Efficient failure probability calculation through mesh refinement

Abstract

We present a novel way of accelerating hybrid surrogate methods for the calculation of failure probabilities. The main idea is to use mesh refinement in order to obtain improved local surrogates of low computation cost to simulate on. These improved surrogates can reduce significantly the required number of evaluations of the exact model (which is the usual bottleneck of failure probability calculations). Meanwhile the effort on evaluations of surrogates is dramatically reduced by utilizing low order local surrogates. Numerical results of the application of the proposed approach in several examples of increasing complexity show the robustness, versatility and gain in efficiency of the method.