Loosely Coupled Time Integration of Fluid- Thermal-Structural Interactions in Hypersonic Flows
Dissertation
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University
By
Brent Adam Miller, M.S.
Graduate Program in Aeronautical and Astronautical Engineering
The Ohio State University 2015
Dissertation Committee: Jack J. McNamara, Advisor Thomas Eason III Datta Gaitonde Sandip Mazumder S. Michael Spottswood Manoj Srinivasan Copyright by
Brent Adam Miller
2015 Abstract
This dissertation describes time integration strategies for fluid-thermal-structural interactions in hypersonic and high speed (Mach 1) flow using separate solvers for each discipline.
The development of reusable hypersonic cruise vehicles requires analysis ca- pability that can capture the coupled, highly-nonlinear interactions between the
fluid flow, structural mechanics, and structural heat transfer. This coupled analy- sis must also be performed over significant portions of the flight trajectory due to the long-term thermal evolution of the vehicle. The fluid and structural physics operate at significantly smaller time scales than the thermal evolution, requiring time marching that can capture the small time scales for time records that encap- sulate the longer time scale of the thermal response. This leads to extreme compu- tational times and motivates research that seeks to maximize efficiency of the time integrations for the coupled problem.
The goal of this dissertation is to develop time integration procedures that sig- nificantly improve computational efficiency while also maintaining time accuracy
ii and stability for fluid-thermal-structural analysis. This is achieved using carefully designed loosely coupled schemes for the fluid, thermal, and structural solvers.
Here, different time integrators are used for the solvers of each physical field, and boundary conditions are exchanged at most once per time step. Coupling schemes for both time-accurate and quasi-steady flow models are considered. Computa- tional efficiency and time accuracy are improved through the use of both extrap- olating predictors and interpolation during the exchange of boundary conditions; the latter of which enables the use of different sized time steps between the solvers, known as subcycling.
The developed coupling procedures are compared to several other schemes, including a basic one that does not use the predictors, and a subiteration-based strongly coupled scheme. Response predictions of multiple configurations of a panel in two dimensional supersonic and hypersonic flow are performed. Using second order implicit time integrators for the individual solvers, the predictor- based and strongly coupled schemes are demonstrated to retain the second order accuracy with and without subcycling, while the others reduced to first order.
Simulations of two panel configurations are performed to investigate the per- formance of the coupling schemes in predicting a stable response for one config- uration, and an unstable flutter response for the other. From the stable response analysis, the predictor-based schemes are found to be the least computationally expensive compared to the strongly coupled and basic loosely coupled schemes.
In the second configuration, the panel is predicted to undergo snap-through and
iii ultimately a dynamically unstable limit cycle response with each scheme. The predictor-based and strongly coupled schemes converge to a consistent limit cy- cle at larger time steps compared to the others. The predictor-based schemes are shown to provide a 3-10 times reduction in computational cost compared to the other schemes.
Finally, a 30 second response of the panel with a flutter instability using time- accurate CFD is performed and compared to the response using quasi-steady sur- rogate aerothermodynamic and quasi-steady analytical aerothermodynamic mod- els. The unsteady CFD and surrogate based responses have excellent agreement throughout the response, with differences under 1%. The time to flutter predicted from the analytical aerothermodynamic models is 15% higher than that predicted by the CFD and surrogate models. However, the post-flutter response shows good agreement between the three responses, indicating that the quasi-steady assump- tion can accurately capture the dynamics of the flow response.
iv Dedication
To my family for their love and support.
v Acknowledgments
I would like to thank my advisor, Prof. Jack McNamara, for providing me with guidance and support throughout my graduate studies. Professor McNamara’s drive for high quality research in fluid-thermal-structural interactions has been vi- tal to my academic and professional growth, and to this work. Without his passion for the field and support for my development as a researcher, I surely would not have stayed to pursue a doctoral degree. I would also like to thank Profs. Datta
Gaitonde, Sandip Mazumder, and Manoj Srinivasan for serving on my dissertation and candidacy committees, and Drs. Thomas Eason and S. Michael Spottswood for serving on my dissertation committee.
I am deeply grateful for the financial support I have received for this work.
This research was performed from government awards from the AFRL-University
Collaborative Center in Structural Sciences (Cooperative Agreement FA8650-13-2-
2347), with Dr. Ravi Penmetsa as program manager; an HPCMPO Frontier PETTT
Project Grant, under the direction of David Bartoe; the DoD Science, Mathematics, and Research for Transformation (SMART) Scholarship for Service Program; and
vi the AFRL/DAGSI Ohio Student-Faculty Research Fellowship Program. I would also like to thank the Ohio Supercomputer Center for providing computational resources, and the NASA Langley Research Center for providing CFL3D.
Technical contributions to this work have been made by researchers at the US
Air Force Research Laboratory, as well as past and current members of Professor
McNamara’s research group at the university. I appreciate and thank Drs. Ravi
Chona, S. Michael Spottswood, and Tom Eason of the AFRL/RQSS Structural Sci- ences Center for their technical insights. I also gratefully acknowledge Drs. Adam
Culler, Andrew Crowell, and Abhijit Gogulapati for their significant contributions that enabled this work. I would like to thank Dr. Culler for providing the original quasi-steady fluid-thermal-structural models and motivating a great deal of this work; Dr. Crowell for providing me with the surrogate fluid models and helping me learn CFL3D; and Dr. Gogulapati for providing routines to use the surrogate models in my solvers and for general guidance.
I am very grateful to all the other graduate students that I have had the pleasure to work with in the office. Thank you Dr. Krista Kecskemety, Dr. Andrew Crowell,
Steve Nogar, Zach Riley, Rohit Deshmukh, Zach Witeof, Joe Connolly, Jonathen
LaFontaine, Kirk Brouwer, Kelsey Shaler, and everyone else for your support and friendship. Finally I would to thank my parents, Robert and Robyn Miller, and my siblings, Shawna Harvey and Bobby Miller, for always being there for me and encouraging me all the way. Without your continued support I would not have made it this far.
vii Vita
2009 ...... B.S. Aeronautical and Astronautical Engineering, The
Ohio State University
2010 ...... M.S. Aeronautical and Astronautical Engineering, The
Ohio State University
2009-2010 ...... DAGSI Fellowship, The Ohio State University
2010-2014 ...... SMART Scholarship, The Ohio State University
2014 - 2015 ...... Graduate Research Assistant, The Ohio State Univer-
sity
Publications
Journal Publications
Miller, B. A. and McNamara, J. J., ”Time Marching Considerations for Response
Prediction of Structures in Hypersonic Flows,” AIAA Journal, Accepted for publi- cation.
viii Miller, B. A., McNamara, J. J., Culler, A. J., and Spottswood, S. M., ”The Impact
of Flow Induced Loads on Snap-Through Behavior of Acoustically Excited, Ther-
mally Buckled Panels,” Journal of Sound and Vibration, Vol. 330, 2011, pg. 5736-5752.
Crowell, A. R., Miller, B. A., and McNamara, J. J, ”Robust and Efficient Treat-
ment of Temperature Feedback in Fluid-Thermal-Structural Analysis”, AIAA Jour-
nal, Vol. 52, No. 11, 2014, pg. 2395-2413.
Crowell, A. R., McNamara, J. J., and Miller, B. A., ”Hypersonic Aerothermoe-
lastic Response Prediction of Skin Panels Using Computational Fluid Dynamics
Surrogates,” Journal of Aeroelasticity and Structural Dynamics, Vol. 2, No. 2, June
2011, pg. 3-30.
Deshmukh, R., Culler, A. J., Miller, B. A., and McNamara, J. J., ”Response of
Skin Panels to Combined Self and Boundary Layer Induced Fluctuating Pressure,”
Journal of Fluids and Structures. In review.
Conference Publications
Miller, B. A., and McNamara, J. J., Loosely Coupled Time-Marching of Fluid-
Thermal-Structural Interactions with Time-Accurate CFD, 56th AIAA/ASCE/AHS/
ASCStructures, Structural Dynamics, and Materials Conference, January 5-9, Kissim- mee, FL, AIAA 2015-0686, 2015.
ix Miller, B. A. and McNamara, J. J., Efficient Time-Marching of Fluid-Thermal-
Structural Interactions, 55th AIAA/ASME/ASCE/AHS/ASCStructures, Structural Dy-
namics, and Materials Conference, January 13-17, National Harbor, MD, AIAA-2014-
0337, 2014.
LaFontaine, J. H., Gogulapati, A., Miller, B. A., and McNamara, J. J., Develop-
ment of a Framework for Study of Fluid-Thermal-Elastic-Plastic Interactions, 55th
AIAA/ASME/ASCE/AHS/ASCStructures, Structural Dynamics, and Materials Confer-
ence, January 13-17, National Harbor, MD, AIAA-2014-1516, 2014.
Miller, B. A., Crowell, A. R., and McNamara, J. J., Loosely Coupled Time-
Marching of Fluid-Thermal-Structural Interactions, 54th AIAA/ASME/ASCE/AHS/
ASCStructures, Structural Dynamics, and Materials Conference, April 8-11, Boston,
MA, AIAA-2013-1666, 2013.
Miller, B. A., Crowell, A. R., and McNamara, J. J., Modeling and Analysis
of Shock Impingements on Thermo-Mechanically Compliant Surface Panels, 53rd
AIAA/ASME/ASCE/AHS/ASCStructures, Structural Dynamics, and Materials Confer- ence, April 23-26, Honolulu, HI, AIAA-2012-1548, 2012.
Miller, B. A., McNamara, J. J., Culler, A. J., and Spottswood, S. M., The Impact
x of Flow Induced Loads on Snap-Through Behavior of Acoustically Excited, Ther- mally Buckled Panels, 51st AIAA/ASME/ASCE/AHS/ASCStructures, Structural Dy- namics, and Materials Conference, April 12-15, Orlando, FL, AIAA-2010-2540, 2010.
LaFontaine, J. H., Gogulapati, A., Miller, B. A., and McNamara, J. J., Effects of Strain Hardening on Fluid-Thermal-Structural Interactions, 56th AIAA/ASCE/
AHS/ASCStructures, Structural Dynamics, and Materials Conference, January 5-9, Kissim- mee, FL, AIAA 2015-1629, 2015.
LeVett, M. A., Liang, Z., Miller, B. A., and McNamara, J. J., Investigation into
Parallel Time Marching of Fluid-Thermal-Structural Interactions, 56th AIAA/ASCE/
AHS/ASCStructures, Structural Dynamics, and Materials Conference, January 5-9, Kissim- mee, FL, AIAA 2015-1632, 2015.
Riley, Z. B., Deshmukh, R., Miller, B. A., and McNamara, J. J., Characteriza- tion of Structural Response to Hypersonic Boundary Layer Transition, 56th AIAA/
ASCE/AHS/ASCStructures, Structural Dynamics, and Materials Conference, January
5-9, Kissimmee, FL, AIAA 2015-0688, 2015.
Crowell, A. R., Miller, B. A., and McNamara, J. J., Robust and Efficient Treat- ment of Temperature Feedback in Fluid-Thermal-Structural Analysis, 54th AIAA/
ASME/ASCE/AHS/ASCStructures, Structural Dynamics, and Materials Conference, April
xi 8-11, Boston, MA, AIAA 20143-1663, 2013.
McNamara, J. J., Crowell, A. R., Miller, B. A., Riley, Z. B., Deshmukh, R., and
Culler, A. J., Structural Response Prediction of Surface Panels in Hypersonic Flow,
International Forum on Aeroelasticity and Structural Dynamics, Bristol, UK, IFASD-
2013-29C, 2013.
Crowell, A. R., Miller, B. A., and McNamara, A. J., Computational Modeling for Conjugate Heat Transfer of Shock-Surface Interactions on Compliant Skin Pan- els, 52nd AIAA/ASME/ASCE/AHS/ASCStructures, Structural Dynamics, and Materi- als Conference, April 4-7, Denver, CO, AIAA 2011-2017, 2011.
Crowell, A. R., McNamara, J. J., and Miller, B. A., Surrogate Based Reduced-
Order Aerothermodynamic Modeling for Structural Response Prediction at High
Mach Numbers, 52nd AIAA/ASME/ASCE/AHS/ASCStructures, Structural Dynamics, and Materials Conference, April 4-7, Denver, CO, AIAA 2011-2014, 2011.
Fields of Study
Major Field: Aeronautical and Astronautical Engineering
xii Table of Contents
Abstract ...... ii
Dedication ...... v
Acknowledgments ...... vi
Vita ...... viii
List of Tables ...... xvii
List of Figures ...... xix
List of Symbols ...... xxiv
Chapter
1 Introduction and Objectives ...... 1 1.1 Introduction ...... 1 1.2 Literature Review ...... 6 1.2.1 Time Integration Frameworks for Multi-Physics Problems 6 1.2.2 Time Integration of Fluid-Structure Interactions ...... 10 1.2.3 Time Integration of Fluid-Thermal Interactions ...... 13 1.2.4 Time Integration of Fluid-Thermal-Structural Interactions 15 1.2.5 Summary of the State of the Art for Fluid-Thermal-Structural Interactions ...... 18
xiii 1.3 Objectives of this Dissertation ...... 18 1.4 Key Novel Contributions of this Dissertation ...... 20
2 Governing Equations and Time Discretization ...... 22 2.1 Governing Equations ...... 22 2.2 Fluid Time Discretization ...... 27 2.2.1 Quasi-Steady Fluid Modeling ...... 28 2.2.2 Geometric Conservation Law ...... 29 2.2.3 Time-Accurate Fluid Modeling ...... 31 2.3 Structural Time Discretization ...... 33 2.3.1 Newmark-β Time Integration ...... 34 2.3.2 Fourth Order Runge Kutta Time Integration ...... 36 2.4 Thermal Discretization ...... 38 2.4.1 Crank-Nicolson Time Integration ...... 38 2.4.2 Forward Euler Time Integration ...... 40
3 Coupling Procedures ...... 41 3.1 Quasi-Steady Aerothermodynamics ...... 42 3.1.1 Predictor Implicit Scheme ...... 44 3.1.2 Predictor-Corrector Implicit Scheme ...... 49 3.1.3 Additional Quasi-Steady Coupling Schemes ...... 50 3.1.3.1 Basic Implicit Scheme ...... 50 3.1.3.2 Conventional Explicit Scheme ...... 52 3.1.3.3 Strong Implicit Scheme ...... 55 3.2 Time-Accurate Aerothermodynamics ...... 57 3.2.1 Predictor Implicit Scheme ...... 59 3.2.2 Predictor-Corrector Implicit Scheme ...... 64 3.2.3 Additional Coupling Schemes for Time-Accurate Aerother- modynamics ...... 65
xiv 3.2.3.1 Basic Implicit Scheme ...... 66 3.2.3.2 Strong Implicit Scheme ...... 67 3.3 Summary of Schemes ...... 69
4 Configurations and Computational Models ...... 71 4.1 Quasi-Steady Analytical Fluid Modeling ...... 73 4.1.1 Piston Theory ...... 74 4.1.2 Eckert’s Reference Enthalpy ...... 74 4.2 Quasi-Steady CFD-Based Surrogate Fluid Modeling ...... 77 4.3 Time-Accurate CFD Fluid Modeling ...... 77 4.4 Thermal Modeling ...... 79 4.5 Structural Modeling ...... 82 4.6 Summary of Configurations ...... 87
5 Order of Accuracy Analysis ...... 88 5.1 Quasi-Steady Aerothermodynamics ...... 89 5.1.1 Without Subcycling ...... 90 5.1.2 Structural Subcycling ...... 93 5.1.3 Summary of Accuracy Analysis ...... 96 5.2 Time-Accurate Aerothermodynamics ...... 100 5.2.1 Without Subcycling ...... 101 5.2.2 Structural Subcycling ...... 103 5.2.3 Fluid Subcycling ...... 103 5.2.4 Fluid and Structural Subcycling ...... 106 5.2.5 Summary of Accuracy Analysis ...... 109
6 Fluid-Thermal-Structural Response: Quasi-Steady Aerothermodynam- ics ...... 111 6.1 Response Study: Configuration 1 ...... 111 6.2 Response Study: Configuration 2 ...... 118
xv 6.3 Computational Expense ...... 126
7 Fluid-Thermal-Structural Response: Time Accurate Aerothermodynam- ics ...... 130 7.1 Initial Stable Response ...... 131 7.2 Flutter Response with Elevated Initial Temperature ...... 135 7.2.1 Initialization: Static Aeroelastic Equilibrium ...... 136 7.2.2 Flutter Analysis ...... 141 7.2.3 Fluid-Structure Work Balance ...... 146 7.3 Long Time Record Analysis ...... 148 7.4 Computational Cost ...... 155
8 Concluding Remarks ...... 161 8.1 Principal Conclusions Obtained in This Study ...... 161 8.2 Recommendations for Future Research ...... 167
Bibliography ...... 171
Appendix
A Least Squares Based Linear Fitting ...... 182
B Temperature-Dependent Material Properties ...... 185
C CFD Mesh Convergence Study ...... 189
D Thermal Solver Verification ...... 193
E Structural Solver Verification ...... 195
xvi List of Tables
3.1 Summary of coupling schemes for quasi-steady aerothermodynam- ics...... 69 3.2 Summary of coupling schemes for time-accurate aerothermody- namics...... 70
4.1 Post-shock flight conditions for each configuration...... 72 4.2 Panel properties of each configuration...... 72 4.3 Rayleigh Damping parameters for each configuration...... 86 4.4 Summary of configurations for quasi-steady aerothermodynamics. 87 4.5 Summary of Configuration 2 for time-accurate CFD...... 87
5.1 Computed order of accuracy at smallest time step for Configura- tion 1...... 99 5.2 Computed order of accuracy at smallest time step for Configura- tion 2...... 100 5.3 Computed orders of accuracies with time-accurate aerothermody- namics...... 110
6.1 Characteristic behavior of responses for Configuration 1...... 118 6.2 Post-flutter behavior of Configuration 2...... 125 6.3 Computational cost parameters (milliseconds) ...... 128 6.4 Computation time for 10 seconds of response for Configuration 1 129 6.5 Computation time for 35 seconds of response for Configuration 2 129
xvii 7.1 Relative errors of mid-panel displacement compared to benchmark. 136 7.2 Post-flutter behavior for each coupling scheme at different time steps...... 146 7.3 Computational cost parameters, in seconds...... 157 7.4 Walltime for one second of stable response, in minutes...... 158 7.5 Walltime for three seconds of flutter response, in hours...... 159
B.1 Coefficient of thermal expansion and Young’s modulus for Al-7075 186 B.2 Thermal capacitance and thermal conductivity for Al-7075 . . . . 187 B.3 Coefficient of thermal expansion and Young’s modulus for Ti-6Al- 2Sn-4Zr-2Mo ...... 187 B.4 Thermal capacitance and thermal conductivity for Ti-6Al-2Sn-4Zr- 2Mo ...... 188
xviii List of Figures
1.1 Falcon HTV-3X Blackswift concept vehicle [14, 17]...... 3 1.2 Schematic of Fluid-Thermal-Structural Interactions...... 4
2.1 Fluid and Structural Domains...... 23
3.1 Sequence of coupling scheme...... 44 3.2 Quasi-Steady Predictor Implicit (PI) coupling flowchart...... 46 3.3 Quasi-Steady Conventional Explicit (CE) coupling flowchart. . . . 54 3.4 Quasi-Steady Strong Implicit (SI) coupling flowchart...... 56 3.5 Generalized sequence for FTSI of time-accurate aerothermodynam- ics...... 58 3.6 Time-Accurate Predictor Implicit (PI) coupling flowchart...... 61 3.7 Time-Accurate Strong Implicit (SI) coupling flowchart...... 68
4.1 Detail of wedge surface...... 71 4.2 Flow computational grid (85,860 cells)...... 79 4.3 Thermal model domain...... 80 4.4 Structural model domain...... 83
5.1 Average temperature, mid-panel displacement, and mid-panel ve- locity of Configuration 1 over one second...... 91 5.2 Average temperature, mid-panel displacement, and mid-panel ve- locity of Configuration 2 over one second...... 92
xix 5.3 Order of Accuracy of panel displacement, velocity, and tempera- ture for both configurations without subcycling (Quasi-steady aerother- modynamics)...... 94 5.4 Order of Accuracy of fluid pressure and heat flux for both configu- rations without subcycling (Quasi-steady aerothermodynamics). . 95 5.5 Order of Accuracy of panel displacement, velocity, and tempera- ture for both configurations with 25 structural steps per thermal step (Quasi-steady aerothermodynamics)...... 97 5.6 Order of Accuracy of fluid pressure and heat flux for both config- urations with 25 structural steps per thermal step (Quasi-steady aerothermodynamics)...... 98 5.7 Average temperature, mid-panel displacement, and mid-panel ve- locity over 0.112 seconds with time-accurate CFD...... 102 5.8 Order of Accuracy of panel displacement, velocity, temperature,
pressure, and heat flux with ∆tT = ∆tS = ∆tF (Time-accurate aerothermodynamics)...... 104 5.9 Order of Accuracy of panel displacement, velocity, temperature,
pressure, and heat flux with ∆tT = 4∆tS = 4∆tF (Time-accurate aerothermodynamics)...... 105 5.10 Order of Accuracy of panel displacement, velocity, temperature,
pressure, and heat flux with ∆tT = ∆tS = 4∆tF (Time-accurate aerothermodynamics)...... 107 5.11 Order of Accuracy of panel displacement, velocity, temperature,
pressure, and heat flux with ∆tT = 4∆tS = 16∆tF (Time-accurate aerothermodynamics)...... 108
6.1 Average temperature, mid-panel displacement, and mid-panel ve- locity of Configuration 1 over 10 seconds...... 113
xx 6.2 Profiles of temperature,displacement, pressure, and heat flux over panel at t = 0 and t = 10s for Configuration 1...... 114 6.3 Examples of stable responses of mode 2 compared to benchmark solution (red dashed line) for Configuration 1...... 116 6.4 Examples of stable responses of mode 2 compared to benchmark solution (red dashed line) for Configuration 1...... 117 6.5 Average temperature (red) and normalized mid-panel displacement (blue) of Configuration 2 over 35 seconds...... 119 6.6 Profiles of temperature,displacement, pressure, and heat flux over panel at t = 0 and t = 24.5s (immediately prior to snap-through) for Configuration 2...... 120 6.7 Center displacement during initial snap-through and final flutter response for Configuration 2...... 121 6.8 Displacement and power spectral density at 3/4 chord length. . . 123 6.9 x-t diagram of post-flutter limit cycle oscillations...... 124
7.1 Average temperature, mid-panel displacement, and mid-panel ve- locity of panel with time-accurate aerothermodynamics from the benchmark solution...... 132 7.2 Relative errors of the mid-panel displacement compared to bench- mark solution for SI, BI, PI, and PCI schemes over select time step sizes...... 134 7.3 Prescribed temperature distribution from quasi-steady analytical aerothermodynamics at t = 17.5 sec...... 138 7.4 Initial panel solution with prescribed temperature distribution in aeroelastic equilibrium, compared to quasi-steady analytical aerother- modynamics...... 139 7.5 Flowfield pressure of static aeroelastic solution ...... 140
xxi 7.6 Temperature rise and panel displacement during snap-through and subsequent flutter...... 142 7.7 Flutter displacement envelope and PSD at 3/4 chord location of SI, BI, PI, and PCI schemes at ∆t = 62.5µs...... 144 7.8 Space-time contours of normalized displacement over .05 seconds for SI (benchmark), BI, PI, and PCI schemes, ∆t = 62.5µs...... 145 7.9 Work violation over one second, normalized by work done by struc- tural solver, for the BI, PI, PCI, and SI schemes...... 149 7.10 Normalized work violation at one second at each time step size for the BI, PI, PCI, and SI schemes...... 150 7.11 Average temperature, mid-panel displacement, and 3/4 chord panel displacement of long-time record analysis with time-accurate CFD. 151 7.12 Power Spectral density of flutter between long time record simula- tion and short time record simulation (SI scheme, ∆t = 62.5µs). . 152 7.13 Temperature, flutter displacement envelope, and power spectral density of panel with unsteady CFD aerothermodynamics (blue), quasi-steady surrogate aerothermodynamics (red), and quasi-steady analytical aerothermodynamics (black)...... 153 7.14 Displacement, temperature, pressure, and heat flux across panel immediately prior to flutter with unsteady CFD (20.8 seconds), quasi- steady surrogates (20.8 seconds), and quasi-steady analytical aerother- modynamics (24.5 seconds)...... 156
C.1 Deformation profiles used to compare mesh convergence...... 190 C.2 Pressure and heat flux comparisons for both deformation profiles. 191 C.3 Generalized aerodynamic force (GAF) over 1.25 cycles of forced oscillation...... 192
D.1 Heat flux profile prescribed to panel upper surface...... 194
xxii D.2 Comparison of current thermal model to Abaqus using a constant heat flux profile...... 194
E.1 Uniform in space temperature loading over time...... 196 E.2 Comparison of current structural solver to Abaqus with a rising temperature loading...... 197 E.3 Random in time, uniform in space, pressure loading applied to panel for transient response verification...... 198 E.4 Comparison of current structural solver to Abaqus (linear and non- linear) with a random pressure loading...... 198
xxiii List of Symbols
An Modal weight for mode n a Least square fit for offset b Least square fit for slope BI Basic Implicit scheme (quasi-steady or time-accurate fluid) [C] Structural damping matrix
[CT] Thermal capacitance matrix C Computational cost per time step CFD Computational Fluid Dynamics CE Conventional Explicit scheme (quasi-steady fluid) c Specific heat cf Local skin friction coefficient D Plate stiffness e Internal energy per unit volume E Young’s Modulus Fa Structural applied load vector FT Structural thermal load vector FTSI Fluid-thermal-structural interactions GCL Geometric Conservation Law H Total enthalpy h Panel thickness B HA Mapping from domain A to domain B
xxiv HCV Hypersonic cruise vehicle J Metric Jacobian [K] Structural stiffness matrix
[KT] Thermal conductivity matrix L2 L2 norm L Panel length M Mach Number [M] Structural mass matrix
M∞ Freestream Mach number
MT Thermal moment ~n Outward unit normal N Number of time steps
NT In-plane thermal load
Nx In-plane load O Order of PCI Predictor-Corrector Implicit scheme (time-accurate fluid) PI Predict Implicit scheme (quasi-steady or time-accurate fluid) P , p Fluid pressure P r Prandtl Number Q Fluid conserved variable vector
QT Thermal discrete heat load vector q Heat flux q2 Post-shock dynamic pressure R Discretized divergence of fluid fluxes
RGCL Discretized Geometric Conservation Law residual r Recovery factor
Rex Reynolds number of panel location x SI Strong Implicit scheme (quasi-steady or time-accurate fluid)
xxv St Stanton Number T Temperature T Discretized temperature
Tinit Initial panel temperature
Tw Wall temperature
T∞ Freestream temperature t Time
t0 Overhead time
U2 Post-shock fluid velocity u Displacement
u0 In-plane displacement u Discretized structural displacement ~v Velocity w Transverse structural displacement W Work violation across fluid-structure interface
WF Work done by fluid
WT Work done by structure X Reference domain ~x Spatial coordinates
α Coefficient of thermal expansion
αR Mass-proportional Rayleigh coefficient
βR Stiffness-proportional Rayleigh coefficient
ΓF −S Fluid-Structure interface γ Ratio of specific heats (1.4) ∆t Time step
δx, δz Linear shape function Error
xxvi tol Error tolerance κ Thermal conductivity µ Dynamic viscosity φ Solution quantity of interest ρ Density σ Stress tensor ν Poisson’s ratio
φn Mode shape n
ΩF Fluid domain
ΩS Solid/structural domain ω Frequency
Subscripts
0 Stagnation property
2 Post-shock property
∞ Freestream property aw Adiabatic wall property e Boundary layer edge property
F Of the fluid domain/model
I Interpolated between time steps ls Evaluated using least-squares fit
P Predictor
S Of the structural domain/model
T Of the thermal domain/model w Evaluated at the wall
Superscripts ∗ Eckert’s Reference enthalpy condition
xxvii i Subcycling counter between thermal steps j Subcycling counter between structural steps k Fluid step counter m Structural time step counter n Thermal time step counter
xxviii Chapter 1
Introduction and Objectives
1.1 Introduction
Sustained flight at hypersonic speeds has been an elusive goal for the aerospace community for the past half century [1]. As early as the 1960s programs were started to develop such capabilities, among the most ambitious being the National
Aerospace Plane (NASP) in the 1980s, which was ultimately canceled before any prototypes were built [2]. Today, there is a continued interest in hypersonic vehi- cles from National Aeronautical and Space Administration (NASA) and the De- partment of Defense [3, 4]. This is evident in successful vehicle programs such as: the X-43 Hyper-X scramjet demonstrator from NASA [2]; the X-51 WaveRider scramjet demonstrator from the Air Force Research Laboratory (AFRL) [5–7]; and the Hypersonic International Flight Research and Experimentation (HIFiRE) hy- personic testbed from the US Air Force (USAF) and Australian Defence Science and Technology Organisation (DSTO) [8, 9]. These programs are less ambitious
1 than the older NASP program, each seeking to demonstrated specific technologies or providing experimental flight testbeds.
The AFRL is particularly interested enabling the development of a HCV that: operates in the Mach 5 - 7 range for prolonged periods of time; flies 2000 nau- tical miles without refueling; operates with a takeoff weight in excess of 300,000 pounds; and is reusable [10–12]. In order to meet these goals, a vehicle would be designed to operate with a lifting body and an air-breathing scramjet engine in an integrated airframe-propulsion framework [2, 4, 13, 14]. Additionally, instead of ablation-based thermal protection systems used on previous vehicle concepts to reduce the heat load of the vehicle structure [2], a thin-gauge metallic structure capable of operating at elevated temperatures is desired in pursuit of minimizing weight and increasing reusability [12, 15]. As an example of such a vehicle, the
Falcon HTV-3X ”Blackswift” concept is illustrated in Fig. 1.1 [14, 16].
To enable the development of such HCVs, there is still an extensive set of tech- nical challenges that must be addressed [1, 10, 11, 15–17]. These challenges par- tially arise from the extreme loading environment of hypersonic flight and sur- vivability and reusability requirements of the HCV. Any vehicle operating in the hypersonic environment is subjected to prolonged intense heating and severe pres- sure loading, including high-magnitude and high-frequency turbulent boundary layer loads and engine noise excitation [10, 11, 15]. These high loadings, combined with thin metallic structures, are expected to produce compliance of the structure to the aerothermodynamics in a coupled manner.
2 Figure 1.1: Falcon HTV-3X Blackswift concept vehicle [14, 17].
3 Fluid Domain Solid Domain
Aerodynamic Temperature Heat Transfer Aerothermal Heating Heat Flux
Fluid Temperature Dynamics
Deformations Aeroelastic Aerodynamic Structural Pressure Mechanics Pressure
Figure 1.2: Schematic of Fluid-Thermal-Structural Interactions.
This coupling results in the fluid-thermal-structural interactions (FTSI) schemat- ically depicted in Fig. 1.2. The fluid domain and vehicle structure domain interact at the vehicle surface. Aerothermal coupling occurs between the fluid heat flux and structural temperatures, the fluid pressure loading and structural deformations are aeroelastically coupled. Both the fluid pressure and heat flux are intrinsically linked within the fluid dynamics. There is a direct dependency on the deforma- tions to the thermal state through thermal stress and material property changes; while heat generation due to deformation is typically negligible. However, note that there is an indirect feedback coupling from the deformation to the thermal state through the modification in the heat flux. Hence, all interactions between the
flow around the vehicle, its thermal state and deformations may be important.
Because of the high aerothermodynamic heating over the flight trajectory, the thermal state of the vehicle evolves over significant portions of the trajectory [12,
4 18]. The long-term thermal evolution, combined with the mutual interactions be- tween the fluid loadings and thermo-structural vehicle state, results in a path- dependence of the structural response [19]. This path dependence requires that accurate simulation of the vehicle be done over the full trajectory instead of spe- cific flight conditions as traditionally done. Due to conflicting similarity require- ments, testing scaled prototypes to capture these multidisciplinary interactions is impractical [13, 20]. Hence, computational modeling to predict the structural state is critical to the design of HCVs.
A fundamental challenge in computational modeling of FTSI is the inherent differences in time scales between the fluid, thermal, and structural physics. Gen- erally, characteristic times of the fluid and the structure are orders of magnitude smaller than the thermal response [21–24]. The characteristic times of the fluid are the smallest, and numerical stability issues generally limit the maximum time step size of computational solvers [25]. The structural time scales are based on the nat- ural frequencies of the structural modes of interest, which can range from O(1) Hz to O(1000) Hz [10]. The thermal time scale is typically much larger than the fluid and structural scales due to characteristic time scales for typical properties, and can change for large durations of a trajectory (O(seconds) to O(minutes)) [12, 18].
This disparity in time scales can lead to conflicting requirements for time step siz- ing of the computational models to accurately capture the dynamics of the fluid and structural responses, over the long-time records that encapsulate the slower thermal response [10, 11]. Time integration of these models must be done in an
5 efficient manner to enable simulation of the FTSI over long time records.
1.2 Literature Review
This dissertation focuses on the time-integration problem for simulating FTSI over
long time records. The current state-of-the-art for this problem is established by
reviewing the following areas: 1) general frameworks for time-marching multi-
physics systems; 2) time integration of fluid-structural interactions; 3) time in-
tegration of fluid-thermal interactions; and 4) time integration of fluid-thermal-
structural interactions.
1.2.1 Time Integration Frameworks for Multi-Physics Problems
A fundamental issue of multi-physics analysis is how to model and integrate the
different physics together. For FTSI, the problem can be thought of as having two
spatial domains and three physical fields [26]. The two domains consist of the fluid and solid (vehicle) domains. The three fields consist of the fluid, thermal, and structural dynamics.
The physical fields may be coupled together through a monolithic or parti- tioned approach. In the monolithic approach, the equations are solved simulta- neously as a single integrated solver using a consistent time integration algorithm
[27, 28]. For explicit time integrators the solutions are exchanged at each time step, and for implicit integrators the cross-coupling terms of the Jacobian matrix
6 are computed and exchanged at each time step. This consistency can provide ex- act conservation of the state variables as well as ensuring energy is maintained through the fluid-structure interface, which are ideal for ensuring convergence and stability [28, 29]. However, because of the tightly integrated nature of monolithic solvers, the approach is less flexible compared to partitioned methods[30]. The monolithic approach requires a dedicated solver for each set of multiphysics inter- actions; e.g. a fluid-structure interactions solver could not be directly used with a thermal solver for fluid-thermal-structural interactions without extensive modi-
fication. Furthermore, for large industry-level simulations monolithic solvers can become overly complex to maintain, and computationally overly-burdensome [31–
33].
In contrast, the partitioned approach is used to couple the physical fields to- gether from separate solvers, each with potentially different time integration al- gorithms [27, 30]. The individual solvers are coupled through a careful exchange of boundary conditions at the interfaces of the domains, and an exchange of state variables for multifield interaction within a single domain (e.g. thermal-structural coupling). Here, only information at the interface conditions and of the time in- tegration approach are required, leading to a non-intrusive means of coupling the different fields [27, 34–36]. Thus, the partitioned approach is attractive because it more easily allows for the integration of different solvers for each physical field, rendering the incorporation of new solver technologies relatively straight forward.
Because of these advantages, this approach is suitable for larger and more complex
7 problems [27, 31, 33], and is more favored over monolithic approaches [27]. How-
ever because each solver has a different time integration algorithm and associated
lags in boundary conditions at interfaces, ensuring convergence and stability is a
challenge [27].
In the partitioned approach, the time integrators for the individual solvers are
generally implicit or a mix of implicit and explicit algorithms [27]. For implicit time
integrators, at a given solution time t the solver requires the solution at time t + ∆t to march forward in time. Within a single solver, this is achieved by inverting the
Jacobian matrix of the system, but a Jacobian for the global system is not available in the partitioned approach. Because of this, each solver cannot directly determine the t + ∆t solution from the other solvers, thus this must be accounted for in the coupling scheme..
Two general classes of coupling procedures for the partitioned approach are strongly coupled schemes and loosely coupled schemes.1 Strongly coupled schemes make use of subiterations between the solvers at each time step to converge the coupled t + ∆t solution [27, 39]. By converging the solutions between solvers, the order of accuracy of the individual solvers is retained, but the use of subiterations make the analysis more computationally expensive. Like monolithic schemes, strongly coupled schemes can maintain a conservation of energy across the fluid-
1There exist some variations of terminology in the literature. Strongly coupled as used here may also be referred to as tightly coupled [27]. Loosely coupled as used here may also be referred to as weakly coupled or staggered [37]. Some texts even refer to monolithic coupling as tight and partitioned coupling as loose [38]. For the purposes of this dissertation, monolithic/partitioned and strongly/loosely will be used exclusively to avoid confusion.
8 structure interface.
In contrast to strong coupling, loosely coupled schemes neglect the subitera-
tion loop between solvers, exchanging information between solvers just once per
time step. These schemes are less computationally expensive per time step com-
pared to strongly coupled schemes, however, they can reduce the accuracy and
stability of the global solution. This is partially due to the fact that loosely coupled
schemes are incapable of exactly conserving energy across the fluid-structure inter-
face. [40, 41] Because of these restrictions, loosely coupled schemes must be care-
fully designed to maintain the order of accuracy of the individual solvers and max-
imize numerical stability. Even with these restrictions, loosely coupled schemes are
attractive since they offer potentially significant computational savings and imple-
mentation simplicity compared to a strongly coupled scheme [42].
If the time scale of one of the physical fields is small enough compared to the
other physical fields, the transients may be negligible on the coupled response.
For high-speed flow, this can be used to avoid time integration, and reduce com-
putational expense, of the fluid solver using the quasi-static or quasi-steady flow assumptions. Both of these assumptions use instantaneous boundary conditions from the other solvers to compute fluid loads with a time history of the flow. For quasi-static flow, the flow is assumed to be completely steady, neglecting motion of surfaces. For quasi-steady flow, the impact of the velocity of the surface motion is included. Note that for pressure, the quasi-steady assumption includes the effects of aerodynamic damping, while the quasi-static assumption neglects it.
9 1.2.2 Time Integration of Fluid-Structure Interactions
Loosely coupled partitioned time integration schemes are used extensively for
fluid-structure interactions. Due to the vast body of work in this field, only a rep- resentative subset is reviewed here.
Even before considering the coupling of fluid and structural solvers together, a significant challenge of fluid-structure interactions using CFD is the deforma- tion of the fluid mesh. Traditional CFD solvers cannot be used for unsteady flow without modification because the mesh motion introduces additional dynamics that must be included, requiring an Arbitrary Lagrangian-Eulerian (ALE) or sim- ilar moving mesh framework [43]. However, time integration schemes suitable for marching the discretized Navier-Stokes equations from an Eulerian framework may not retain their accuracy and stability in an ALE framework with mesh motion
[44, 45]. A significant source of this loss of accuracy is from non-conservation of the state variables in the presence of a moving mesh from the computation of time derivatives on changing mesh volumes. The Geometric Conservation Law (GCL),
first coined in [46], is a condition that is satisfied if a flow solver can exactly repro- duce uniform flow in the presence of arbitrary mesh motion. Violating the GCL has been found to reduce the accuracy and stability of the flow solver [47, 48]. The
GCL can be satisfied through a variety of ways including: computing the mov- ing mesh metrics to satisfy a discrete form of the GCL [45, 48]; reformulating the
Navier-Stokes equations to avoid computing time derivatives on changing cells
10 [49]; or adding source terms to the discretized Navier-Stokes equations that auto-
matically satisfy the GCL [50, 51].
For loosely coupled schemes, the use of extrapolating predictors for the struc-
ture or aerodynamic loads have been found to improve the solution accuracy. The
surface displacement must be predicted at the next time step for the fluid solver
if it is updated first; conversely the aerodynamic loads must be predicted if the
structure is updated first. Structural predictors will conserve forces across the
fluid-structure boundary exactly; however the displacement and velocity of the
fluid-structure boundary cannot be exactly matched [42]. The converse is true for
fluid load predictors.
Edwards et al. [52] compared several coupling schemes coupling a linear modal
structure to a transonic nonlinear flow solver, and the choice of a linear extrapola-
tion of the fluid loads as a predictor were advocated in [53, 54]. Later studies [50]
added a corrector step to the formulation in [53] to update the structure with the
aerodynamic loads computed at the end of the time step. Aerodynamic predictors
have also been used more recently in [55] using linear and cubic pressure extrap-
olations with an implicit multi-step structural time integrator and an explicit fluid
solver.
In a series of studies [40, 42, 56–59], several coupling schemes were developed
with structural predictors for fluid-structure interaction. The generalized serial stag-
gered (GSS) scheme was specifically designed to maintain the second order tem- poral accuracy of the individual fluid and structural solvers using a second order
11 extrapolation of the structural displacement as a prediction for the fluid solver.
A modified version of the GSS, the Improved Serial Staggered (ISS) scheme offsets the structural solver by half a time step compared to the fluid solver, which was shown to have superior conservation of energy transfer with the time integrators used. The schemes were also formulated to satisfy the geometric conservation law with a specific fluid time integrator [42]. These schemes were contrasted with a conventional serial staggered (CSS) scheme that did not make use of such predictors and suffered in accuracy and stability compared to the GSS and ISS schemes. The schemes were also found to have better energy conservation across the interface than the CSS scheme [40, 42], which can be a significant factor in numerical stabil- ity.
Another type of loosely coupled scheme was developed in [60] based on a combined interface boundary condition (CBIC), in which an additional differen- tial equation is solved along the fluid-structure interface to predict a higher-order correction to the interface velocity and fluid pressure. The CIBC scheme was com- pared favorably to a CSS scheme in both relative accuracy and stability using an implicit structural time integrator and explicit fluid time integrator for low speed
flow.
One potentially useful feature of loosely coupled partitioned approaches for the class of fluid-thermal-structural interactions considered in this work is the ability to use different time step sizes for the individual solvers. Denoted as subcycling, the disparity of time scales between solvers may be exploited, reducing computational
12 expense by calling the thermal or structural solver and exchanging information be- tween solvers less frequently [61]. Loosely coupled schemes with subcycling have also been successfully applied to high-order DNS flow solvers modeling panel re- sponse in supersonic flow [62] using a fourth order explicit fluid time integrator and a second order implicit structural integrator; however the scheme used to cou- ple the solvers resulted in first order time accuracy.
1.2.3 Time Integration of Fluid-Thermal Interactions
In fluid-thermal interactions, the thermal state of the structure is coupled to the aerothermodynamics through the heat transfer and temperature at the fluid-structure boundary. Unlike fluid-structure interactions, mesh motion is generally not re- quired, and existing CFD codes can be used without significant modification.
Schemes for partitioned fluid-thermal analysis have been investigated by Birken et al. for loosely and strongly coupled time integration using an implicit first or- der scheme [63] and a second order implicit Runge-Kutta scheme [64]. In [63], a basic loosely coupled scheme was found to predict the cooling of a panel exposed to transonic flow with similar error to a strongly coupled scheme that was twice as expensive. In [64], the same analysis is performed using second order implicit schemes with loose and strong coupling with adaptive time stepping. The loosely coupled scheme reduced to first order while the strongly coupled scheme main- tained second order accuracy, with computational expense favoring the strongly
13 coupled scheme with adaptive time stepping.
Combined interface boundary condition methods have also been used for fluid- thermal analysis by Roe et al. in [65], and the accuracy and stability was compared favorably to the fluid-thermal analogue of the CSS scheme. The scheme was shown to retain the stability boundary of the individual solvers. However, the individual solvers required low CFL numbers for stability, limiting the size of the global time step.
Predictors has been used successfully in fluid-thermal analysis of rocket nozzle heating in hypersonic flows in [66]. The coupling scheme was a based on a predic- tor and corrector of the thermal state, in which an extrapolation of the fluid loads was used to compute the thermal predictor before updating the fluid. Note this is similar to the predictor-corrector scheme used in [50] for fluid-structural coupling.
Because the fluid response time is orders of magnitude smaller than the ther- mal response time for high-speed vehicles, the quasi-static flow assumption may be used to neglect the time integration of the fluid solver [22, 24, 67]. With this assumption, the thermal problem is still treated as transient, but the instantaneous heat flux is computed from steady-state flow solutions. In [24] the laminar heat- ing of a spherical dome in hypersonic flow was modeled using a transient thermal solver and a high-order flow solver marched to steady state. In [67], a similar anal- ysis was performed, except the thermal solver was marched forward in time for several steps before the quasi-static flow solution was updated at a coarser step.
An adaptive time stepping scheme was used to determine the coarseness the flow
14 solution stepping based on a tolerance criteria. In [22], a similar scheme without adaptive time stepping was used to analyze the heating of metallic panels in hy- personic flow. Instead of applying the quasi-static heat flux directly, a heat trans- fer coefficient was applied between the fluid solution updates to provide a linear change in heat flux in response to temperature changes. In general the coupling used in these analyses are formally first order accurate in time because the fluid solutions used for the next thermal time step are computed using the previous thermal step.
1.2.4 Time Integration of Fluid-Thermal-Structural Interactions
A number of studies have emerged on fluid-thermal-structural interactions (FTSI) of vehicles in high-speed flows, with varying levels of simplification to the prob- lem. Many studies on aerothermoelasticity model the thermal field as a prescribed temperature rise decoupled from the fluid-structural interactions [68–72]. How- ever, the mutual coupling between the thermal and fluid-structural systems can have a significant influence on the temperature distribution and ultimately the structural state [19, 73, 74].
In order to capture the prolonged thermal evolution coupled with the fluid and structural response, the quasi-static assumption can be applied to both the fluid and structural solvers. With this assumption, the transients of the structural heat transfer is modeled, but the time histories of the structural and fluid solvers are
15 neglected. In [75, 76] a partitioned approach to couple finite element flow ther- mal and structural models was used in a quasi-static analysis for panels [75] and leading edges [76]. The thermal evolution was modeled in [75] using steady-state solutions from the flow, and updates to the static structural state at set intervals. In
[76] the flow and thermal solvers were integrated together in a transient analyses with a static structural response; a similar method was also used in [77]. Due to the static structure assumption, any dynamic response or instabilities of the struc- ture cannot be captured with these frameworks. Note that the procedures in these studies did not make use of predictors, and the global solution is generally only
first order accurate in time.
Quasi-steady or quasi-static flow can also be modeled using simplified aerother- modynamic models or CFD-based reduced order models in conjunction with tran- sient structural and thermal solvers. Simplified quasi-steady flow models were used in [73, 78] to couple fluid-thermal-structural interactions of the cylindrical bending of simply supported panels [78] and a stiffened composite panel from the
NASP program [73]. In [73], the long-term thermal evolution was modeled us- ing static structural deformations, and the transient aeroelastic stability was tested at discrete points in time. In [78], the thermal and structural solvers were both time integrated using a loose coupling algorithm. Structural subcycling was also used to decrease the computational cost, with boundary conditions from the struc- tural solver passed to the thermal solver in a time-averaged sense. A fourth order explicit time integrator was used for the structure, however the scheme is only for-
16 mally first order accurate due to the coupling with the thermal solver. This loosely
coupled time integration framework was used in several other studies for CFD-
based surrogate models [79], reduced order thermal and structural models [80],
and stochastic analysis based on uncertainty in the turbulent heating [81]. A simi-
lar methodology was also used in [82] to model the aerothermoelastic response of
a composite panel. In general these coupling schemes are only first order accurate.
Limited studies have also been performed on the coupling of unsteady fluid,
thermal, and structural solvers. Lohner¨ et al. [83] discussed coupling CFD, com- putational thermal dynamics, and computational structural dynamics codes in a monolithic or partitioned manner. Tran and Farhat [84] considered the aerother- moelastic response of a flat panel by coupling a CFD solver with finite element thermal and structural solvers using a CSS-like scheme. This scheme was not predictor-based, and was formally first order accurate in time. However only small time records (0.1 seconds) were considered. In [85] a framework for full aerothermoelastic coupling in a monolithic solver using finite elements was devel- oped. However, this restricted the fluid, thermal, and structural problems to be discretized with the same time integrator and marched at the same time step.
17 1.2.5 Summary of the State of the Art for Fluid-Thermal-Structural
Interactions
From the literature, it is evident that loosely coupled schemes offer promise in maintaining accuracy in fluid-structural analysis and fluid-thermal analysis. While such an approach is used almost exclusively for fluid-thermal-structural analysis, there has not been a rigorous evaluation for approaches used. The loose coupling techniques already in use for FTSI generally degrade the accuracy to first order because they do not make use of the accuracy-preserving techniques from fluid- structure and fluid-thermal coupling procedures. This loss of accuracy can also lead to a loss in stability, requiring smaller time steps, and increasing the com- putational expense of coupled simulations. Furthermore, studies of FTSI have al- most exclusively used the quasi-steady and quasi-static flow assumptions, with only very limited studies with unsteady flow. While the quasi-steady/quasi-static assumptions are attractive to avoid the small fluid time scales for computational efficiency, the capability to perform simulations with time-accurate aerothermody- namics is important to test the limits of these assumptions in a coupled analysis.
1.3 Objectives of this Dissertation
The primary goal of this dissertation is the development and application of loosely coupled time integration schemes for fluid-thermal-structural interactions in hy-
18 personic and high speed (M1) flow. These schemes are to be designed to main- tain the second order accuracy used in individual solvers through the use of ex- trapolating predictors, and exploit the disparities of time scales through the use of subcycling. Schemes for both time-accurate and quasi-steady flow are considered.
The specific objectives are:
1. Develop second-order accurate loosely coupled time integration procedures
for fluid-thermal-structural interactions using either time-accurate CFD or
quasi-steady flow models.
2. Assess the accuracy and stability of the coupling procedures through numer-
ical studies.
3. Benchmark the various coupling schemes at different time step levels against
a strongly-coupled solution for panels operating in supersonic/hypersonic
flows.
4. Compare the computational expense of the various time marching proce-
dures.
The remainder of this dissertation is arranged as follows: the governing equa- tions and time discretization of the individual solvers are outlined in Chapter 2; the different time marching schemes developed and examined are detailed in Chap- ter 3; the computational configurations for the panels considered are described in
Chapter 4; order of accuracy comparisons are provided in Chapter 5; a study of
19 panel responses to quasi-steady flow models and comparison of computational expense are discussed in Chapter 6; a study of panel responses to time-accuracy
CFD and comparison of computational expense are discussed in Chapter 7; and the principal conclusions and suggested future work are provided in Chapter 8.
1.4 Key Novel Contributions of this Dissertation
The principal contributions to the state-of-the-art made in this dissertation are summarized below:
1. Development of two second-order accurate loosely coupled schemes for fluid-
thermal-structural interactions for quasi-steady or quasi-static aerothermo-
dynamics. These schemes are based on predictors of the fluid loads with
structural subcycling.
2. Extensions of the two second-order accurate loosely coupled schemes to in-
clude time-accurate aerothermodynamics with both structural and fluid sub-
cycling.
3. A 30 second transient analysis of the heating and deformation of a panel in
high-speed flow undergoing thermally-induced flutter using a time-accurate
CFD solver.
4. Assessment of the impact of the quasi-steady flow assumption on a panel
in high-speed flow compared to unsteady flow, including both CFD-based
20 quasi-steady modeling strategies and lower fidelity analytical aerothermo- dynamic models.
21 Chapter 2
Governing Equations and Time
Discretization
2.1 Governing Equations
Fluid-thermal-structural interactions consist of two spatial domains, as illustrated
in Fig. 2.1. Although only two spatial domains exist, the problem is divided into
the three physical fields [26]: fluid dynamics; structural dynamics; and thermal
dynamics. The fluid domain, denoted as ΩF , is made up of the physical fluid
field. The structural domain, denoted as ΩS, is shared between the structural and thermal fields. The two domains interact at the fluid-structure interface, ΓFS.
The fluid, structural, and thermal fields are all governed by the equations of continuity of mass, momentum, and energy. Without loss of generality, these
fields can be fundamentally represented as the same equations using an Arbitrary
22 ΓF-S Fluid-Structure Interface
ΩF Fluid Domain (Fluid field)
ΩS Structural Domain (Thermal and Structural Fields)
Figure 2.1: Fluid and Structural Domains.
Lagrangian-Eulerian (ALE) frame of reference [43]:
1 ∂ h i (Jρ) + ∇~ · ρ ~v − ~x˙ = 0 (2.1a) J ∂t X
1 ∂ h i (Jρ~v) + ∇~ · ρ~v ⊗ ~v − ~x˙ − σ = ~0 (2.1b) J ∂t X
1 ∂ h i (Je) + ∇~ · e ~v − ~x˙ − σ · ~v + ~q = 0 (2.1c) J ∂t X
where X is a reference domain, J is the metric Jacobian with respect to the reference
domain, ~x˙ is the motion of the reference domain relative to the spatial domain, ρ is the density, ~v is the velocity of the medium with respect to the spatial domain, e is the specific internal energy, σ is the stress tensor, and ~q is the heat flux.
The compressible Navier-Stokes equations can be represented by Eqs. 2.1 di-
23 rectly. If mesh motion is used, the reference domain X is the initial undeformed
fluid domain, or the initial undeformed mesh when discretized. If mesh motion is
not used, X represents the unchanging spatial domain, J = 1 and ~x˙ = ~0, reducing
Eqs. 2.1 to:
∂ρ + ∇~ · (ρ~v) = 0 (2.2a) ∂t ∂ (ρ~v) + ∇~ · (ρ~v ⊗ ~v − σ) = ~0 (2.2b) ∂t ∂e + ∇~ · (e~v − σ · ~v + ~q) = 0 (2.2c) ∂t
The structural field in ΩS consists of the conservation of mass and momentum in
Eqs. 2.1(a-b). Generally the Lagrangian frame of reference is used for structural mechanics, except for very large deformations. Using the Lagrangian framework, the frame of reference X is the material itself, so ~v = ~x˙. The mass and momentum
equations reduce to:
∂(Jρ) = 0 (2.3a) ∂t
ρ~u¨ − ∇~ · σ = ~0 (2.3b)
where ~u is the deformation of the material from the initial configuration. Note that the density ρ is usually assumed constant in time for most structural applica- tions, reducing J to 1, and making Eq. (2.3)(a) trivial. Equation (2.3) represents the structural dynamics equations of motion.
24 The thermal field in ΩS consists of the conservation of energy in Eq. 2.1(c). The internal energy of a solid is defined in Eq. (2.4) [86], and the heat flux by Cauchy’s law in Eq. 2.5 [87]:
de ≡ cρdT (2.4)
~q ≡ −κ∇~ T (2.5) where c is the temperature-dependent specific heat, T is the temperature, and κ is the temperature dependent thermal conductivity. Using these definitions, the thermal heat transfer equation is reduced to:
ρcT˙ − ∇~ · σ · ~u˙ + κ∇~ T = 0 (2.6)
Note the second term with the stress tensor in Eq. (2.6) is the heat generated from internal strain rates. This term is generally negligible and can be neglected in aerospace applications [23], resulting in the familiar heat transfer equation:
ρcT˙ − ∇~ · κ∇~ T = 0 (2.7)
The fluid domain ΩF and structural domain ΩS interact at the at the fluid-
25 structure interface ΓFS:
σF · ~n = σS · ~n (2.8a)
~ ~qF · ~n = −(κ∇T )S · ~n (2.8b)
~xF = (~u + ~x)S (2.8c)
v˙F = ~u˙ S (2.8d)
TF = TT (2.8e)
where ~n is the unit normal of ΓFS in the direction of ΩF . The subscripts T , S, and F represent variables from the thermal, structural, and fluid fields, respec- tively. Equation (2.8a) specifies that the fluid stresses must be equal to the struc- tural stresses at the fluid-structure boundary. This requirement is usually relaxed by considering only the pressure for the fluid stresses, since shear stresses gener- ally do not have a significant impact on the structural response. Equation (2.8b) specifies that the heat flux between the structure and fluid must be equal at the boundary. Equations (2.8c-d) specify that the structural displacements and veloc- ities must be equal to the boundary motion of the fluid. Equation (2.8e) specifies that the fluid and thermal field temperatures must be equal at the boundary.
In addition to the interactions at ΓFS, the thermal and structural fields interact
26 in ΩS via:
TT = TS (2.9a)
(σ · ~u˙)T = (σ · ~u˙)S (2.9b)
If the strain-induced heating is neglected and Eq. (2.7) is used, Eq. (2.9b) is not needed, and the thermo-structural coupling reduced to a matching of the internal temperature field.
In practice, the coupling is applied through the use of prescribed Dirichlet boundary conditions and Neumann-type boundary loadings. The fluid stresses are applied to the structure as loading, satisfying Eq. (2.8a). The fluid heat flux is applied to the thermal field as a heat loading for Eq. (2.8b). The fluid mesh mo- tion and fluid velocities at the surface are defined by the structural deformations in
Eqs. (2.8c-d). The fluid temperature at the boundary is set from the thermal field for Eq. (2.8e). Finally, the structural temperatures are treated as a field loading from the thermal field to satisfy Eq. (2.9b).
2.2 Fluid Time Discretization
The fluid may be modeled as unsteady flow, solving Eqs. 2.1 directly, or as quasi- steady flow, ignoring the history effects of the fluid.
27 2.2.1 Quasi-Steady Fluid Modeling
The quasi-steady assumption for fluid modeling has been used with success in
fluid-thermal-structural analyses of structures subject to high speed flows[24, 73,
75, 76, 78, 88]. Common examples of this assumption in high speed flows are sim- plified theoretical models (e.g., Piston Theory[89]), semi-empirical models (e.g.,
Eckert’s Reference Enthalpy [90]), or CFD-based models (e.g., CFD surrogates [79,
91]). Quasi-steady aerothermodynamic models are expected to play an exten- sive role in the design and life prediction of structures in high-speed flow due to the potential for significantly smaller computational expense compared to time- integrating an unsteady flow solver. Thus, this study explores these types of mod- els in the analysis of the different time marching procedures.
With quasi-steady flow, transient mesh motion is not considered, and the Eule- rian framework for the Navier-Stokes equations from Eq. 2.2 can be used instead of the more general ALE framework in Eq. (2.1). Furthermore, since the time- history is not modeled, the temporal derivatives are all assumed to be zero, reduc- ing Eqs. (2.2) to:
∇~ · (ρ~v) = 0 (2.10a)
∇~ · (ρ~v ⊗ ~v − σ) = ~0 (2.10b)
∇~ · (e~v − σ · ~v + ~q) = 0 (2.10c)
28 These equations can then be discretized in space using CFD-based techniques, or approximated using simplified theoretical models. Because there are no time- history effects, any model based on the quasi-steady assumption can be repre- sented as a function of the instantaneous boundary conditions. For FTSI applica- tions, these models are functions of the instantaneous surface displacements, ve- locities, and temperatures, with outputs being the surface pressure and heat flux: