Research on Deadlock-Free Scheduling of Flexible Manufacturing Systems Based on Reinforcement Learning

Flexible manufacturing system can meet the demand of personalized and small-lot production due to its flexibility and resource sharing characteristics, but the deadlock problem that may occur during its scheduling process will seriously affect the production efficiency, so effective deadlock-free scheduling is called the key challenge of flexible manufacturing system re. Traditional deadlock-free scheduling methods have problems such as high scheduling complexity and inability to fully utilize resources, etc. In this paper, Deep Reinforcement Learning (DRL) is applied to the deadlock-free problem of flexible manufacturing system, and a DF-DQN model is proposed, which combines the structural features of Petri net, enhances the model’s ability to express the state of the system, and introduces a deadlock recovery mechanism, which further improves the scheduling reliability and efficiency, and guarantees the deadlock-free scheduling. While ensuring deadlock-free scheduling, it improves resource utilization and productivity. Comparative experimental results show that the model outperforms the traditional method in reducing the deadlock rate and the maximum completion time, which verifies its effectiveness in deadlock prevention and scheduling optimization.