A Study on the Implementation and Verification of Autonomous Control System for Machining Process based on Intelligent Digital Twin and AAS
In modern manufacturing, transitioning from manual to autonomous control in CNC machining processes is crucial for improving productivity and product quality. While data-driven artificial intelligence (AI) technologies have been introduced, conventional black-box deep learning models often fail to reflect physical machining constraints, causing reliability issues on actual shop floors. This study proposes an autonomous control system for CNC machining processes integrating Asset Administration Shell (AAS)-based data standardization and an intelligent digital twin. To overcome the limitations of existing AI approaches, a physics-guided hybrid optimization model (integrating PINN-lite with 2-stage XGBoost) for deriving optimal cutting conditions and a vision AI-based predictive model for tool wear and remaining useful life (RUL) were developed. The proposed system was successfully deployed and verified in a real-world manufacturing environment. Performance evaluation certified by an authorized testing agency (KOTCA) demonstrated a 97.85% accuracy in optimal condition derivation and a 100% accuracy in tool wear prediction. The proposed closed-loop framework provides a highly reliable and practical solution for realizing level-3 autonomous factories in multi-variety, small-batch production environments.