An intelligent monitoring system for robotic milling process based on transfer learning and digital twin

Robotic milling is becoming widely used in aerospace and auto manufacturing due to its high flexibility and strong adaptability. However, the practical challenges including complex and time-consuming robot trajectory planning, insufficient monitoring, and lacking three-dimensional visualization limits its further application. To address these challenges, an intelligent monitoring system for robotic milling process based on transfer learning and digital twin was proposed and developed in this paper. Firstly, the fundamental framework of this system was conducted based on a five-dimensional digital twin model for motion simulation, visualization, and tool wear prediction during the robotic milling process. Secondly, the parsing algorithm converting NC code to robot's machining trajectory and material removal algorithm based on bounding box and mesh deformation were proposed for robotic dynamic milling simulation. Thirdly, a novel transfer learning algorithm named CNN-LSTM-TrAdaBoost.R2 was developed by integrating CNN-LSTM with TrAdaBoost.R2 for automated feature extraction and real-time prediction of tool wear. Finally, the effective and accuracy of tool wear prediction algorithm is verified by ablation experiment and the robotic milling simulation is validated by real milling experiment, as well. The results indicate that the proposed monitoring system for robotic milling process demonstrates great virtual-real mapping. It can offer new insights and technical support for constructing sophisticated digital twin frameworks and enhancing operational monitoring in manufacturing systems.