Statistical inference for imperfect maintenance models with missing data

Yann Dijoux (1) , Mitra Fouladirad (1) , Dinh Tuan Nguyen (1)

1 LM2S - Laboratoire Modélisation et Sûreté des Systèmes

The paper considers complex industrial systems with incomplete maintenance history. A corrective maintenance is performed after the occurrence of a failure and its efficiency is assumed to be imperfect. In maintenance analysis, the databases are not necessarily complete. Specifically, the observations are assumed to be window-censored. This situation arises relatively frequently after the purchase of a second-hand unit or in the absence of maintenance record during the burn-in phase. The joint assessment of the wear-out of the system and the maintenance efficiency is investigated under missing data. A review along with extensions of statistical inference procedures from an observation window are proposed in the case of perfect and minimal repair using the renewal and Poisson theories, respectively. Virtual age models are employed to model imperfect repair. In this framework, new estimation procedures are developed. In particular, maximum likelihood estimation methods are derived for the most classical virtual age models. The benefits of the new estimation procedures are highlighted by numerical simulations and an application to a real data set.

Domains: Performance [cs.PF]

Yann Dijoux, Mitra Fouladirad, Dinh Tuan Nguyen. Statistical inference for imperfect maintenance models with missing data. Reliability Engineering and System Safety, 2016, 154, pp.84-96. ⟨10.1016/j.ress.2016.05.017⟩. ⟨hal-02362384⟩