Health index construction and remaining useful life prediction of rolling bearings

Abstract

The problems in the remaining useful life (RUL) prediction of rolling bearings are difficulty in determining the failure threshold and large prediction error with a single prediction model. To solve them, based on the adaptive boosting integrated relevance vector machine model (AdaBoost_RVM), a method for constructing health indices and predicting RUL is proposed. The experimental results show that the failure thresholds of all the different bearings are 1 using the proposed method, and compared with the single RVM model, it yields a bearing RUL prediction that has a smaller error and is closer to the true value.