A Novel Two-Stage Reliability Analysis Method Combining Improved Cross-Entropy Adaptive Sampling and Relevant Vector Machine

The computational burden becomes unbearable when reliability analysis involves time-consuming finite element analysis, especially for rare events. Therefore, reducing the number of performance function calls is the only way to improve computing efficiency. This paper proposes a novel reliability analysis method that combines relevant vector machine (RVM) and improved cross-entropy adaptive sampling (iCE). In this method, RVM is employed to approximate the limit state surface and iCE is performed based on the constructed RVM. To guarantee the precision of RVM, the first level samples and the last level samples of iCE are used as candidate samples and the last level samples are regenerated along with the RVM updates. To prevent unnecessary updates of RVM, the proposed method considers the positions of the samples in the current design of experiment. In addition, based on the statistical properties of RVM and iCE, an error-based stopping criterion is proposed. The accuracy and efficiency of the proposed method were validated through four benchmark examples. Finally, the proposed method is applied to engineering problems which are working in extreme environment.