An efficient sequential anisotropic RBF reliability analysis method with fast cross-validation and parallelizability

Reliability analysis is crucial to ensure the safety and reliability of a system with uncertainty. In reliability analysis, the metamodeling method is usually engaged to reduce the related computation. The metamodel-based reliability analysis methods need to continuously improve the accuracy of important areas, which is often accompanied by an increase in the sample number. Therefore, to improve the metamodel's accuracy while ensuring its calculation efficiency, an RBF-based reliability analysis method is presented in this paper. Firstly, an anisotropic technology is proposed to further improve the local accuracy of RBF. Secondly, a fast cross-validation method is developed to construct the model variance, which can also significantly reduce the computation intensity in model training. Furthermore, a novel weighted clustering method is contrived to sample in parallel when proceeding with reliability analysis to effectively reduce the number of iterations. Calibrated against the currently prevailing methods, the superior efficiency and accuracy of the proposed methods are demonstrated by exemplifications with popular benchmark problems and a complex reliability analysis problem of a pintle nozzle.