Joint optimization of serial system based on deep reinforcement learning

While the serial manufacturing system is a productive mode of production, rework and maintenance are inevitable aspects of real-world operations, and their promptness and effectiveness are crucial for both production effic iency and product quality. This study begins by analyzing the manufacturing system to understand the impact of unqualified work-in-progress (WIP) on the overall system. Maintenance tasks can enhance machine processing quality and reduce the production of u nqualified WIP items; however, they also increase maintenance costs, consume system resources, and decrease production output. Conversely, rework can lower manufacturing costs by converting unqualified WIPs into qualified ones, but processing unqualified W IPs can negatively impact machine quality, leading to further unqualified WIP production. To simulate and optimize the rework and maintenance process, thereby maximizing system production efficiency and product quality, this study proposes a combined optimization model based on the Double Deep Q-Network (DDQN) algorithm. Experimental findings demonstrate that the proposed DDQN algorithm offers significant practical value and a broad range of application prospects. It effectively optimizes rework and maintenance decisions, reduces the number of unqualified WIPs, lowers production costs, and enhances production efficiency.