A novel physics-informed model for tool wear monitoring under multi-machining conditions

Tool wear monitoring (TWM) is essential for ensuring workpiece surface quality and machining efficiency. Existing TWM techniques include physics-based models and data-driven methods. However, physics-based models built under simplified or ideal conditions cannot guarantee the accuracy and reliability of TWM. In addition, over-reliance on data-driven models fails to address the interpretability and generalizability of prediction results. To address these issues, a novel physics-informed model is proposed. First, a physics-guided feature adaptive filtering reconstruction strategy is proposed to eliminate the interference and redundancy in the measured signals, and then construct robust features under multi-machining condition; Second, a physics-aware TimesNet multi-scale features is established, which introduces a priori knowledge of the model structure to achieve comprehensive extraction of multi-feature cycles. On this basis, a Physics-guided Cross Attention (PGCA) module is proposed. A priori knowledge learned from the TW physical model is combined with the data features to guide the feature extraction of the model in high-dimensional space. In addition, considering the physical degradation law, the domain invariant features under MSD are introduced to quantitatively assess the physical inconsistency. Finally, the effectiveness of the proposed method is verified by high-speed milling experiments under multi-machining conditions. According to the experimental results, compared with advanced models, the proposed model reduced MAPE and RMSE by 51.16% and 43.11%, respectively, in terms of TW monitoring accuracy and robustness.