Tool wear state recognition method based on multiscale feature fusion

It is very important to recognize the state of machining tools in the machining process, and the untimely replacement of broken tools will inevitably have a serious impact on the quality of machined parts. In recent years, deep learning-based tool wear state recognition has received increasing attention. Such methods usually use neural network models to directly extract deep-level features from different signals to evaluate the tool wear state. However, the existing methods ignore the interactive fusion process of different levels of features in the model, and the lack of reliable constraints on the features of different degrees of tool wear states leads to unsatisfactory model classification results. In this paper, we propose a multiscale feature fusion-based tool wear state recognition framework MSF. It extracts richer feature information by fusing low-level, medium-level and deep-level features, while incorporating a spatial attention mechanism to focus the model on important regions, and finally introducing metric learning to enable the model to learn more discriminative features. The proposed method is validated on the publicly available dataset PHM2010, and achieves the best tool wear state recognition results compared to the other methods.