An end-to-end deep learning approach for tool wear condition monitoring

It is important to establish a real-time and accurate tool wear monitoring system for improving machining quality, tool utilization, and reducing cost. In this paper, an end-to-end tool wear condition monitoring algorithm is proposed by combining 1D convolutional neural network (1DCNN) with residual block and Transformer. Firstly, the original sensor signal is processed directly by one-dimensional convolution, the local features of the signal are extracted and the dimensionality of the signal is reduced. Then, Transformer is applied to model the sequence and capture the global feature relationship. After threefold cross-validation, the average index of the presented method: accuracy, F1 score, precision, and recall rate on the PHM2010 milling dataset were 97.41%, 97.4%, 97.43%, and 97.4%, respectively. It can be seen that the proposed algorithm can complete the task of tool wear classification and also showed a strong generalization ability on each validation set.