A hybrid active learning Kriging model for time-dependent reliability analysis

Time-dependent reliability analysis (TDRA) aims to measure the probability that a product will successfully perform its intended functions over a specified period under various sources of uncertainty. In practical engineering, it is computationally prohibitive to propagate uncertainty based on expensive and complex numerical simulation models. Currently, Kriging surrogate model-based reliability analysis has gained considerable attention. However, achieving a balance between high accuracy and computational efficiency remains challenging. Therefore, a hybrid active learning Kriging model is proposed in this paper for TDRA. Initially, a new learning function, the expected least global (ELG) function, is introduced to draw learning samples of random variables to refine the surrogate model. Under given random inputs, the expected improvement (EI) function is employed to find the optimal time instance. A practical physics-based simulation case of a two-component parallel Daniels system is provided to demonstrate the proposed method. This study contributes to advancing active learning Kriging models for TDRA.