A Hybrid Method Combining Importance Sampling and Ensemble Learning for Structural Reliability Analysis

ABSTRACT This paper proposes a hybrid structural reliability analysis (SRA) method by integrating importance sampling (IS) with ensemble learning (EL) surrogate models. The uncertainty space is defined as a joint distribution that involves uncertainties of dynamic load and structural properties. Different from general strategies which develop surrogate models to approximate limit state functions (LSF), this study applies surrogate models to fit the probability function of conditional failure events for which the structural properties are assumed deterministic. The regression problem between the structural samples and the conditional failure probability (CFP) is analyzed and the EL models are used as surrogates to fit this problem. Moreover, the estimation of the CFPs is accelerated by an IS technique, which is particularly efficient in rare events analysis. The EL model is built to predict the CFPs on a prediction set, and the overall failure probability is estimated as the expectation of these CFPs. Numerical simulations under assumptions of Gaussian/non‐Gaussian uncertainties of the structures are designed and implemented to demonstrate the efficiency and accuracy of the proposed method.