Robinson Social Discourse and Moral Judgement

1. Auflage 2013
ISBN: 978-0-08-088580-3
Verlag: Elsevier Science & Techn.
Format: PDF
Kopierschutz: 1 - PDF Watermark

E-Book, Englisch, 260 Seiten, Web PDF

This edited work presents a unique and authoritative look at morality - its development within the individual, its evolution within society, and its place within the law. The contributors represent some of the foremost authorities in these fields, and the book represents a collection of essays presented at a symposium on social constructivism and morality.

Autoren/Hrsg.

Robinson, Daniel N.

Inhaltsverzeichnis

1;Front Cover;1
2;Parallel Computational Fluid Dynamics: Trends and Applications;2
3;Copyright Page ;3
4;Table of Contents;8
5;Part 1: Invited Papers;14
5.1;Chapter 1. Perspectives and Limits of Parallel Computing for CFD Simulation in the Automotive Industry;16
5.2;Chapter 2. Application of Navier-Stokes Methods to Predict Votex-Induced Vibrations of Offshore Structures;26
5.3;Chapter 3. Dynamics Controlled by Magnetic Fields: Parallel Astrophysical Computations;44
5.4;Chapter 4. A Software Framework for Easy Parallelization of PDE Solvers;56
5.5;Chapter 5. Parallel Computing of Non-equilibrium Hypersonic Rarefied Gas Flows;66
5.6;Chapter 6. Large-Eddy Simulations of Turbulence: Towards Complex Flow Geometries;78
5.7;Chapter 7. Direct Numerical Simulations of Multiphase Flows;90
5.8;Chapter 8. Aerodynamic Shape Optimization and Parallel Computing Applied to Industrial Problems;98
6;Part 2: Affordable Parallel Computing;110
6.1;Chapter 9. Accurate Implicit Solution of 3-D Navier-Stokes Equations on Cluster of Work Stations;112
6.2;Chapter 10. Performance of a Parallel CFD-Code on a Linux Cluster;120
6.3;Chapter 11. Utilising Existing Computational Resources to Create a Commodity PC Network Suitable for Fast CFD Computation;128
6.4;Chapter 12. Use of Commodity Based Cluster for Solving Aeropropulsion Applications;136
6.5;Chapter 13. Using a Cluster of PC's to Solve Convection Diffusion Problems;144
6.6;Chapter 14. Building PC Clusters: An Object-oriented Approach;152
6.7;Chapter 15. The Solution of Pitching and Rolling Delta Wings on a Beowulf Cluster;160
7;Part 3: Performance Issues;168
7.1;Chapter 16. Serial and Parallel Performance Using a Spectral Code;170
7.2;Chapter 17. On the Design of Robust and Efficient Algorithms that Combine Schwartz Method and Multilevel Grids;178
7.3;Chapter 18. 2-D To 3-D Conversion for Navier-Stokes Codes: Parallelization Issues;186
8;Part 4: Load Balancing;194
8.1;Chapter 19. Dynamic Load Balancing in International Distributed Heterogeneous Workstation Clusters;196
8.2;Chapter 20. Dynamic Load Balancing for Unstructured Fluent;204
8.3;Chapter 21. Parallel Computing and Dynamic Load Balancing of ADPAC on a Heterogeneous Cluster of Unix and NT Operating Systems;212
8.4;Chapter 22. Efficient Techniques for Decomposing Composite Overlapping Grids;220
9;Part 5: Tools and Environments;228
9.1;Chapter 23. Computer Load Measurement for Parallel Computing;230
9.2;Chapter 24. Numerical Algorithms and Software Tools for Efficient Meta-computing;238
9.3;Chapter 25. Mixed C++/Fortran 90 Implementation of Parallel Flow Solvers;246
9.4;Chapter 26. COUPL+ : Progress Towards an Integrated Parallel PDE Solving Environment;254
9.5;Chapter 27. Implementations of a Parallel 3D Thermal Convection Software Package;262
9.6;Chapter 28. Development of a Common CFD Platform-UPACS-;270
10;Part 6: Numerical Schemes and Algorithms;278
10.1;Chapter 29. Numerical Investigation of Viscous Compressible Gas Flows by Means of Flow Field Exposure to Acoustic Radiation;280
10.2;Chapter 30. A New Low Communication Parallel Algorithm for Elliptic Partial Differential Equations;288
10.3;Chapter 31. Parallel Multigrid on Cartesian Meshes with Complex Geometry;296
10.4;Chapter 32. Parallelisation of a CFD Code: The Use of Aztec Library in the Parallel Numerical Simulation of Extrusion of Aluminium;304
10.5;Chapter 33. An Efficient Highly Parallel Multigrid Method for the Advection Operator;312
10.6;Chapter 34. A Parallel Robust Multigrid Algorithm for 3-D Boundary Layer Simulations;320
10.7;Chapter 35. Parallel Computing Performance of an Implicit Gridless Type Solver;328
10.8;Chapter 36. Efficient Algorithms for Parallel Explicit Solvers;336
10.9;Chapter 37. Parallel Spectral Element Atmospheric Model;344
11;Part 7: Optimization Dominant CFD Problems;352
11.1;Chapter 38. Domain Decomposition Methods Using GAs and Game Theory for the Parallel Solution of CFD Problems;354
11.2;Chapter 39. A Parallel CFD Method for Adaptive Unstructured Grids with Optimum Static Grid Repartitioning;362
11.3;Chapter 40. Parallel Implementation of Genetic Algorithms to the Solution for the Space Vehicle Reentry Trajectory Problem;370
12;Part 8: Lattice Boltzmann Methods;378
12.1;Chapter 41. Perspectives of the Lattice Boltzmann Method for Industrial Applications;380
12.2;Chapter 42. Parallel Efficiency of the Lattice Boltzmann Method for Compressible Flow;388
12.3;Chapter 43. Turbomachine Flow Simulations with a Multiscale Lattice Boltzmann Method;396
12.4;Chapter 44. Parallel Simulation of Three-dimensional Duct Flows using Lattice Boltzmann Method;404
12.5;Chapter 45. Parallel Computation of Rising Bubbles Using the Lattice Boltzmann Method on Workstation Cluster;412
12.6;Chapter 46. Performance Aspects of Lattice Boltzmann Methods for Applications in Chemical Engineering;420
13;Part 9: Large Eddy Simulation;428
13.1;Chapter 47. PRICELES: A Parallel CFD 3-Dimensional Code for Industrial Large Eddy Simulations;430
13.2;Chapter 48. Large Eddy Simulations of Agitated Flow Systems Based on Lattice-Boltzmann Discretization;438
13.3;Chapter 49. Direct Numerical Simulation of Three-dimensional Transition to Turbulence in the Incompressible Flow Around a Wing by a Parallel Implicit Navier-Stokes Solver;446
13.4;Chapter 50. Preliminary Studies of Parallel Large Eddy Simulation using OpenMP;454
13.5;Chapter 51. MGLET: A Parallel Code for Efficient DNS and LES of Complex Geometries;462
13.6;Chapter 52. Large Eddy Simulation (LES) on Distributed Memory Parallel Computers Using an Unstructured Finite Volume Solver;470
13.7;Chapter 53. LES Applications on Parallel Systems;478
14;Part 10: Fluid-Structure Interaction;486
14.1;Chapter 54. Parallel Application in Ocean Engineering. Computation of Vortex Shedding Response of Marine Risers;488
14.2;Chapter 55. Experimental and Numerical Investigation into the Effect of Vortex Induced Vibrations on the Motions and Loads on Circular Cylinders in Tandem;496
14.3;Chapter 56. Meta-computing for Fluid-Structure Coupled Simulation;504
15;Part 11: Industrial Applications;512
15.1;Chapter 57. A Parallel Fully Implicit Sliding Mesh Method for Industrial CFD Applications;514
15.2;Chapter 58. Using Massively Parallel Computer Systems for Numerical Simulation of 3D Viscous Gas Flows;522
15.3;Chapter 59. Explosion Risk Analysis - Development of a General Method for Gas Dispersion Analyses on Offshore Platforms;530
15.4;Chapter 60. Parallel Multiblock CFD Computations Applied to Industrial Cases;538
15.5;Chapter 61. Parallel and Adaptive 3D Flow Solution Using Unstructured Grids;546
16;Part 12: Multiphase and Reacting Flows;554
16.1;Chapter 62. Interaction Between Reaction Kinetics and Flow Structure in Bubble Column Reactors;556
16.2;Chapter 63. Parallel DNS of Autoignition Processes with Adaptive Computation of Chemical Source Terms;564
16.3;Chapter 64. Application of Swirling Flow in Nozzle for CC Process;572
17;Part 13: Unsteady Flows;580
17.1;Chapter 65. Computational Fluid Dynamic (CFD) Modellling of the Ventilation of the Upper Part of the Tracheobronchial Network;582
17.2;Chapter 66. Parallel Computing of an Oblique Vortex Shedding Mode;588
17.3;Chapter 67. Three-dimensional Numerical Simulation of Laminar Flow Past a Tapered Circular Cylinder;594

Leseproben

Perspectives and Limits of Parallel Computing for CFD Simulation in the Automotive Industry

H. Echtle; H. Gildein; F. Otto; F. Wirbeleit; F Klimetzek    DaimlerChrysler AG, HPC E222, D70546 Stuttgart, Germany

1 ABSTRACT

To achieve shorter product development cycles, the engineering process in the automotive industry has been continuously improved over the last years and CAE techniques are widely used in the development departments. The simulation of the product behaviour in the early design phase is essential for the minimisation of design faults and hence a key factor for cost reduction.

Parallel computing is used in automotive industry for complex CFD simulations since years and can be considered as state of the art for all applications with non-moving meshes and a fixed grid topology. The widely used commercial CFD packages (e.g. Fluent, StarCD etc.) show an acceptable performance on massively parallel computer systems.

Even for complex moving mesh models, as they are used for the simulation of flows in internal combustion engines excellent speed-ups were demonstrated recently on MPP systems and a parallel efficiency of 84% on 96 nodes of a Cray T3E-900 was achieved within the ESPRIT Project 20184 HPSICE.

In the near future parallel computing will allow a nearly instantaneous solution for selected 3d simulation cases. Within the ESPRIT Project 28297 ViSiT Virtual Reality based steering techniques for the simulation are already tested and developed. This allows the intuitive steering of a 3d simulation running on a MPP system through a direct interaction with the simulation model in VR.

Keywords

CFD

combustion

spray

grid generation visualisation

HPC

VR

parallel computing

engine simulation

computational steering

MPP

3 PROCESS CHAIN ENGINEERING SIMULATION

Due to the requirements of the market, car manufacturers are currently faced with the situation to develop more and more products for small and profitable niche markets (e.g. sport utility vehicles). This requires the development of hardware in a shorter time. In addition the development costs must be decreased to remain competitive.

In order to achieve these contradictory goals the behaviour of the new product has to be evaluated in the early design phase as precise as possible. The digital simulation of the product in all design stages is a key technology for the rapid evaluation of different designs in the early design phase, where as shown in Figure 1 the largest impact on production costs can be achieved. The costs associated with a design adjustment should be kept small by minimising changes in the pre-production or production phase. Ideally no design changes should be required after job # 1, when the first vehicle leaves the factory.

Figure 1 Typical cost Relationships for Car Development

4 CFD SIMULATION CYCLE

CFD applications are beside crash simulation the most demanding and computationally intensive application in automotive development. CFD is used for a wide range of problems including external aerodynamics, climate systems, underhood flows and the flow and combustion process in engines. In the past the usage of CFD as a regular design tool was limited mainly due to the extremely long CPU time and complex mesh generation.

A typical simulation sequence starting from CAD data and valid for in-cylinder analysis is given in Figure 2. The different steps of the entire engine simulation are depicted including the names of the simulation software used (in grey boxes). STAR-HPC is the parallel version of the numerical simulation code STAR-CD from Computational Dynamics (CD). The programs ProICE and ProSTAR are the pre-processing tools from ADAPCO used for the benchmark results shown in the figures below. Similar tools from other companies e.g. ICEM-CFD are available as well. The visualisation package COVISE is developed at the University of Stuttgart and is commercialised by VirCinity. To complete such a cycle it typically took 12 only 3 years ago and takes now one week by using advanced mesh generation tools, parallel computers and new post processing techniques.

Figure 2 CFD Simulation Cycle

Most commercially available CFD codes are implemented efficiently on MPP systems at least for non moving meshes This reduced the computer time by nearly two orders of magnitude as shown in. Figure 3 for a non moving mesh and Figure 4 for a moving mesh case. Using the implementation strategy for the coupling of StarHPC and ProICE shown in Figure 6 a parallel efficiency of 84 percent on 96 processors for moving grid problems with a reasonable grid size of 600000 cells was demonstrated and a typical simulation can be done within a day or two now, instead of several weeks. Recently similar improvements in the parallelisation of two phase flows with a lagrangian spray simulation could be shown (Figure 5) and parallel computing can be efficiently used for the design of direct injected engines with low fuel consumption as well.

Figure 3 Speed-up steady state, non moving mesh case

Figure 4 Speed-up transient, moving mesh case

Figure 5 Speed-up transient spray simulation

Figure 6 Scalable Implementation of starCD for Moving Grid Problems

The speed-up achieved in simulation automatically shifted the bottlenecks in the simulation process to the pre- and post-processing.(Figure 7) Although considerable achievements were made in the pre-processing with semi-automatic mesh generators for moving mesh models, further improvements in this domain and a closer integration with existing CAD packages are required

Figure 7 Turnaround time for engine simulation

5 ENGINE SIMULATION

An overview of the physics, which are simulated in a typical spark ignited engine configuration, is shown in Figure 8. Due to the moving valves and piston the number of cells and the mesh structure is changed considerably during a simulation run. Beside the cold flow properties the fuel spray and the combustion process has to be simulated. Spray and fluid are tightly coupled and the correct prediction of mixture formation and wall heat transfer are essential for an accurate combustion simulation. In particular the combustion process and the spray fluid interaction are still a matter of research.

Figure 8 Engine Configuration

5.1 Mathematical Method and Discretisation

The implicit finite volume method which is used in STAR-HPC discretises the three dimensional unsteady compressible Navier-Stokes equations describing the behaviour of mass, momentum and energy in space and time.

All results for engines shown here, were done with:

- k-e turbulence model with a wall function to model the turbulent behaviour of the flow,

- combustion modelling (e.g. premixed version of the 2-equation Weller model),

- several scalar transport equations to track the mixture of fresh and residual gas and the reactants.

The fuel injection is modelled by a large number of droplet parcels formed by droplets of different diameter. The number of parcels has to be large enough to represent the real spray in a statistical sense. An ordinary differential equation for every parcel trajectory has to be solved as a function of the parcel and flow properties (mass, momentum, energy, drag, heat conduction). Each droplet is considered as a sphere and based on this geometric simplification droplet drag and vaporisation rates are evaluated. In addition collision and break-up models for droplet-droplet and droplet-wall interaction are used to describe the spray and its feedback on the flow realistically.

5.2 Domain Decomposition and Load Balancing

To get scalability of a parallel application for a high number of processors it is necessary to balance the load and restrict the memory address space locally for each processor. A standard domain decomposition is used for non moving grid problems and the grid is decomposed in different parts. MPI or PVM is used for inter-processor communication in StarHPC.

For moving grid problems with sprays, as in engine simulation, an adapted decomposition strategy is required to account for:

- the number of cells in the grid, changing due to the mesh movement,

- the computational effort, depending on the complexity of physics in a cell (number of droplets, chemical reactions etc.),

Currently this problem is not yet solved in general terms.

5.3 Results

Figure 9 shows the mixing process of fresh air (blue) and residual gas (yellow) in a cross section of an engine, which is a typical result of a transient cold flow simulation. It can be seen how the piston is going down from top dead centre (step 1) to bottom dead centre (step 4). The gray surface below the intake valve at the right side is an iso-surface of a constant residual gas concentration. This type of simulation can be used to optimise valve timings or port...