Inverse Reliability Analysis with EGRA

摘要

This paper presents an extension to the Efficient Global Reliability Analysis (EGRA) method to enable it to perform inverse reliability analysis (i.e., to determine the response value that corresponds to a specified probability level). EGRA makes use of Gaussian process modeling to construct a surrogate to the true performance function. The method is based on the idea that an accurate surrogate model can be constructed efficiently by focusing the training data in the vicinity of the limit state. The algorithm is naturally amenable to forward reliability analysis, since the limit state contour for which it searches is known a priori. This work presents an extension of the method for solving the inverse reliability analysis problem by continuously updating the state of knowledge about the target limit state as the surrogate model is built. Additional discussion is presented on assessing convergence of the analysis based on reliability confidence bounds due to surrogate model uncertainty. The approach is demonstrated on a small collection of example problems.