Tool wear state prediction based on feature-based transfer learning

Accurate identification of the tool wear state during the machining process is of great significance to improve product quality and benefit. The wear states of the same tool type and machining material have similarities during the machining process. By mining the data value of the historical machining process and analyzing the similarity of the procedure, the subsequent machining process can be predicted with the help of transfer learning. Therefore, this study proposes a tool wear prediction scheme based on feature-based transfer learning to realize the accurate prediction of the tool wear state. The genetic algorithm (GA) is used to select a subset of sensor features that are highly correlated with tool wear. Then, the source domain and target domain are constructed on the basis of the selected sensor features of the historical tool and the new tool during the machining process, respectively. In addition, features in the life cycle of the new tool are completed by feature-based transfer learning. After feature transfer, the maximum mean square discrepancy (MMD) method is used to evaluate the similarity of features, and the optimal feature subset is selected according to the evaluation result. Finally, the particle swarm-optimized support vector machine (PSO-SVM) model is applied to predict the tool wear states during the new tool machining. The effectiveness of the proposed tool wear scheme is verified by the cutting force and wear data of the tool life cycle under three different milling parameter combinations. Results with high accuracy show the advantages of the feature-based transfer learning method for tool wear state prediction.