Harnessing bidirectional gated spatio-temporal fusion and multi-scale wavelet transformation for robust tool wear classification

Abstract Accurate tool wear monitoring is critical to ensuring machining quality and equipment efficiency in mechanical processing. A novel monitoring method for tool wear conditions is proposed in this paper, which integrates the bidirectional gated spatiotemporal fusion (BGSTF) module with multi-scale wavelet transform (MSWT) to achieve high-precision classification of tool wear states. The BGSTF module effectively captures spatiotemporal dependencies in raw tool signals, while the MSWT extracts multi-scale features from vibration signals across different frequency bands. To further enhance feature representation and classification, an attention fusion mechanism and Transformer encoder are developed. Experimental results show that the proposed method effectively distinguishes between initial wear, normal wear, and severe wear conditions, achieving excellent performance in classification accuracy and noise robustness. This research provides an effective solution for tool health monitoring in industrial manufacturing.