Failure Sampling with Optimized Ensemble Approach for Structural Reliability Analysis of Complex Problems

Abstract

Failure sampling is a structural reliability method based on modified conditional expectation suitable for complex problems for which reliability index–based approaches are inapplicable and simulation is needed. Such problems tend to have nonsmooth limit-state boundaries or are otherwise highly nonlinear. Previous studies recommended implementation of failure sampling with an extrapolation technique using Johnson’s distribution or the generalized lambda distribution. However, what implementation method works best is problem-dependent. The uncertainty of which approach provides best results for a particular problem limits the potential effectiveness of the method. In this study, a solution is proposed to this issue that eliminates this uncertainty. The proposed approach is an optimized ensemble that forms a uniquely weighted solution by utilizing the predictive capability of multiple curves to maximize accuracy for any particular problem. It was found that the proposed approach produces solutions superior to the methods of implementing failure sampling previously presented in the literature.