Abstract:

Constructing accurate surrogate models is an effective solution to addressing the problem of multi-dimensional design variables and implicit nonlinear responses in the reliability design of complex structures. However， using experiment design based on a predetermined sample size to construct surrogate models may face challenges of inefficiency or insufficient accuracy. Therefore， an active learning PC-Kriging model for reliability analysis is proposed， which combines the advantages of Polynomial Chaos Expansion for enhancing global approximation accuracy and Kriging for capturing local features. The active learning strategy is utilized to adaptively select the optimal sample points to minimize the training sample size， reducing computational cost of structural performance analysis， and improving analysis efficiency. Further， an active learning PC-Kriging model-driven multi-software co-design framework is constructed. Secondary development of pre-processing and post-processing software is conducted to enable seamless integration of parametric modeling， performance analysis， and post-processing， forming a comprehensive automated analysis workflow. Finally， reliability analysis is performed using a battery pack structure as a case study to verify the efficiency and accuracy of the proposed method.

Key words: structural reliability analysis, active learning, surrogate model, PC-Kriging, multi-software collaboration