Enhancing efficiency of reliability analysis via a DU learning function-based approach for Kriging surrogate models

Abstract In recent years, Kriging models utilizing active learning strategies have been widely applied to the failure probability analysis of complex structures. However, traditional Kriging models often struggle to considerably minimize the count of performance function calls or the Finite Element Method (FEM). In this study, we propose an improved DU learning function that integrates a hierarchical sampling strategy with the traditional U learning function. By introducing a sample distance optimization criterion, the proposed method enhances sample selection efficiency and guides sample points to concentrate in regions with high failure probability. This method substantially enhances both the precision and computational performance of reliability assessments. To demonstrate its efficacy, two case studies are provided, along with a probabilistic evaluation of the automotive front I-beam structure. A comparative study with existing failure probability analysis methods is conducted. The results suggest that the proposed approach substantially reduces computational cost while maintaining high accuracy, offering an efficient solution for the reliability analysis of structures.