpm, flat starts at 15°, (Rotor at top, stator at bottom).
Flat region
Flat region
y y
Ramp region
Ramp region
15◦ 15◦
x x
(a) Air: rotor (b) Air: stator
Flat region Flat region
y y
Ramp region Ramp region
15◦ 15◦
x x
(c) CO2 : rotor (d) CO2 : stator
Figure 5.13: Approximate streamlines at 90% (rotor) and 10% (stator) film thickness (radial component is
exaggerated).
Effect of Operating Conditions Section 5.6 125
to correctly analyse film-riding bearings when operating with dense gases. When using light gases
(air), the solutions agree to within 4 %. However for dense gas cases, significant differences exist,
which are caused by the fact that Reynolds equation cannot account for the irregular radial velocity
profiles that are created by the strong inertial effects.
5.6 Effect of Operating Conditions

For the small sCO2 test rig at SNL, leakage flow passes around the compressor and turbine and
through labyrinth shaft seals to provide lubrication to the foil bearings [5] as shown in Figure 5.14.
The leakage flow is continuously pumped out of this region using a scavenging pump to maintain
a reduced pressures, which is around 1.4 MPa and 300 K. This is taken as the operating condition
for the foil thrust bearings in SNL’s tests [8]. This turbomachinery architecture incorporating foil
bearings can eliminate system contamination from oil and enable high speed operation. Therefore,
the architecture of the turbomachinery system is simplified by using foil bearings.
Inlet Outlet
Bearing
Leakage Leakage
Outlet Turbine Shaft Compressor Inlet

Seal Seal
Figure 5.14: The architecture of a supercritical CO2 turbomachinery system.
5.6.1 Selection of Operating Conditions

The effect of operating conditions on the performance of foil thrust bearings for sCO2 power cycles
is unknown. With different CO2 ambient pressure and temperature, the fluid exhibits changing prop-
erties as shown in Figure 5.15 (dynamic viscosity and density). Dynamic viscosity simply varies as
a function of temperature, allowing the investigation to be de-coupled from viscosity by maintain-
ing a fixed temperature. More importantly, as shown in Figure 5.15, CO2 exhibits a wide range of
densities (1 to 90 kg/m3 ), which results in very different fluid dynamics and centrifugal inertia effects.
4 90 80 70
20
60
3.5
16.5
17.5
18.5
19.5
17
18
19
50
3
Pressure, MPa
40
2.5
2 30
1.5 20
1
16
15.5
10
0.5
300 320 340 360 380 400

Temperature, K
Figure 5.15: Density (solid line, in kg/m3 ) and dynamic viscosity (dashed line, in ×106 kg/ms) at different
operating conditions.
As indicated in Section 5.5 (rigid bearing simulation) the highly dense CO2 generates a higher peak
pressure in the ramp region and a higher pressure drop in the flat region compared to what is expected
at more traditional operating conditions.
Due to the high density, the flow can also be in the turbulent regime [8]. The transition from
laminar to turbulent flow can be determined by the rotational Reynolds number Rer , defined as,
ρ ω r h2
Rer = . (5.8)
µ
No experimental study has been undertaken to determine the critical Reynolds number Rec for
CO2 , however, Souchet [131] recommended 900 to 1600 as the critical Reynolds number for fluid
flow in bearings. Below 900 the flow is purely laminar, above 1600 the flow is fully turbulent and
the flow is regarded as transitional between these Reynolds numbers. The turbulent flow can increase
load capacity and friction torque for foil thrust bearings, due to the effects of eddy viscosity in the
turbulent regime [132].
The bearing studied in this Chapter has the geometry from Dickman [73], summarised in Table 2.1.
However, the inner and outer radii of the foil thrust bearing have been adapted to suit a 100 kW ra-
dial inflow turbine currently being designed at the authors institution, which are 8 mm and 32 mm,
respectively. A uniform distribution of bump foil stiffness is prescribed over the entire pad (ramp
and flat regions) and the magnitude of bump stiffness per unit area is set to 6.44 N/mm3. The bump
Reynolds number
4 00 00
40 35 300
0 0
3.5 250
0
3 200
Pressure, MPa
2.5
1500
2
1.5 1000
1
500
0.5
300 320 340 360 380 400

Temperature, K
Figure 5.16: Rotational Reynolds number at different operating conditions (based on 5 µm initial gap and
50 000 rpm).
foil distribution and stiffness should be studied to maximise the bearing performance in future. The
rotational speed is set to 50 000 rpm and the initial rotor and top foil separation is set to 5 µm.
In the SNL experiment, foil thrust bearings are placed downstream of the seals [5]. To match these
conditions, the operating pressure range was selected as 0.1 to 4.0 MPa and the operating temperature
range was selected as 300 to 400 K. As shown in Figure 5.16, the rotational Reynolds number for
these operating conditions ranges from 200 to 4000, resulting in different flow regimes across the
range of operating conditions.
5.6.2 Centrifugal Effects

Using the laminar flow solvers, the foil thrust bearing exhibits largely varying performance character-
istics across the operating range as shown by the contours of load capacity and torque in Figure 5.17.
The results are obtained by evaluating the foil thrust bearing performance at multiple points across
the operating pressure (0.1 MPa, 0.4 MPa, 0.7 MPa, 1.0 MPa, 1.2 MPa and 1.4 MPa) and the operating
temperature (300 K, 320 K, 340 K, 360 K, 380 K and 400 K) range. For some cases, localised trans-
ition to turbulence might take place close to the outer radius. As no transition model is implemented
in Eilmer at the moment and, as this transition is localised, all cases have been simulated as fully
laminar.
For laminar simulations with a fixed operating pressure, both the load capacity and friction torque
increase with increasing operating temperature, this can be easily explained by the conventional the-
1.2 76
1.0
8.4
Pressure, MPa
72
10.0
9.6
9.2
0.8
8.8 68
0.6
0.4 64
.4
10
0.2 60
56
300 320 340 360 380 400
Temperature, K
Figure 5.17: Performance comparison at different operating conditions from laminar fluid solver, solid line:
load capacity in N, dashed line: torque in Nmm, Eilmer-laminar. Rotational speed: 50 000 rpm.
70
Ramp region Flat region
60 Rotational direction
Pressure increase, kPa
50 13.9 kPa
40
6.77 kPa
30
0.1MPa and 400K
20 0.4MPa and 400K
0.7MPa and 400K
1.0MPa and 400K
10 1.2MPa and 400K
1.4MPa and 400K
0 5 10 15 20 25 30 35 40 45
Circumferential angle, degree
Figure 5.18: Pressure increase (p − pa ) at the different operating pressures. Operating temperature fixed at
400 K, Eilmer-laminar. Rotational speed: 50 000 rpm.
0 100 m/s 0 100 m/s 0 100 m/s

Rotor Rotor Rotor
Vr=0 Vr=0 Vr=0
Stator Stator Stator

0 0.33 0.67 1 0 0.33 0.67 1 0 0.33 0.67 1
Inner Outer Inner Outer Inner Outer
radius radius radius radius radius radius
(a) 0.1 MPa and 400 K (b) 0.7 MPa and 400 K (c) 1.4 MPa and 400 K
ρ = 1.33 kg m−3 ρ = 9.38 kg m−3 ρ = 19.01 kg m−3
Figure 5.19: Radial velocity profile in the ramp region (circumferential angle: 5°) for different operating
conditions, bold line: Vr = 0, Eilmer-laminar.
ory for predicting the bearing performance, where viscous forces are dominant. The increased oper-
ating temperature and resulting high dynamic viscosity results in larger viscous forces. Hence, both
the load capacity and friction torque increase as operating temperature increases. On the other hand,
for the simulations with the fixed operating temperature, the friction torque is almost constant for the
different operating pressure as shown in Figure 5.17. However as operating pressure increases the
load capacity of the foil thrust bearing increases also. Variation can be as high as 40 %.
To provide more insight to this increase in load capacity, the pressure increase relative to the
operating pressure at the medium radius for the different operating pressure is shown in Figure 5.18.
For these operating conditions, the ambient temperature is fixed at 400 K, implying constant viscous
forces under the same rotational speed and bearing geometry. Figure 5.18 shows that there is a higher
peak pressure in the ramp region for the high operating pressures. In the flat region, the pressure
then decreases, but there is a comparatively larger pressure drop for the highest operating pressure
(1.4 MPa) as shown in Figure 5.18. The pressure drop is approximately 13.9 kPa, roughly twice the
amount observed for the lowest operating pressure (0.1 MPa).
To explain this difference, Figures 5.19 and 5.20 show the radial velocity profile for three radial
slices (ramp and flat region). The arrowed contours in these figures indicate the local radial velo-
city component at the different position. For the current bearing geometry and rotational speed the
maximum radial velocity is approximately 30 m/s at the outer radius, while the maximum tangential
velocity varies between 42 m/s and 168 m/s. The bold line corresponds to the location of zero radial
velocity. On the left of the zero radial velocity line, CO2 flows towards the inner radius, while to right
CO2 flows towards the outer radius.
In the ramp region, as shown in Figure 5.19 close to the rotor the zero radial velocity line
0 100 m/s 0 100 m/s 0 100 m/s

Rotor Rotor Rotor
Vr=0 Vr=0 Vr=0
Stator Stator Stator

0 0.33 0.67 1 0 0.33 0.67 1 0 0.33 0.67 1
Inner Outer Inner Outer Inner Outer
radius radius radius radius radius radius
(a) 0.1 MPa and 400 K (b) 0.7 MPa and 400 K (c) 1.4 MPa and 400 K
ρ = 1.33 kg m−3 ρ = 9.38 kg m−3 ρ = 19.01 kg m−3
Figure 5.20: Radial velocity profile in the ramp region (circumferential angle: 25°) for the different operating
conditions, bold line: Vr = 0. Note, the inner radius velocity profile are so small that they are difficult to see,
Eilmer-laminar.
moves towards the inner edges as operating pressure increases. At a pressure of 1.4 MPa the region
closest to the rotor actually experiences flow with a radially outward direction entering the bearing.
The location of the zero radial velocity close to the stator remains almost fixed at approximately
0.3 rinner + 0.7 router . In the flat region, the same trend exists as shown in Figure 5.20. The zero radial
velocity line close to the rotor also moves towards the inner edge. So does the zero radial velocity
line close to the stator, but at a lower rate. To aid the understanding of the flow pattern, stream-
lines close to the rotor and stator (at 90 % and 10 % film thickness) for the two operating conditions
(0.1 MPa, 400 K and 1.4 MPa, 400 K) are plotted in Figures 5.21 and 5.22. Close to the rotor very
similar streamlines are shown, as tangential velocities are substantially larger than the radial velocit-
ies. However, a zoomed-in view at the inner radius, close to the rotor highlights the different flow
pattern. The streamlines are parallel to the tangential direction for the high density condition, while
the flow moves towards the inner radius at the low density condition. This shows that for the high
density case centrifugal effects have altered the flow pattern. The contours of radial velocity close
to the rotor, Figures 5.21(a) and 5.22 (d), further highlight the difference in flow pattern and how
increased outwards flow exists for the dense gas. For the streamlines at the stator, a backward flow is
shown in the ramp region, which is due to the pressure forces being higher than the shear stress. The
corresponding top foil deflection for these two operating pressures is illustrated in Figure 5.23. Both
operating conditions show the highest deflection in the bearing centre, while the minimum deflection
is indicated at the surrounding edges.
At the low operating pressure (0.1 MPa, 400 K, ρ = 1.33 kg m−3 ), fluid is drawn into the bearing
from the ramp start, and pressurised as the viscous forces drag the fluid up the ramp into a convergent
gap. The highest pressure is generated along a constant radius line (approximately at 0.3 rinner +
0.7 router ) and fluid flows radially inwards and radially outwards towards the bearing edges (fixed
pressure boundary conditions). The same process continues after the ramp, where fluid continues to
flow from this line towards the inner and outer edge. Effectively a fixed amount of CO2 is drawn into
the bearing and this depletes as CO2 leaks to the inner and outer edge as illustrated by the streamline
in Figure 5.21. This resembles a fluid flow that is dominated by the pressure and viscous forces.
The flow-field for the high operating pressure (1.4 MPa, 400 K, ρ = 19.01 kg m−3) is different.
Again CO2 is drawn into the bearing from the ramp start. However the fluid close to the rotor now
experiences a significant radial force due to centrifugal inertia effects. Hence while the flow-field
close to the stator is similar to the flow experienced with the low operating pressure, predominantly
outwards flow exists close to the rotor, see Figure 5.19(c). Only towards the end of the ramp, when
dP
the pressure inside the bearing is much higher than the surrounding pressure (large dr
) does the flow
direction at the bearing inner radius reverse. This illustrates that for bearings operating with highly
dense CO2 , inertia effects have a significant impact on the flow-field. In addition to changing the flow,
the inertial effect also creates a mechanism, which enhances pressure increase and lift generation. The
overall effects are illustrated by the streamlines shown in Figure 5.22. The strong inertia force acting
on the fluid close to the bearing inner edge balances the pressure gradient, which manifests a radial
velocity, Vr ≈ 0 in the top sixth of the film close to the inner edge as shown in Figure 5.19(c). This
means, contrary to the low density case, where fluid spills out of the inner edge over the entire film
height, now fluid is actually pumped into the bearing close to the rotor (approximately a sixth of
the film height). Consequently the amount of gas that is drawn into the bearing from the ramp start
depletes less quickly.
After the ramp, the high operating pressure flow continues to be different. Here the centrifugal
effects continue to influence the flow direction as shown in Figures 5.20 and 5.22. The bearing now
experiences increased leakage at the outer edge and most of the flow is drawn outwards. This is
particularly the case close to the rotor. Effectively for most of the flat region the centrifugal force
dominates the pressure gradient, leading to a mainly outwards flow. The result is a mass loss through
the bearing outer edge. The same effect has also been identified in Ref. [127, 169], which studied
inertia effects in film riding bearings. Consequently, rather than keeping the higher pressure created
in the ramp region, the flat region now experiences an increased pressure drop as shown in Figure 5.18.
The effects described above, and particularly the ability to create a high peak pressure at the end
p p
0.02 0.02
150000 150000
145000 145000
140000 140000
135000 135000
130000 130000
0.015 125000 0.015 125000
120000 120000
Y, m
Y, m
115000 115000
110000 110000
105000 105000
0.01 0.01
0.005 0.005
0 0
0.01 0.015 0.02 0.025 0.03 0.01 0.015 0.02 0.025 0.03
X, m X, m
(a) stator, pressure (in Pa) contour and streamline (b) rotor, pressure (in Pa) contour and streamline
12 12
20 20
9 9
Position y, mm
Position y, mm
15 15
6 6
10 10
3 3
5 0 5 0
10 15 20 25 30 −3 10 15 20 25 30 −3
Position x, mm Position x, mm
(c) stator, radial velocity (in m/s) (d) rotor, radial velocity (in m/s)
Figure 5.21: Streamlines and radial velocity close to stator and rotor. Operating condition: 0.1 MPa and 400 K,
ρ=1.33 kg m−3 . the rotational direction is anti-clockwise, Eilmer-laminar.
p p
0.02 0.02
1470000 1470000
1465000 1465000
1460000 1460000
1455000 1455000
1450000 1450000
0.015 1445000 0.015 1445000
1440000 1440000
Y, m
Y, m
1435000 1435000
1430000 1430000
1425000 1425000
0.01 1420000 0.01 1420000
1415000 1415000
1410000 1410000
1405000 1405000
0.005 0.005
0 0
0.01 0.015 0.02 0.025 0.03 0.01 0.015 0.02 0.025 0.03
X, m X, m
(a) stator, pressure (in Pa) contour and streamline (b) rotor, pressure (in Pa) contour and streamline
12 12
20 20
9 9
Position y, mm
Position y, mm
15 15
6 6
10 10
3 3
5 0 5 0
10 15 20 25 30 −3 10 15 20 25 30 −3
Figure 5.22: Streamlines and radial velocity close to stator and rotor. Operating condition: 1.4 MPa and 400 K,
ρ=19.01 kg m−3 , the rotational direction is anti-clockwise, Eilmer-laminar.
3. 4.500
20 00 20
0
Position y, mm
Position y, mm
15 15
6.0
00
10 10
3.000
6.00
0
4.5 7.500
5 4.500 5 00
1.5 3.000 1.5
00 00
0 10 15 0.000 20 25 30 0 10 15 0.000
20 25 30
(a) 0.1 MPa and 400 K (b) 1.4 MPa and 400 K
Figure 5.23: Local deflection (in µm) for two operating conditions, rotational direction: anti-clockwise, Eilmer-
laminar.
of the ramp is also counteracted by the motion of the foils. Effectively in the high pressure regions,
the top foil is deflected most, which results in a large gap and diminishing viscous effects. The
corresponding top foil shape for the two simulations is shown in Figure 5.23. The general shape is
similar, however the deflection for the high pressure operating point (higher density) are increased by
almost 50 %. The largest increases in deflection exist at the location of peak pressure (approximately
at 0.3 rinner + 0.7 router ) and close to the end of the ramp. While a convergent shape from the peak
pressure region towards the bearing edge is preferable to maximise pressures within the gap, the
ability to retain mass is governed by the final restriction. Consequently the increased gap height at
the outer radius of the bearing in combination with centrifugal forces that generate and increased
outwards flow leads to substantial mass loss from the bearing and simultaneous decrease in pressure
as observed in Figures 5.21 and 5.22. Therefore, the observed pressure drop at the high operating
pressure is a combined effect of the centrifugal force and the top-foil shape.
These results highlight the mechanism by which centrifugal inertia forces influence the perform-
ance of the foil thrust bearing. For high density, high pressure applications the centrifugal inertia
effect can cause additional fluid to be entrained into the bearing in the ramp region close to the inner
edge of the rotor. This can augment the peak pressure magnitude. However in the flat region the
inertial effect leads to increased fluid leakage through the bearing outer edge, which reduces pressure
and total load capacity. Contrary to rigid bearings, the deflected top foil can prevent the mass loss
through the radial direction. It is recommended that the use of stiffer bump foils close to outer radius
can further prevent mass loss.
5.6.3 Turbulence Effects

Due to the high density of CO2 , the flow regime transitions to turbulence at some operating conditions
as shown in Figure 5.16. The addition of eddy viscosity in the turbulent flow has a significant impact
on the performance of the foil thrust bearings. To investigate this influence the results from different
operating pressures in the range 0.1 MPa to 4.0 MPa are compared. The operating temperature and
the rotational speed are fixed at 400 K and 50 000 rpm and the initial minimum film thickness is set
to 5 µm. In the absence of a transition model in Eilmer, the simulations are divided into laminar and
turbulent cases. For operating pressure less than 1.64 MPa (Re=900), the flow is modelled as laminar,
while the turbulence model is used for operating pressures larger than 2.89 MPa (Re=1600).
200 200
150 150
Load capacity, N
Power loss, W
Load capacity Power loss
100 100
50 Transition 50
Laminar Turbulent
0 0
0.5 1 1.5 2 2.5 3 3.5 4 4.5
Operating pressure, MPa
Figure 5.24: Performance comparison for different flow regimes, the operating temperature is fixed as 400 K.
The results of power loss and load capacity at the different operating pressures are compared in
Figure 5.24. For the laminar case power loss is almost constant with at approximately 55 W for the
current bearing geometry and operating conditions. This is due to the linear velocity distribution
between rotor and stator in the laminar flow regime and the fact that viscosity is largely independent
of pressure. The small reduction with pressure can be attributed to increasing bump foil deflections
caused by the increasing load being supported. However, in the turbulent regime the power loss
increases to 190 W. As a consequence of turbulence the velocity profile in the boundary layer close to
4.
20 50
0
6.
00
0
00
Position y, mm
15
7.5
10.
9.000
500
4.500
10
5 6.000
1.5
00
3.000
0 10 15 0.000
20 25 30
Position x, mm
Figure 5.25: Local deflection (in µm) at an operating pressure of 1.4 MPa and operating temperature of 400 K.
rotor and stator becomes steeper. This results in increased wall shear stress and subsequently a larger
power loss as shown in Figure 5.24.
Load capacity shows a near linear relationship in terms of operating pressure as shown in Fig-
ure 5.24. Figure 5.27 shows the pressure and streamline, at the operating condition of 4.0 MPa and
400 K, with a Reynolds number 2260. The streamlines close to the rotor are very similar to an equi-
valent laminar case (e.g. Figure 5.22(b)). However, close to the stator substantial differences exist
compared to the laminar cases shown in Figure 5.21(a) and Figure 5.22(a). In the turbulent flow re-
gime, only the fluid closest to the wall within the laminar sub-layer is strongly affected by viscous
effects. This is confirmed by the steep velocity gradient as shown in Figure 5.26 compared to laminar
flows. The fluid further away, in the transition layer is much more susceptible to pressure gradients
and other effects. Consequently a larger portion of the flow is now affected by the pressure gradients,
resulting in the substantial reversed flow on the ramp and increased flow towards the foil edges. The
deflected shape shown in Figure 5.25 is still similar to the laminar simulation (see Figure 5.23), but
the maximum deflection has doubled to 12.0 µm. It is expected that the load capacity and rotor torque
can be improved through optimisation of the bump foil stiffness and top foil thickness.
5.7 Revisit of Computational Domain: Steady State Performance

In literature, typically only the ramp and flat regions of foil thrust bearings are considered as the
computational domain [8, 26, 76, 84, 85], and a fixed static pressure boundary condition is set at the
Revisit of Computational Domain: Steady State Performance Section 5.7 137
1
Laminar flow
Turbulent flow
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
Figure 5.26: Comparison of tangential velocity profile at laminar and turbulent flow at the medium radius with
the circumferential angle of 24.20°.
leading and trailing edges (see Figure 5.28(a)). Thus, the groove region between two adjacent bearing
pads is neglected, and the pressure in this region is commonly assumed as ambient. However, this
groove region or the trailing edge shape can play a significant effect on bearing performance. For
example, the chevron shaped trailing edge proposed by Bruckner [170] provides enhanced mixing
and a substantial passive cooling mechanism. This allows a twofold increase in bearing load capacity
and enhances the damping as confirmed by experimental results.
This section is devoted to the selection of the computational domain for rotordynamic analysis of
foil thrust bearings operating with CO2 . The selection is performed by comparing the steady state and
dynamic performances of foil thrust bearings. The bearing geometry is defined in Table 2.1. The data
of this prototype was released by the National Aeronautics and Space Administration (NASA) [23].
The ambient static pressure and temperature are 1.4 MPa and 300 K, respectively. Two different
computational domains are selected,
• Domain 1: ramp and flat regions as shown in Figure 5.28(a);
• Domain 2: ramp, flat and groove regions as shown in Figure 5.28(b).
For domain 1, the ambient pressure and temperature are set at the leading and trailing edges. For
domain 2, the leading edge is connected with the groove region by a cyclic boundary condition. The
top surface is regarded as the rotor, where a rotational speed is prescribed. The bottom is modelled
as a fixed temperature wall and connected to the structural deformation solver. The boundaries at
p p
0.02 0.02
4090000 4090000
4080000 4080000
4070000 4070000
4060000 4060000
4050000 4050000
0.015 4040000 0.015 4040000
4030000 4030000
Y, m
Y, m
4020000 4020000
4010000 4010000
4000000 4000000
0.01 0.01
0.005 0.005
0 0
0.01 0.015 0.02 0.025 0.03 0.01 0.015 0.02 0.025 0.03
Y, m X, m
(a) stator, pressure contour and streamline (b) rotor, pressure contour and streamline
12 12
20 10 20 10
8 8
Position y, mm
Position y, mm
15 15
6 6
10 4 10 4
2 2
5 5
0 0
10 15 20 25 30 −3 10 15 20 25 30 −3
Figure 5.27: Pressure and radial velocity close to stator and rotor. Operating condition: 4.0 MPa and 400 K,
the rotational direction in anti-clockwise, Eilmer-turbulent.
Steady State Performance Section 5.8 139
Trailing edge
Groove region
Trailing edge Flat region
Flat region Ramp region
Ramp region
Cyclic
Leading edge
y z
Leading edge x
(a) Domain 1. (b) Domain 2.
Figure 5.28: Schematic diagram of different computational domains for foil thrust bearings.
the inner and outer radii are modelled as the fixed static pressure and temperature. Both cases are
investigated for operation with a fixed rotational speed of 40 000 rpm and for three load cases (200 N,
300 N and 400 N). To rule out the effects of the deformed shape and structural parameters, e.g.
structural stiffness and damping, a rigid computational domain is used for the comparison.
The steady state gauge pressure contours at a load of 200 N, and gauge pressure along the medium
radius are depicted in Figure 5.29. Here, the gauge pressure is the static pressure with respect to the
ambient pressure (1.4 MPa in this paper).
As observed in Figure 5.29, the gauge pressure contours are similar for the two domains, but
a difference in magnitude is observed close to the leading and trailing edges. The computational
domain 2 now predicts a pressure above ambient at the leading edge (circumferential angle equals to
0°), and a sub-ambient pressure at the trailing edge (circumferential angle equals to 45°) as shown
in Figure 5.29(c). The pressure increases when the flow enters from the groove into the ramp region
and decreases when exiting from the trailing edge. This highlights the importance of simulating the
groove to ensure accurate pressure distributions.
5.8 Steady State Performance

Using the full computational domain, the steady state performance (load capacity and power loss)
is investigated. Results are obtained by evaluating performances of the foil thrust bearing operating
with CO2 at multiple rotational speeds (30 000 rpm, 40 000 rpm and 50 000 rpm) and at different rotor
to top foil separations (from 10 µm to 26 µm, intervals of 2 µm). A mesh with 48×72×15 cells is used
190 190
40 40
160 160
35 35
130 130
30 30
25 100 25 100
Y, m m
Y, m m
20
kPa 20
kPa
70 70
15 15
40 40
10 10
10 10
5 5
0 − 20 0 − 20
15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50
X, m m X, m m
(a) Domain 1. (b) Domain 2.

200
Domain 1
Domain 2
150
Gauge presssure, kPa
100
50
-50
0 10 20 30 40 50 60
Circumferential angle, deg
(c) Gauge pressure at the medium radius.
Figure 5.29: Comparison of pressure distribution at different computational domains, rotational speed is
40 000 rpm and load is 200 N, notable differences in pressure at start of ramp and end of flat regions.
Steady State Performance Section 5.8 141
for the fluid film and 48×24×105 cells is used for the groove. The meshes have been shown to give
grid independent.
The film pressures and deflection contours at different rotor to top foil separations (10 µm, 18 µm
and 26 µm) are shown in Figure 5.30, for a rotational speed of 30 000 rpm. The slices of gauge pres-
sure and local deflection at the medium radius are plotted in Figure 5.30(g) and (h). The maximum
deflection (approximately 70 µm) is in the ramp region close to the outer radius. The reason is that
there are no bump foils underneath the top foil within the ramp. Thus no structural support is provided
for the top foil. The deflection is between 0 to 10 µm in the flat region. The peak pressure is observed
at the end of the ramp region, which is attributed to the convergent shape. The pressure then reduces
in the flat region due to leakage through the sides and the centrifugal inertia effects as highlighted in
Refs. [171, 172]. However, pressure increases again close to the trailing edge (circumferential angle
equals to 45°) for the high load case (initial rotor to top foil separation of 10 µm), shown in Fig-
ure 5.30(g). Since the bump foils exhibit a hardening effect [173], especially at high load conditions,
the change of the local deflection is not proportional to the load. This is confirmed in Figure 5.30(h),
showing that the deflection at the trailing edge is almost the same for the cases with the initial rotor
to top foil separations of 18 µm and 26 µm, despite the different pressure (see Figure 5.30(g)). For the
high load, a second convergent shape is generated close to the trailing edge and provides a mechanism
to increase the pressure here.
The results of load capacity and power loss at different rotor to top foil separations and rotational
speeds are depicted in Figure 5.31. The maximum load of 516 N is obtained with a power loss
of 1648 W (rotational speed of 50 000 rpm). The magnitudes of load and power loss are consistent
with work by Conboy [8] using turbulent Reynolds equation to solve film pressure. As shown in
Figure 5.31(a), the load capacity of foil thrust bearings exhibits a roughly linear relationship in terms
of rotor to top foil separations. This is attributed to the increased compression of fluid film when the
rotor to top foil separation decreases.
The power loss shown in Figure 5.31(b) also increases linearly as a function of the rotor to top
foil separation, but at a slower rate compared to load capacity. The power loss is mainly attributed to
the turbulent Couette flow and is inversely proportional to the gas film thickness. It is noted that the
power loss is less sensitive to the rotor to top foil separation. This can be explained by the increase in
the top foil deflection, which increase as load increases as shown in Figure 5.30. Only a 20 % change
140 80
40 40 70
120
35 35 60
100
30 30 50
80
25 25
Y, m m
Y, m m
40
60 kPa m
20 20 30
40
15 15 20
10 20 10 10
5 0 5 0
0 − 20 0 − 10
15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50
X, m m X, m m
(a) Gauge pressure, 10 µm. (b) Deflection, 10 µm.

140 80
40 40 70
120
35 35 60
100
30 30 50
80
25 25
Y, m m
Y, m m
40
60 kPa m
20 20 30
40
15 15 20
10 20 10 10
5 0 5 0
0 − 20 0 − 10
15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50
X, m m X, m m
(c) Gauge pressure, 18 µm. (d) Deflection, 18 µm.

140 80
40 40 70
120
35 35 60
100
30 30 50
80
25 25
Y, m m
Y, m m
40
60 kPa μm
20 20 30
40
15 15 20
10 20 10 10
5 0 5 0
0 − 20 0 − 10
15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50
X, m m X, m m
(e) Gauge pressure, 26 µm. (f) Deflection, 26 µm.

200 80
Initial rotor to top foil separation, 10 µm Initial rotor to top foil separation, 10 µm
Initial rotor to top foil separation, 18 µm 70 Initial rotor to top foil separation, 18 µm
150 Initial rotor to top foil separation, 26 µm Initial rotor to top foil separation, 26 µm
Gauge pressure, kPa
60
100 50
40
50 30
20
0
10
-50 0
0 10 20 30 40 50 60 0 5 10 15 20 25 30 35 40 45
(g) Gauge pressure. (h) Deflection.
Figure 5.30: Gauge Pressure (unit in kPa) and deflection (unit in µm) contours at different rotor to top foil
separation, rotational speed of 30 000 rpm.
600 2000
500
1500
Power loss, W
400
Load, N
300 1000
200
500
100 30000 rpm 30000 rpm
40000 rpm 40000 rpm
50000 rpm 50000 rpm
0 0
5 10 15 20 25 30 35 5 10 15 20 25 30 35
Initial rotor to top foil separation, µm Initial rotor to top foil separation, µm
(a) Load. (b) Power loss.
Figure 5.31: Steady state performance of foil thrust bearings under different rotational speed and initial film
thickness.
is observed across the range of rotor to top foil separations at each speed. However, the power loss
increases by almost 100% when the rotational speed increases from 30 000 rpm to 40 000 rpm. This
confirms that rotational speed is the main driver behind power loss and heat generation, much more
than load.
5.9 Conclusions
The modified fluid-structure simulation tool is used to identify the challenges of operating foil thrust
bearings with high pressure CO2 . It is found that at SNL’s operating point for foil thrust bearings
(1.4 MPa and 300 K), there is the noticeable deviation of properties between the ideal gas model and
look-up table and a slightly non-ideal gas behaviour was also indicated. It is recommended that the
look-up table approach is used in the subsequent analysis. With dense CO2 as the operating fluid,
there is a different pressure distribution resulting from CFD and Reynolds equation analyses. The
Reynolds equation analysis cannot account for the irregular radial velocity profiles that are driven by
strong inertial effects.
The effect of operating conditions on the performance of foil thrust bearings were investigated. It
is found that a ten-fold increase of CO2 pressure and thus density can significantly alter foil bearings
performance. The density change can alter load capacity by up to 40 %. This is due to the additional
mass flow at the inner edge of the ramp region created by centrifugal inertia effects. But centrifugal
inertia effects also cause an increased leakage towards the outer edges in the flat region which is
detrimental to lift generation.

Due to the high densities foil thrust bearings can also operate in the turbulent flow regime. As a
consequence of the additional eddy viscosity, the foil thrust bearings now experience different velocity
profiles across the film height, which result in increased power consumption. In contrast, load capacity
only shows a slight increase for turbulent flows. Due to the compliance of the bump foils, which
deflect almost twice as much under turbulent conditions, the potential to increase lift is decreased.
Load capacity and power loss of the foil thrust bearings for operation with CO2 can be enhanced
through optimisation of bump foil and top foil stiffness in future.
For the selected rotor to top foil separations and rotational speeds, the maximum load can be up to
516 N, whilst a maximum power loss of 1648 W is also generated. Both load and power loss increase
linearly as the rotor to top foil separations decrease at different rotational speeds, but a slower rate is
obtained for power loss compared to load. The rotational speed has a larger effect on the power loss
compared to the rotor to top foil separations.
Chapter 6
Foil Bearing Dynamic Performance
6.1 Introduction
Dynamic performance is also a key parameter to consider selecting appropriate bearings. These
parameters influence the vibration of the shaft. Typically, bearings with high stiffness and damping
coefficients are desirable. These features are favourable to reduce the shaft vibration amplitude. In this
Chapter, the procedure to obtain stiffness and damping coefficients from fluid-structure simulations
is introduced in Section 6.2. Section 6.3 details the suitability of the method to calculate stiffness and
damping by comparing to a verification case. Section 6.4 revisits the computational domain including
the groove by highlighting dynamic performances.
6.2 Dynamic Performance Evaluation
Once the steady state solution for foil thrust bearings using fluid-structure simulation tool is obtained,
a sinusoidal rotor excitation is applied, as shown in Figure 6.1 to obtain the dynamic stiffness and
damping. The conventional rotordynamic model to quantify reaction forces, for small axial rotor
motions is presented in Equation 6.1 [174]. Here all coefficients are assumed frequency dependent
and added mass effects are neglected,
− fz (t) = K(Ω) ∆z(t) + C(Ω) ∆ż(t) . (6.1)
K is the stiffness coefficient, C is the damping coefficient, Ω is the excitation frequency, fz is the
reaction force and ∆z is the perturbed rotor position. The reaction forces fz acting on the rotor are
146 Chapter 6 Foil Bearing Dynamic Performance
Figure 6.1: Schematic diagram of rotor perturbation motion.
obtained by integrating the instantaneous pressure,

Z
fz (t) = (p(t) − pa ) dA , (6.2)
where p is the pressure acting on the rotor surface, while pa is the ambient pressure, and A is the rotor
area. The reaction forces have the same frequency as the prescribed rotor movement but are shifted
in phase. The rotor position is perturbed using a uni-directional harmonic function,
∆z(t) = δ sin(Ω t) , (6.3)
where the amplitude δ is defined as a fraction of the clearance between rotor and bearing and the
excitation frequency Ω is chosen as a fraction of the rotational speed of the rotor.
The rotordynamic coefficients can be determined by analysing the reaction forces due to the pre-
scribed rotor motion. To solve the frequency-dependent rotordynamic coefficients in Equation. 6.1,
Dynamic Performance Evaluation Section 6.2 147
the Laplace transform is performed over the interval [0, T ], written as,
F (s) = D(s) Z(s) . (6.4)
The time-dependent component are,

Z T
F (s) = f (t) est+φ dt , (6.5)
0
D(s) = K (Ω) + s C(Ω) , (6.6)
Z T
Z(s) = ∆z(t) est dt , (6.7)
0
√
where s = i Ω, i = −1 and φ is the phase lag between the rotor motion and the reaction forces. The
phase lag φ between the rotor position and the reaction force is obtained by measuring the peak-to-
peak time delay ∆T between the harmonic motion and force data from the unsteady CFD simulations.
This is turned into a phase lag using φ = 2 π Ω ∆T . As all the component are complex variables, F (s)
and Z(s) can be written as,
F (s) = Fr (Ω) + i Fi (Ω) , (6.8)
Z(s) = Zr (Ω) + i Zi (Ω) . (6.9)
The real and imaginary parts of reaction force and rotor motion are computed as [175],
n
1X
Zr (Ω) = ∆t[∆z(ti ) cos(Ω ti ) + ∆z(ti−1 ) cos(Ω ti−1 )] , (6.10)
2 i=1
n
1X
Zi (Ω) = ∆t[∆z(ti ) sin(Ω ti ) + ∆z(ti−1 ) sin(Ω ti−1 )] , (6.11)
2 i=1
n
1X
Fr (Ω) = ∆t[f (ti ) cos(Ω ti ) + f (ti−1 ) cos(Ω ti−1 )] , (6.12)
2 i=1
n
1X
Fi (Ω) = ∆t[f (ti ) sin(Ω ti ) + f (ti−1 ) sin(Ω ti−1 )] . (6.13)
2 i=1
100 2
Reaction force
Rotor motion
95
1
Reaction Force, N
Rotor motion, µm
90
0
85
Phase lag
-1
80
Cycle 1 Cycle 2 Cycle 3
75 -2
0 1 2 3 4
Iteration loop
Figure 6.2: History plot of reaction force and rotor motion.
By substituting the above equations into Equation 6.4, the following equations are obtained,
Fr = K(Ω) Zr − Ω C(Ω) Zi , (6.14)
Fi = K(Ω) Zi + Ω C(Ω) Zr . (6.15)
Hence, it is now straightforward to calculate the stiffness and damping coefficients as,
Fr Zr + Fi Zi
K(Ω) = − , (6.16)
Zr2 + Zi2
Fr Zi − Fi Zr
C(Ω) = . (6.17)
Ω (Zr2 + Zi2 )
6.3 Verification of Dynamic Analysis Capability

The structural deformation solver is executed with a comparatively large time step ∆ts during fluid-
structure simulations to obtain the steady state bearing performance. To accurately predict the dy-
namic performance of foil thrust bearings, ∆ts has to be selected appropriately. To investigate this,
the rotational speed of the foil thrust bearing is set to 21 000 rpm, matching the operating conditions
from the experiment presented by Dickman [73]. The rotordynamic performance of this foil thrust
bearing was investigated by San Andrés [26], and his results are used for verification.
The optimal mesh (48×96×15) for foil thrust bearings, as attained in Chapter 5 is used. In order to
Revisit of Computational Domain: Dynamic Performance Section 6.4 149
9 1800
Stiffness, MN/m
Damping, Ns/m
Stiffness coefficient
8.5 Damping coefficient 1600
8 1400
7.5 1200
7 1000
6.5 800
0 1 2 3 4
Cycle
Figure 6.3: Comparison of the calculated stiffness and damping coefficient at different cycles.
establish time-step independence, first the number of oscillations required to reach a steady response
has to be determined. The time history of the reaction force and rotor motion for a time step ∆ts =2 µs
for the structural deformation solver is depicted in Figure 6.2. The phase lag is approximately 10°
(80 µs). The rotordynamic coefficients are compared in Figure 6.3. It is found that the results converge
after two periods. This matches the observations from Ref. [175].
Three different time steps, ∆ts (2 µs, 1 µs and 0.5 µs) are used for a sensitivity study to select a
suitable time step for fluid-structure simulations. The results are shown in Figure 6.4(a) and are all
calculated during the second harmonic motion of the rotor. Insensitivity to the size of the time step
is observed for all cases and 1 µs is selected for the subsequent analysis. The effect of perturbation
amplitude (5 %, 10 % and 15 % of the clearance) is compared in Figure 6.4(b). This indicates the in-
sensitivity of results to excitation amplitude. A 10 % perturbation amplitude is used for the following
studies.
San Andrés et al. [26] predicted the synchronous rotordynamic performance of air foil thrust
bearings at a rotational speed of 21 000 rpm. His results are based on small perturbations of Reynolds
equation. These results can be used as a the verification case. The results obtained using the current
code are compared to results for San Andrés [26] in Figure 6.5. Good agreement is achieved for
damping coefficients at the various load conditions. A slight difference is found for stiffness coeffi-
cients. This can be attributed to the use of different top & bump foil models. However, the trend in
terms of load capacity is the same.
6.4 Revisit of Computational Domain: Dynamic Performance

This section is a follow-up of Section 5.7. The synchronous stiffness and damping coefficients at
different loads are shown in Figure 6.6. The computational domain 1 overpredicts rotordynamic coef-
Stiffness, MN/m
Damping, Ns/m
Stiffness coefficient 1360
7.6
1200
7.4
1120
7.2
1040
7
0 5e-07 1e-06 1.5e-06 2e-06 2.5e-06
Time step, s
(a) Time step for structural deformation solver.
8
Stiffness, MN/m
Damping, Ns/m
Stiffness coefficient 1360
7.6
2
7.4
2
7.2
7
0 0.05 0. . .2
Amplitude/Clearance
(b) Perturbation amplitude.
Figure 6.4: Comparison of the calculated stiffness and damping coefficient at different parameters.
20
Sti ness, MN/m
San Andres
15 CFD
10
5
0
40 60 80 100 120 140 160 180 200
Load, N
(a) Stiffness coefficient
2500
Damping, Ns/m
San Andres
2000 CFD
1500
1000
500
0
40 60 80 100 120 140 160 180 200
Load, N
(b) Damping coefficient
Figure 6.5: Comparison of CFD and numerical results from San Andrés [26], rotational speed: 21 000 rpm.
60 2500
Domain 1 Domain 1
Domain 2 Domain 2
Equivalent Stiffness, MN/m
Equivalent Damping, Ns/m

50 2000
40 1500
30 1000
20 500
10 0
150 200 250 300 350 400 450 150 200 250 300 350 400 450
Load, N Load, N
(a) (b)
Figure 6.6: Comparison of foil thrust bearings performance at different computational domain, (a): synchron-
ous stiffness; (b): synchronous damping.
ficients. At a load of 200 N, the synchronous stiffness decreases by approximately 20 % compared

to that of computational domain 1, while a reduction of 39 % is observed for synchronous damping.
Moreover, these differences increase with increasing load. The reason is that the fixed static pressure
condition at the leading and trailing edges cannot provide accurate transient performance, as pressure
is truncated at these boundaries. Therefore, pressure waves do not propagate to the next thrust pad
correctly. The groove may act as a destabilising force to the rotor as revealed by results from compu-
tational domain 2. Since a notable difference is found between these two computational domains in
terms of the steady state and dynamic performances, domain 2 shown in Figure 5.28, is utilised for
further analysis.
6.5 Conclusions
In this Chapter, the dynamic performances of foil thrust bearings are investigated. The computa-
tional domain including the groove should be utilised, since rotordynamic coefficients are different
compared to those obtained from a computational domain consisting of only ramp and flat regions.
Dynamic performances of foil thrust bearings operating with CO2 in terms of load and rotational
speeds will be investigated in future.
Chapter 7
Fluid-Structure-Thermal Simulations
7.1 Introduction
The necessity to predict of thermal behaviour of foil bearings has been discussed in Chapter 2. This
chapter presents a computational framework for a three-dimensional fluid-structure-thermal simula-
tion of foil thrust bearings. First, the basic computational framework for fluid-structure-thermal sim-
ulations is provided including a description of the fluid, structural deformation, and heat conduction
solvers in Section 7.2. Next, the coupling strategies between these solvers are introduced, followed
by two test cases. Section 7.3 details the heat transfer model applied to foil thrust bearings. Finally,
the fluid-structure-thermal simulation results for air and CO2 foil thrust bearings are presented and
discussed in Section 7.4.
7.2 Computational Framework

The computational model includes three aspects: fluid, structural deformation, and heat conduction
solvers. The corresponding fluid and structural deformation solvers are discussed in Chapter 4. Only
the heat conduction solver is explained in this Chapter.
7.2.1 Heat Conduction Solver
As the fluid solver Eilmer currently is not capable of three-dimensional heat conduction analysis, the
solver LaplacianFoam from the open source CFD toolbox FOAM-Extend-3.0 [139] is used to model
heat conduction through the solid parts of the foil bearing. The governing equation for conduction in
LaplacianFoam is,
∂T
− ∇2 (DT · T ) = 0 . (7.1)
∂t
154 Chapter 7 Fluid-Structure-Thermal Simulations
Since LaplacianFoam is already a mature solver for solving the transient Laplace’s equation, veri-
fication or validation cases are not provided here.
7.2.2 Coupling Method: Fluid-Thermal

For fluid-thermal interactions, only the steady-state performance of the foil thrust bearings is of in-
terest due to the large time constant τheat for the heat conduction problem. Effectively the time it
takes for foil bearing components to adjust temperature is much longer than the time it takes for the
foils to change shape or for fluid properties to adjust. Thus the temperature will affect steady state
performance and not be directly linked to dynamic performance. In the final, steady state, heat flux
between domains (fluid and solid) and temperature at the fluid-structure interface has to agree. The
condition at the interface, i can be mathematically expressed as,
qi,f = qi,s , (7.2)
Ti,f = Ti,s . (7.3)
where q is the heat flux, T is the temperature and subscripts f and s indicate the interface of fluid and
solid domain, respectively. Considering only conduction at the interfaces, the flux condition can be
expanded as,
∂T ∂T
qi,f = −λi,f · ni = qi,s = −λi,s · ni (7.4)
∂n i,f ∂n i,s

where λ is thermal conductivity. The spatial derivative ∂T for the fluid domain is approximated
∂n i,f
using a one-sided difference between the wall-adjacent cell centre temperature and the interface tem-
perature. For turbulent simulations with wall functions, the heat flux qi,f is extracted from the wall

function. The temperature gradient at the solid domain ∂T , is defined as
∂n i,s

∂T qi,f
= , (7.5)
∂n i,s λi,s

Since only the steady state of the foil thrust bearings is of interest, a weak coupling strategy is
proposed for the fluid-thermal simulations. Relaxation at each iteration is necessary to increase the
stability of the numerical scheme. A simple relaxation strategy is given by,
Tk+1 k k
i,f = Ti,f + β fi,s , (7.6)
Computational Framework Section 7.2 155
where β is the under-relaxation factor and k is the index of the iteration loop. During the iteration
loop k, a nonlinear operation is applied to an input vector Tki,f generating an output vector Tki,s of the
k
same size. The residual vector is defined as fi,s = Tki,s − Tki,f . Typically β is set to a constant between
0 and 1. This relaxation factor is consistent for each cell but can be different for each fluid-thermal
iteration loop.
Alternatively, an optimal value of the relaxation factor can be selected for each iteration. Different
dynamic relaxation techniques are reported in literature. In this study, we use the variant of vector
Aitken △2 method formulated in Ref. [176], because it is easy to be implemented and does not
require too much memory during computations. This algorithm starts with a known sequence of two
input/output pairs of vectors and calculates an optimum relaxation parameter as,
T
∆f k f k
βk = βk−1 + βk−1 , (7.7)
k f k k22
where the residual difference vector is ∆f k = f k−1 − f k . The first iteration is carried out using a
pre-selected relaxation constant β0 . This method is straightforward to implement, needs minimum
processor and storage resources, and only takes a single residual evaluation per iteration. The entire
procedure is as follows
1. The fluid domain is solved by Eilmer with an imposed initial temperature Tki,f until an equilib-
rium fluid state is achieved;
2. Heat flux q at the fluid-solid interfaces in Eilmer is evaluated. The temperature gradient at the
solid side is calculated based on the continuity of heat flux across the fluid-structure boundary
(Equation 7.4 and 7.5);
3. The temperature gradient is taken as the boundary condition for the heat conduction analysis.
The simulation is conducted with LaplacianFoam until the temperature solution in the solid is
converged;
4. The surface temperature Tki,s at the interfaces from LaplacianFoam is extracted. The new Tki,f
is set as the boundary condition for Eilmer. In the first iteration loop, a pre-defined under-
relaxation factor is used in Equation 7.6, thereafter β is calculated with Equation 7.7;
5. Repeat steps 1-4 using the new Tki,f until convergence.

Moving wall with fixed temperature, T0
Fluid
y
Solid
x Stationary wall with fixed temperature, T1
Figure 7.1: Schematic diagram for conjugate Couette flow.
0.8
Analytical solution
Height, m
0.6 Eilmer
K=10
0.4
K=0.1
0.2
0
0 0.2 0.4 0.6 0.8 1
Non-dimensional temperature
Figure 7.2: Comparison of temperature distribution between fluid-thermal simulation and analytical solution.
For air foil thrust bearings, it typically takes 60 iteration loops to obtain a converged solution. 30 loops
are usually sufficient for CO2 foil thrust bearings. Two validation cases for the proposed fluid-thermal
coupling method are discussed next.
7.2.3 Conjugate Couette Flow
The first validation case is the problem of conjugate Couette flow in a parallel plate channel, a typical
shear driven flow for conjugate heat transfer analysis. The schematic diagram is depicted in Figure 7.1
with a fluid between a hot upper wall with temperature T0 that moves at a constant velocity and a
stationary conducting solid at the bottom. The side of the conducting solid away from the fluid is
maintained at a constant temperature T1 . The solid part has a height of 0.25 m, while the height of the
fluid domain is 0.5 m.
The computed solutions are compared with the analytical solutions [177] in Figure 7.2. This
shows excellent agreement with the analytical solutions for two conducting ratios, K = λs /λf . The
relative error between numerical results and analytical solution is less than 0.04 %.
7.2.4 Conjugate Nozzle Flow
This test case is used to validate turbulent conjugate heat transfer. The supersonic flow inside a cooled
axisymmetric convergent divergent nozzle is investigated. The analysis is based on the experimental
data reported by Back et al. [178]. The test nozzle is axisymmetric, has a throat diameter of 0.0458 m,
and contraction area ratio of 7.75:1, an expansion area ratio of 2.68:1, a convergent half-angle of
30 deg, and a divergent half-angle of 15° deg as shown in Figure 7.3. The detailed geometry and
operating condition are available in Ref. [179, 180].
The temperature distribution for the outer wall of the nozzle from the experiment is shown in
Figure 7.4(a). This is used as the nonuniform temperature boundary for the outer wall in the numerical
simulation. The uncertainty on the temperature measurements is approximately 2 % [178]. The inflow
condition for this axisymmetric nozzle is, T0 =843.3 K and p0 =517.1 kPa.
This case of conjugate nozzle flow has previously been simulated by Marineau et al. [180] and Liu
et al. [181]. The wall material was not specified by Back et al. [178], however the thermal conduct-
ivity of the material, λs , can be determined from the temperature gradient and the heat flux provided
by Back et al. [178]. Marineau et al. [180] concluded that the thermal conductivity of the nozzle
wall material is approximately 27 W/mK. The conjugate nozzle heat transfer problem is solved by
imposing the temperature profile shown Figure 7.4(a) along the outside wall. The temperatures at the
solid sides that correspond to the nozzle inlet and outlet are specified as 299 and 283 K, respectively,
as suggested by the experimental data. The problem was simulated using the fluid thermal coupling
strategy discussed in Section 7.2.2. The grid was refined until a grid independent solution was ob-
tained. Turbulence was modelled using the k-ω model without wall functions. The schematic diagram
is shown in Figure 7.3.
Figure 7.4(b) shows the comparison between predicted inner wall temperatures and experimental
data. The predicted wall temperatures are close to the experimental values. Due to unknown paramet-
ers from the experiment, there are some deviations between numerical results and experiment data,
especially inaccuracies close to the nozzle inlet and exit. This is also reported in Ref. [180, 181] and
most likely due to the solid thermal boundary condition applied to the axial faces. The suitability of
the proposed coupling strategy for fluid-thermal simulation has been tested. This coupling strategy is
further refined to include couple structural deformation solver in Section 7.2.5.
T
400 500 600 700 800
305 842
Wall, fluid-structure interface
Inflow
Exrapolate outlet
Symmetric
Figure 7.3: Boundary conditions for conjugate nozzle flow, coloured by temperature.
330
Temperature, K
320
310
300
290
280
-0.4 -0.3 -0.2 -0.1 0 0.1 0.2
X, m
(a)
600
Temperature, K
Fluid-Thermal Simulation
500 Experiment
400
300
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
X, m
(b)
Figure 7.4: (a) Temperature distribution along outer wall of nozzle [178], used as the nonuniform temperature
boundary condition for numerical simulation, (b) comparison of temperature distribution along inner wall of
nozzle between numerical simulation and experiment [178].
7.2.5 Coupling Method: Fluid-Structure-Thermal
The three individual solvers for the current problem (transient fluid flow, structural deformation and
heat conduction) within foil thrust bearings have been described as well as the individual coupling
methods for fluid-structure and fluid-thermal simulations. However, these solvers have to be coupled
for a full fluid-structure-thermal simulation. The magnitude of the structural deflection is typically in
microns, thus the effect of the deflection is minimal when solving the heat diffusion problem in the
solid. Therefore, the effect of the deformation is not considered when solving the thermal equations.
However the effect of temperature is considered by the structural solver.
e
Structure
Sturcture
St
u ctu e
mation
T e
Figure 7.5: Schematic diagram for fluid-structure-thermal simulations.
The schematic diagram showing information exchange between the solvers for the fluid-structure-
thermal simulations is shown in Figure 7.5. A tight coupling is implemented between the fluid and
structure solvers, as the structural deformation reacts to the pressure increase in the fluid domain
quickly and to allow rotordynamic analysis. Transient simulations of the coupled fluid-structure sys-
tem, starting from stationary fluid case, have shown that the dynamically coupled fluid structure sys-
tem can reach a steady operating point in less than 1 ms. However, as noted prior, the time constants
in the heat conduction solver system are large, usually of the order of minutes. Hence, the temperat-
ure field takes a long time to achieve an equilibrium state. As shown in Ref. [112], it usually takes
3000 s for a three-pad journal bearings to reach the thermal equilibrium state. Hence, a weak coupling
between the fluid and heat conduction solvers is selected.
Th m n
t r T
n m )
Sh t
Figure 7.6: Schematic diagram for layout of foil thrust bearings.
7.3 Application to Foil Thrust Bearings

In this section, the detailed thermal analysis of a foil thrust bearing is described. The layout of a foil
thrust bearing in a typical turbomachinery system is shown in Figure 7.6. It consists of shaft, rotor,
stator (bump and top foils) and bearing housing. For the thermal analysis, the simulation domain is
extended to include the rotor, stator and housing to allow a fluid-structure-thermal simulation of the
foil thrust bearings while taking account of the surroundings. This provides an accurate temperature
distribution within the foil thrust bearing components.
7.3.1 Heat Transfer Model for Rotor
The computational domain for the rotor is depicted in Figure 7.7. It uses a computational mesh of
64×120×40 cells. Although foil thrust bearings consist of several sectors (usually 6), only one sector
is simulated with a periodic boundary condition to reduce computational cost. The computational
domain for the rotor is different from that of the thin film. The inner radius is extended to the shaft as
shown in Figure 7.6 and the outer radius is the same as the fluid domain.
For boundary conditions, the north and south boundaries (shaded surfaces) are connected with
periodic boundary conditions. The bottom boundary (z = zmin ) is modelled as the fluid-solid inter-
face, and the coupling method for fluid-thermal simulation is used here. As the rotor is spinning, a
mixing-plane or tangentially strip averaged boundary conditions is applied when mapping the heat
Application to Foil Thrust Bearings Section 7.3 161
y
x
Figure 7.7: Computational domain for the rotor.
flux from fluid domain to the rotor. In this way, heat flux at the fluid domain is averaged in the
tangential direction when passed to the solid interface. Since the west boundary (the inner radius) is
connected with the shaft, a fixed operating temperature is imposed by considering the large heat trans-
fer area between the shaft and the operating fluid. At the east (the outer radius) and top (z = zmax )
surfaces, convective heat transfer is assumed. For this purpose the rotor is modelled as a simple ro-
tating disk exposed to an infinite quiescent medium. The following empirical correlations for heat
transfer on rotating discs [182] are used.
Nu = 0.33 Re0.5 P r 1/3 Re < 1.95 × 105 ,
Nu = 10 × 10−20 Re4 P r 1/3 1.95 × 105 < Re < 2.5 × 104 , (7.8)
Nu = 0.0188 Re0.8 P r 1/3 Re > 2.5 × 105 .
7.3.2 Heat Transfer Model for Stator
The computational domain of the stator is shown in Figure 7.8. The stator geometry is much more
complex than the rotor, and consists of top foil, bump foils and housing.
The upper surface of the top foil is the fluid-thermal interface, and the coupling method for fluid-
thermal simulation is implemented at this boundary. Since the stator is a non-rotating part, and as
there is a slight pressure difference in the radial direction, a natural convective boundary condition is
applied at the inner and outer radius, within the bump foil channels and on the backside of the top
foil. This approach was also employed by Lee and Kim [116]. In typical foil bearing applications
Ro
ai
Top Foil o
a
Dir
ect
i on
Bump Foil
Housing
z Attachment Edge
y
x
(a)
Ro
t at
ion
al
Dir
ect
i on
Bump Foil
Housing
z
y
x
(b)
Figure 7.8: Computational domain for the stator, (a): Stator in three dimensional, (b): Stator with top foil
removed in three dimensional.
Application to Foil Thrust Bearings Section 7.3 163
Adiabatic Wall Contact Resistance

op Foil
Adiabatic Wall Housing Bump Foil
Figure 7.9: Schematic diagram of heat transfer from top foil to thrust plate.
forced convective cooling, achieved by applying a pressure gradient between the bearing inner and
outer edge, is employed to cool the rear of the top foil and bump foil structure by enhanced heat
transfer. The pressure gradient is set to suit the application (rotational speed and load). To remove
the additional dependency on pressure gradient, the current study employs natural convection within
the bump channel, rear of top foils, and inner and outer surfaces of the housing. The heat transfer
coefficient is calculated by the method described in [183],
0.518(F r P r)1/4
Nu = 0.36 + Gr P r < 109 , (7.9)
(1 + (0.559/P r)9/16)4/9
where Gr is the Grashof number, P r is the Prandtl number, and Nu is the Nusselt number. However,
the presented modelling framework has the flexibility to incorporate a forced convection model, as
required.
In addition, the back surface of the top foil and the stator housing experience thermal and mechan-
ical contact with the bump foil. Figure 7.9 illustrates the thermal resistances at the bump contacts with
the top foil and bearing housing. The bump foils link the top foils, which are heated by the viscous
shear in the thin film, to the bearing housing via these thermal contact resistances. Thermal contacts
are complicated in nature due to the thermal constriction and spreading of heat flux lines as well as
the random distribution and the unknown boundary condition of micro-contacts [109]. For nominally
flat and rough surfaces, the thermal contact resistance (m2 K/W) obtained from experimental results
Adiabatic wall
∆θ
z
Δy
Heat Flux
Figure 7.10: Schematic diagram of the additional geometry to account for the contact resistance.
op Foil Bump Compression, σB
RB RB-σB
Bump Foil
rvature centre
Figure 7.11: Schematic diagram of the deflection of the bump foil.
in [113, 116] is expressed as,
if Pbump <= 0.9

4 3 2 4
Rcont = (7.115Pbump − 16.159Pbump + 13.08Pbump − 4.503Pbump + 1.207) × 10−3 ,
if Pbump > 0.9
Rcont = 0.633 × 10−3 , (7.10)
where Pbump is the local gauge pressure expressed in bar. In the present study, these contact resistances
are modelled as additional structures within the foil thrust bearings as shown in Figure 7.10 that are
inserted into computational model. For these additions, boundaries that are not connected with the
top foil or bump foil or housing are modelled as adiabatic walls as shown in Figure 7.9. The height
of the additional structure △y is calculated as,
∆y = λs Rcont . (7.11)
Comparison Between Air and CO2 Foil Thrust Bearings Section 7.4 165
Figure 7.12: Schematic diagram of computational mesh of stator.
The bump contact area (Acont ) between the bump and top foil varies depending on the applied
pressure and is accompanied by beam-like deformation of the bump foils, as illustrated in Figure 7.11.
For simplicity, the bump contact area is approximated using the model shown in Figure 7.11. The
bump foil deflection σB is obtained from the fluid-structure simulation. The trigonometric relation
shown in Figure 7.11 determines the contact distance using the bump arc radius of curvature (RB )
and the calculated bump foil deflection, resulting in a contact width, given by,
q
2 2
∆θ = 2 RB − (RB − σB )2 . (7.12)
The resulting computational domain of stator incorporating the contact resistances is shown in Fig-
ure 7.12.
7.4 Comparison Between Air and CO2 Foil Thrust Bearings

To assess the relative performance and effects of heat generation of the bearing operating with air and
CO2 , the bearing is simulated with an axial load of 110 N applied to the rotor. The remaining operating
conditions are summarised in Table 7.1. The fluid domain is replotted and shown in Figure 7.13. The
computational mesh of 48×96×15 was shown to be grid independent in Chapter 5. It is recommended
by Bruckner [170, 184] that the groove between adjacent thrust pads is used for passive thermal
Cyclic Ro
tat
ion
al
Dir
ect
ion
Groove
z
y Cyclic
Figure 7.13: Computational domain of a foil thrust bearing.
control, hence this part is also modelled. The leading and trailing edges of a thrust pad are connected
with cyclic boundary conditions. The stator geometry shown in Figure 7.8 is meshed with an in-
house tool and converted into an OpenFOAM format mesh [185]. The studied bearing geometry is in
accordance with the bearing presented by NASA [23], summarised in Table 7.1. Not all data for this
geometry is released by NASA. The parameters labelled with ”a” are estimated based on the authors’
experience.
At these operating conditions, the power loss is 47.94 W and 146.51 W for air and CO2 respect-
ively. The higher loss for CO2 is attributed to this bearing operating in a turbulent flow regime as
confirmed by Reynolds number is Table 7.1. The temperature distributions within the fluid, close to
the fluid-structure interface are depicted in Figures 7.14 and 7.15. The rotor temperatures are smeared
circumferentially due to the rotor motion. At the same time there is an increase in temperature in the
radial direction, due to increase in relative velocity and outward convection of the fluid. Contrary to
this on the stator, a non-uniform temperature develops, exhibiting cold regions immediately above the
bump foil contacts. This is due to better heat transfer from the rear of the top foil to the bump foils,
compared to natural convection that is applied to the remainder of the top foil rear surface.
Experimental data for thrust bearing foil temperature distributions are limited. Furthermore,
the thermal boundary condition and structural parameters of the rotor and stator are generally not
provided. These parameters are essential for the thermal analysis and to thermally anchored simu-
lations [115, 116]. Also, data in literature relates to journal style bearings (for example [79]), these
Table 7.1: Geometry and parameters for the foil thrust bearing.
Geometry Value Operating Condition Value

Number of pads 6 Working fluid air and CO2
Outer diameter 101.6 mm air 0.1 MPa and 300 K
Inner diameter 50.8 mm CO2 1.4 MPa and 300 K
Shaft diametera 46 mm Rotor-to-stator sep. 10.12 µm (air) and 17.82 µm (CO2 )
Pad arc extend 45 deg Reynolds number 73 (air) and 3533 (CO2 )
Pad taper extend 15 deg Load 110 N
Groove extend 60 deg Load per unit area 18.1 kPa
Top foil thickness 0.15 mm Power loss 47.94 W (air) and 146.51 W (CO2 )
Rotor thicknessa 16 mm
Stator thicknessa 20 mm
Groove deptha 0.6 mm
Rotational speed 21 000 rpm
Pad area 6080.5 mm2
have only limited value in regards to thrust bearings, which have substantially different rotor and
stator shapes and boundary conditions. Therefore, verification and analysis relies on the investigation
of energy conservation and flows.
Heat is generated within the thin gas film due to a combination of viscous effects and compression
work on the fluid. This heat, which equals the power loss, can be removed by the following processes:
1. Advection by the gas being pumped through the film in the radial and tangential directions, this
includes the exchange and replacement of the fluid in the grooves between discrete pads;
2. Convection into the rotor. Heat conducts through the rotor, either to the shaft (a fixed temper-
ature boundary) or to the rear and outside surface, where forced convection takes place to an
infinite fluid reservoir at 300 K;
3. Convection into the top foil. From here heat is removed either by natural convection within
the bump foil structure (from top foil, bump foils and bearing housing) or conduction into the
bearing housing through the bump foils. Natural convection boundary conditions are set using
a bulk fluid temperature of 300 K.
The net heat flowcharts for air and CO2 are shown in Figures. 7.14(c) and 7.15(c). In both cases
the majority of the heat is convected to the rotor (90 % and 67 % for air and CO2 ). This is due to good
T T
322 322
320 320
310 310
300 300
(a) (b)
Air foil thrust bearing
43.17 W Heat conduction

90.05% into shaft 25.26 W 58.51%
Rotor
Heat convection into

outside cooling fluid
17.91 W 41.49%
Conduction into bearing

housing 2.97 W 88.66%
47.94 W 3.35 W
Convection to the channel
100% 6.99% 0.29 W 8.66%
through top foil
Drag dissipation
Stator
power loss
through bump foil 0.08 W 2.39%

through bearing housing 0.01 W 0.30%
1.42 W
2.96%
Advection of heat
by gas film flow
(c)
Figure 7.14: Air foil thrust bearings interface temperatures, (a): rotor; (b): stator; (c): heat flow chart for
complete bearing.
T T
318 318
316 316
312 312
308 308
304 304
300 300
(a) (b)
* foil thrust bearing
Heat conduction
into shaft
Rotor
Heat convection into
outside cooling fluid
Conduction into bearing

housing

through top foil
Drag dissipation
Stator
power loss
through bump foil

through bearing housing
Advection of heat
by gas film flow
(c)
Figure 7.15: CO2 foil thrust bearings interface temperatures, (a): rotor; (b): stator; (c): heat flow chart for
complete bearing.
thermal conductivity of the rotor, which is able to maintain a surface temperature substantially below
the stator. Once heat enters the rotor, the split between conduction to shaft and convection on the
rear surface is substantially different (see relative heat fluxes in Figures 7.14 and 7.15). In the case
of CO2 , the turbulent flow regime, provides a much higher heat transfer coefficient (2314 W/m2 K)
on the rotor outside surface, approximately 10 times bigger than the corresponding coefficient for
air (238 W/m2 K). This high heat transfer coefficient, and corresponding lower surface temperature
highlight the ability of CO2 to provide effective coupling if the rotor operates in a CO2 environment.
In contrast, the top foil surface reaches a significantly higher temperature and only a much smaller
portion of the heat is convected in this direction. This is caused by the much higher thermal resistance
of the bump foil structure. This is due to the contact resistance and also small conduction areas within
the bump foils. Considering the split of heat flux downstream of the top foil, it can be seen that this is
similar for air and CO2 . The majority of heat ( 88 %) is conducted into the bearing housing structure.
The remainder is convected to the fluid present within the bump channels. Most convection takes
place from the top foil rear surface ( 9 %), followed by the bump foils ( 2.5 %), and a bit from the
bearing housing surface ( 0.3 %). However, the total amount of removed energy is low because heat
transfer relies on natural convection. These power splits within the bump channel are consistent with
the results from Ref. [116].
A further aspect to consider is advection of heat by the gas in the film and grooves which is driven
by the radial pumping of the bearing within the film and fluid exchange in the grooves as shown in
Figure 7.16. The pumping mass flow rates are 0.004 g/s and 0.155 g/s for air and CO2 , respectively.
The higher flow rate for CO2 is caused by increased density and centrifugal effects as discussed in
Chapter 5. With the increased radial mass flow, the CO2 bearing is able to advect a substantially larger
portion of the generated heat. Through this process, the CO2 bearing is able to remove approximately
30 % of the generated heat, whereas the air bearing is only capable of removing 3 %. This illustrates
the favourable impact of radial pumping on bearing cooling. The improved fluid exchange in the
groove is shown in Figure 7.16. The radial transportation is significant in Figure 7.16(b).
In future CO2 applications, it is expected that foil thrust bearings will be required to operate with
substantially higher loads due to the substantially higher operating pressures. While the comparison
to air bearings has shown a favourable cooling performance, in particular due to the advection within
in the film, more substantial cooling improvements will be required to maintain low, uniform top foil
Z
X
(a)
Z
X
(b)
Figure 7.16: Comparison of streamline between air and CO2 thrust bearings in the groove, (a) air; (b) CO2 .
1.2 360
Maximum top foil temperature, K

Heat transfer from top foil to channels
Heat transfer from bump foil to channels
1 Heat transfer from housing to channels 350
Heat transfer rate, W
Heat conduction to housing

0.8
340
0.6
330
0.4
0.2 320
0 310
0 10 20 30 40 50 0 10 20 30 40 50
2 2
Convective coefficient, W/m K Convective coefficient, W/m K
(a) (b)
Figure 7.17: Comparison of (a) heat flow and maximum top foil temperature (b) for different convection
coefficients within the bump channels.
temperatures. A mechanism to achieve this is to force fluid through the bump channels, for example,
by applying a pressure difference in the radial direction. To provide insight towards this approach, a
thermal investigation was conducted to analyse the effect of increasing heat transfer coefficient within
the bump channels. Effectively this is the same as adding a convetive cooling flow. For this analysis,
only the stator side is considered. A uniform heat flux, corresponding to 1 W per pad (164 W/m2 ),
is applied to the fluid side of the top foil. The resulting distributions of heat flow and maximum
top foil temperature for different convective coefficients are shown in Figure 7.17. The lowest heat
transfer coefficient of 5 W/m2 K corresponds to the natural convection coefficient used previously.
These data show that increasing heat transfer in the bump channels is an effective way to extract a
larger portion of the supplied heat directly from the rear of the top foil. The heat transferred to the
fluid in the channel increases almost 5-fold. This corresponds to a 39 % reduction of the maximum
top foil temperature as shown in Figure 7.17(b). At the same time, the requirement to conduct heat to
the bearing housing through the bump foils diminishes.
7.5 Preliminary Performance Investigation

Radil and Zeszotek [79] performed an experimental investigation into the temperature profile of a
foil journal bearings. Both journal rotational speed and radial load were key factors determining the
heat generation. But rotational speed played a more significant role in heat generation and resulting
Preliminary Performance Investigation Section 7.5 173
high peak temperature. This experimental data are typically used as the validation data for developed
models [115, 116]. The deflection due to thermal stresses is found to be significant, and results in
the operation failure [186] because of mechanical contact between rotor and top foil. Linear thermal
expansion theory is usually employed to model thermal growth induced by the increased temperat-
ure [113, 115, 116].
In this section, the solver DisplacementFoam from the open source CFD toolbox FOAM-Extend-
3.0 [139] is used to model heat conduction and deflections due to thermal stresses. Since Displace-
mentFoam is already a mature solver for solving thermal stresses, verification or validation cases are
not provided here. The coupling is still the same as mentioned in Section 7.2.5. The difference in
the modelling approach is that the deflection due to thermal stresses is calculated. For numerical
simulations, CO2 is used as the working fluid, a convective heat transfer coefficient of 50 W/m2 K is
prescribed on the bump foils. The bearing geometry is defined in Table 7.1.
First, the temperature and deflection contours at different rotor to top foil separations are com-
pared in Figure 7.18 to 7.22. The rotational speed is set to 30 000 rpm. Figures 7.18 and 7.21 show
the temperature increase relative to 295 K, which is the reference temperature when calculating the
deflection due to thermal stresses. For the rotor, the higher power loss generated at lower separations
creates a comparatively high temperature as shown in Figure 7.18 and subsequently higher deflection
as shown in Figure 7.19. The deflection and temperature profiles along the radial direction at the
rotor are depicted in Figure 7.20. The temperature and deflection does not indicate a relationship
that can be modelled with the linear thermal expansion theory for thrust bearings. This implies that
the more accurate modelling approach is needed for thermal deflections. For the stator, temperature
and deflection contours are also plotted in Figures 7.21 and 7.22. Again, the highly loaded cases and
associated high power loss and rotor to top foil separations generates the largest temperatures and
thermal deflections.
The effect of rotational speed is compared in Figures 7.23 to 7.26. The performance is compared
at two different rotational speeds: 40 000 rpm and 50 000 rpm. It has been confirmed that rotational
speed has more impact on the power loss compared to rotor to top foil separations. For thermal
behaviour, the differences in peak temperature and thermal deflection are almost 100 %. This is
applied to both the rotor and stator.
To assess the need for fluid-structure-thermal prediction, comparison between two modelling ap-
Figure 7.18: Comparison of rotor temperature (in K) at power loss, (a) 318 N and 417 W, (b) 100 N and 352 W,
rotational speed is 30 000 rpm.
Figure 7.19: Comparison of rotor deflection (in m) at different power loss, (a) 318 N and 417 W, (b) 100 N and
352 W, rotational speed is 30 000 rpm.
40 6
Temperature
35 Deflection 5
Temperature, K 30 4
Deflection, µm
25 3
20 2
15 1
10 0
5 -1
0 -2
30 35 40 45 50
Radius, mm
Figure 7.20: Temperature and deflection at the rotor, load of 318 N, power loss of 417 W, rotational speed is
30 000 rpm.
Figure 7.21: Comparison of stator temperature (in K) at different load and power loss, (a) 318 N and 417 W,
(b) 100 N and 352 W, rotational speed is 30 000 rpm.
Figure 7.22: Comparison of stator deflection (in m) at different power loss, (a) 318 N and 417 W, (b) 100 N
and 352 W, rotational speed is 30 000 rpm.
Figure 7.23: Comparison of rotor temperature (in K) at different rotational speed, (a) 40 000 rpm, load is 275 N,
power loss is 815 W, (b) 50 000 rpm, load is 318 N, power loss is 1273 W.
Figure 7.24: Comparison of rotor deflection (in m) due to thermal stress at different rotational speed, (a)
40 000 rpm, load is 275 N, power loss is 815 W, (b) 50 000 rpm, load is 318 N, power loss is 1273 W.
Figure 7.25: Comparison of rotor temperature (in K) at different rotational speed, (a) 40 000 rpm, load is 275 N,
power loss is 815 W, (b) 50 000 rpm, load is 318 N, power loss is 1273 W.
Figure 7.26: Comparison of stator deflection (in m) due to thermal stress at different rotational speed, (a)
40 000 rpm, load is 275 N, power loss is 815 W, (b) 50 000 rpm, load is 318 N, power loss is 1273 W.
proaches (fluid-structure and fluid-structure-thermal) is depicted in Figure 7.27. Load capacity and
power loss are comparable from these approaches at the rotational speed of 30 000 rpm. Therefore,
the inclusion of thermal modelling is not necessary in the prediction of bearing performance at the
low rotational speed. At the rotational speed of 40 000 rpm, the deviation is found between load and
power loss. This deviation becomes apparent at the rotational speed of 50 000 rpm. A reduction of
70 N in load is observed, while a reduction in power loss is also spotted. This is due to the large de-
flection at the rotor as shown in Figure 7.24. Thus, thermal solver has to be included at the rotational
speed larger than 40 000 rpm for foil bearings operating with CO2 , since the predicted performance
differs between two modelling approaches. When operating at high rotational speeds, the negative
deflection (towards top foil) shown in Figure 7.24 might cause the mechanical contact between the
rotor and top foil. This is also reported in Ref. [186] as the failure of foil bearings operating in high
temperature.
7.6 Conclusions
In this chapter, a computational framework for the fluid-structure-thermal simulations of foil thrust
bearings is developed. Individual solvers and their coupling strategies are detailed together with val-
idation cases. A detailed description of the implemented foil bearing heat transfer models is provided.
The numerical tool is used to conduct a comparison study between a foil thrust bearing of the same
geometry operating with air and CO2 . This comparison highlights a number of differences in bear-
ing operation, in particular the heat fluxes and cooling requirements. Power loss and heat generation
2500
30000 rpm, fluid-structure 30000 rpm, fluid-structure
600 30000 rpm, fluid-structure-thermal 30000 rpm, fluid-structure-thermal
500 40000 rpm, fluid-structure-thermal 2000 40000 rpm, fluid-structure-thermal
Power loss, W
50000 rpm, fluid-structure-thermal 50000 rpm, fluid-structure-thermal
Load, N
400
1500
300
1000
200
100 500
5 10 15 20 25 30 35 5 10 15 20 25 30 35
Initial rotor to top foil separation, µm Initial rotor to top foil separation, µm
Figure 7.27: Comparison of bearing performance for different modelling approaches.
increases three-fold for the CO2 bearing at the same load. However, due to improved cooling, peak
temperatures are maintained below the air case. The CO2 bearing significantly benefits from increased
convective cooling on the rear surface of the rotor, if the rotor operates in a high pressure CO2 envir-
onment. This allows substantially more energy to be extracted through the rotor. In addition, almost
a third of the generated heat is advected with the fluid in the CO2 case, compared to only 3 % in the
air case. This effect, is caused by the centrifugal pumping that naturally occurs in CO2 bearings due
to the high fluid density [187], the higher heat capacity of the working fluid and better fluid exchange
in the groove. This is a highly advantageous mechanism for foil bearing cooling. Heat transfer to
the stator is similar in both cases. Here heat flow is limited due to the high thermal resistance im-
posed by the bump structure. Only a comparatively small portion of the heat flux entering the top
foil is extracted via convection to fluid in the bump channels when natural convection is assumed. A
separate numerical investigation showed that applying force convection in the bump channels is an
effective approach to enhance cooling, which will be essential for bearings operating at high loads.
This chapter has provided some new insights to the heat flux distribution in foil bearings, how this
is affected by the working fluid, and was identified mechanisms to enhance this heat transfer. At the
rotational speed less than 30 000 rpm, the predicted bearing performance between fluid-structure and
fluid-structure-thermal modelling approaches is comparable. However, at higher rotational speeds,
the predicted performance differs due to large deflections due to thermal stresses. Therefore, thermal
solver has to be included for foil thrust bearings operating at rotational speeds larger than 40 000 rpm.
Chapter 8
Conclusions and Future Work
8.1 Conclusions
The inspiration for this work was to investigate the possibility of using foil bearings for turboma-
chinery in the supercritical CO2 cycle. The work presented in this thesis provides new insight into
foil thrust bearings operated with high rotational speed and using highly dense CO2 as the working
fluid. The performance of CO2 foil thrust bearings was investigated by systematically increasing
the complexity of the foil bearing flow physics in each chapter, beginning with a rigid bearing, and
progressing to fluid-structure interactions, fluid-structure-thermal modelling and finally rotordynamic
analysis. In each chapter we improve our physical insight and understanding, as new layers of com-
plexity are added. This approach has allowed a rigorous contribution to the knowledge to be con-
structed. The use of CFD for foil thrust bearings allowed the flow field to be probed in high detail to
provide new significant insights into the operation of foil thrust bearings with CO2 . The intention is
that these new insights can be used in future CO2 foil bearing designs to provide a greater degree of
freedom to enable high speed turbomachinery system for supercritical CO2 cycles. A brief summary
of each chapter with major findings is given here:
• In Chapter 3, the in-house CFD code Eilmer is first modified for foil bearing simulations by
adding new boundary conditions and reconstruction method. Next, to allow fast and stable
turbulent simulations in thin film film bearings, two new features are added: compressible wall
function and fourth-order artificial dissipation. These new additions are verified and validated
with different test cases representative of foil bearing flows. The compressible wall function
from Nichols et al. [137] is able to be applied for the hybrid pressure and shear driven flow
within foil bearings. A y + value in the first cell from the wall of less than 20 is required to
182 Chapter 8 Conclusions and Future Work
accurately calculate the wall shear stress for the studied geometry. A comparison of turbulence
models show that the Wilcox’s 2006 k − ω model is most accurate for the hybrid pressure and
shear driven flow compared to other models including: SST k − ω, Spalart Allmaras, standard
k − ǫ and Nonlinear k − ǫ Shih. Incorporating the fourth-order artificial dissipation allows a
speed up of the the turbulent simulations for foil thrust bearings, while ensuring a stable and
accurate computation without contaminating the fluid flow. For the test case used in this thesis,
a speed-up of 6.57 times is achieved at a rotational speed of 40 000 rpm. The minimum k4 value
for stabilising turbulent simulations increases in terms of the rotational speed and cell aspect
ratio. It is recommended that k4 larger than 0.01 is selected.
• In Chapter 4, the moving grid capability is added to Eilmer to allow the multiphysics simu-
lations. This capacity is validated with the different test cases including inviscid, viscous and
turbulent simulations. In addition, a bespoke solver based on Kirchhoff plate theory was de-
veloped for the structural deformation of the top foil within foil bearings, and verified with the
commercial software ANSYS. The coupling strategy is proposed to provide the steady state and
time-accurate transient simulations. A mapping algorithm is used when exchanging pressure
and deflection between two solvers, the difference is less than 1 %.
• In Chapter 5, the developed fluid-structure simulation tool for the steady state performance
of foil thrust bearings was validated with the experimental data from literature. Using the
developed fluid-structure solver, the steady state performance of foil bearings operating at dif-
ferent conditions is also investigated. It is found that a ten-fold increase of CO2 pressure and
thus density can significantly alter foil bearings performance. The density change can alter load
capacity by up to 40 %. This is due to additional mass flow at the inner edge of the ramp region,
caused by increased centrifugal inertia forces acting on the more dense CO2 . But centrifugal
inertia effects also cause an increased leakage towards the outer edges in the flat region which is
detrimental to load generation. This work has highlighted the significance of the centrifugal in-
ertia forces on the bearing performance and provided new insight to the underlying effects that
influence bearing performance. Due to the high densities of CO2 , the foil thrust bearings can
experience turbulent flow within the film. As a consequence of the additional eddy viscosity,
the foil thrust bearings now experience different velocity profiles across the film height, which
result in increased power consumption. In contrast, load capacity only shows a slight increase
as turbulence levels increase. This is due to the compliance of the bump foils, which deflect
almost twice as much under turbulent conditions, thereby decreasing the potential to increase
load.
• In Chapter 6, the procedure to obtain the dynamic performance of foil thrust bearings using the
fluid-structure simulation tool is discussed and validated. The computational domain including
the groove has to be utilised, since the steady state and rotordynamic coefficients are different
by the presence of the groove.
• In Chapter 7, a computational framework for the fluid-structure-thermal simulations of foil

thrust bearings is developed by adding LaplacianFoam to the fluid-structure simulation tool.
The numerical tool is used to conduct a comparative study between a foil thrust bearing of the
same geometry operating with air and CO2 . It was found that power loss and heat generation
increases three-fold for the CO2 bearing. However, due to improved cooling, peak temperat-
ures are maintained below the air case. The CO2 bearing significantly benefits from increased
convective cooling on the rear surface of the rotor. This allows substantially more energy to be
extracted through the rotor. In addition, almost a third of the generated heat is advected with the
fluid in the CO2 case, compared to only 3 % in the air case. This effect, caused by the centrifu-
gal pumping and improved fluid exchange in the groove that naturally occurs in CO2 bearings
due to the high fluid density. This provides a new and effective cooling mechanism for CO2
bearing. Heat transfer to the stator is similar for both cases. Here heat flow is limited due to
the high thermal resistance imposed by the bump structure. Only a comparatively small portion
of the heat flux entering the top foil is extracted via convection to fluid in the bump channels
when natural convection is assumed. A separate numerical investigation showed that applying
force convection in the bump channels is an effective approach to enhance cooling, which will
be essential for bearings operating at high loads. At the rotational speed less than 30 000 rpm,
the predicted bearing performance between fluid-structure and fluid-structure-thermal model-
ling approaches is comparable. The fluid-structure simulation tool can then be used. However,
at higher rotational speeds, the predicted performance differs due to large deflections due to
thermal stresses. Therefore, thermal solver has to be included for foil thrust bearings operating
at rotational speeds larger than 40 000 rpm.
8.2 Recommendations for Foil Bearing Designs

Based on the results from this thesis, recommendations for foil bearing design for CO2 operation
include:
• Add bump foils underneath the ramp, this results in a more pad area to support the pressure.
• Cooling through the bump foil channels: The forced cooling in the bump foil channels can
reduce the temperature increase by 39 %, and enable the bearing to operate at higher rotational
speed or with increased load.
• Adding radially varying bump foil stiffness to minimise fluid loss from flat region.
• Optimise radial and tangential bump foil stiffness.
• Use groove to increase fluid exchange.
8.3 Future Research

Recommendations for future work include:
• Dynamic Performance: Dynamic performances of foil thrust bearings operating with CO2 in
terms of load and rotational speed will be investigated.
• Further Improvement of Conjugate Heat Transfer: The Neumann and Dirichlet conditions
between fluid and solid interface require a long time to converge. Suitable acceleration methods
have to be implemented.
• Experimental Validation: There is a lack of experimental data of foil thrust bearings operating
with the high dense gas. Experimental validation should be conducted to provide relevant
validation. This includes steady state performance at different rotational speeds and rotor to
top foil separations. Temperature data is required to provide validation of the fluid-structure-
thermal modelling approach. Dynamic stiffness and damping are also of interest, especially at
high load condition, where the negative equivalent damping is predicted.
• Parametric Study: The geometry of foil thrust bearings investigated in this thesis is based
on a geometry from NASA. However this is only a preliminary design. It was experimentally
Future Research Section 8.3 185
confirmed that changing the trailing edge can increase the load capacity two-fold. At the same
time, damping is also largely increased. Parametric studies have to be conducted to investig-
ate the key parameter influencing bearing performance. Key parameters might include groove
shape and depth, ramp height, top foil thickness and bump stiffness distribution.
• Heat Transfer: In this thesis, the heat transfer coefficient at the rear surface of the rotor is
calculated from empirical correlations. A large heat transfer coefficient is obtained compared
to air, if the rotor is submerged in CO2 environment. The experimental validation has to be
conducted to test the suitability of the high dense CO2 as the efficient heat transfer fluid. Such
work is currently underway at the University of Queensland.
• Cooling Methods: The deflection of the rotor due to thermal stresses is significant, leading to
a substantial reduction in load capacity at high rotational speeds. Similarly, heat transfer to the
stator is minimal due to the thermal resistance of the bump structure. To reduce the thermal
deflection, the new cooling methods that minimise temperature non-uniformities within the foil
thrust bearings should be investigated. This will decrease the thermal deflection and recover the
load capacity.
References
[1] Feher, E. G., “The Supercritical Thermodynamic Power Cycle,” Energy Conversion, Vol. 8,
1968, pp. 85–90.
[2] Angelino, G., “Perspectives for the Liquid Phase Compression Gas Turbine,” ASME Journal
of Engineering for Power, Vol. 89, 1967, pp. 229–236.
[3] Angelino, G., “Carbon Dioxide Condensation Cycles for Power Production,” ASME Journal of
Engineering for Power, Vol. 90, 1968, pp. 287–295.
[4] Dostal, V., A Supercritical Carbon Dioxide Cycle for Next Generation Nuclear Reactors, PhD
Thesis, Massachusetts Institute of Technology, Cambridge, MA, 2004.
[5] Conboy, T., Wright, S., Pasch, J., Fleming, D., Rochau, G., and Fuller, R., “Performance Char-
acteristics of an Operating Supercritical CO2 Brayton Cycle,” ASME Journal of Engineering
for Gas Turbines and Power, Vol. 134, 2012, pp. 111703.
[6] Qi, J., Reddell, T., Qin, K., Hooman, K., and Jahn, I. H., “Supercritical CO2 Radial Turbine
Design Performance as a Function of Turbine Size Parameters,” ASME Journal of Turboma-
chinery, Vol. 139, No. 8, 2017, pp. 081008.
[7] Wright, S., Radel, R., Vernon, M., Rochau, G., and Pichard, P., “Operation and Analysis of a
Supercritical CO2 Brayton Cycle,” Technical report SAN2010-0171, Sandia National Laborat-
ories, September 2010.
[8] Conboy, T., “Real-Gas Effects in Foil Thrust Bearings Operating in the Turbulent Regime,”
ASME Journal of Tribology, Vol. 135, No. 3, March 2013, pp. 031703.
[9] Harnoy, A., Bearing Design in Machinery: Engineering Tribology and Lubrication, CRC
press, 1st ed., 2002.
188 References
[10] Preuss, J. L., “Application of Hydrostatic Bearings in Supercritical CO2 Turbomachinery,” The
5th International Supercritical CO2 Power Cycles Symposium, San Antonio, Texas, 2016, pp.
1–15.
[11] Heshmat, H., Walowit, J., and Pinkus, O., “Analysis of Gas Lubricated Compliant Thrust Bear-
ings,” ASME Journal of Tribology, Vol. 105, No. 4, October 1983, pp. 638–646.
[12] Song, Y., Ren, X., Gu, C.-w., and Li, X.-s., “Experimental and Numerical Studies of Cavitation
Effects in a Tapered Land Thrust Bearing,” ASME Journal of Tribology, Vol. 137, No. 1, 2015,
pp. 011701.1–011701.9.
[13] Missana, A., Booser, E., and Ryan, F., “Performance of Tapered Land Thrust Bearings for
Large Steam Turbines,” ASLE TRANSACTIONS, Vol. 14, No. 4, 1971, pp. 301–306.
[14] Khonsari, M. M. and Booser, E. R., Applied Tribology: Bearing Design and Lubrication, John
Wiley & Sons, New Jersey, 2nd ed., 2008.
[15] Dimond, T., Younan, A., and Allaire, P., “A Review of Tilting Pad Bearing Theory,” Interna-
tional Journal of Rotating Machinery, Vol. 2011, 2011, pp. 1–23.
[16] Shalash, K. and Schiffmann, J., “On the Manufacturing of Compliant Foil Bearings,” Journal
of Manufacturing Processes, Vol. 25, 2017, pp. 357–368.
[17] San Andrés, L. and Chirathadam, T. A., “Performance Characteristics of Metal Mesh Foil
Bearings: Predictions Versus Measurements,” ASME Journal of Engineering for Gas Turbines
and Power, Vol. 135, No. 12, 2013, pp. 122503.
[18] Feng, K., Liu, W., Yu, R., and Zhang, Z., “Analysis and Experimental Study on A Novel
Gas Foil Bearing with Nested Compression Springs,” Tribology International, Vol. 107, 2017,
pp. 65–76.
[19] Ao, H., Jiang, H., Wei, W., and Ulanov, A., “Study on the Damping Characteristics of MR
Damper in Flexible Supporting of Turbo-pump Rotor for Engine,” Systems and Control in
Aerospace and Astronautics, 2006. ISSCAA 2006. 1st International Symposium on, IEEE, Har-
bin, China, 2006, pp. 1–5.
References 189
[20] San Andrés, L. and Chirathadam, T. A., “A Metal Mesh Foil Bearing and a Bump-type Foil
Bearing: Comparison of Performance for Two Similar Size Gas Bearings,” ASME Journal of
Engineering for Gas Turbines and Power, Vol. 134, No. 10, 2012, pp. 102501.
[21] San Andrés, L., Chirathadam, T. A., and Kim, T.-H., “Measurement of Structural Stiffness and
Damping Coefficients in a Metal Mesh Foil Bearing,” ASME Journal of Engineering for Gas
Turbines and Power, Vol. 132, No. 3, 2010, pp. 032503.
[22] San Andrés, L. and Chirathadam, T. A., “Metal Mesh Foil Bearing: Effect of Motion Amp-
litude, Rotor Speed, Static Load, and Excitation Frequency on Force Coefficients,” Journal of
Engineering for Gas Turbines and Power, Vol. 133, No. 12, 2011, pp. 122503.
[23] Dykas, B., Bruckner, R., DellaCorte, C., Edmonds, B., and Prahl, J., “Design, Fabrica-
tion, and Performance of Foil Gas Thrust Bearings for Microturbomachinery Applications,”
ASME Journal of Engineering for Gas Turbines and Power, Vol. 131, No. 1, October 2008,
pp. 012301.
[24] Hannon, W. M., The Generalized Universal Reynolds Eqaution for Variable Property Fluid-
Film Lubrication and Variable Geometry, Phd thesis, University of Akron, Akron, Ohio, 2006.
[25] Aksoy, S. and Aksit, M., “A Fully Coupled 3D Thermo-elastohydrodynamics Model for a
Bump-type Compliant Foil Journal Bearing,” Tribology International, Vol. 82, 2015, pp. 110–
122.
[26] San Andres, L. and Diemer, P., “Prediction of Gas Thrust Foil Bearing Performance for Oil-
Free Automotive Turbochargers,” ASME Journal of Engineering for Gas Turbines and Power,
Vol. 137, No. 3, September 2014, pp. 032502.
[27] Thatte, A., Loghin, A., Shin, Y., and Ananthasayanam, B., “Performance and Life Character-
istics of Hybrid Gas Bearing in a 10 MW Supercritical CO2 Turbine,” ASME Turbo Expo 2016:
Turbomachinery Technical Conference and Exposition, American Society of Mechanical En-
gineers, Seoul, South Korea, 2016, pp. V009T36A018–V009T36A018.
[28] Kim, D., “Design Space of Foil Bearings for Closed-Loop Supercritical CO2 Power Cycles
Based on Three-Dimensional Thermohydrodynamic Analyses,” ASME Journal of Engineering
for Gas Turbines and Power, Vol. 138, No. 3, 2016, pp. 032504.
190 References
[29] Moullec, Y. L., “Conceptual Study of a High Efficiency Coal-fired Power Plant with CO2
Capture using a Supercritical CO2 Brayton Cycle,” Energy, Vol. 49, 2013, pp. 32 – 46.
[30] Turchi, C., Ma, Z., and Wagner, M., “Thermodynamic Study of Advanced Supercritical Carbon
Dioxide Power Cycles for Concentrating Solar Power Systems,” ASME Journal of Solar Energy
Engineering, Vol. 135, No. 4, June 2013, pp. 041007.
[31] Gurgenci, H., “Challenges for Electrical Power Generation from EGS,” Proceedings World
Geothermal Congress 2010, Bali, Indonesia, 2010.
[32] Besarati, S. M. and Goswami, D. Y., “Analysis of Advanced Supercritical Carbon Dioxide
Power Cycles with a Bottoming Cycle for Concentrating Solar Power Applications,” ASME
Journal of Solar Energy Engineering, Vol. 137, 2014, pp. 011020.
[33] Kim, Y. M., Sohn, J. L., and Yoon, E. S., “Supercritical CO2 Rankine cycles for waste heat
recovery from gas turbine,” Energy, Vol. 118, 2016, pp. 893–905.
[34] Kim, Y., Kim, C., and Favrat, D., “Transcritical or Supercritical CO2 Cycles using Both Low-
and High-Temperature Heat Sources,” Energy, Vol. 43, No. 1, 2012, pp. 402 – 415, 2nd In-
ternational Meeting on Cleaner Combustion (CM0901-Detailed Chemical Models for Cleaner
Combustion).
[35] “Direct normal irridation in Australia,” http://solargis.com/assets/graphic/free-

map/DNI/Solargis-Australia-DNI-solar-resource-map-en.png, Accessed: 2016-10-18.
[36] Singh, R., Miller, S. A., Rowlands, A. S., and Jacobs, P. A., “Dynamic Charateristics of a
Direct-heated Supercritical Carbon-dioxide Brayton Cyle in a Solar Thermal Power Plant,”
Energy, Vol. 50, 2013, pp. 194–204.
[37] Singh, R., Rowlands, A. S., and Miller, S. A., “Effects of Relative Volume-ratios on Dynamic
Performance of a Directe-heated Supercritical Carbon-dioxide Closed Brayton Cyle in a Solar-
thermal Power Plant,” Energy, Vol. 55, 2013, pp. 1025–1032.
[38] Singh, R., Kearney, M. P., and Manzie, C., “Extremum-seeking Control of a Supercrtical
Carbon-dioxide Closed Brayton Cycle in a Direct-heated Solar Thermal Power Plant,” Energy,
Vol. 60, 2013, pp. 380–387.
References 191
[39] Singh, R., Dynamics and Control of a Closed Carbon-Dioxide Brayton Cycle, Phd thesis,
University of Queensland, Brisbane, Australia, 2013.
[40] Twomey, B., Nagy, A., Russell, H., Rowlands, A., Czapla, J., Singh, R., de M Ventura,
C. A., and Jahn, I., “The University of Queensland Refrigerant and Supercritical CO2 Test
Loop,” ASME Turbo Expo 2016: Turbomachinery Technical Conference and Exposition,
American Society of Mechanical Engineers, Seoul, South Korea, 2016, pp. V003T25A013–
V003T25A013.
[41] Inverson, B. D., Conboy, T. M., Pasch, J. J., and Kruizenga, A. M., “Supercritical CO2 Brayton
Cycles for Solar-thermal Energy,” Applied Energy, Vol. 111, 2013, pp. 957–970.
[42] T., C., Pasch, J., and Fleming, D., “Control of a Supercritical CO2 Recompression Brayton
Cycle Demonstration Loop,” ASME Journal of Engineering for Gas Trubines and Power,
Vol. 135, 2013, pp. 111701.
[43] Kang, Y.-S., Huh, J.-S., Cho, J., Shin, H., and Baik, Y.-J., “Design and Performance Assess-
ments of a Partial Admission Axial Turbine Using Supercritical Carbon Dioxide,” ASME 2016
Fluids Engineering Division Summer Meeting collocated with the ASME 2016 Heat Trans-
fer Summer Conference and the ASME 2016 14th International Conference on Nanochannels,
Microchannels, and Minichannels, American Society of Mechanical Engineers, Seoul, South
Korea, 2016, pp. V01AT09A010–V01AT09A010.
[44] Zhang, H., Zhao, H., Deng, Q., and Feng, Z., “Aerothermodynamic Design and Numerical
Investigation of Supercritical Carbon Dioxide Turbine,” ASME Turbo Expo 2015: Turbine
Technical Conference and Exposition, American Society of Mechanical Engineers, Montreal,
Quebec, Canada, 2015, pp. V009T36A007–V009T36A007.
[45] Ventura, C. A., Jacobs, P. A., Rowlands, A. S., Petrie-Repar, P., and Sauret, E., “Preliminary
Design and Performance Estimation of Radial Inflow Turbines: An Automated Approach,”
ASME Journal of Fluids Engineering, Vol. 134, No. 3, 2012, pp. 031102.
[46] Shanechi, M. M., Veidt, M., and Hooman, K., “Forced Response Analysis of Supercritical CO2
Radial Inflow Turbine Designed for Concentrating Solar Power,” ASME Turbo Expo 2016:
192 References
Turbomachinery Technical Conference and Exposition, American Society of Mechanical En-

gineers, Seoul, South Korea, 2016, pp. V009T36A021–V009T36A021.
[47] Kus, B. and Nekså, P., “Development of One-dimensional Model for Initial Design and Eval-
uation of Oil-free CO2 Turbo-compressor,” International Journal of Refrigeration, Vol. 36,
No. 8, 2013, pp. 2079–2090.
[48] Lee, J., Lee, J. I., Yoon, H. J., and Cha, J. E., “Supercritical Carbon Dioxide Turboma-
chinery Design for Water-cooled Small Modular Reactor Application,” Nulcear Engineering
and Design, Vol. 270, 2014, pp. 76–89.
[49] Chu, X. and Laurien, E., “Direct Numerical Simulation of Heated Turbulent Pipe Flow at
Supercritical Pressure,” ASME Journal of Nuclear Engineering and Radiation Science, Vol. 2,
No. 3, June 2016, pp. 031019.
[50] Baltadjiev, N. D., Lettieri, C., and Spakovszky, Z. S., “An Investigation of Real Gas Effects
in Supercritical CO2 Centrifugal Compressors,” ASME Journal of Turbomachinery, Vol. 137,
No. 9, 2015, pp. 091003.
[51] Monje, B., Sánchez, D., Savill, M., Pilidis, P., and Sánchez, T., “1D and 3D Tools to Design
Supercritical CO2 Radial Compressors: A Comparison,” Proceedings of the 2nd International
Seminar on ORC Power Systems, Rotterdam, Netherlands, 2013.
[52] Furusawa, T., Miyazawa, H., and Yamamoto, S., “Numerical Method for Simulating Flows
of Supercritical CO2 Compressor With Nonequilibrium Condensation,” ASME Turbo Expo
2016: Turbomachinery Technical Conference and Exposition, American Society of Mechanical
Engineers, Seoul, South Korea, 2016, pp. V009T36A003–V009T36A003.
[53] Pecnik, R., Rinaldi, E., and Colonna, P., “Computational Fluid Dynamics of a Radial Com-
pressor Operating with Supercritical CO2 ,” ASME Journal of Engineering for Gas Turbines
and Power, Vol. 134, No. 12, October 2012, pp. 122301.
[54] Rinaldi, E., Pecnik, R., and Colonna, P., “Computational Fluid Dynamic Simulation of a Super-
critical CO2 Compressor Performance Map,” ASME Journal of Engineering for Gas Turbines
and Power, Vol. 137, No. 7, July 2015, pp. 072602.
References 193
[55] Kim, S. G., Lee, J., Ahn, Y., Lee, J. I., Addad, Y., and Ko, B., “CFD Investigation of a
Centrigual Compressor Derived from Pump Technology for Supercritical Carbon Dioxide as a
Working Fluid,” Journal of Supercritical FLuid, Vol. 86, 2014, pp. 160–171.
[56] Behafarid, F. and Podowski, M. Z., “Modeling and Computer Simulation of Centrifugal CO2
Compressors at Supercritical Pressures,” ASME Journal of Fluids Engineering, Vol. 138, No. 6,
2016, pp. 061106.
[57] Guo, J., “Design Analysis of Supercritical Carbon Dioxide Recuperator,” Applied Energy,
Vol. 164, 2016, pp. 21–27.
[58] Sarkar, J., “Performance of a Flat-plate Solar Thermal Collector using Supercritical Carbon
Dioxide as Heat Transfer Fluid,” International Journal of Sustainable Energy, Vol. 32, No. 6,
2013, pp. 531–543.
[59] Kim, T. H., Kwon, J. G., Yoon, S. H., Park, H. S., Kim, M. H., and Cha, J. E., “Numerical Ana-
lysis of Air-foil Shaped Fin Performance in Printed Circuit Heat Exchanger in a Supercritical
Carbon Dioxide Power Cycle,” Nuclear Engineering and Design, Vol. 288, 2015, pp. 110–118.
[60] Dang, C. and Hihara, E., “In-tube Cooling Heat Transfer of Supercritical Carbon Dioxide. Part
2. Comparison of Numerical Calculation with Different Turbulence Models,” International
Journal of Refrigeration, Vol. 27, No. 7, 2004, pp. 748–760.
[61] He, S., Kim, W., and Bae, J., “Assessment of Performance of Turbulence Models in Predicting
Supercritical Pressure Heat Transfer in a Vertical Tube,” International Journal of Heat and
Mass Transfer, Vol. 51, No. 19, 2008, pp. 4659–4675.
[62] Liao, S. and Zhao, T., “Measurements of Heat Transfer Coefficients from Supercritical Car-
bon Dioxide Flowing in Horizontal Mini/Micro Channels,” ASME Journal of Heat Transfer,
Vol. 124, No. 3, 2002, pp. 413–420.
[63] Dang, C. and Hihara, E., “In-tube Cooling Heat Transfer of Supercritical Carbon Dioxide. Part
1. Experimental Measurement,” International Journal of Refrigeration, Vol. 27, No. 7, 2004,
pp. 736–747.
194 References
[64] Oh, H.-K. and Son, C.-H., “New Correlation to Predict the Heat Transfer Coefficient in-tube
Cooling of Supercritical CO2 in Horizontal Macro-tubes,” Experimental Thermal and Fluid
Science, Vol. 34, No. 8, 2010, pp. 1230–1241.
[65] Kruizenga, A., Anderson, M., Fatima, R., Corradini, M., Towne, A., and Ranjan, D., “Heat
Transfer of Supercritical Carbon Dioxide in Printed Circuit Heat Exchanger Geometries,”
ASME Journal of Thermal Science and Engineering Applications, Vol. 3, No. 3, 2011,
pp. 031002.
[66] Peeters, J. W., Pecnik, R., Rohde, M., van der Hagen, T., and Boersma, B., “Turbulence Atten-
uation in Simultaneously Heated and Cooled Annular Flows at Supercritical Pressure,” Journal
of Fluid Mechanics, Vol. 799, 2016, pp. 505–540.
[67] Fairuz, Z. and Jahn, I., “The Influence of Real Gas Effects on the Performance of Supercritical
CO2 Dry Gas Seals,” Tribology International, Vol. 102, 2016, pp. 333–347.
[68] Zakariya, M. F. and Jahn, I. H., “Performance of Supercritical CO2 Dry Gas Seals Near the Crit-
ical Point,” ASME Turbo Expo 2016: Turbomachinery Technical Conference and Exposition,
American Society of Mechanical Engineers, Seoul, South Korea, 2016, pp. V009T36A007–
V009T36A007.
[69] Thatte, A. and Dheeradhada, V., “Coupled Physics Performance Predictions and Risk Assess-
ment for Dry Gas Seal Operating in MW-Scale Supercritical CO2 Turbine,” ASME Turbo Expo
2016: Turbomachinery Technical Conference and Exposition, American Society of Mechan-
ical Engineers, Seoul, South Korea, 2016, pp. V009T36A017–V009T36A017.
[70] Bidkar, R. A., Sevincer, E., Wang, J., Thatte, A. M., Mann, A., Peter, M., Musgrove, G.,
Allison, T., and Moore, J., “Low-Leakage Shaft-End Seals for Utility-Scale Supercritical CO2
Turboexpanders,” ASME Journal of Engineering for Gas Turbines and Power, Vol. 139, No. 2,
2017, pp. 022503.
[71] Xu, F. and Kim, D., “Three-Dimensional Turbulent Thermo-Elastohydrodynamic Analyses of

Hybrid Thrust Foil Bearings Using Real Gas Model,” ASME Turbo Expo 2016: Turboma-
chinery Technical Conference and Exposition, American Society of Mechanical Engineers,
Seoul, South Korea, 2016, pp. V07BT31A030–V07BT31A030.
References 195
[72] Kim, D., “Design Space of Foil Bearings for Closed-Loop Supercritical CO2 Power Cycles
Based on Three-Dimensional Thermohydrodynamic Analyses,” ASME Journal of Engineering
for Gas Turbines and Power, Vol. 138, No. 3, October 2015, pp. 032504.
[73] Dickman, J., An Investigation of Gas Foil Thrust Bearing Performance and Its Influence
Factors, MS Thesis, Case Western Reserve University, Cleveland, OH, May 2010.
[74] Balducchi, F., Arghir, M., Gauthier, R., and Renard, E., “Experimental Analysis of the Start-up
Torque of a Mildly Loaded Foil Thrust Bearing,” ASME Journal of Tribology, Vol. 135, No. 3,
2013, pp. 031702.
[75] Balducchi, F., Arghir, M., and Gauthier, R., “Experimental Analysis of the Dynamic Character-
istics of a Foil Thrust Bearing,” ASME Journal of Tribology, Vol. 137, No. 2, 2015, pp. 021703.
[76] Feng, K., Liu, L.-J., Guo, Z.-Y., and Zhao, X.-Y., “Parametric Study on Static and Dynamic
Characteristics of Bump-type Gas Foil Thrust Bearing for Oil-free Turbomachinery,” Proceed-
ings of the Institution of Mechanical Engineers, Part J: Journal of Engineering Tribology,
Vol. 229, No. 10, 2015, pp. 1247–1263.
[77] Dykas, B. and Radil, K., “Experimental Measurement of Thrust Foil Bearing Temperature Pro-
files,” World Tribology Congress III, American Society of Mechanical Engineers, Washington,
D.C., USA, 2005, pp. 93–94.
[78] Dykas, B. D., Factors Influencing the Performance of Foil Gas Thrust Bearings for Oil-free
Turbomachinery Applications, Ph.D. thesis, Case Western Reserve University, 2006.
[79] Radil, K. and Zeszotek, M., “An Experimental Investigation into the Temperature Profile of a
Compliant Foil Air Bearing,” Tribology Transactions, Vol. 47, No. 4, 2004, pp. 470–479.
[80] Kim, T. H., Park, M. S., Lee, J., Kim, Y. M., Ha, K.-K., Park, J., Lee, C., and Kim, C., “Identi-
fication of Dynamic Characteristics of Gas Foil Thrust Bearings Using Base Excitation,” ASME
Turbo Expo 2016: Turbomachinery Technical Conference and Exposition, American Society
of Mechanical Engineers, Seoul, South Korea, 2016, pp. V07BT31A006–V07BT31A006.
196 References
[81] Reynolds, O., “On the Theory of Lubrication and Its Application to Mr. Beauchamp Tower’s
Experiments, Including an Experimental Determination of the Viscosity of Olive Oil,” Philo-
sophical Transactions of the Royal Society of London, Vol. 177, January 1886, pp. 157–234.
[82] Bruckner, R., Simulation and Modeling of the Hydrodynamic, Thermal, and Structural Beha-
vior of Foil Thrust Bearings, PhD Thesis, Case Western Reserve University, Cleveland, OH,
2004.
[83] Wang, N., Chang, S., and Huang, H., “Comparison of Iterative Methods for the Solution
of Compressible Fluid Reynolds Equation,” ASME Journal of Tribology, Vol. 133, 2011,
pp. 021702.
[84] Gad, A. M. and Kaneko, S., “Performance Characteristics of Gas-lubricated Bump-type Foil
Thrust Bearing,” Proceedings of the Institution of Mechanical Engineers, Part J: Journal of
Engineering Tribology, Vol. 229, No. 6, 2015, pp. 746–762.
[85] Gad, A. M. and Kaneko, S., “Tailoring of the Bearing Stiffness to Enhance the Performance of
Gas-lubricated Bump-type Foil Thrust Bearing,” Proceedings of the Institution of Mechanical
Engineers Part J Journal of Engineering Tribology, Vol. 230, 2015.
[86] Lee, D. and Kim, D., “Design and Performance Prediction of Hybrid Air Foil Thrust Bearings,”
ASME Journal of Engineering for Gas Turbines and Power, Vol. 133, No. 4, 2011, pp. 042501.
[87] Ng, C. W. and Pan, C. H. T., “A Linearized Turbulent Lubrication Theory,” ASME Journal of
Fluids Engineering, Vol. 87, 1964, pp. 675–682.
[88] Constantinescu, V., “Basic Relationships in Turbulent Lubrication and Their Extension to In-
clude Thermal Effects,” ASME Journal of Tribology, Vol. 95, No. 2, Apirl 1973, pp. 147–154.
[89] Hirs, G., “A Bulk-Flow Theory for Turbulence in Lubricant Films,” ASME Journal of Tri-
bology, Vol. 95, No. 2, Apirl 1973, pp. 137–145.
[90] Patanka, S., Numerical Heat Transfer and Fluid Flow, Hemisphere Publishing Co., New York,
1st ed., 1980.
References 197
[91] Chen, P. Y. K. and Hahn, E. J., “Use of Computational Fluid Dynamics in Hydrodynamic
Lubrication,” Proceedings of the Institution of Mechanical Engineers, Part J: Journal of En-
gineering Tribology, Vol. 212, 1998, pp. 427–436.
[92] Brajdic-Mitidieri, P., Advanced Modelling of Elastohydrodynamic Lubrication, Ph.D. thesis,

Imperial College London, London, UK, 2005.
[93] Guo, Z., Hirano, T., and Kirk, R. G., “Application of CFD Analysis for Rotating Ma-
chinery—Part I: Hydrodynamic, Hydrostatic Bearings and Squeeze Film Damper,” ASME
Journal of Engineering for Gas Turbines and Power, Vol. 127, No. 2, 2005, pp. 445–451.
[94] Papadopoulos, C. I., Kaiktsis, L., and Fillon, M., “Computational Fluid Dynamics Thermo-
hydrodynamic Analysis of Three-Dimensional Sector-Pad Thrust Bearings With Rectangular
Dimples,” ASME Journal of Tribology, Vol. 136, 2013, pp. 011702.
[95] Zhang, X., Yin, Z., Jiang, D., and Gao, G., “The Design of Hydrodynamic Water-lubricated
Step Thrust Bearings Using CFD Method,” Mechanics & Industry, Vol. 15, 2014, pp. 197–206.
[96] Iordanoff, I., “Analysis of an Aerodynamic Compliant Foil Thrust Bearing: Method for a Rapid
Design,” ASME Journal of Tribology, Vol. 121, No. 4, October 1999, pp. 816–822.
[97] Lez, S. L., Arghir, M., and Frene, J., “A New Bump-Type Foil Bearing Structure Analytical
Model,” ASME Journal of Engineering for Gas Turbines and Power, Vol. 129, No. 4, 2007,
pp. 747–757.
[98] Hryniewicz, P., Wodtke, M., Olszewski, A., and Rzadkowski, R., “Structural Properties of
Foil Bearings: A Closed-form Solution Validated with Finite Element Analysis,” Tribology
Transactions, Vol. 52, No. 4, 2009, pp. 435–446.
[99] Feng, K. and Kaneko, S., “Analytical Model of Bump-Type Foil Bearings Using a Link-Spring
Structure and a Finite Element Shell Model,” ASME Journal of Tribology, Vol. 132, No. 2,
Apirl 2010, pp. 021706.
[100] Gad, A. and Kaneko, S., “A New Structural Stiffness Model for Bump-Type Foil Bearings:
Application to Generation II Gas Lubricated Foil Thrust Bearing,” ASME Journal of Tribology,
Vol. 136, No. 4, May 2014, pp. 041701.
198 References
[101] Rubio, D. and San Andres, L., “Structural Stiffness, Dry Friction Coefficient, and Equivalent
Viscous Damping in a Bump-type Foil Gas Bearing,” ASME Journal of engineering for gas
turbines and power, Vol. 129, No. 2, 2007, pp. 494–502.
[102] Agrawal, G. L., “Foil Air/gas Bearing Technology—an Overview,” International Gas Turbine
and Aeroengine Congress and Exhibition, American Society of Mechanical Engineers, Or-
lando, Florida, June 1997.
[103] Ku, C.-P. R. and Heshmat, H., “Structural Stiffness and Coulomb Damping in Compliant Foil
Journal Bearings: Theoretical Considerations,” Tribology transactions, Vol. 37, No. 3, 1994,
pp. 525–533.
[104] Le Lez, S., Arghir, M., and Frene, J., “Static and Dynamic Characterization of a Bump-type
Foil Bearing Structure,” ASME Journal of Tribology, Vol. 129, No. 1, 2007, pp. 75–83.
[105] Andres, L. S. and Kim, T., “Analysis of Gas Foil Bearings Integrating FE Top Foil Models,”
Tribology International, Vol. 42, 2009, pp. 111–120.
[106] Lee, D. and Kim, D., “Three-Dimensional Thermohydrodynamic Analyses of Rayleigh Step
Air Foil Thrust Bearing with Radially Arranged Bump Foils,” Tribology Transacations,
Vol. 54, No. 3, March 2011, pp. 432–448.
[107] Heshmat, C., Xu, D., and Heshmat, H., “Analysis of Gas Lubricated Foil Thrust Bearings
Using Coupled Finite Element and Finite Difference Methods,” ASME Journal of Tribology,
Vol. 122, No. 1, June 1999, pp. 199–204.
[108] Lehn, A., Mahner, M., and Schweizer, B., “Elasto-gasdynamic Modeling of Air Foil Thrust
Bearings with a Two-dimensional Shell Model for Top and Bump Foil,” Tribology Interna-
tional, Vol. 100, 2016, pp. 48–59.
[109] Sim, K. and Kim, T. H., “Thermohydrodynamic Analysis of Bump-type Gas Foil Bearings
using Bump Thermal Contact and Inlet Flow Mixing Models,” Tribology International, Vol. 48,
2012, pp. 137–148.
[110] Salehi, M., Swanson, E., and Heshmat, H., “Thermal Features of Compliant Foil Bearings–
Theory and Experiments,” ASME Journal of Tribology, Vol. 123, No. 3, 2001, pp. 566–571.
References 199
[111] Kim, D., Ki, J., Kim, Y., and Ahn, K., “Extended Three-Dimensional Thermo-Hydrodynamic
Model of Radial Foil Bearing: Case Studies on Thermal Behaviors and Dynamic Character-
istics in Gas Turbine Simulator,” ASME Journal of Engineering for Gas Turbines and Power,
Vol. 134, No. 5, May 2012, pp. 052501.
[112] Shrestha, S. K., Kim, D., and Kim, Y. C., “Experimental Feasibility Study of Radial Injection
Cooling of Three-Pad Air Foil Bearings,” ASME Journal of Tribology, Vol. 135, No. 4, 2013,
pp. 041703.
[113] Lee, D. and Kim, D., “Three-dimensional Thermohydrodynamic Analyses of Rayleigh Step
Air Foil Thrust Bearing with Radially Arranged Bump Foils,” Tribology Transactions, Vol. 54,
No. 3, 2011, pp. 432–448.
[114] DellaCorte, C., Lukaszewicz, V., Valco, M., Radil, K., and Heshmat, H., “Performance and
durability of high temperature foil air bearings for oil-free turbomachinery,” Tribology trans-
actions, Vol. 43, No. 4, 2000, pp. 774–780.
[115] San Andrés, L. and Kim, T. H., “Thermohydrodynamic Analysis of Bump-type Gas Foil Bear-
ings: a Model Anchored to Test Data,” ASME Journal of Engineering for Gas Turbines and
Power, Vol. 132, No. 4, 2010, pp. 042504.
[116] Lee, D. and Kim, D., “Thermohydrodynamic Analyses of Bump Air Foil Bearings with De-
tailed Thermal Model of Foil Structures and Rotor,” ASME Journal of Tribology, Vol. 132,
No. 2, 2010, pp. 021704.
[117] Gollan, R. and Jacobs, P., “About The Formulation, Verification and Validation of The Hyper-
sonic Flow Solver Eilmer,” International Journal for Numerical Methods in Fluids, Vol. 73,
No. 1, September 2013, pp. 19–57.
[118] Jacobs, P., Gollan, R., Jahn, I., and Potter, D., “The Eilmer3 code: user guide and example book
2015 edition,” Technical report 2015/07, The University of Queensland, Brisbane, Australia,
August 2015.
[119] Ventura, C., Sauret, E., Jacobs, P., Petrie-Repar, P., and der Laan, P. V., “Adaption and Use of
a Compressible Flow Code for Turbomachinery Design,” 5th European Conference on Com-
200 References
putational Fluid Dynamics ECCOMAS CFD 2010, European Community on Computational

Methods in Applied Sciences, Lisbon, Portugal, 2010.
[120] Lemmon, E., Huber, M., and McLinden, M., “NIST Standard Reference Database 23: Refer-
ence Fluid Thermodynamic and Transport Properties-REFPROP, Version 9.1,” May 2013.
[121] Jacobs, P., Gollan, R., Denman, A., O’Flaherty, B., Potter, D., Petrie-Repar, P., and Johnston, I.,
“Eilmer’s Theory Book: Basic Models for Gas Dynamics and Thermochemistry,” Tech. Rep.
2010/09, The University of Queensland, 2012.
[122] Larignon, B. D., Marr, K., and Goldstein, D. B., “Monte Carlo and Navier-Stokes Simulations
of Compressible Taylor-Couette Flow,” Journal of thermophysics and heat transfer, Vol. 20,
No. 3, 2006, pp. 544–551.
[123] Schlichting, H., Boundary-layer Theory, McGraw-Hill, New York, seventh ed., 1979.
[124] DiPrima, R., Eagles, P., and Ng, B., “The Effect of Radius Ratio on the Stability of Couette
Flow and Taylor Vortex Flow,” Physics of Fluids (1958-1988), Vol. 27, No. 10, 1984, pp. 2403–
2411.
[125] Jones, C., “The Transition to Wavy Taylor Vortices,” Journal of Fluid Mechanics, Vol. 157,
1985, pp. 135–162.
[126] Wereley, S. and Lueptow, R., “Azimuthal Velocity in Supercritical Circular Couette Flow,”
Experiments in fluids, Vol. 18, No. 1-2, 1994, pp. 1–9.
[127] Garratt, J. E., Hibberd, S., Cliffe, K. A., and Power, H., “Centrifugal Inertia Effects in High-
speed Hydrostatic Air Thrust Bearings,” Journal of Engineering Mathematics, Vol. 76, No. 1,
October 2012, pp. 59–80.
[128] Zhao, H., Choy, F., and Braun, M., “Modeling and Analysis of a Wavy Thrust Bearing,” Tri-
bology Transactions, Vol. 45, No. 1, 2002, pp. 85–93.
[129] Peng, Z.-C. and Khonsari, M., “Hydrodynamic Analysis of Compliant Foil Bearings with Com-
pressible Air Flow,” ASME Journal of Tribology, Vol. 126, No. 3, 2004, pp. 542–546.
References 201
[130] Brunetiere, N., Tournerie, B., and Frene, J., “Influence of Fluid Flow Regime on Performances
of Non-Contacting Liquid Face Seals,” ASME Journal of Tribology, Vol. 124, No. 3, May 2002,
pp. 515–523.
[131] Souchet, D., Comportement Thermohydrodynamique des butées á Patins Oscillants en régime
Laminaire et Turbulent, PhD Thesis, University of Poitiers, Poitiers, France, 1991.
[132] Wilcox, D., Turbulence Modeling for CFD, DCW Industries, CA, 2nd ed., 2006.
[133] Chan, W., Jacobs, P., and Mee, D., “Suitability of the k − w turbulence model for scramjet
flowfiled simulations,” International Journal for Numerical Methods in Fluids, Vol. 70, No. 4,
October 2011, pp. 493–514.
[134] EL Telbany, M. and Reynolds, A., “Velocity Distribution in Plane Turbulent Channel Flows,”
Journal of Fluid Mechanics, Vol. 100, 1980, pp. 1–29.
[135] EL Telbany, M. and Reynolds, A., “Turbulence in Plane Channel Flows,” Journal of Fluid
Mechanics, Vol. 111, 1981, pp. 283–318.
[136] White, F., Viscous Fluid Flow, McGraw-Hill, New York, 3rd ed., 2006.
[137] Nichols, R. H. and Nelson, C., “Wall Function Boundary Conditions Including Heat Transfer
and Compressibility,” AIAA Journal, Vol. 42, No. 6, 2004, pp. 1107–1114.
[138] Davidson, L. and Farhanieh, B., “CALC-BFC: A Finite Volume Code Employing Collocated
Variable Arrangement and Cartesian Velocity Components for Computation of Fluid Flow and
Heat Transfer in Complex Three-dimensional Geometries,” Technical report 91/14, Chalmers
University of Technology, Sweden, 1991.
[139] Weller, H., Tabor, G., Jasak, H., and Fureby, C., “A Tensorial Approach to Computational Con-
tinuum Mechanics using Object-oriented Techniques,” Computers in Physics, Vol. 12, No. 6,
November 1998, pp. 620–631.
[140] Gao, Z., Jiang, C., and Lee, C., “Improvement and Application of Wall Function Bound-
ary Condition for High-speed Compressible Flows,” Science China Technological Sciences,
Vol. 56, No. 10, 2013, pp. 2501–2515.
202 References
[141] Veiser, W., Esch, T., and Menter, F., “Heat Transfer Predictions Using Advanced Two-Equation
Turbulence Models,” Technical report CFX-VAL10/0602, CFX Technical Memorandum, ON,
Canada, November 2002.
[142] Van Driest, E. R., “On Turbulent Flow Near a Wall,” Journal of the Aeronautical Sciences,
Vol. 23, No. 11, 1956, pp. 1007–1011.
[143] Dawes, W., “A Computer Program for the Analysis of Three Dimensional Viscous Compress-
ible Flow in Turbomachinery Blade Rows,” Technical report, Whittle Laboratory, University
of Cambridge, UK, 1988.
[144] Macrossan, M. N., “FCT Filter: Using The Diffusion And Anti-Diffusion Equations As A
Filter,” Technical report 12/00, The University of Queensland, December 2000.
[145] Jameson, A., Schmidt, W., and Turkel, E., “Numerical Solutions of the Euler Equations by
Finite Volume Methods using Runge-Kutta Time-stepping Schemes,” 14th Fluid and Plasma
Dynamics Conference, American Institute of Aeronautics and Astronautics, Palo Alto, CA,
June 1981.
[146] Zhang, S., Yu, A., and Zulli, P., “Preconditioned Multigrid Method for Fluid Flow in Por-
ous Media,” Second International Conference on CFD in the Minerals & Process Industries,
CSIRO, Melbourne, Australia, June 1999.
[147] Goncalves, E. and Houdeville, R., “Reassessment of the Wall Functions Approach for RANS
Computations,” Aerospace Science and Technology, Vol. 5, No. 1, 2001, pp. 1–14.
[148] Wang, J., Gollan, R., and Veeraragavan, A., “Verification of RANS Turbulence Model in Eilmer
using the Method of Manufactured Solutions,” Proceedings of the 20th Australasian Fluid
Mechanics Conference, Australasian Fluid Mechanics Society, Perth, WA, Australia, Decem-
ber 2016, pp. 1–4.
[149] Petrie-Repar, P., Numerical Simulation of Diaphragm Rupture, PhD Thesis, University of
Queensland, Brisbane, QLD, December 1997.
[150] Johnston, I., Simulation of Flow Around Hyersonic Blunt-Nosed Vehicles for the Calibration
of Air Data System, PhD Thesis, University of Queensland, Brisbane, QLD, January 1999.
References 203
[151] Liou, M. and Steffen, C., “A New Flux Splitting Scheme,” Journal of Computational Physics,
Vol. 107, No. 1, July 1993, pp. 23–39.
[152] Wada, Y. and Liou, M., “An Accurate and Robust Flux Splitting Scheme for Shock and Contact
Discontinuities,” Journal of Scientific Computing, Vol. 18, No. 3, May 1997, pp. 633–657.
[153] Macrossan, M., “The Equilibrium Flux Method for The Calculation of Flows with Non-
equilibrium Chemical Reactions,” Journal of Computational Physics, Vol. 80, No. 1, January
1989, pp. 204–231.
[154] Ambrosi, D., “Full Potential and Euler Solutions for Transonic Unsteady Flow,” Aeronautical
Journal, Vol. 98, No. 1, pp. 340–348.
[155] Grandy, J., “Efficient Computation of Volume of Hexahedral Cells,” Technical report UCRL-
ID-128886, Lawrence Livermore National Laboratory, October 1997.
[156] Trepanier, J., Reggio, M., Paraschivoiu, M., and Camarero, R., “Unsteady Euler Solutions for
Arbitrarily Moving Bodies and Boundaries,” AIAA Journal, Vol. 31, No. 10, October 1993,
pp. 1869–1876.
[157] Trepanier, J., Reggio, M., Zhang, H., and Camarero, R., “A Finite-Volume Method for The
Euler Equation on Arbitrary Langrangian-Eulerian Grids,” Computers and Fluids, Vol. 20,
No. 4, 1991, pp. 399–409.
[158] Sod, G. A., “A Survey of Several Finite Difference Methods for Systems of Nonlinear Hyper-
bolic Conservation Laws,” Journal of Computational Physics, Vol. 27, No. 1, 1978, pp. 1–31.
[159] Gollan, R., Johnston, I., O’Flaherty, B., and Jacobs, P., “Development of Casbar: a Two-
phase Flow Code for the Interior Ballistics Problem,” Proceedings of the 16th Australasian
Fluid Mechanics Conference, edited by P. Jacobs, T. McIntyre, M. Cleary, D. Buttsworth,
D. Mee, R. Clements, R. Morgan, and C. Lemckert, The University of Queensland, Gold Coast,
Queensland, Australia, December 2007, pp. 295–302.
[160] Jacobs, P., “L1d: A Computer Program for the Simulation of Transient-Flow Facilities,” Tech-
nical report 1/99, The University of Queensland, January 1999.
204 References
[161] AGARD, “Compendium of Unsteady Aerodynamic Measurements,” Technical report

AGARD-R-702, 1982.
[162] Jahangirian, A. and Hadidoolabi, M., “Unstructured Moving Grids for Implicit Calculation
of Unsteady Compressible Viscous Flows,” International Journal for Numerical Methods in
Fluids, Vol. 47, No. 10, 2005, pp. 1107–1113.
[163] Barakos, G. and Drikakis, D., “An Implicit Unfactored Method for Unsteady Turbulent Com-
pressible Flows with Moving Boundaries,” Computers & Fluids, Vol. 28, No. 8, November
1999, pp. 899–922.
[164] Timoshenko, S. and Woinowsky-Krieger, S., Theory of Plates and Shells, McGraw-Hill, New
York, 5th ed., 1959.
[165] Reddy, J., Theory and Analysis of Elastic Plates and Shells, Taylor and Francis, Philadelphia,
PA, 1st ed., 2007.
[166] ANSYS, Inc, ANSYS Mechanical User’s Guide, 15th ed., 2014.
[167] Joshi, O. and Leyland, P., “Stability Analysis of a Partitioned Fluid–Structure Thermal Coup-
ling Algorithm,” AIAA Journal of Thermophysics and Heat Transfer, Vol. 28, No. 1, January
2014, pp. 59–67.
[168] Span, R. and Wagner, W., “A New Equation of State for Carbon Dioxide Covering the Fluid
Region from the Triple Point Temperature to 1100 K at Pressures up to 800 MPa,” Journal of
Physical and Chemical Reference Data, Vol. 25, No. 6, 1996, pp. 1509–1596.
[169] Pinkus, O. and Lund, J., “Centrifugal Effects in Thrust Bearings and Seals under Laminar Con-
ditions,” ASME Journal of Lubrication Technology, Vol. 103, No. 1, Januaray 1981, pp. 126–
136.
[170] Bruckner, R. J., “Passive Thermal Management of Foil Bearings,” June 2015, US Patent
9,062,712.
[171] Qin, K., Jahn, I., and Jacobs, P., “Effect of Operating Conditions on the Elastohydrodynamic
Performance of Foil Thrust Bearings for Supercritical CO2 Cycles,” ASME Journal of Engin-
eering for Gas Turbines and Power, Vol. 139, No. 4, November 2017, pp. 042505.
References 205
[172] Qin, K., Jahn, I., Gollan, R., and Jacobs, P., “Development of a Computational Tool to Simulate
Foil Bearings for Supercritical CO2 Cycles,” ASME Journal of Engineering for Gas Turbines
and Power, Vol. 138, No. 9, March 2016, pp. 092503.
[173] Rubio, D. and San Andrés, L., “Bump-type Foil Bearing Structural Stiffness: Experiments
and Predictions,” Journal of engineering for gas turbines and power, Vol. 128, No. 3, 2006,
pp. 653–660.
[174] Childs, D. W., Turbomachinery Rotordynamics: Phenomena, Modeling, and Analysis, John
Wiley & Sons, New Jersey, 1st ed., 1993.
[175] Yan, X., He, K., Li, J., and Feng, Z., “A Generalized Prediction Method for Rotordynamic
Coefficients of Annular Gas Seals,” ASME Journal of Engineering for Gas Turbines and Power,
Vol. 137, No. 9, 2015, pp. 092506.
[176] Küttler, U. and Wall, W. A., “Fixed-point Fluid-structure Interaction Solvers with Dynamic
Relaxation,” Computational Mechanics, Vol. 43, No. 1, 2008, pp. 61–72.
[177] Malatip, A., Wansophark, N., and Dechaumphai, P., “Combined Strealine Upwind Petro Galer-
kin Method and Segregated Finite Element Algorithm for Conjugate Heat Transfer Problems,”
Journal of Mechanical Science and Technology, Vol. 20, No. 10, 2006, pp. 1741–1752.
[178] Back, L., Massier, P., and Gier, H., “Convective Heat Trasnfer in a Convergent-Divergent
Nozzle,” International Journal of Heat and Mass Transfer, Vol. 7, 1964, pp. 549–568.
[179] DeLise, J. and Naraghi, M., “Comparative Studies of Convective Heat Transfer Models
for Rocket Engines,” 31st AIAA/ASME/SAE/ASEE Joint Propulsion Conference and Exhibit,
American Institute of Aeronautics and Astronautics, San Deigo, CA, July 1995.
[180] Marineau, E., Schetz, J., and Neel, R., “Turbulent Navier-Stokes Simulations of Heat Trans-
fer with Complex Wall Temperature Variations,” 9th AIAA/ASME Joint Thermophysics and
Heat Transfer Conference, American Institute of Aeronautics and Astronautics, San Francisco,
California, June 2006.
206 References
[181] Liu, Q., Luke, E., and Cinnella, P., “Coupling Heat Transfer and Fluid Flow Solvers for Mul-
tidisciplinary Simulations,” AIAA Journal of Thermophysics and Heat Transfer, Vol. 19, No. 4,
0ctober-December 2005, pp. 417–427.
[182] aus der Wiesche, S. and Helcig, C., Convective Heat Transfer From Rotating Disks Subjected
To Streams Of Air, Springer, 1st ed., 2016.
[183] Holman, J., Heat Transfer, McGraw Hill Higher Education, tenth ed., 2010.
[184] Bruckner, R., “The Chevron Foil Thrust Bearing: Improved Performance Through Passive
Thermal Management and Effective Lubricant Mixing,” World Tribology Congress, Torino,
Italy, September 2013.
[185] Jahn, I. and Qin, K., “e3prepToFoam: a Mesh Generator for OpenFOAM,” Technical report
2015/04, The University of Queensland, June 2015.
[186] Ryu, K. and San Andrés, L., “On the Failure of a Gas Foil Bearing: High Temperature Op-
eration without Cooling Flow,” ASME Journal of Engineering for Gas Turbines and Power,
Vol. 135, No. 11, 2013, pp. 112506.
[187] San Andrés, L., Ryu, K., and Diemer, P., “Prediction of Gas Thrust Foil Bearing Performance
for Oil-Free Automotive Turbochargers,” ASME Journal of Engineering for Gas Turbines and
Power, Vol. 137, No. 3, 2015, pp. 032502.
Appendix A
Discretisation of Reynolds equation
The generalized Reynolds equation is given by [8],
∂ p 1 ∂p 1 ∂ p ∂(ρ h)
(ρ h3 ) + ρ h3 + 2 (ρ h3 ) = 6 η ω (A.1)
∂r r r ∂r r ∂θ θ ∂θ
µ0
where η = p0
( Rh22 )2 . The above equation can be expanded to:
• ∂
∂r
(ρ h3 pr )
1 ∂p
• r
ρ h3 ∂r
• 1 ∂
r 2 ∂θ
(ρ h3 θp )
∂(ρ h)
• 6ηω ∂θ
Each part can be discretized as follows,
∂ p ∂ 2 p p ∂(ρ h3 )
(ρ h3 ) = ρ h3 2 +
∂r r ∂r r ∂r
2
∂ p p ∂ρ h
= ρ h3 2 + (h3 + 3 h2 ρ )
∂r r ∂r r
2
∂ p ∂p
= a1 2 + a2 (A.2)
∂r ∂r
1 3 ∂p ∂p
ρh = a3 (A.3)
r ∂r ∂r
208 Appendix A Discretisation of Reynolds equation
2
1 ∂ 3 p 1 3 ∂ p p ∂(ρ h3 )
(ρ h ) = (ρ h + )
r 2 ∂θ θ r2 ∂θ2 θ ∂θ
1 ∂2p ∂p 3 ρ ∂h
= 2 (ρ h3 2 + (h + 3 h2 ρ ))
r ∂θ ∂θ ∂θ ∂θ
∂2p p
= a4 2 + a5 (A.4)
∂θ θ
∂(ρ h) ∂ρ ∂h
6ηω = 6 η ω (h + ρ )
∂θ ∂θ ∂θ
= a6 (A.5)
where ,
a1 = ρ h3 (A.6)
∂ρ h
a2 = h3 + 3 h2 ρ (A.7)
∂r r
1 3
a3 = ρh (A.8)
r
1
a4 = 2 ρ h3 (A.9)
r
1 ∂ρ ∂h
a5 = 2 (h3 + 3 h2 ρ ) (A.10)
r ∂θ ∂θ
∂ρ ∂h
a6 = 6 η ω (h +ρ ) (A.11)
∂θ ∂θ
The generalized Reynolds equation is,
∂2p ∂p ∂2p ∂p
a1 + (a2 + a3 ) + a4 + a5 = a6 (A.12)
∂r 2 ∂r ∂θ2 ∂θ
Using the central differencing method, and the following discretise partial derivatives,
∂2p p(i,j+1) + p(i,j−1) − 2 + p(i,j)

= , (A.13)
∂r 2 ∆r 2
∂p p(i,j+1) − p(i,j−1)
= , (A.14)
∂r 2∆r
∂2p p(i+1,j) + p(i+1,j) − 2 + p(i,j)
= , (A.15)
∂θ2 θr 2
∂p p(i+1,j) − p(i−1,j)
= . (A.16)
∂θ 2∆θ
Discretisation of Reynolds equation Appendix A 209
the discretised Reynolds equation tales the form,
A1 p(i,j) = A2 p(i,j) + A3 p(i,j−1) + A4 p(i+1,j) + A5 p(i−1,j) + A6 (A.17)
where,
2 a1 2 a4
A1 = + , (A.18)
∆r 2 ∆θ2
a1 a2 + a3
A2 = + , (A.19)
∆r 2 2 ∆r
a1 a2 + a3
A3 = − , (A.20)
∆r 2 2 ∆r
a4 a5
A4 = + , (A.21)
∆θ2 2 ∆θ
a4 a5
A5 = − , (A.22)
∆θ2 2 ∆θ
A6 = −a6 . (A.23)
This equation can be solved over a grid in the r and θ space using the Gauss-Seidel iteration method.
210 Appendix A Discretisation of Reynolds equation
Appendix B
Description of Eilmer
The compressible Navier-Stokes equations implemented in Eilmer can be expressed as [117],
∂
Z I Z
UdV = − (Fi − Fv ) · n̂dA + QdV . (B.1)
∂t V S V
For the thermal equilibrium models with a single-species, the conserved quantities U are defined as,
 
ρ
 
 ρux 
 
 
U =  ρuy  . (B.2)
 
 
 ρuz 
 
 
ρE
The convective flux Fi consists of mass flux G, momentum flux L and energy flux H,
   
G ρu · n̂
   
Fi =  L  =  ρuu · n̂ + pn̂  . (B.3)
   
   
H ρEu · n̂ + pu · n̂
212 Appendix B Description of Eilmer
The components of the viscous flux Fv are,

 
0
 
τxx
 
 
 
Fv =  τyx  î +
 
 
τzx
 
 
 
τxx ux + τyx uy + τzx uz + qx
 
0
 
τxy
 
 
 
τyy  ĵ +
 

 
τzy
 
 
 
τxy ux + τyy uy + τzy uz + qy
 
0
 
τxz
 
 
 
τyz  k̂ . (B.4)
 

 
τzz
 
 
 
τxz ux + τyz uy + τzz uz + qz
In three dimensions, the finite-volume cells are hexahedral, with 6 (possibly-nonplanar) quadri-
lateral surfaces interfacing to neighbouring cells. Flux values are estimated at midpoints of the cell
interfaces and the integral conservation Equation B.1 is approximated by the algebraic expression,
dU 1 X
=− (Fi − Fv ) · n̂ dA + Q . (B.5)
dt V cell−surf ace
where U and Q now represent cell-average values.

The convective flux vector is estimated using a reconstruction-evolution approach. The viscous
flux calculation is based on an estimate of spatial derivatives of flow quantities at the midpoint of
cell-interfaces,
1
I
∇φ = φ n̂ dA . (B.6)
V S
Here, the surface S defines a secondary cell surrounding a primary-cell vertex, as shown in Ref. [117].
The discretised equations are integrated in a time-accurate manner using one of several explicit
Description of Eilmer Appendix B 213
update schemes: Euler, predictor-corrector or a 3-stage Runge-Kutta update. Within each stage of the
selected time-integration process, Eilmer uses an operator-splitting approach, with the contributions
from the convective and viscous transport processes being computed sequentially.
214 Appendix B Description of Eilmer
Appendix C
Source Code
C.1 Moving Wall Boundary Condition
110 c a s e EAST :
111 i = bd . imax ;
112 f o r ( k = bd . kmin ; k <= bd . kmax ; ++k ) {
113 f o r ( j = bd . j m i n ; j <= bd . jmax ; ++ j ) {
114 c e l l = bd . g e t c e l l ( i , j , k ) ;
115 I F a c e = c e l l −>i f a c e [ EAST ] ;
116 F l o w S t a t e &f s = * ( I F a c e −>f s ) ;
117 f s . c o p y v a l u e s f r o m ( * ( c e l l −>f s ) ) ;
118 f s . v e l = c r o s s ( r o m eg a , I F a c e −>p o s − c e n t r e ) + v t r a n s ;
119 if ( Twall flag )
120 {
121 f o r ( s i z e t imode = 0 ; imode < nmodes ; ++ imode ) f s . g as−>T [
imode ] = T w a l l ;
122 }
123 fs . tke = 0.0;
124 f s . omega = i d e a l o m e g a a t w a l l ( c e l l ) ;
125 i f ( bd . bcp [ EAST]−>wc bc ! = NON CATALYTIC) {
126 cw−>a pply ( * ( c e l l −>f s −>g a s ) , f s . g as−>m a s s f ) ;
127 }
128 } / / end j l o o p
129 } / / for k
130 break ;
216 Appendix C Source Code
C.2 Periodic Boundary Condition
129 c a s e BOTTOM:
130 k = bd . kmin ;
131 f o r ( i = bd . i m i n ; i <= bd . imax ; ++ i ) {
132 f o r ( j = bd . j m i n ; j <= bd . jmax ; ++ j ) {
133 r e o r i e n t v e c t o r q u a n t i t i e s i n c e l l ( bd . g e t c e l l ( i , j , k −1) , R m a t r i x ) ; / / g h o s t
cell 1.
134 r e o r i e n t v e c t o r q u a n t i t i e s i n c e l l ( bd . g e t c e l l ( i , j , k −2) , R m a t r i x ) ; / / g h o s t
cell 2.
135 } / / end j l o o p
136 } / / for i
137 } / / end s w i t c h . . .
138
139 r e t u r n SUCCESS ;
140 }
141
142 / / Helper f u n c t i o n s
143 v o i d a p p l y m a t r i x t r a n s f o r m ( c o n s t s t d : : v e c t o r <d o u b l e>& R m a t r i x ,
144 c o n s t s t d : : v e c t o r <d o u b l e>& o l d v ,
145 s t d : : v e c t o r <d o u b l e > &newv )
146 {
147 / / W r i t e o u t t h e m a t r i x m u l t i p l i c a t i o n , lo ng−h an d .
148 newv [ 0 ] = R m a t r i x [ 0 ] * o l d v [ 0 ] + R m a t r i x [ 1 ] * o l d v [ 1 ] + R m a t r i x [ 2 ] * o l d v [ 2 ] ;
151 }
152
153 v o i d r e o r i e n t v e c t o r q u a n t i t i e s i n c e l l ( F V C e l l * c , c o n s t s t d : : v e c t o r <d o u b l e>&
Rmatrix )
154 {
155 g l o b a l d a t a &G = * g e t g l o b a l d a t a p t r ( ) ;
156 s t d : : v e c t o r <d o u b l e >o l d v ( 3 ) ;
157 s t d : : v e c t o r <d o u b l e >newv ( 3 ) ;
158 o l d v [ 0 ] = c−>f s −>v e l . x ; o l d v [ 1 ] = c−>f s −>v e l . y ; o l d v [ 2 ] = c−>f s −>v e l . z ;
Source Code Appendix C 217
159 a p p l y m a t r i x t r a n s f o r m ( R m a t r i x , o l d v , newv ) ;
160 c−>f s −>v e l . x = newv [ 0 ] ; c−>f s −>v e l . y = newv [ 1 ] ; c−>f s −>v e l . z = newv [ 2 ] ;
161 i f ( G .MHD ) {
162 o l d v [ 0 ] = c−>f s −>B . x ; o l d v [ 1 ] = c−>f s −>B . y ; o l d v [ 2 ] = c−>f s −>B . z ;
163 a p p l y m a t r i x t r a n s f o r m ( R m a t r i x , o l d v , newv ) ;
164 c−>f s −>B . x = newv [ 0 ] ; c−>f s −>B . y = newv [ 1 ] ; c−>f s −>B . z = newv [ 2 ] ;
165 }
166 }
C.3 Reconstruction
260 if ( interpolate in local frame ) {

261 / / P a u l P e t r i e −Rep ar and J a s o n Qin h a v e n o t e d t h a t t h e v e l o c i t y n e e d s
262 / / t o be r e c o n s t r u c t e d i n t h e i n t e r f a c e −l o c a l f r a m e o f r e f e r e n c e s o t h a t
263 / / t h e n o r m a l v e l o c i t i e s a r e n o t m essed up f o r m i r r o r −im ag e a t w a l l s .
264 / / PJ 21− f e b −2014
265 cL1 . f s −>v e l . t r a n s f o r m t o l o c a l ( I F a c e . n , I F a c e . t 1 , I F a c e . t 2 ) ;
266 cL0 . f s −>v e l . t r a n s f o r m t o l o c a l ( I F a c e . n , I F a c e . t 1 , I F a c e . t 2 ) ;
267 cR0 . f s −>v e l . t r a n s f o r m t o l o c a l ( I F a c e . n , I F a c e . t 1 , I F a c e . t 2 ) ;
268 cR1 . f s −>v e l . t r a n s f o r m t o l o c a l ( I F a c e . n , I F a c e . t 1 , I F a c e . t 2 ) ;
269 }
218 Appendix C Source Code
Appendix D
Description of Wilcox’s 2006 k-ω model
Mass Conservation:
∂ ρ̄ ∂
+ (ρ̄ũi ) = 0 (D.1)
∂t ∂xi
Momentum Conservation:
∂ ∂ ∂P ∂
(ρ̄ũi ) + (ρ̄ũj ũi ) = − + [t̄ji + ρ̄τji ] (D.2)
∂t ∂xj ∂xi ∂xj
Energy Conservation:

∂ ũi ũi ∂ ũi ũi
ρ̄ ẽ + +k + ρ̄ũj h̃ + +k =
∂t 2 ∂xj 2
" #
∂ µ µT ∂ h̃ ρ̄k ∂k ∂
+ + µ + σ∗ + [ũi (t̄ij + ρ̄τij )] (D.3)
∂xj P rL P rT ∂xj ω ∂xj ∂xj
Molecular and Reynolds-Stress Tensors:
2 1 ∂ ũk
t̄ij = 2µS̄ij ρ̄τij = 2µT S̄ij − ρ̄kδij S̄ij = Sij − δij (D.4)
3 3 ∂xk
Eddy Viscosity:
 s 
ρ̄k  2S̄ij S̄ij  7
µT = ω̃ = max ω , Clim Clim = (D.5)
ω̃  β∗  8
Turbulence Kinetic Energy (k):

∂ ∂ ∂ ũi ∂ ∗ ρ̄k ∂k
(ρ̄k) + (ρ̄ũj k) = ρ̄τij − β ∗ ρ̄kω + µ+σ (D.6)
∂t ∂xj ∂xj ∂xj ω ∂xj
220 Appendix D Description of Wilcox’s 2006 k-ω model
Specific Dissipation Rate (ω):

∂ ∂ ω ∂ ũi ρ̄ ∂k ∂ω ∂ ρ̄k ∂ω
(ρ̄ω) + (ρ̄ũj ω) = α ρ̄τij − β ρ̄ω 2 + σd + µ+σ (D.7)
∂t ∂xj k ∂xj ω ∂xj ∂xj ∂xj ω ∂xj
Closure Coefficients:
13 9 1 3 1
α= β = βo fβ β∗ = σ= σ∗ = σdo = (D.8)
25 100 2 5 8

 0, ∂k ∂ω
8 ∂xj ∂xj
≤0
βo = 0.0708 P rT = σd = (D.9)
9  σ , ∂k ∂ω > 0
do ∂xj ∂xj

1 + 85χω Ω Ω Ŝ
ij jk ki 1 ∂ ũm
fβ = χω = 3 Ŝki = Ski − δki (D.10)
1 + 100χω 2 ∂xm

(β ω)
∗

s4310658 Final Thesis

Uploaded by

Copyright:

Available Formats

s4310658 Final Thesis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

s4310658 Final Thesis

Uploaded by

Copyright:

Available Formats

Development and Application of Multiphysics Simulation Tools

for Foil Thrust Bearings Operating with Carbon Dioxide

A thesis submitted for the degree of Doctor of Philosophy at

The University of Queensland in 2017

School of Mechanical and Mining Engineering,

Contributor Statement of contribution

Contributor Statement of contribution

Contributor Statement of contribution

Contributor Statement of contribution

Problem specification (50 %)

Contributor Statement of contribution

Contributor Statement of contribution

Contribution by others to the thesis

Australian and New Zealand Standard Research Classifications (ANZSRC)

091307 Numerical Modelling and Mechanical Characterisation, 50%

Fields of Research (FoR) Classification (FOR)

0913 Mechanical Engineering, 100%

List of Figures xvii

List of Tables xxvi

4 Fluid-Structure Coupled Simulations 79

6 Foil Bearing Dynamic Performance 145

7 Fluid-Structure-Thermal Simulations 153

8 Conclusions and Future Work 181

A Discretisation of Reynolds equation 207

B Description of Eilmer 211

C Source Code 215

D Description of Wilcox’s 2006 k-ω model 219

1.1 Schematic of sCO2 turbo-alternator-compressor, taken from Ref. [8]. . . . . . . . . . 2

4.16 Grid convergence study for the pitching NACA64A010 airfoil. . . . . . . . . . . . . 96

4.18 Schematic diagram for forces and moments. . . . . . . . . . . . . . . . . . . . . . . 98

4.22 Structural damping coefficient, CB . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

5.1 Side view of foil thrust bearings, not to scale. . . . . . . . . . . . . . . . . . . . . . 110

5.4 Comparison of measurements and numerical simulation at fixed rotational speed of

5.6 Pressure-Enthalpy diagram for CO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

5.10 Comparison of non-dimensional pressure distribution at the medium radius between

5.14 The architecture of a supercritical CO2 turbomachinery system. . . . . . . . . . . . . 125

5.16 Rotational Reynolds number at different operating conditions (based on 5 µm initial

5.18 Pressure increase (p − pa ) at the different operating pressures. Operating temperature

6.1 Schematic diagram of rotor perturbation motion. . . . . . . . . . . . . . . . . . . . . 146

7.1 Schematic diagram for conjugate Couette flow. . . . . . . . . . . . . . . . . . . . . 156

Figure 1.1: Schematic of sCO2 turbo-alternator-compressor, taken from Ref. [8].

Figure 1.4: Tilting pad bearing, taken from Ref. [15].

Figure 1.5: Metal mesh bearing, taken from Ref. [21].

1.2 Aims and Objectives

1.3 Thesis Overview

Chapter 5 Foil Bearing Steady State Performance

2.2 Background on Supercritical CO2 Power Cycle

Figure 2.5: Photograph of the SNL sCO2 Brayton loop [42].

2.2.1 Components for Supercritical CO2 Cycles

2.3 Experimental Work on Foil Thrust Bearings

Parameters Magnitude Parameters Magnitude

2.4 Fluid Simulation for Gas Bearings

2.4.1 Classical Reynolds Equation

4. The body forces are negligible;

5. The flow is laminar;

7. The fluid is Newtonian;

8. Non-slip boundaries exist at the rotor and stator.

2.4.2 Turbulent Reynolds equation

2.4.3 Computational Fluid Dynamics