(An) efficient sampling method for multi-fidelity Kriging model based reliability analysis

Abstract

The reliability analysis is demanded to quantify the probability of failure in the engineering systems, the sampling-based reliability analysis is commonly used because of its high accuracy. A surrogate model is utilized to reduce the cost of the sampling-based method, the accuracy and efficiency can be improved through the sequential sampling. However, when a high-fidelity sample needed to construct a surrogate model is time-consuming, an accurate model cannot be made because the use of samples would be limited. To alleviate the cost problem, the multi-fidelity surrogate model, which also use the low-fidelity sample having a cheap cost, can be employed. In this study, a new sampling method for the efficient reliability analysis based on the multi-fidelity surrogate mode is proposed. First, it carries out the sampling to reduce the variance of the probability of failure, so that the surrogate modeling becomes to focus on the reliability analysis. And, the candidate samples to add can be selected by using the uncertainty information corresponding to each fidelity level. Those improvements enable to prevent the waste of sampling and to make an efficient surrogate modeling. The feasibility of that the proposed method can take an accurate and efficient reliability analysis is demonstrated by application to the numerical examples and the engineering example.