Tool wear prediction using multi-sensor data fusion and attention-based deep learning

Tool wear is significant in the machining process due to its direct relationship with part accuracy and quality. A precise prediction of tool wear can reduce downtime and improve product quality. Conventional methods of measuring tool wear are not feasible in the age of Industry 4.0. Therefore, this paper introduces a new attention-based deep learning model that accurately predicts tool wear. The model achieves multiscale feature fusion, focusing on the most relevant features. We proposed a channel Attention mechanism that integrates with residual connections, considering the weight of each feature map to improve the model performance. We conducted lathe-turning experiments under varying cutting conditions and collected multi-sensor data to assess the performance of the model and compare it with existing models such as Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), and Support Vector Regression (SVR). Model performance was evaluated using root mean square error (RMSE), mean absolute error (MAE), and the coefficient of determination (R²). The results demonstrated that the proposed model outperformed the others across all evaluation metrics.