A system reliability analysis method combining active learning Kriging model with adaptive size of candidate points

This paper investigates the improvement of system reliability analysis (SRA) methods which combine active learning Kriging (ALK) model with Monte Carlo simulation. In this kind of methods, a number of Monte Carlo samples are treated as the candidate points of the ALK models, and the size (or the number) of candidate points vitally affects the efficiency. However, the existing strategies fail to build the Kriging model with the optimal size of candidate points. Therefore, a certain quantity of training points was wasted. To circumvent this drawback, a strategy with an adaptive size of candidate points (ASCP) is exploited and seamlessly integrated into one of the recently proposed ALK model-based SRA method. In this strategy, the optimal size is iteratively predicted and updated according to the predicted information of component Kriging models. After several iterations, the optimal size can be approximately obtained, and the learning process can be executed with an optimal size of candidate points hereafter. Three numerical examples are investigated to demonstrate the efficiency and accuracy of the proposed method.