A general approach to optimal imperfect maintenance activities of a repairable equipment with imperfect maintenance and multiple failure modes

TL;DR

This paper introduces a flexible, data-driven method for optimizing imperfect maintenance of repairable equipment with multiple failure modes, accounting for model uncertainty and individual component differences.

Contribution

It proposes a novel approach that selects the best failure and maintenance model per component using maximum likelihood and model selection criteria.

Findings

Effective model selection improves maintenance scheduling.

The approach reduces costs and enhances reliability.

Validated with real-world data example.

Abstract

In this paper we describe a general approach to optimal imperfect maintenance activities of a repairable equipment with independent components. Most of the existing works on optimal imperfect maintenance activities of a repairable equipment with independent components. In addition, it is assumed that all the components of the equipment share the same model and the same maintenance intervals and that effectiveness of maintenance is known. In this paper we take a different approach. In order to formalize the uncertainty on the occurrence of failures and on the effect of maintenance activities we consider, for each component, a class of candidate models obtained combining models for failure rate with models for imperfect maintenance and let the data select the best model (that might be different for the different components). All the parameters are assumed to be unknown and are jointly estimated via maximum likelihood. Model selection is performed, separately for each component, using standard selection criteria that take into account the problem of over-parametrization. The selected models are used to derive the cost per unit time and the average reliability of the equipment, the objective functions of a Multi-Objective Optimization Problem with maintenance intervals of each single component as decision variables. The proposed procedure is illustrated using a real data example.