An adaptive divergence-based method for structural reliability analysis via multiple Kriging models

This paper presents a novel multiple-surrogate method to compute the probability of failure. Recently, some adaptive methods using Kriging surrogate model have been developed for structural reliability assessment. However, a suitable regression trend in the Kriging model is so important to reduce the number of calls to performance function by adaptive Kriging method. Hence, this paper develops the original adaptive Kriging method to use different regression trends. The proposed method is based on a machine-learning algorithm, namely “query by committee”, in which adaptive training samples are selected based on the maximal disagreement between multiple surrogate models. The proposed method has a low sensitivity to the type of regression trends, because the algorithm starts with different regression trends, and inappropriate regression trends are filtered out in the next iterations. Therefore, the use of multiple surrogates can provide an efficient tool to estimate the probability of failure. In addition, two new approaches of prediction with multiple surrogates are employed in the proposed method. These approaches are based on the local and global surrogate models that provided by the learning function. The performance of the proposed method is evaluated through three analytical and two structural problems. The results show the efficiency and accuracy of the proposed method.