A flexible manufacturing assembly system with deep reinforcement learning. (January 2022)

• Record Type: Journal Article
• Authors: Li, Junzheng; Pang, Dong; Zheng, Yu; Guan, Xinping; Le, Xinyi
• Is Part Of: Control engineering practice. Volume 118(2022)
• Publication Date: 2022-01
• Subjects: Reinforcement learning -- Digital twin -- Flexible manufacture -- Assembly line; Automatic control -- Periodicals; 629.89
• DOI: 10.1016/j.conengprac.2021.104957
• Languages: English
• ISSNs: 0967-0661

Abstract

Abstract: Traditional assembly line requires a significant amount of designs from engineers, especially in the case of multi-species and small-lot production. Recently, intelligent algorithms based on reinforcement learning are proposed to address this issue. However, the lower success rate and safety reasons limit their industrial applications. In this article, we proposed a systematic solution, including the automatic planning of assembly motions and the monitoring system of the production lines. In the planning stage, we built the digital twin model of the assembly line, then trained a deep reinforcement learning agent to assembly the workpieces. In the production stage, the digital twin model is used to monitor the assembly lines and predict failures. To validate the system we proposed, we conducted a peg-in-hole assembly experiment, and reached a 90% success rate for a single assembly attempt. During the whole experiment, no collision happens in the real world.