A Parallel Deep Reinforcement Learning Framework for Controlling Industrial Assembly Lines

Decision-making in a complex, dynamic, interconnected, and data-intensive industrial environment can be improved with the assistance of machine-learning techniques. In this work, complex instance assembly line control is formalized parallel deep reinforcement learning approach presented. We consider an problem which set tasks (e.g., vehicle tasks) needs to planned controlled during their execution, aim optimizing given key performance criteria. Specifically, will that planning task order minimize total time taken execute all (also called cycle time). Tasks run on workstations line. To run, need specific resources. Therefore, tackled optimally mapping resources workstations, deciding optimal execution times tasks. doing so, several constraints respected precedence among tasks, needed deadlines, etc.). The proposed uses learn tasks/resources policy effective minimizing resulting time. method allows us explicitly take into account constraints, and, once training complete, used real dynamically Another motivation for work ability also scenarios, presence uncertainties. As matter fact, use neural networks model problem, contrast with, e.g., optimization-based techniques, require writing equations problem. speed up phase, we adopt scheme more agents are trained parallel. Simulations show provide real-time decision support operators scheduling rescheduling activities, achieving goal tasks’