Multiple related degradation model based on improved composite kernel function and relevance vector regression algorithm

The study of system reliability modeling and life prediction considering degradation is one of the current hot research directions in the field of reliability. However, the correlation between different failure processes and the correlation shown by the failure time among system components have brought difficulties to system degradation modeling. The current multiple related degradation modeling research generally exists for deterministic factors leading to poor generality, computational complexity and other shortcomings. Therefore, this paper introduces a multiple related degradation modeling research with multiple related degradation products as the object of research, based on the relevance vector regression algorithm and the introduction of composite kernel function. The research breakthroughs in modeling product degradation under a combination of uncertainties, and addresses the lack of computational complexity through algorithmic improvements. First, the method of constructing relevance vector regression model based on composite kernel function is studied. On the one hand, multiple kernel functions are selected and composed into a composite kernel function based on information entropy. On the other hand, the Simulated Annealing algorithm is used to solve the global optimal parameter values. Then, the research of optimization algorithm for solving the parameters of relevance vector regression is carried out, and the sequential sparse Bayesian learning algorithm is used to solve the parameter values in the relevance vector regression model. Finally, a multiple related degradation model is established. Through the case study, the remaining life prediction results of multiple related degradation products obtained by using this model are better.