Reliability analysis traditionally relies on deterministic simulators, where identical inputs yield identical outputs. However, many real-world systems exhibit stochastic behavior, producing non-repeatable outcomes even under identical conditions. Stochastic simulators account for this behavior by representing the response as a random variable, whose intrinsic variability must be considered in reliability analysis. While Monte Carlo simulation can address this problem, its computational cost is often prohibitive. Stochastic emulators have therefore been introduced as surrogate models capable of reproducing the random simulator response at reduced cost. Recent studies have shown their potential for reliability analysis, but accurate estimates may still require relatively large training sets, which can be impractical for expensive models. In this work, we propose an active learning framework to further reduce the computational effort. Focusing on stochastic polynomial chaos expansions (SPCE), we introduce a learning function that identifies regions relevant to reliability estimation where the emulator exhibits high predictive uncertainty. We further exploit the asymptotic normality of the maximum likelihood estimator to quantify local prediction uncertainty. The resulting methodology, termed active learning stochastic polynomial chaos expansions (AL-SPCE), is validated on three types of problems. In all cases, AL-SPCE significantly improves computational efficiency compared with previous surrogate-based approaches and direct Monte Carlo simulation, while maintaining accurate reliability estimates.
翻译:暂无翻译