Structural reliability analysis by a Bayesian sparse polynomial chaos expansion

Accurate computation of failure probability considering uncertain input parameters is very challenging within limited computational cost. An efficient surrogate model, referred to here as sparse variational Bayesian inference based polynomial chaos expansion (SVB-PCE), is formulated in this paper for reliability analysis. The sparsity in the polynomial basis terms is introduced by the automatic relevance determination (ARD) algorithm and the coefficients corresponding to the sparse polynomial bases are computed using the VB framework. The reliability analysis is performed on four typical numerical problems using the SVB-PCE model. The failure probability and the reliability index for all the examples are assessed accurately by the SVB-PCE model using fewer number of model evaluations as compared to the state-of-art methods. Further, the ARD enables to capture the most important terms in the polynomial bases which also reduces the computational cost in assessing the failure probability.

Automated summary

An efficient surrogate model, SVB-PCE, is proposed for accurate computation of failure probability and reliability index, achieving accurate assessment with fewer model evaluations. The model reduces computational cost by capturing the important terms in the polynomial bases using the automatic relevance determination (ARD) algorithm.