New paper published in Structural Safety

In our work entitled “Reliability Analysis for Non-Deterministic Limit States Using Stochastic Emulators”, we propose a methodology for the reliability analysis of systems modelled by stochastic simulators. These are models that produce different outputs for identical inputs due to internal randomness. We show that, in this setting, failure becomes a probabilistic concept conditioned on the input, and, as consequence, different estimators of the failure probability can be defined. One such estimator relies on the conditional failure probability, which quantifies the probability of failure given a specific input. We approximate this quantity using two recently proposed surrogate modelling approaches: generalized lambda models (GLaM) and stochastic polynomial chaos expansions (SPCE). Once trained, these emulators yield closed-form approximations of the conditional failure probability, enabling more efficient failure probability estimation.

For more information, please visit the external page journal website.external page