Real-time rescheduling for smart shop floors: an integrated method

Rescheduling is an important means of production and operation management to realise efficient production in dynamic environments. In smart shop floors, diverse manufacturing resources, flexible product processes, and complex information flows bring new challenges to real-time rescheduling. To address the problem of real-time rescheduling decision-making and optimization in smart shop floors under disturbances, this work proposes a novel rescheduling integrated method. In this method, the real-time rescheduling control on the continuous time scale is modelled as a closed-loop sequence decision process. The rescheduling agent is then trained by Deep Recurrent Q-Network to achieve autonomous and adaptive rescheduling which integrates decision-making and optimization. To balance passive deterioration and active optimization in the production process, composite repair actions based on the critical path and reward-shaping function based on the resilience index of the production system are developed. Additionally, a recurrent Long Short Term Memory network is introduced to integrate temporal information and improve the rescheduling efficiency for partial observability of real production environments. Results of the case study indicate that the proposed method can reduce makespan by up to 19.76% and variation by up to 87.93%, outperforming existing well-known rescheduling strategies, offering a valuable methodology for cooperative operational control of smart shop floors in dynamic environments.