Warnings of Malfunction: The Decision to Inspect and Maintain Production Processes on Schedule or on Demand
用马尔可夫模型刻画生产系统退化过程，比较了定期预防维护和按需维护两种基本策略，并进一步按观察对象（过程本身或产出）细分，分析四种策略的长期效用，数值例子表明最优维护策略可显著节约成本。
In this paper we study different maintenance policies for the control of production systems that are subject to deterioration over time. A Markov model is used to characterize the deterioration process. Two types of basic maintenance policies are considered: scheduled preventive maintenance and maintenance on demand. The former deals with scheduled maintenance procedures performed on the process at fixed intervals, regardless of its status. In the latter, maintenance procedures are performed in response to signals observed from the production process, and are thus a function of the condition of the process over time. For both approaches the policies can be further classified, based on the mode of observation: the process itself or the outputs of the process. Thus we can have preventive maintenance of the process, scheduled inspection of the outputs of the process, maintenance in response to signals given by the process, and maintenance in response to signals as indicated by the quality of the outputs. We analyze the performances of these four policies, and derive their long-run disutilities. Numerical examples show that, by monitoring the process with the appropriate optimal maintenance policy, significant savings can be achieved.