关键词: 预测性维护决策, 流水线, 深度强化学习, 数字孪生车间

Abstract: Predictive maintenance is designed to perform maintenance activities based on the condition of equipment,which can improve the business bottom line by reducing maintenance cost and improving production performance.The modeling,analysis and decision-making of serial production lines with machine degradation process were studied.A Markov chain model was developed by analyzing the dynamics of a serial production line with machine failures and predictive maintenance,and the analytical formulas of transient performance measures were derived.A predictive maintenance decision model was established as a Markov decision process to minimize the work-in-process,backlog and maintenance costs.A deep reinforcement learning method was utilized to explore optimum maintenance policies,which was obtained through the training of neural network with dataset generated from Markov chain model.Case study was performed to validate the effectiveness of the proposed decision model.The results indicated that the maintenance and production related costs were significantly reduced.

Key words: predictive maintenance decision-making, serial production line, deep reinforcement learning, digital twin workshop