A novel active learning method for profust reliability analysis based on the Kriging model
Xufeng Yang,Xin Cheng,Zeqing Liu,Tai Wang
DOI: https://doi.org/10.1007/s00366-021-01447-y
IF: 8.083
2021-06-20
Engineering With Computers
Abstract:Profust reliability analysis, in which the failure state of a load-bearing structure is assumed to be fuzzy, is investigated in this paper. A novel active learning method based on the Kriging model is proposed to minimize the number of function evaluations. The new method is termed ALK-Pfst. The sign of performance function at a given random threshold determines the profust failure probability. Therefore, the expected risk function at an arbitrary threshold is derived as the learning function of ALK-Pfst. By making full use of the prediction information of Kriging model, the prediction error of profust failure probability is carefully derived into a closed-form expression. Aided by the prediction error, the accuracy of Kriging model during the learning process can be monitored in real time. As a result, the learning process can be timely terminated with little loss of accuracy. Four examples are provided to demonstrate the advantages of the proposed method.
engineering, mechanical,computer science, interdisciplinary applications