A new uncertainty reduction-guided single-loop Kriging coupled with subset simulation for time-dependent reliability analysis

Time-dependent reliability analysis (TRA) of rare failure events is an imperative task in structural engineering, and the single-loop Kriging method is considered one of the most promising methods when dealing with time-consuming performance functions. However, the adaptive learning strategy still has improvement potential to achieve a better balance of accuracy and efficiency. To fill this gap, a new uncertainty reduction-guided single-loop Kriging coupled with subset simulation (UR-SLK-SS) method is proposed for TRA. Firstly, the uncertainty estimation of each intermediate failure probability is derived to measure the contribution of each time trajectory, which includes not only the individual contribution of the time trajectories but also the correlated contribution between the time trajectories. Then, a learning function for simultaneously selecting new random samples and time nodes is built. In the time trajectories with the largest uncertainty, the time node with the largest reduction in uncertainty is selected for adaptive updating. Further, a two-stage stopping criterion is proposed to guarantee the accuracy of the results as well as higher efficiency. Seven case studies are presented to demonstrate the better capability of the proposed method.