Parallel ResNet-BiGRU tool wear prediction model based on attention mechanism

Cutting tools play a vital role in CNC machining processes. Effective tool wear prediction is essential for lowering machining costs and boosting productivity. The current mainstream research is to mine the spatiotemporal correlation between sensor data and tool wear degree using deep learning combinatorial models of CNN and RNN. However, most of the combined models tend to ignore the problem of feature interference when mining spatial and temporal features, leading to a decrease in prediction accuracy. To address the above issues, three sets of complete life cycle milling experiments were conducted on 40Cr quenched and tempered steel to collect three-axis cutting force signals fed into the parallel model for prediction. The experimental results show that the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination ( R 2 ) of this method, which reaches 4.668, 5.979, and 0.935 in the average of three test sets, respectively, are better than the comparison model. Finally, the transferability of the proposed model under different working conditions is verified on a public dataset using a transfer learning method.