This study provides a comprehensive analysis of tool wear in precision machining on CNC machine tools, introduces relevant theories on tool wear, and collects two milling experiment datasets from PHM and NASA. Based on deep residual networks and attention mechanisms, an improved deep residual contraction network tool wear monitoring model is designed to map tool wear features into a high-dimensional space, thereby enhancing the model’s recognition performance. Experiments demonstrate the feasibility of the proposed model in tool wear monitoring. The improved model achieves an accuracy rate of approximately 96% on both the PHM and NASA datasets during training, with loss values fluctuating around 0.12 and 0.17, respectively. Evaluation parameters such as recognition accuracy and sensitivity under different wear conditions are significantly higher than those of the comparison methods. Additionally, the model can accurately predict rear face wear conditions, with a maximum absolute recognition error of only 8.5 \(\mu\)m, significantly lower than the 20.7 \(\mu\)m of the comparison model. The coefficient of determination exceeds 97% on both datasets, indicating good generalization capability. Under different sensor signals, this model can also accurately identify tool wear states, with the highest recognition accuracy for initial and severe wear stages.