Robotic disassembly sequence dynamic planning under uncertain irremovable condition using dueling deep Q-network based on digital twin

As an important step in recycling the end-of-life products, robotic disassembly can reduce human labor costs and robotic disassembly sequence planning helps to improve efficiency. The irremovable condition of components is uncertain and must be recognized during the robotic disassembly process. This uncertainty leads to the inapplicability of the optimal disassembly solution generated by pre-planning method, which impossibly considers the accurate irremovable condition prior to disassembly. To address the robotic disassembly sequence dynamic planning problem under the uncertain irremovable condition, this paper leverages a digital twin model and a dueling deep-Q network to dynamically generate the optimal solutions according to the recognized irremovable condition. First, a digital twin framework is proposed and the digital twin of the robotic disassembly process is built. Next, a dueling deep-Q network is utilized to solve the proposed problem. Case studies on a camera and a gear pump are conducted to validate the proposed method. Experimental analyses include the connection tests, the error assessments of the digital twin, and the performance evaluations of the algorithm under different scenarios. Results demonstrate that the trained model dynamically generates superior disassembly sequences tailored to the recognized irremovable condition within a reasonable time compared with the other algorithms.