A Bayesian approach for predictive maintenance policy with imperfect monitoring

In the traditional preventive maintenance policy, the periodic maintenance activities are scheduled on the basis of the a-priori information about the failure behaviour of the population which the component belongs to, by assuming a probability distribution function and by estimating the involved statistical parameters. On the contrary, with the predictive approach, the maintenance activity is scheduled on the basis of the real degradation level of the component. So, it is possible to reduce the failure probability and, at the same time, to use the component for almost all its useful life. For this reason, the predictive maintenance policy makes possible the reduction of the maintenance costs with respect to the preventive approach and it is particularly effective for those components that must work with a high required degree of reliability in systems where failures can produce dramatic consequences. To apply the predictive approach, it is necessary to monitor the component degradation behaviour by using sensors. Nevertheless, before implementing a predictive policy, it is necessary to take into account the costs and the uncertainty of the monitoring system. In this paper we compare by simulation the effectiveness of the predictive maintenance policy with the traditional preventive one when the component must operate with a fixed reliability level. It is shown how the convenience of the predictive maintenance approach depends both on the parameters characterizing the stochastic degradation process and on the uncertainty of the monitoring system. For the preventive policy the a-priori information on the population is considered while, for the predictive one, this information is updated by a Bayesian approach using the data coming from the monitoring system.