Parallel Computational Fluid Dynamics 2006 (eBook)

Parallel Computing and its Applications

A. Ecer, Pat Fox, Jang-Hyuk Kwon, Jacques Periaux, N. Satofuka (Herausgeber)

eBook Download: EPUB

2007 | 1. Auflage
318 Seiten
Elsevier Science (Verlag)
978-0-08-055004-6 (ISBN)

Infos

The proceedings from Parallel CFD 2006 covers all aspects of parallel computings and its applications. Although CFD is one of basic tools for design procedures to produce machineries, such as automobiles, ships, aircrafts, etc., large scale parallel computing has been realized very recently, especially for the manufactures. Various applications in many areas could be experienced including acoustics, weather prediction and ocean modeling, flow control, turbine flow, fluid-structure interaction, optimization, heat transfer, hydrodynamics.

- Report on current research in the field in an area which is rapidly changing
- Subject is important to all interested in solving large fluid dynamics problems
- Interdisciplinary activity. Contributions include scientists with a variety of backgrounds

Inhalt

Front Cover 1
Parallel Computational Fluid Dynamics 4
Copyright Page 5
Table of Contents 8
Preface 6
Acknowledgements 7
Part 1. Invited Talk 12
Chapter 1 Parallel Hybrid Particle-Continuum (DSMC-NS) Flow Simulations Using 3-D Unstructured Mesh 12
1. Introduction 12
2. Hybrid DSMC-NS Scheme Using 3D Unstructured Mesh 13
3. Results and Discussions 16
4. Conclusions 20
REFERENCES 20
Part 2. Parallel Algorithm 22
Chapter 2 A Parallel CFD-CAA Computation of Aerodynamic Noise for Cylinder Wake-Airfoil Interactions 22
1. Introduction 22
2. Computational Grid 23
3. Hydrodynamics 24
4. Acoustics 25
5. Conclusions 29
REFERENCES 29
Chapter 3 CFD Problems Numerical Simulation and Visualization by means of Parallel Computation System 30
1. Validation 30
2. Actual problems 30
3. On-the-fly visualization on parallel systems 33
4. S-VR structure 34
5. Hybrid UNIX-Windows versions 35
6. GDT package effectiveness 36
7. Conclusion 37
REFERENCES 37
Chapter 4 Markov Prefetching in Multi-Block Particle Tracing 38
1. Introduction 38
2. Previous Work 39
3. Multi-Block Topology 39
4. Markov Prefetching 40
5. Results 42
6. Conclusion and Future Work 44
Acknowledgements 45
REFERENCES 45
Chapter 5 A Parallel 2-D Explicit-Implicit Compressible Navier-Stokes Solver 46
1. INTRODUCTION 46
2. MATHEMATICAL FORMULATION 48
3. TEST CASE 51
4. RESULTS 51
5. SUMMARY 53
REFERENCES 53
Chapter 6 A Numerical Analysis on the Collision Behavior of Water Droplets 54
1. Introduction 54
2. Theories of droplet collision 55
3. Numerical method 56
4. Numerical results 57
5. Conclusions 58
REFERENCES 59
Chapter 7 Parallel URANS Simulations of an Axisymmetirc Jet in Cross-flow 62
INTRODUCTION 62
SOLUTION METHODOLOGY 63
IMPLEMENTATION AND CASE SPECIFIC DETAILS 63
SIMULATION RESULTS AND DISCUSSION 64
PARALLELIZATION 65
CONCLUSIONS 66
REFERENCES 69
Chapter 8 Parallel Performance Assessment of Moving Body Overset Grid Application on PC cluster 70
1. Introduction 70
2. Parallel Implementation of Grid Assembly 71
3. Test results 74
4. Conclusion 76
REFERENCES 77
Chapter 9 Stream Function Finite Element Method for Magnetohydrodynamics 78
1. MHD and its streamfunction approach 78
2. Finite element discretization 79
3. Nonlinear and linear solvers 80
4. Numerical experiments: Tilt Instability 81
REFERENCES 83
Chapter 10 Parallelization of Phase-Field Model to Simulate Freezing in High-Re Flow-Multiscale Method Implementation 86
1. INTRODUCTION 86
2. MULTISCALE METHOD 87
3. TWO-SCALE PHASE-FIELD MODEL IN FLOW FIELD 89
4. NUMERICAL METHODS AND RESULTS 91
5. APPROACH TO PARALLELIZATION AND RESULTS 92
REFERENCES 93
Chapter 11 Parallel Property of Pressure Equation Solver with Variable Order Multigrid Method for Incompressible Turbulent Flow Simulations 94
1. INTRODUCTION 94
2. NUMERICAL METHOD 95
3. MULTIGRID PROPERTY OF PRESSURE EQUATION SOLVER 97
4. TURBULENT CHANNEL FLOW SIMULATION 99
5. CONCLUDING REMARKS 100
REFERENCES 101
Chapter 12 Parallel Numerical Simulation of Shear Coaxial LOX/GH2 Jet Flame in Rocket Engine Combustor 102
1. INTRODUCTION 102
2. FLOW CONDITIONS 103
3. NUMERICAL METHOD 104
4. RESULTS 104
5. SUMMARY 108
REFERENCES 109
Part 3. Parallel Environment 110
Chapter 13 Construction of Numerical Wind Tunnel on the e-Science Infrastructure 110
1. INTRODUCTION 110
2. e-AIRS Portal 111
3. Computational Simulation Service 112
4. Remote Experiment Service 115
5. Collaboration Environment 116
6. Conclusions 116
References 117
Chapter 14 Efficient Distribution of a Parallel Job Across Different Grid Sites 118
Abstract 118
1 Introduction 118
2 Block distribution across Grid sites 119
3 Test Case 121
4 Conclusions 124
5. Acknowledgement 125
6. References 125
Chapter 15 New Cooperative Parallel Strategy for Massive CFD Computations of Aerodynamic Data on Multiprocessors 126
1. ABSTRACT 126
2. INTRODUCTION 126
3. NEW COOPERATIVE STRATEGY FOR MASSIVE PARALLEL ITERATIVE COMPUTATIONS 127
4. IMPLEMENTATION OF THE COOPERATIVE STRATEGY FOR PARALLEL CFD COMPUTATIONS 128
5. ANALYSIS OF RESULTS 130
REFERENCES 132
Chapter 16 Performance Analysis of Fault Tolerant Algorithms for the Heat Equation in Three Space Dimensions 134
1. Introduction 134
2. Definition of the Application Benchmark 135
3. Fault Tolerant Approaches 135
4. Performance Recovery Results 138
5. Conclusion 139
REFERENCES 140
Chapter 17 Parallel Deferred Correction method for CFD Problems 142
1. Motivation of Time Decomposition in the Parallel CFD context 142
2. ODE approach for CFD problem 143
3. Spectral Deferred Correction Method 144
4. Parallel SDC: a time domain decomposition pipelined approach 145
5. conclusions 146
REFERENCES 147
Chapter 18 Performance Evaluation and Prediction on a Clustered SMP System for Aerospace CFD Applications with Hybrid Paradigm 150
1. Introduction 150
2. System Overview 151
3. Performance evaluation for JAXA aerospace CFD applications on the CeNSS 151
4. Performance prediction for the JAXA CFD applications with hybrid programming 154
5. Summary 155
References 156
Chapter 19 Non-Intrusive Information Collection for Load Balancing of Parallel Applications 158
1. INTRODUCTION 158
2. MEASURING TIMING INFORMATION USING A MPI PROFILING LIBRARY 159
3. MEASURE TIMING INFORMATION USING PROC 161
4. HOW THE LOAD BALANCER COMMUNICATE WITH THE APPLICATION PROGRAM 161
5. EXPERIMENTAL RESULTS 162
6. CONCLUSION 164
REFERENCES 164
Chapter 20 Reuse Procedure for Open Source Software 166
1. Introduction 166
2. Related Work 167
3. Reuse Procedures of OSSs 167
4. Case Study 172
5. Conclusion 174
References 174
Chapter 21 Design of CFD Problem Solving Environment based on Cactus Framework 176
1. INTRODUCTION 176
2. CACTUS FRAMEWORK 177
3. NUMERICAL RESULTS 179
4. CONCLUSION 183
ACKNOWLEDGEMENT 183
References 183
Part 4. Multi-Disciplinary Simulation 184
Chapter 22 Prediction of Secondary Flow Structure in Turbulent Couette-Poiseuille Flows inside a Square Duct 184
1. Introduction 184
2. Governing Equations and Modeling 185
3. Numerical and parallel Algorithms 185
4. Results 186
5. Conclusion 188
6. Acknowledgments 188
REFERENCES 189
Chapter 23 The Prediction of the Dynamic Derivatives for the Separated Payload Fairing Halves of a Launch Vehicle in Free Falling 192
1. Introduction 192
2. Numerical Approach 193
3. Code Validation 194
4. Numerical Results 195
5. Conclusion 199
REFERENCES 199
Chapter 24 Variability of Mesoscale Eddies in the Pacific Ocean Simulated by an Eddy Resolving OGCM 200
1. INTRODUCTION 200
2. MODEL 201
3. PARALLEL COMPUTATION 202
4. RESULTS 202
5. CONCLUSION 203
REFERENCES 203
Chapter 25 Sensitivity Study with Global and High Resolution Meteorological Model 208
1. Introduction and motivation 208
2. The test cases 209
3. Characteristic of numerical model and performed runs 211
4. The spin up 212
4. Analysis of model performances 213
5. Future developments 216
Acknowlegment 216
Reference 217
Chapter 26 Computational Performance Evaluation of a Limited Area Meteorological Model by using the Earth Simulator 218
INTRODUCTION 218
1. THE GLOBAL/REGIONAL NON-HYDROSTATIC ATMOSPHERIC MODEL 219
2. THE ES SYSTEM 220
3. IMPLEMENTATION DETAILS AND COMPUTATIONAL EXPERIMENTS 220
4. COMPUTATIONAL RESULTS 221
5. CONCLUSIONS 224
REFERENCES 225
Chapter 27 A Scalable High-Order Discontinuous Galerkin Method for Global Atmospheric Modeling 226
1. INTRODUCTION 226
2. CONSERVATIVE DISCONTINUOUS GALERKIN MODEL 227
3. NUMERICAL TEST 229
4. PARALLEL IMPLEMENTATION 229
5. CONCLUSION 231
REFERENCES 232
Chapter 28 Weather Prediction and Computational Aspects of Icosahedral-Hexagonal Gridpoint Model GME 234
1. Introduction 234
2. Brief Description of DWD GME Model 235
3. Computational Performance of GME on Xeon Cluster (KISTI HAMEL) 236
4. GME Results: NWP Approach 237
5. Seasonal Prediction Experiments with GME 237
6. Summary 238
Acknowledgement 239
REFERENCES 239
Chapter 29 Numerical Investigation of Flow Control by a Virtual Flap on Unstructured Meshes 242
Introduction 242
Numerical Method 242
Results and Discussion 243
Conclusions 244
REFERENCES 245
Chapter 30 Implicit Kinetic Schemes for the Ideal MHD Equations and their Parallel Implementation 250
NOMENCLATURE 250
INTRODUCTION 250
IMPLICIT KINETIC SCHEMES FOR THE EULER EQUATIONS 251
IMPLICIT KINETIC SCHEMES FOR THE IDEAL MHD EQUATIONS 252
NUMERICAL RESULTS AND DISCUSSIONS 253
CONCLUSIONS 255
REFERENCES 255
Chapter 31 Parallel Simulation of Turbulent Flow in a 3-D Lid-Driven Cavity 256
1. INTRODUCTION 256
2. ANALYSIS 257
3. RESULTS 259
4. SUMMARY/CONCLUSIONS 262
REFERENCES 262
Chapter 32 Numerical Analysis of Supersonic Jet Flow from Vertical Landing Rocket Vehicle in Landing Phase 264
1. INTRODUCTION 264
2. EXPERIMENTAL CONFIGURATION 265
3. NUMERICAL METHODS 267
4. RESULTS AND DISCUSSIONS 267
5. SUMMARY 271
ACKNOWLEDGEMENT 271
References 271
Chapter 33 Parallel Computation of a Large Number of Lagrangian Droplets in the LES of a Cumulus Cloud 272
1. INTRODUCTION 272
2. MODEL 273
3. PARALLELIZATION TECHNIQUE 274
4. RESULTS 276
5. CONCLUSION 277
REFERENCES 277
List of Figures 277
Chapter 34 A Fluid-Structure Interaction Problem on Unstructured Moving-Grid using Open MP Parallelization 280
1. INTRODUCTION 280
2. THREE-DIMENSIONAL UNSTRUCTURED MOVING-GRID FINITE-VOLUME METHOD 281
3. A FLUID-STRUCTURE INTERACTION PROBLEM 285
4. CONCLUSIONS 287
REFERENCES 287
Chapter 35 Investigation of Turbulence Models for Multi-Stage Launch Vehicle Analysis Including Base Flow 288
1. INTRODUCTION 288
2. GOVERNING EQUATIONS AND NUMERICAL TECHNIQUES 289
3. NUMERICAL RESULTS 291
4. CONCLUSION 294
ACKNOWLEDGEMENT 294
References 295
Chapter 36 Path Optimization of Flapping Airfoils based on NURBS 296
1. INTRODUCTION 296
2. PERIODIC PATH BASED ON NURBS 297
3. NUMERICAL METHOD 297
4. RESULTS AND DISCUSSION 299
5. CONCLUSIONS 301
6. ACKNOWLEDGMENT 303
REFERENCES 303
Chapter 37 Aerodynamic Optimization Design System for Turbomachinery Based on Parallelized 3D Viscous Numerical Analysis 304
1. INTRODUCTION 304
2. AERODYNAMIC OPTIMIZATION PLATFORM 305
3. PARALLEL COMPUTATION PLATFORM 307
4. OPTIMIZATION CASES AND DISCUSSION 308
5. CONCLUSION AND PROSPECT 310
Acknowledgements 311
REFERENCES 311
Chapter 38 Genetic Algorithm Optimization of Fish Shape and Swim Mode in Fully-Resolved Flow Field 312
1. INTRODUCTION 312
2. GOVERNING EQUATIONS AND NUMERICAL METHODS 313
3. SPECIFIC STUDY OF PROPULSIVE STAGE 315
4. OPTIMISATION WITH GENETIC ALGORITHM 316
5. CONCLUDING REMARKS 319
REFERENCES 319
ACKNOWLEDGEMENT 319

Textauszug

Parallel hybrid particle-continuum (DSMC-NS) flow simulations using 3-D unstructured mesh

J.-S. Wua; Y.Y. Liana; G. Chengb; Y.-S. Chenc    a Mechanical Engineering Department, National Chio Tung University, Hsinchu 30050, TAIWAN
b Mechanical Engineering Department. University of Alabama, Birmingham, AL 35294, USA
c National Space Organization. Hsinchu 30048, TAIWAN

1 Introduction

Several important gas flows involve; flow fields having continuum and rarefied regions, e.g., hypersonic flows [2]. vacuum-pump flows with high compression ratio [3], expanding RCS (reaction control system) nozzle plumes in aerospace and space applications [4], physical vapor deposition processes with heated sources [5], pulsed-pressure chemical vapor deposition processes [6], among other things. Understanding of the underlying physics through simulation and modeling in the above flows are important for the success of these disciplines, in addition to the usual experiemental studies. In general. these flow problems arc governed by the Boltzmann equation, which is very difficult to solve numerically or analytically due to the existence of collision integral and high number of phase-space dimensions. It is well known the direct simulation Monte Carlo (DSMC) method [7] can provide more physically accurate results in flows having rarefied and non-equilibrium regions than the continuum flow models such as the NS equations. However, the DSMC method is extremely computational expensive especially in the near-continuum and continuum regions, which prohibits its applications to practical problems with large domains. In contrast, the computational fluid dynamics (CFD) method, employed to solve the Navier-Stokes (NS) or Euler equation for continuum flows, is computationally efficient in simulating a wide; variety of flow problems. But the use of continuum theories for the flow problems involving the rarefied gas or very small length scales (equivalently large Knudsen numbers) can produce inaccurate results due to the breakdown of continuum assumption or thermal equilibrium. A practical approach for solving the flow fields having near-continuum to rarefied gas is to develop a numerical model combining the CED method for the continuum regime with the DSMC method for the rarefied and thermal non-equilibrium regime. A well-designed hybrid scheme is expected to take advantage of both the computational efficiency and accuracy of the NS solver in the continuum regime and the physical accuracy of the DSMC method in the rarefied or thermal non-equilibrium regime.

In the past, there were several efforts in developing the hybrid particle-continuum scheme. Most studies employed structured grid for both the particle and continuum solvers [8-12], in which the location of breakdown interfaces between continuum and rarefied regions was specified in advance [8,9,11,12] or identified after “one-shot” CFD simulation [10]. One immediate disadvantage by employing structured grid is that the pre-specified breakdown interface does not follow faithfully the interface deterrninde by some breakdown parameters [7,13]. which may in turn either increases the runtime or induce inaccuracies of the solution. In addition, some techniques, such as particle cloning [12], overlapped region [8,12] and iteration [8] between particle and continuum regions, arc used to reduced the statistical uncertainties in coupling the two solvers. Among these, the hybrid schemes developed by Wang et al. [11] and Roveda et al. [12] are potentially suitable for simulating unsteady flows, while the others were only designed for simulating steady flows. In the above, only one-dimensional and two-dimensional flows were demonstrated and extension to parallel or three-dimensional simulation has not been reported to the best knowledge of the authors. Recently, Wu et al. [1] has developed a parallel hybrid DSMC-NS scheme using 3D unstructured mesh. Parallel implementation is realized on distributed-memory parallel machines, e.g., PC-cluster system. In this method, a domain overlapping strategy, taking advantage of unstructured data format, with Dirichlet-Dirichlet type boundary conditions based on two breakdown parameters, was used iterativcly to determine the choice of solvers in the spatial domain. The selected breakdown parameters for this study include: 1) a local maximum Knudsen number, proposed by Wang and Boyd [13], defined as the ratio of the local mean free path and local characteristic length based on property gradient and 2) a thermal non-equilibrium indicator defined as the ratio of the difference between translational and rotational temperatures to the translational temperature. A supersonic nitrogen flow (M∞ = 4) over a quasi-2-D 25° wedge and a nitrogen flow, which two near-continuum parallel orifice jets underexpand into a near-vacuum environment, were simulated to verify its validity in simulating gas flows involving rarefied and continuum regions. In the present paper, this recently developed hybrid DSMC-NS scheme is described briefly for completeness. Improvement of the NS solver by using a pressure-based solver is described next. Then, several practical experiences learned from the development and verifications of the implementation are shown in detail. Finally, simulation results for a RCS (reaction control system) nozzle plume are presented to demonstrate its capability in simulating realistic flow problems.

2 Hybrid DSMC-NS Scheme Using 3D Unstructured Mesh

In the proposed coupled DSMC-NS method [1], steady-state flow calculation is assumed. There were two numerical flow solvers included: one is the 3-D DSMC code for rarefied or continuum breakdown and thermally non-equilibrium regions, named PDSC (Parallel Direct Simulation Monte Carlo Code), developed by Wu’s group [14, eg] and the other is HYB3D, a density-based 3-D Euler and Navier-Stokes solver for continuum regions, developed by Koomullil [15, eg]. In the present study for simulating the RCS (reaction control system) nozzle plume issuing from the spacecraft, a pressure-based NS solver, named UNIC-UNS, using 3D unstructured mesh, developed by Chen and his coworkers [16,17, eg] that is applicable at all speeds, is used for continuum regions. It is rather straightforward to exchange the information between the PDSC and the UNIC solver in the proposed DSMC-NS scheme because both methods use the unstructured grid topology with parallel computing. However, proposed coupling procedures between DSMC and NS solvers are not limited to any specific codes, and selection of these two solvers is only for the purpose of demonstration. Both codes are introduced briefly in the following, respectively, for completeness.

2.1 DSMC Solver (PDSC)

Details of the features of the PDSC code can be found in the reference [14] and are only briefly described here for brevity. PDSC features 3-D unstructured grid, parallel computing with dynamic domain decomposition using MPI, variable time-step scheme with adaptive mesh refinement and treatment of high-temperature chemical reacting flows. In addition, iterative pressure boundary treatment is also available for treating internal flows. It can be implemented efficiently on general distributed-memory parallel machines, such as PC-cluster system.

2.2 Navier-Stokes Solvers

Details of the HYB3D that is an density-based NS code have been described in detail elsewhere [15] and are skipped here for brevity. The present Navier-Stokes solver (UNIC-UNS), developed by Chen and his coworkers [16,17] employs an unstructured-grid topology and has the following important features: 1) Cell-centered finite-volume for the numerical integration of governing equations, 2) An upwind method with linear reconstruction scheme for convective flux evaluation, 3) Modified pressure-velocity-density coupling algorithm of the SIMPLE type with added pressure damping term, 4) Parallel computing based on domain decomposition with message passing interface (MPI), 5) Turbulent flow simulation capability with the standard and extended k-e turbulence models, and 6) General chemical reacting flow treatment. One of the most important features of this NS solver is the use of pressure-based method which allows accurate simulation of the flows at all speeds. Either implicit first-order Euler time-marching or second-order time-centered scheme can be used for time integration. A second-order spatial accuracy is achieved using Taylors series expansion and the gradients of the flow properties are computed using a least-square method. The creation of local extreme during the higher order linear reconstruction is eliminated by the application of limiter proposed by Barth and Jespersen [18]. Parallel computing of the NS solver also incorporates the graph-partition tool, METIS, which is the same as that in the PDSC.

2.3 Breakdown Parameters

A continuum breakdown parameter, proposed by Wang and Boyd [13] for hypersonic flows, is employed in the present, hybrid DSMC-NS method as one of the criteria for selecting proper solvers. The continuum breakdown parameter Klimax is defined as

nmax=max[KnD,KnV,KnT]

  (1)

where KnD KnV and KnT are the local Knudsen numbers based on density, velocity and temperature, respectively. They can be calculated from the following general...

Produkt-Details

• Erscheint lt. Verlag: 12.9.2007
• Sprache: englisch
• Themenwelt: Mathematik / Informatik ► Informatik ► Theorie / Studium
• Themenwelt: Naturwissenschaften ► Physik / Astronomie ► Strömungsmechanik
• Themenwelt: Technik ► Bauwesen
• Themenwelt: Technik ► Maschinenbau
• ISBN-10: 0-08-055004-5 / 0080550045
• ISBN-13: 978-0-08-055004-6 / 9780080550046