A piecewise method for bearing remaining useful life estimation using temporal convolutional networks

Most existing research on the prediction of bearing remaining useful life (RUL) focus on building a whole-lifecycle degradation model, which may affect prediction performance due to different states during the bearing degradation process. In addition, traditional network models such as CNN and RNN have limitations in directly dealing with time series problems. This paper proposes an adaptive degradation stage division strategy and a temporal convolutional network (TCN)-based RUL piecewise estimation method, called ADSD-TCNPE. This method mainly includes three steps. (1) Extract features from different domains and select the features that are highly correlated with bearing degradation. (2) Adaptively divide the whole lifecycle of bearing into different degradation stages. (3) Establish a TCN-based piecewise degradation model for different degradation stages to accurately predict bearing RUL. Finally, experimental verification and result analysis demonstrate the effectiveness and superiority of the proposed method.