Motor Failure Prediction Using Hybrid Entropy and Combined Forecasting Model

Abstract

The fault prognosis of the motor plays a key role in reducing unplanned maintenance and improving machine reliability and safety. The main problem of industrial applications lies in usually only a small amount of operation data of motors is available. Establishing an effective forecasting model is a challenging task. A novel prognostics approach based on the hybrid entropy of motor current signal and a combined forecasting model is proposed. First, the wavelet packet energy entropy and Renyi spectrum entropy are extracted from the online motor current signal and then are integrated into a unified one. Since the hybrid entropy describes the change in current signals from the views of concentration degree of time-frequency-domain energy and the uniformity degree of spectrum distribution systematically, it represents motor working conditions accurately. Next, a hybrid approach based on wavelet transform, autoregressive integrated moving average (ARIMA), and improved GM(1, 1) model is employed. The time series of entropy values was decomposed into different trend items by wavelet transform, and the growth trend and random trend are described by the background value optimization GM(1, 1) model and ARIMA model, respectively. Finally, the prediction output was obtained by wavelet reconstruction. Industrial experiment results demonstrate the effectiveness of the proposed approach for motor fault prediction based on small amounts of data.