Assessing the impact of preventive maintenance based on censored data

Recent work in the field of competing risks enables us to start an assessment of the impact of preventive maintenance on the failure characteristics of a piece of equipment. Competing risk is the term given when more than one factor conspires to take a piece of equipment out of service. A simple example of this is given by preventive maintenance and failure. Each could occur to a piece of equipment but only actually can. The preventive maintenance time may censor the unobserved latent failure time, thus ensuring that the statistical analysis of the failure times is made more difficult. Maintenance optimization models require knowledge of the underlying failure distribution. The Kaplan–Meier estimator (and other similar techniques) is commonly used to remove the effect of the censoring variable on the estimate of the distribution of the variable of interest. When the censoring variable is preventive maintenance, however, it is most unlikely that the assumptions of the Kaplan–Meier estimator (independence between the latent failure and preventive maintenance times) are valid. In this paper we give an overview of some of the different methods which have recently emerged in this area. These make different assumptions about the relation between PM and failure to allow the PM censoring to be ‘removed’. Potentially, these methods would allow us to look at changing maintenance strategies. Although they are in the early stages of development they do already allow us to quantify the effect of some changes.