A Kriging-based important region sampling method for efficient reliability analysis

Abstract

ABSTRACT Reliability analysis methods such as active learning Kriging surrogate model combined with simulation-based methods have been paid much attention in recent years. These techniques can reduce the computational cost to a certain extent. However, the computational burden may still be heavy for complex engineering problems. To address these issues, a Kriging-based important region sampling method is proposed for efficient reliability analysis. The new method is an improvement on the original active learning reliability method combining Kriging and Monte Carlo simulation (AK-MCS), and three strategies are developed to enhance the original method: 1) a new strategy, which is called the key point method, is utilized to define the initial design of experiment (DoE) instead of the Latin hypercube sampling; 2) the concept of dynamic important region/uncertain region and importance factor is proposed to avoid adding useless sample points to the DoE, which have little effect on the accuracy improvement of the Kriging model; 3) the redundant region is introduced to make the distance between the new and existed sample points be larger than a certain value and avoid the information redundancy caused by too close sample points. Five examples are utilized to demonstrate the efficiency and accuracy of this new method.