Knowledge Graph-Enhanced Digital Twin System for Tool Wear Detection Using GNN-MSCNN Architecture

Computer Numerical Control (CNC) machining is one of the core technologies in the manufacturing industry, and tool wear during the cutting process directly affects machining accuracy and product quality. Traditional wear monitoring approaches exhibit inherent limitations in data interoperability and relational reasoning, falling short of intelligent manufacturing requirements. To address this, we developed a Digital Twin (DT) system for tool wear detection that integrates a knowledge graph using multi-modal data, enhancing the capabilities of DT system in terms of spatio-temporal feature extraction, multi-source data fusion and correlation analysis. Considering that the constituted knowledge graph has abundant isolated nodes and weakly correlated nodes, we proposed a Graph Neural Network Multi-Scale Convolutional Neural Network (GNN-MSCNN) method to complete the regression task and solve the problem of unsatisfactory detection results of a single GNN network. It achieves accurate tool wear detection, significantly improving the detection accuracy of tool wear. The experimental results demonstrate that the proposed algorithm exhibits higher accuracy in tool wear detection compared with existing deep learning detection methods.