Active learning-based structural reliability evaluation Kriging model and sequential importance sampling

Abstract

The efficient analysis of structural system reliability with multiple failure modes and small failure probabilities presents significant challenges for active Kriging methods. This paper proposes a novel method, PAK-SEIS, which combines the Parallel Active learning Kriging model and the Sequential Importance Sampling method to overcome these challenges. The proposed method includes a new sequential importance sampling method that integrates the sequential Monte Carlo simulation (MCS) and kernel density estimation to construct the optimal importance density. Additionally, the proposed parallel learning strategy allows for the selection of multiple new training samples, reducing performance function calls and refining iterations of Kriging models. The PAK-SEIS method gradually approximates the optimal importance density by updating multiple new training samples and failure modes in parallel at each iteration based on a small number of candidate samples. Candidate samples generated in the final sequence can cover all limit state surfaces, and all Kriging models are accurately constructed. The PAK-SEIS method can effectively address system reliability problems with small failure probability, multiple failure regions, and implicit functions. Two examples demonstrate the efficiency and accuracy of the proposed PAK-SEIS method.