Reinforcement Learning Based Maintenance Scheduling for a Two-Machine Flow Line with Deteriorating Quality States

This paper investigates the maintenance policy of Two-Machine-One-Buffer flow line two machines with an intermediate buffer and analyzes effects of parameters on maintenance policy and system performance. The system produces good products with gradually decreasing probability when no maintenance is executed. A maintenance policy of control limit form having been testified in many literatures is adopted and a discounted reward Q-P-learning algorithm based on policy iteration is used to find out the optimal maintenance triggering states and also verify the model.