Abstract: Predictive maintenance based on prognostic information is an emerging maintenance mode which can decrease life cycle cost and increase operational availability effectively. This paper focuses on the prognostic approach based on canaries which can be divided into two categories: line replaceable unit(LRU)-independent canaries and LRU-dependent canaries. Under the perfect replacement assumption, a predictive maintenance decision model is proposed based on the renewal reward theorem, which can evaluate the benefit of the use of canary devices from the long-run average cost rate, average operational availability and average effectiveness-cost ratio. For LRU-independent canaries, the prognostic distance and shape parameter of Weibull distribution are chosen to optimize the maintenance decision, while for LRU-dependent canaries, the accumulated damage factor and stochastic standard deviation are chosen as decision variables. Finally, this model is demonstrated with a numerical implementation example using Monte Carlo simulation. The results show that the predictive maintenance policy with LRU-independent canaries is better than the corrective maintenance policy but worse than the age replacement policy, which exhibits an essential characteristic of the method. Furthermore, the predictive maintenance policy with LRU-dependent canaries is better than the age replacement policy in some conditions and the optimal parameters of canaries for maintenance decisions are chosen.

Key words: condition monitoring, prognostics and health management, canaries, maintenance optimization, renewal process, Monte Carlo methods