A two-stage method based on extreme learning machine for predicting the remaining useful life of rolling-element bearings

Abstract Rolling-element bearing is one of the main parts of rotating equipment. In order to avoid the mechanical equipment damage caused by the sudden failure of rolling-element bearings, it is necessary to monitor the condition of bearing and predict its life. Therefore, a two-stage prediction method based on extreme learning machine is proposed to predict the remaining useful life of rolling-element bearings quickly and accurately. This method uses the relative root mean square value (RRMS) to divide the operation stage of the bearing into two stages: normal operation and degradation. Starting from the normal operation stage, according to the principle of univariate prediction, a feedback extreme learning machine model is constructed for real-time short-term prediction of bearing degradation trend. Once the predicted value shows that the bearing has entered the degradation stage, the sensitive features are selected as the input by correlation analysis, and the multi variable feedback extreme learning machine model, which takes into account the dual advantages of multivariable regression and small sample prediction, is constructed to predict the remaining useful life. The experimental results show that the proposed method has higher short-term prediction accuracy and faster operation speed in the case of limited learning sample size.