Parallel Computing Techniques for Sensitivity Analysis in Optimum Structural Design

Abstract

Among different activities of the optimum structural design using the gradient-based optimization approaches, design sensitivity analysis is the most time-consuming computational process. By introducing parallel computing techniques for sensitivity computation, significant speedup has been obtained in optimum structural design. Computation of design sensitivities is characteristically uncoupled, thus opening the door to parallelization. In this paper, two types of approaches viz. single-level and multilevel parallelisms are pursued for design sensitivities. The design sensitivities are computed using analytical and finite-difference methods. Numerical studies show that the performance of the parallel algorithms for design sensitivities on message passing systems is very good. Good speedups have been achieved in parallel multilevel sensitivity calculation. The parallel algorithms for design sensitivity analysis have been implemented on message passing parallel systems within the software platform of Parallel Computer Adaptive Language.