A novel ensemble multiple kernel relevance vector regression modelling for reliability analysis

Reliability analysis is a crucial aspect of evaluating and enhancing product quality. A novel ensemble multiple kernel relevance vector regression (EMKRVR) modelling approach is introduced in this context. This approach incorporates adaptive learning strategies, effectively combining kernel functions to reduce the number of calls made to the performance function, thus improving efficiency and accuracy in reliability analysis. The EMKRVR model is further strengthened by the integration of Monte Carlo simulation (MCS), further enhancing its precision. Notably, an active learning function that focuses on areas with significant prediction errors is adopted. This allows the model to refine its predictions continuously. A hybrid efficient stopping criteria is employed for automatic termination. The results from three illustrative examples validate that our approach provides accurate failure probability estimates with fewer performance function evaluations compared to traditional methods. This method shows great effectiveness in the realm of quality improvement and product reliability analysis.