The well-established Bayesian failure probability inference (BFPI) framework offers a solid foundation for developing new Bayesian active learning reliability analysis methods. However, there remains an open question regarding how to effectively leverage the posterior statistics of the failure probability to design the two key components for Bayesian active learning: the stopping criterion and learning function. In this study, we present another innovative Bayesian active learning reliability analysis method, called ‘Weakly Bayesian Active Learning Quadrature’ (WBALQ), which builds upon the BFPI framework to evaluate extremely small failure probabilities. Instead of relying on the posterior variance, we propose a more computationally feasible measure of the epistemic uncertainty in the failure probability by examining its posterior first absolute central moment. Based on this measure and the posterior mean of the failure probability, a new stopping criterion is devised. A recently developed numerical integrator is then employed to approximate the two analytically intractable terms inherent in the stopping criterion. Furthermore, a new learning function is proposed, which is partly derived from the epistemic uncertainty measure. The performance of the proposed method is demonstrated by five numerical examples. It is found that our method is able to assess extremely small failure probabilities with satisfactory accuracy and efficiency. • Another novel Bayesian active learning method is proposed for structural reliability analysis. • An alternative measure for capturing the epistemic uncertainty in the failure probability is derived. • A new stopping criterion is developed based on the measure and the posterior mean of the failure probability. • A new learning function is proposed based partly on the measure. • Our method can evaluate extremely small failure probabilities on the order of 1 0 − 9 – 1 0 − 7 .
Stored only when both papers are in the corpus — 29 citations exist upstream.