Method for Predicting the Remaining Useful Life of Cutting Tools Based on an Improved Tcn Network

The accurate prediction of the remaining useful life of Computer Numerical Control cutting tools has a direct impact on the quality of the processed products. Currently, data-driven deep learning methods are widely used for predicting tool remaining useful life. As the cutting process is achieved by the rapid rotation of the cutting head, the collected dataset has a prominent periodicity. However, most deep learning-based prediction models fail to utilize this feature for tool life prediction. Therefore, this study proposed an effective method to extract and apply the periodicity feature. Firstly, the original data is partitioned according to the data collection period, and the covariance between each period distribution and the remaining data distribution is calculated to obtain the most relevant periodic section for data pre-processing. Secondly, to better capture the periodicity feature in the data, the input data is decomposed into two components: periodic and trend, and the Temporal Convolutional Network module is used to extract temporal features to predict the remaining useful life of the tools. Finally, the proposed method is validated using the publicly available PHM2010 dataset, and it is shown to have better prediction performance compared with other compared models, proving the effectiveness of our proposed approach.