Sliding window-based real-time remaining useful life prediction for milling tool

Traditional remaining useful life (RUL) prediction methods do not make full use of the time-series information of the sensor monitoring data, hence the prediction accuracy can not be satisfied. To deal with this issue, a real-time RUL prediction method of tool based on sliding windows is proposed in this paper. First, the time-frequency domain features are extracted from multi-channel signals collected by sensors. Considering the influence of the previous wear value data, the tool wear value data is extracted in the form of sliding windows and being put into the long short-term memory (LSTM) network together with the time-frequency domain features for model training. Finally, in the prediction stage, we similarly extract the tool wear value data using the previous predicted wear values instead of the real wear values. In this manner, the real-time RUL prediction of tools is achieved. IEEE PHM 2010 challenge data has been used to validate the effectiveness of the method. The main advantage of the method is that the time-series characteristic of the data is considered, hence the prediction accuracy is improved and real-time prediction is achieved.