Abstract
Stochastic simulators, which exhibit intrinsic output variability even for fixed inputs, are increasingly encountered in the modelling of complex engineering systems such as wind turbines, or in performance-based engineering under wind or seismic loading. Developing structural reliability methods for such systems, however, remains in its infancy [1]. In this talk, we first formulate the reliability problem for noisy and stochastic simulators, highlighting their distinct nature and implications. We then show that brute-force Monte Carlo simulation quickly becomes computationally intractable for estimating failure probabilities. In a second part, we introduce stochastic emulators, and in particular stochastic polynomial chaos expansions (SPCE), which have recently emerged as an appealing surrogate framework that explicitly represents latent model stochasticity. Once calibrated on limited training data, such surrogates enable failure probability estimation, albeit at a still significant sampling cost [2]. In a third part, we present an active learning strategy based on SPCE that adaptively focuses sampling on regions of the input space where output variability and proximity to failure are most critical [3]. By generating ensembles of plausible SPCE models through likelihood-based sampling of the emulator coefficients, epistemic uncertainty is quantified and exploited to guide model enrichment. The proposed methodologies are illustrated on analytical benchmark problems as well as on a realistic, data-driven wind turbine reliability assessment. [1] A. Pires, M. Moustapha, S. Marelli, and B. Sudret, Reliability analysis for data-driven noisy models using active learning, Structural Safety, vol. 112, no. 102543, pp. 1–12, 2025. [2] A. Pires, M. Moustapha, S. Marelli, and B. Sudret, Reliability analysis for non-deterministic limit-states using stochastic emulators, Structural Safety, vol. 117, no. 102621, pp. 1–14, 2025. [3] A. Pires, M. Moustapha, S. Marelli, and B. Sudret, AL-SPCE – Reliability analysis for nondeterministic models using stochastic polynomial chaos expansions and active learning, 2026, https://arxiv.org/abs/2507.04553.