An Adaptive Sampling Strategy to Minimize Uncertainty in Reliability Analysis using Kriging Surrogate Model

摘要

Kriging surrogate model is widely used for reliability analysis because of its accuracy and low cost. Despite its advantages, Kriging model contains epistemic uncertainty in prediction, which is expressed as prediction variance. This uncertainty in prediction propagates to the uncertainty in reliability analysis. Additional samples can reduce the prediction variance and thus the uncertainty in the reliability analysis. When a limited amount of resource is available, it is important to And the locations of additional samples to minimize the epistemic uncertainty. In this paper, a new adaptive sampling method is developed to update a Kriging surrogate model in order to reduce the uncertainty in the reliability analysis. Based on efficient global reliability analysis, the proposed method also considers the input aleatory randomness in order to approximate the limit state function near the design point. The result shows that the proposed method can obtain more accurate reliability analysis result with less number of samples compared to the existing method.