Toward high-performance machining of thin-walled parts: Fusion of parallel spatial-temporal information in high-speed milling for monitoring tool wear

The complex dynamic characteristics of milling systems for thin-walled parts lead to non-stationary changes in cutting signals, which makes it difficult to capture key information at the spatial-temporal level, and thus weakens the performance of tool wear-sensitive features. To solve this problem, unlike the traditional cascade network, a parallel spatial-temporal feature extraction method is proposed, which ensures the integrity of the temporal information. Based on the attention mechanism, it improves the ability of recognizing key features. Then, to improve the performance of weak features, a spatial-temporal feature adaptive fusion strategy is established, which provides a new way to solve the problem of artificial experience dependence in feature selection. Experiments demonstrate that the proposed method is sensitive to both early and late stages of tool wear. Compared with the existing methods, the RMSE and MAE are reduced by 9.22 and 5.46, respectively, while the R2 is improved by 0.08.