Robotic Disassembly Sequence Planning Considering Robotic Movement State Based on Deep Reinforcement Learning

Remanufacturing provides an alternative way to realize natural resources saving and environment protection. Disassembly, as a key step in remanufacturing, has been attracted much attention in recent years. To make up for the deficiency of manual disassembly which is low efficiency and high cost, industry robots have great advantages in handling large volume and repeatable disassembly activities. Besides, proper disassembly sequence planning helps to improve the disassembly efficiency. In this paper, the framework of robotic disassembly sequence planning using deep reinforcement learning (DRL) is proposed to solve robotic disassembly sequence planning (RDSP) problem. Considering the smoothness of starting and stopping in robotic movement, dynamic moving speed model is built for moving time in disassembly. Firstly, a disassembly precedence matrix (DPM) is constructed according to the structure of disassembly products. After that, RDSP is modeled as Markov decision process and the state, action and reward of the agent in DRL environment are designed. The deep reinforcement learning network model is trained to obtain the optimal disassembly sequence in RDSP. Finally, case study based on double coupling shaft with 21 components proves that the DRL algorithm used in this paper can obtain a disassembly sequence for better performance compared with other two meta-heuristic methods.