Imperfect maintenance with delayed corrective action: modelling, statistical inference for censored data and industrial application

Keywords :

• données réelles
• fiabilité
• vieillissement
• censure à gauche
• censure par intervalle
• estimation statistique

The thesis proposes a new maintenance model for a repairable system. The system is inspected at deterministic times. If a failure occurs between two successive inspections, a minimal repair is performed. At inspection times, two types of preventive maintenances can be performed (MP or MPC). The first originality of this thesis is to consider that the choice between these two types of preventive maintenances depends on the fact that there has been a failure or not since the previous inspection. Both types of maintenances are assumed to be imperfect, of the ARA (Arithmetic Reduction of Age) type, with different efficiency parameters.

We present several statistical inference methods for estimating the parameters of this new maintenance model. The classical case of complete data is first studied. The second originality of this thesis lies in the estimation of the parameters of this model for interval censored and left censored data, as well as for data combining these two types of censoring. The third original feature also concerns the estimation of the parameters of this model, but applied to heterogeneous data, using covariates. The quality of the proposed methods is assessed on simulated data.

Finally, real data, both censored and heterogeneous, from passive systems are studied using the methods developed.

President:

Mr Christophe Berenguer (Université Grenoble Alpes)

Directors:

• Mr Olivier Gaudoin (Université Grenoble Alpes )
• Mr Laurent Doyen (Université Grenoble Alpes )
• Mr Florent Brissaud

Raporteurs:

• Mr Jean-Yves Dauxois (INSA Toulouse )
• Mme Mitra Fouladirad (Centrale Marseille - Aix Marseille Université )

Examinators:

• Mr Vincent Couallier (Université de Bordeaux )
• Mr Christophe Berenguer (Université Grenoble Alpes )