The advent of the Fourth Industrial Revolution brought about a major change in the decision-making process and production management system of the manufacturing environment. With the introduction of technologies such as Internet of Things(IoT), artifi...
The advent of the Fourth Industrial Revolution brought about a major change in the decision-making process and production management system of the manufacturing environment. With the introduction of technologies such as Internet of Things(IoT), artificial intelligence(AI), and big data, a large amount of data is being collected in real time from production equipment and logistics systems.
Smart Factory is an intelligent manufacturing system that ensures connectivity between equipment, processes, and logistics and makes autonomous decisions based on such data. Material handling systems that manage transport routes directly affect productivity and delivery performance. However, traditional static routing methods cannot adequately reflect the dynamically changing environment, as they
determine routes based on predefined information.
To address this limitation, this work proposes a distributed routing system based on deep reinforcement learning. In the proposed system, each workpiece and material handling device is defined as an independent agent, and reinforcement learning allows each agent to interact with the environment to autonomously determine the optimal routing path. This approach alleviates computational load and dependency problems inherent in centralized control architectures and enables flexible and fast response in dynamic manufacturing environments.
The proposed system is validated by simulation. Two reinforcement learning algorithms, Q-learning and DQN, which aim to minimize the average lead time, are applied. Simulation results show that DQN-based models outperform traditional Dijkstra and Q-learning-based models. In particular, DQN models show stable results even in unseen environments, demonstrating their generalization ability and adaptability to dynamic conditions.
In conclusion, the deep reinforcement learning based distributed routing system proposed in this study enables autonomous and efficient material transport in smart factory environments while overcoming the limitations of traditional centralized control methods. In addition, this study verifies that the distributed control architecture using reinforcement learning is effective in optimizing logistics paths and improving production efficiency of smart factories. Future studies will focus on verifying the proposed model in an environment that integrates uncertainties such as equipment failures and defects.