IE-AK: A novel adaptive sampling strategy based on information entropy for Kriging in metamodel-based reliability analysis

Abstract

This article focuses on the adaptive Kriging metamodel-based reliability analysis for reducing a sequential number of calls of the complex original functions. To avoid the repetitive and tedious deterministic response analysis with stochastic simulation method (including Monte Carlo Simulation and its various improvement, such as importance sampling, subset simulation) in reliability analysis, herein a novel sequential sampling strategy related to Kriging metamodel is proposed, which is implemented based on information entropy theory. In addition, the generalized F-discrepancy method is simultaneously quoted to further optimize the candidate pool to improve the effectiveness of the training metamodel. Finally, a new structural reliability analysis method is proposed, which continuously reduces the number of deterministic analysis of structures without sacrificing accuracy. To highlight the applicability of the method and verify its accuracy and effectiveness, a series of typical examples are tested and compared, including highly nonlinear limit state functions, high-dimension performance function with analytic expressions and dynamic reliability analysis of nonlinear engineering structures subject to seismic excitation with implicit performance function. Numerical results show that significant computational savings and desired accuracy can be achieved when dealing with different reliability analysis cases.