An active-learning method based on hierarchical Kriging model for multi-fidelity reliability analysis

Abstract

An active-learning method based on hierarchical Kriging model (AHK-MCS) for multi-fidelity reliability analysis is proposed. It is a two-stage method that includes: building the low-fidelity (LF) model using LF samples firstly, and constructing the high-fidelity (HF) model based on the LF model using HF samples secondly. Additionally, an experimental design method based on the LF model is proposed. The two-stage framework facilitates the high-precision LF model and initial HF samples selection, consequently enhancing the efficiency of the HF active-learning process. The AHK-MCS method is compared with two one-stage methods through eight numerical examples. The results show that the proposed method is capable of providing an accurate estimation of failure probability with fewer HF samples. Moreover, the proposed experimental design method is evaluated and demonstrated to result in a reduction of HF samples. The influence of the LF model is assessed, indicating that the utilisation of a precise LF model can diminish the number of HF samples required for the construction of the HF model. The performance of the AHK-MCS method under different cost ratios is also investigated.