Optimal maintenance decisions on the basis of uncertain failure probabilities

Abstract

Due to a lack of data, many maintenance optimisation models have to be initialised on the basis of expert judgment. Rather than eliciting the parameters of a continuous lifetime distribution, experts give more reliable answers when assessing a discrete lifetime distribution. If the prior uncertainty in the probabilities of failure per unit time is expressed in terms of a Dirichlet distribution, Bayes estimates can be obtained of three cost‐based criteria to compare maintenance decisions over unbounded time‐horizons: first, the expected average costs per unit time; second, the expected discounted costs over an unbounded horizon; and third, the expected equivalent average costs per unit time. Illustrates the maintenance model by determining optimal age replacement and lifecycle costing policies, which optimally balance both the failure cost against the preventive repair cost, and the initial cost against the future cost.