A Subdomain uncertainty-guided Kriging method with optimized feasibility metric for time-dependent reliability analysis

Active learning strategy combined with single-loop Kriging method has attracted much attention for Time-dependent Reliability Analysis (TRA). Large sample pool across all time instants is required to comprehensively cover the failure surface. Identifying training samples with most existing methods requires assessing the response of the entire sample pool, leading to high time costs. Similarly, to assist in identifying critical sampling regions, evaluating failure probability with a large sample pool in each iteration is also time-consuming. Additionally, higher-order uncertainty information is typically ignored, impairing the sampling efficiency. To address these issues, a Subdomain Uncertainty-guided Kriging (SUK) Method is proposed. Stochastic processes are first equivalently converted to random variables. By simultaneously sampling random variables and time parameter, an equivalent sample pool with significantly reduced size is generated for adaptive sampling. With a concise subdomain uncertainty assessment function, critical sampling region, i.e. sensitive subdomain, is distinguished efficiently. By comprehensively considering both the expectation and standard deviation of feasibility function, a novel Optimized Feasibility Metric (OFM) is then proposed for active learning. The proportion of misclassified samples is analytically deduced as stopping criterion. Finally, comparison results on four examples demonstrate the good performances of the proposed subdomain uncertainty-guided Kriging method.