Integrated multivariate degradation prediction by RVM

Degradation prediction is important for safety related products to avoid failures. When the degradations of multiple parameters of a product is taken into account, traditional univariate degradation prediction method is not applicable, especially when the parameters are correlated. To cope with this problem, Mahalanobis distance is proposed, to combine multiple parameters into one unified index. Then healthy baselines of the product are determined based on the unified index. Finally, the method of Relevance Vector Machines is applied to predict the change trend of the unified index and find the failure time. A case study is presented to prove the validity of our proposed method.