Optimization of Circular Disassembly Lines With Human-Assisted Robotic Workstations Using Two-Stage Greedy PPO Algorithm

摘要

Disassembly is a critical step in the recycling and reusing of end-of-life products. As Industry 5.0 emerges, manufacturing is shifting from a system-oriented approach to a human-centered paradigm, advancing human–robot collaboration to a new stage. However, existing studies on human–robot collaboration in disassembly lines generally overlook the mobility of workers. To fill the research gap, this work proposes a novel human–robot collaboration mode that considers both the mobility of workers during disassembly and the flexibility of collaboration time in human–robot interaction. Based on this model, this work proposes the human-assisted robotic circular disassembly line balancing problem and establishes a profit-oriented spatiotemporal decomposition mixed-integer programming model. A two-stage greedy proximal policy optimization algorithm is designed to solve it. To validate the effectiveness of the proposed model and algorithm, ten sets of benchmark instances are generated with different scales based on real product structure data. Comparative experiments with reinforcement learning algorithms and classical heuristic methods demonstrate the feasibility and significant superiority of the proposed algorithm in solving this type of problem.