A Laplace asymptotic integral-based reliability analysis method combined with artificial neural network

Reliability analysis aims to estimate the exceedance probability of structural response over a prescribed threshold value, which is essentially a multi-dimensional integral problem. In this paper, a novel and efficient reliability analysis method combining Laplace asymptotic integral and artificial neural network is proposed. Laplace asymptotic integral is employed to approximate the multi-dimensional integral to calculate the failure probability, and an active artificial neural network is taken as the surrogate model. Different from the existing surrogate model-based reliability methods, the proposed method focuses on approximating the limit state function in the vicinity of the target design point, instead of that in the whole sample space. No candidate sample population for active learning is needed, which significantly reduces the requirement for computer memory and improves the calculation efficiency. A novel learning function based on an optimization formulation is proposed in this paper to select the most informative samples for local approximation. Some numerical examples as well as a practical engineering issue are studied, and the superiority of the proposed method over other surrogate model-based reliability analysis methods in terms of efficiency and accuracy are verified. (c) 2022 Elsevier Inc. All rights reserved.

Automated summary

In this paper, a novel and efficient reliability analysis method combining Laplace asymptotic integral and artificial neural network is proposed. The method approximates the multi-dimensional integral using Laplace asymptotic integral and selects the most informative samples for local approximation, improving calculation efficiency and accuracy.