Physics-based intelligent prognosis for rolling bearing with fault feature extraction

Abstract

Successful condition monitoring of rotating machinery relies on the accurate prediction of the remaining life of the rotating components. This paper proposes a physics-based prognostic model for rolling element bearings using the realized volatility (RV) and wavelet neural network (WNN) to predict the remaining life of bearings. The proposed method overcomes the difficulties of forecasting the failure of bearings under various degradation patterns and improves the accuracy of prediction. The faulty signal is extracted using an AR filter to reveal the degradation trend. The prognosis is performed by calculating the RV and energy ratio, which are used to identify the abnormality of vibration in the system and degradation patterns before catastrophic failure occurs. Then, the WNN incorporated with physical inputs from the properties of the bearings predicts the failure cycles. The prediction yields a satisfactory result between the experimental and predicted failure cycle. The prognostic accuracy of the physics-based WNN model is compared with a widely used time-series model, the auto-regressive integrated moving average (ARIMA) model. The WNN model has an advantage of reducing error over the ARIMA model in various degradation patterns and overcomes the disadvantage of models requiring additional data after significant damage is present in the bearings.