High-fidelity Simulation-based Optimum Design Utilizing Computing Grid Technology

This paper describes optimization framework developed for simulation-based design which utilizes newly emerging supercomputing platform called Grid. Also, experiments with structural design based on parallel flnite element analysis of detailed models with approximately 1 million degrees-of-freedom using the optimization framework is presented. For simulation based design tasks, Grid-enabled genetic algorithm and particle swarm optimization algorithm are implemented. Both methods are attractive and efiective for distributed computing environments. The algorithms are inherently easy to parallelize and especially for Grid environments characterized by large amount of resources with difierent computational power and networks, cooperative algorithm such as the GA and PSO can be regarded as the most suitable. Also, to efiectively implement optimization algorithms on highly dynamic Grid platform, GA and PSO are implemented with asynchronous fltness evaluations. Several design experiments are described to validate the efiectiveness of the approach and to demonstrate the potential of utilizing Grid technologies for high-fldelity simulationbased design. These include optimum design of a composite blade system of a novel VTOL vehicle and structural design of satellite system. Each design case is composed of the sequential execution of the automatic mesh generation, parallel flnite element structural analysis for both static and natural frequency responses. and post-processing for the evaluation of the fltness of the design case. For the parallel flnite element analysis IPSAP, an in-house, parallel, flnite element program based on high-performance parallel multifrontal solver is employed.