Online Tool Wear Monitoring Based on Multi-sensor Feature Fusion and Domain Adaptation

Abstract Machine tool condition monitoring is of great significance for machine tool health management and machining quality. There exist a series of problems in the existing models, such as a large error in the prediction of machine tool wear, and the possible differences in the distribution of the training set and testing set are not considered. In this work, the residual connection block and channel attention block are sequentially stacked as the feature extractor. The introduction of domain adaptation enables the model to extract domain-independent features. Through the weight discrepancy restriction, the feature vectors between different sensors are driven to be orthogonal to improve the effectiveness and comprehensiveness of feature extraction. Finally, the fusion features are mapped to the tool wear value through the fully connected layer. To verify the model effect, experiments were conducted using the PHM 2010 milling cutter wear dataset. The experiment results indicate that the average RMSE and average MAE of this model are 6.97 and 6.29 on the three tools C1, C4, and C6, respectively. There is more than a 12% improvement compared with TCN, TDConvLSTM, and other models.