摘要
在生产过程中,设备状态的衰变会影响产品质量,尽管设备仍能运行,但其成品率水平逐渐下降.针对由两台具有衰变质量状态的设备和一个库存缓冲组成的2M1B流水线系统,研究衰变设备的预防维护策略.每台设备可视为一个Agent,其预防维护问题被描述成半马氏决策过程模型,并与另一台设备的维护模型相关.以考虑系统全局即时成本为前提,提出了一种分布式的多Agent强化学习方法,获得两台设备在缓冲库存水平下的维护策略.学习所得的维护策略是典型的控制限型形式,即对于给定库存水平,当设备衰变至等于或劣于其相应的控制极限状态时,便触发维护行动.
In manufacturing systems, the deterioration of machine states will influence the quality of the produced parts. The machine might be operating with a lower yield level. This paper investigates the maintenance policy for the machines with deteriorating quality states in a flow line system consisting of two series machines with a finite buffer in between. Each of the machines is considered as an agent. The maintenance problem for the machine is modeled as a semi-Markov decision process, which is related with the maintenance decision process of the other machine. On the premise of considering the global system immediate costs, a distributed multi-agent reinforcement learning algorithm is proposed to obtain the maintenance policy for the machines associated with a given buffer level. The learned policy appears to be a typical control limit form, which means the maintenance action should be triggered whenever the state of the machine deteriorates to a certain control limit.
出处
《系统工程学报》
CSCD
北大核心
2013年第5期702-708,共7页
Journal of Systems Engineering
基金
国家自然科学基金资助项目(60904075)
国家杰出青年科学基金资助项目(71125001)
