A novel reliability analysis method for engineering problems: Expanded learning intelligent back propagation neural network

Estimating the failure probability of highly reliable structures in practice engineering, such as aeronautical components, is challenging because of the strong-coupling and the small failure probability traits. In this paper, an Expanded Learning Intelligent Back Propagation (EL-IBP) neural network approach is developed: firstly, to accurately characterize the engineering response coupling relationships, a high-fidelity Intelligent-optimized Back Propagation (IBP) neural network metamodel is developed; furthermore, to elevate the analysis efficacy for small failure assessment, a novel expanded learning strategy for adaptive IBP metamodeling is proposed. Three numerical examples and one typical practice engineering case are analyzed, to validate the effectiveness and engineering application value of the proposed method. Methods comparison shows that the EL-IBP method holds significant efficiency and accuracy superiorities in engineering issues. The current study may shed a light on pushing the adaptive metamodeling technique deeply toward complex engineering reliability analysis.