A reliable technique for remaining useful life estimation of rolling element bearings using dynamic regression models

Abstract Induction motors most often fail due to faults in the rolling element bearings. Such failures can cause long and unscheduled downtime in a production facility, which can result in huge economic losses. The prediction of imminent failures and estimation of a bearing's remaining useful life (RUL) is vital for avoiding abrupt shutdowns and scheduling maintenance. In this paper, a reliable technique for the health prognosis of rolling element bearings is proposed, which infers a bearing's health through a dimensionless health indicator (HI) and estimates its RUL using dynamic regression models. The HI measures the instantaneous vibration level of the bearing with respect to a normal baseline value. The regression models are recursively updated to capture the evolving trend in the bearing's health indicator and are then used to project the future values of the health indicator and estimate the RUL of the bearing. The RUL of a bearing is estimated after determining the time to start prediction (TSP) using a new approach. The proposed algorithm is tested and validated on the PRONOSTIA dataset, and its prognostic performance is compared with two state-of-the-art techniques that are based on the extended Kalman filter and an exponential model that is improved using particle filters. The experimental results demonstrate the excellent prognostic performance of the proposed method due to its ability to determine an appropriate TSP and dynamic calibration of the regression models to adopt to the evolving trend in the bearing health indicator.