Bearing Failure Prediction Technique Using Exponential Moving Average Crossover Threshold with Support Vector Regression and Kernel Regression

Abstract

An early accurate prediction of remaining useful life (RUL) is essential for improving the machine reliability and prevents system failure. This study proposes an efficient technique to evaluate the health state of bearings and estimate the RUL. It employs the Exponential Moving Average (EMA) crossover technique to actively anticipates an upcoming failure trend of the bearing and the support vector regression (SVR) to constantly predict the RUL of the bearing while its health state is still within the EMA crossover threshold. Once the health state of the bearing exceeds the EMA crossover threshold, Kernel Regression (KR) technique along with SVR will be utilized to instantly predict the failure point and estimate the RUL of the bearing. The effectiveness of the model is validated by experimental data collected from the Center for Intelligent Maintenance Systems (IMS). The proposed prognostic technique shows an effective early failure prediction with great accuracy in comparison to the common model.