Modeling of Condition-based Maintenance by Competing Risks

Abstract

Condition-based maintenance is a preventive maintenance strategy where intervention times are determined based on knowledge of actual item condition provided by condition monitoring. In this report condition-based maintenance is modelled by a special case of competing risks where only two risks are present. That is maintenance and failure. The two risks are thought to compete to end the service sojourn of a technical item, and maintenance is considered to be censoring of the latent failure times. As maintenance intervention times are based on indications of emerging failures, then maintenance and failure are assumed to be dependent competing risks. Also, the probability of detecting indications of emerging failures and with that successfully perform maintenance is assumed to be independent of the age of the item. In this thesis the above described case is modelled in particular by a random signs censoring model. Several methods are proposed in order to simulate data from the desired model, and two methods are chosen for further study. That is simulation of desired competing risks data by first hitting times in Wiener processes, and by a method where failures occur at random and maintenance can be performed some time prior to this. The two chosen methods are implemented in the statistical software R in order to make use of them in a practical case study. The case study originates from the oil and gas industry and aims at assessing the effect on availability by applying condition-based maintenance to a subsea pump.