Fault prognostic algorithm based on relevance vector machine regression

To solve a kind of fault prognostic problem, an algorithm based on relevance vector machine (RVM) regression is presented. The algorithm employs a relevance vector machine to learn the hidden information about system fault evolution from historical datasets. Then it uses the learned models to predict the future trend of system fault. The algorithm adopts the ideas from recursive calculation process of time series multi-step ahead prediction. Besides, it fully takes into account the prediction uncertainty transfer problem of the recursive computation process. Monte Carlo sampling approach is introduced into above recursive prediction process, which avoids the limitation of choosing kernel functions of relevance vector machine. The prediction outputs of the algorithm use the form of random distributions of targeted system remaining useful lifetime, which is more realistic as opposed to the form of certainty values traditional algorithms used. Compared with several traditional fault prognostic algorithms, the simulation result demonstrates that RVM is superior to the traditional ones.