摘要
预测性维护决策旨在提高维护效率的同时,降低维护停机对生产的影响。预测性维护根据设备的实际健康状态进行维护决策,能有效避免过度维护造成的浪费和维护不及时造成的设备随机故障。然而类似于设备故障停机,预测性维护需要关闭设备进行维护作业。如果维护时机选择不当,维护过程会引起生产线的饥饿和阻塞,造成生产损失。因此,预测性维护不仅需要关注设备自身的维护需求,还需要与实际生产进行联动。以考虑机器状态劣化的航空产品流水线为研究对象,在流水线产出损失分析的基础上,研究流水线预测性维护决策问题。首先,针对缺料停机、设备故障等扰动停机事件和预测性维护事件,分析停机事件对流水线产出的影响,量化造成的流水线产出损失。其次,考虑流水线产出损失和维护成本构建奖励函数,建立基于马尔可夫决策过程的流水线预测性维护决策模型,结合深度Q网络算法求解模型获得优化决策方案。最后,通过仿真试验对比其他三种维护方法,验证了所提出决策模型的有效性。
In production lines,predictive maintenance is adopted to improve the maintenance efficiency while achieving a desirable throughput.Predictive maintenance makes maintenance decisions based on the status of equipment,which is important to reduce the waste resulted from excessive maintenance and the unexpected equipment breakdown.However,it is not uncommon that machines need to be shut down for maintenance.Similar to random machine failure,predictive maintenance can cause significant production loss if it is not planned appropriately.Therefore,this paper proposes a predictive maintenance decision-making model for a serial production line based on a thorough understanding of production dynamics.Firstly,the impact of disruption events and maintenance events are evaluated in terms of the permanent production loss.Then,a Markov decision process model is established for predictive maintenance decision making.The model considers the penalty cost by permanent production loss and the maintenance cost.Deep Q network is adopted to exploit the optimal maintenance policies.Lastly,case studies are performed to validate the effectiveness of the proposed predictive maintenance decision-making model by comparing with other three maintenance decision-making models.
作者
张文沛
崔鹏浩
李洋
延爽
Zhang Wenpei;Cui Penghao;Li Yang;Yan Shuang(Northwestern Polytechnical University,Xi’an 710072,China)
出处
《航空科学技术》
2022年第5期53-62,共10页
Aeronautical Science & Technology
基金
国家自然科学基金(52175485)
航空科学基金(2019ZG053001)。
关键词
预测性维护
流水线
决策优化
马尔可夫决策过程
深度Q网络
predictive maintenance
serial production line
decision-making optimization
Markov decision process
deep Q network