Frailty Models for Survival Data applied to Maintenance Management

Abstract

Due to large number of equipment in industrial companies, maintenance management may not always be able to identify and quantify assets heterogeneity concerning failure data. Several statistical and mathematical methods are being developed in the last years looking for an improvement on the system reliability. We’ve been working on survival analysis applied to equipment failure data using Cox proportional hazards, however, we notice that individualization with respect to the risk must be considered. Frailty models allowed us to identify equipment not following patterns of identical distributed times to failure. Although there exist some controversy around frailty models, claiming its poor contribution to the model, the truth is that when clustering our equipment by its id with a non-observed frailty component, we quickly identify those with higher influence on the hazard. This is particularly useful whether the intention is to find a homogeneous sample or work with all samples considering frailty component, contributing either way, to an enhanced and adaptive maintenance management.