A scheduling support system for large-scale facilities using reinforcement learning in consideration of skill educations and working conditions

• Published in: Journal of Advanced Mechanical Design, Systems, and Manufacturing Vol. 8; no. 5; p. JAMDSM0069
• Main Authors: TATEYAMA, Takeshi, TATENO, Toshitake, KAWATA, Seiichi
• Format: Journal Article
• Language: English
• Published: The Japan Society of Mechanical Engineers 01.01.2014
• Subjects: Dynamic work planning; Maintenance; On the job training (OJT); Reinforcement learning; Relearning; Scheduling; Simulation; Working condition
• ISSN: 1881-3054; 1881-3054
• DOI: 10.1299/jamdsm.2014jamdsm0069

This paper deals with a scheduling method of manual work in consideration of On-the-Job Training (OJT). As of skillful works such as maintenance works of airplanes, workers have to learn many kinds of works with OJT. However, it is difficult to balance between working efficiency and keeping the number of skilled workers by performing OJT. The purpose of this study is to develop a method for adjusting the value of EC automatically by considering the balance between working efficiency and skill education and to find effective rules for adjusting the value of EC. To overcome this problem, the authors previously proposed an education coefficient (EC) as a parameter that adjusts the frequency of OJT and a scheduling support system for the long-term scheduling of OJT. However, it is difficult to find effective values of EC that give a good balance between working efficiency and skill education to maintain a certain number of skilled workers. In this paper, the authors propose a simulation-based scheduling support system using reinforcement learning. The objective of this system is to adjust the value of EC automatically based on the specific situations (deadline and current number of skilled workers) by considering the balance between working efficiency and skill education. This paper also shows that the learning system generates suitable schedules when the working conditions (the number of workers) change during the progress of working by using relearning methods. The experimental results show that the proposed learning system generates adequate schedules to obtain as many skilled workers as possible within the fixed time limit.