Multi-scenario digital twin-driven human-robot collaboration multi-task disassembly process planning based on dynamic time petri-net and heterogeneous multi-agent double deep Q-learning network

To reduce the environmental impacts and resource utilization of End-of-Life (EOL) product recycling, it is imperative to achieve the high efficiency of EOL product recycling and reutilization, including disassembly. However, the disassembly of EOL products is being faced with huge challenges due to the uncertainties of EOL product recycling and dynamic disassembly requirements. Therefore, this paper proposes a digital twin (DT)-assisted multi-agent human-robot collaboration (HRC) disassembly system with multi-scenario data simulations to achieve multi-agent disassembly operations and process optimization. In addition, the dynamic disassembly structure based on dynamic Time Petri Net (TPN) model represents the real-time disassembly information and associated disassembly relationships, which incorporates the digital twin technology to simulate the application environment of HRC disassembly operations. By integrating the multi-agent Dueling-Double deep Q-learning network (MADDQN) algorithm to determine the optimal disassembly sequence and associated task strategy in the DT-assisted HRC disassembly platform. Similarly, it is essential to evaluate the performance of the proposed algorithm for multi-task disassembly planning based on HRC disassembly operations. By conducting an in-depth analysis of the NEV-P50 battery pack from the Weilai ES8 as a case study, the practical implementation of the MADDQN algorithm is demonstrated to optimize the dynamic disassembly sequence and uncertain task allocation with DT data, which provides an effective and flexible approach to the complex disassembly tasks in multi-scenario HRC disassembly processes. • Multi-scenario DT-assisted HRC disassembly platform determines multi-agent multi-task disassembly planning. • Dynamic disassembly graph based on TPN model represents the information and relationships of unstructured variations. • MADDQN algorithm solves optimal disassembly strategy than other algorithm by comparing their performance. • The implementation of disassembly procedure demonstrates the feasibility of decision-making of multi-scenario tasks.