摘要：AbstractThe paper aims at proposing a card controlling model to improve the standard CONWIP procedure, granting a similar system throughput while reducing Work In Progress (WIP) levels. To achieve this objective, the authors developed a Digital Twin-based production control system including a reinforcement learning algorithm (i.e. Q-Learning). The Digital Twin is responsible for short term predictions of the behavior of the system aimed at a what-if analysis with different numbers of cards. As there is lack of evidence of research related to Digital Twin applications for production control and for order release systems in particular, we aim at proposing this as an initial work to start the exploration of problems in this control area. The proposed model has been tested both in a Job Shop and in a Flow Shop systems with promising results.
