Reinforcement learning-based selective disassembly sequence planning for the end-of-life products with structure uncertainty

Selective disassembly sequence planning (SDSP) is regarded as an efficient strategy to determine optimal disassembly sequences for extracting target parts (TP) from complex end-of-life (EOL) products. Previous research assumes that all EOL products have the same structure and the optimal selective disassembly sequences are given before the EOL products are removed. However, the products have different operation states during their use stage, which results in high structure uncertainty of EOL products. The structure uncertainty of EOL products often makes the predetermined selective disassembly sequences impractical for minimizing disassembly time and maximizing disassembly profit. This letter undertakes this challenge by integrated reinforcement learning (RL) to determine the optimal disassembly sequences adaptive to the structure uncertainty of the EOL products. Firstly, a multi-level selective disassembly hybrid graph model (MSDHGM) is developed to illustrate the contact, precedence, and level relationships among parts. Then, the SDSP is formulated as a finite Markov Decision Process and a deep Q-network based selective disassembly sequence planning (DQN-SDSP) is proposed. Finally, extensive comparative experiments are conducted to verify the proposed method compared with NSGA-II and ABC algorithms.