MPP-based Latin Hypercube Sampling Approach for Efficient Reliability Analysis

Abstract

This paper introduces a MPP-based Latin Hypercube (MBLHC) sampling approach for efficient reliability analysis. This new approach updates a MPP-based Monte Carlo (MBMC) simulation approach which was developed to increase the confidence on probability estimate of MPP-based reliability methods. MBLHC combines the MPP-based importance sampling method and Latin Hypercube sampling method in the failure region. This approach first computes a most probable point (MPP) and builds a conservative failure region after the MPP is adjusted. Then importance samples are randomly generated in the region, which provides conditional probability given a failure region. This approach tries to improve the MBMC approach by replacing regular Monte Carlo random samples with stratified LHC samples in the approximate failure region. Enhanced performance of this approach is demonstrated using two examples of nonlinear limit-state functions.