Direct Multi-objective Optimization through LS-OPT® using Small Number of Crashworthiness Simulations

Genetic algorithms typically require a large number of simulations, which would be economically prohibitive for crash simulations without the advent of today’s cost-effective multi-core computers. A study is conducted to seek improvements while restricting the number of simulations and exploiting the ability to use parallelization. The parallelization, achieved by simultaneously running multiple simulations for each GA generation on a HP quad- core cluster, resulted in a significant time savings. Furthermore, the optimal distribution of computational effort to achieve the greatest improvement in performance was explored. A crashworthiness simulation of a vehicle with 58,000 element finite element model was used as a test example. Various population sizes and numbers of generations were tried while keeping the total number of simulations constant. The optimization performance is also compared with Monte-Carlo and space filling sampling methods. It is observed that using GA, one can find many feasible and trade-off solutions. It is beneficial to allow a greater number of generations to get good trade-off solutions. Significant improvements in the performance were observed.