PRA–MFCS: Performance Reliability Analysis Based on Multi–Fidelity Simulation and Clustering Surrogate Model Under Multiple Failure Modes

Surrogate models are widely utilized in performance reliability analysis due to their superior abilities in balancing computational cost and accuracy. However, few of the existing relevant studies effectively integrate multi-fidelity information, leading to inefficiency in estimating small failure probabilities. To address this, a clustering-based surrogate modeling method under multiple failure modes is proposed, systematically incorporating both high- and low-fidelity data. Specifically, an active learning strategy guided by clustering selectively retains high-value sample points to promote surrogate accuracy during iteration. Besides, it integrates mixed-weight importance sampling to evaluate system failure probability while reflecting the contribution of individual failure modes, with optimal model parameters via particle swarm optimization algorithm. The proposed approach enhances the efficiency of small failure probability analysis by innovatively integrating clustering-based multi-fidelity data and quantifying failure mode interactions with mixed weights. Numerical and engineering studies demonstrate that PRA-MFCS achieves superior accuracy and efficiency compared to traditional channels, providing a reliable tool for the refined design of complex mechanical systems.