Architecture of digital twin management and control system for production workshops driven by multi-agent collaboration

Digital twin technology offers the feasibility of real-time control of production processes, significantly improving equipment health management, production process optimization, and resource scheduling efficiency. However, current production control still relies on manual experience-based decision-making, which is limited by response lag, strong subjectivity, and poor adaptability, making it difficult to cope with the multi-objective conflicts and dynamic disturbances in complex manufacturing environments. This research aims to overcome the bottlenecks of traditional experience-based control by constructing a digital twin control architecture driven by multi-agent collaboration, thereby upgrading the control mode from passive loss prevention to proactive intervention. This architecture comprises a physical perception layer, a digital twin interaction layer, and a multi-agent collaborative decision-making layer. The method agent is responsible for real-time response to disturbance events such as equipment failures and emergency orders, generating prioritized control plans based on case-based reasoning. The mechanism agent optimizes the system's anti-interference rules through deep reinforcement learning, enhancing system resilience. The resource agent strategically restructures production factors, forming a three-level adaptive closed loop consisting of instantaneous disturbance event response, short-term anti-interference rule iteration, and long-term resource reorganization. Finally, the effectiveness of the proposed architecture is demonstrated by a case study of a conveyor jam.