Reliability analysis and joint assessment of Brown―Proschan preventive maintenance efficiency and intrinsic wear-out
Résumé
An imperfect maintenance model for repairable systems is considered. Corrective Maintenances (CM) are assumed to be minimal, i.e. As Bad As Old (ABAO). They are considered to be left and right censored. Preventive Maintenances (PM) are assumed to be done at predetermined planned times and to follow a Brown-Proschan (BP) model, i.e. they renew (As Good As New, AGAN) the system with probability p, and they are minimal (ABAO) with probability 1-p. BP PM effects (AGAN or ABAO) are assumed to be not observed. In this context, different methods (maximum likelihood, moment estimation, and expectation-maximization algorithm) are considered in order to estimate jointly the PM efficiency parameter p and the parameters of the first time to failure distribution corresponding to the new unmaintained system. A method to individually assess the effect of each PM is also proposed. Finally, some reliability characteristics are computed: failure and cumulative failure intensities, reliability and expected cumulative number of failures. All the corresponding algorithms are detailed and applied to a real maintenance data set from an electricity power plant.