Q-learning based Dynamic Scheduling for No-wait Flow Shop with Maintenance Window and Minimization of Total Tardiness

This paper discusses the dynamic no-wait permutation flow shop scheduling Problem (PFSP). Orders arrive in the manufacturing system dynamically while the preventative maintenance (PM) is considered on the last machine. Q-learning is applied to provide the real-time intelligent scheduling for the dynamic PFSP to minimize the total tardiness cost for all orders entering the system. To facilitate this, we introduce a vast array of instances to train our Q-table. The learning progression demonstrates that the trained agent can select the optimal dispatching rule at each decision juncture. Upon completion of the training, we retain the trained model and compare it with SPT, FIFO, LPT, EDD. The comparative analysis indicates a substantial performance advantage of our trained scheduling agent over these rules.