Predictive maintenance for aircraft components using proportional hazard models

Wim J.C. Verhagen*, Lennaert W.M. De Boer

*Corresponding author for this work

Research output: Contribution to journalArticleScientificpeer-review

30 Citations (Scopus)
225 Downloads (Pure)


Unscheduled maintenance can contribute significantly to an airline's cost outlay. Reliability analysis can help to identify and plan for maintenance events. Reliability analysis in industry is often limited to statistically based approaches that incorporate failure times as the primary stochastic variable, with additional strict assumptions regarding independence of events and underlying distributions of failure phenomena. This foregoes the complex nature of aircraft operations, where a whole range of operational factors may influence the probability of occurrence of a maintenance event. The aim of this research is to identify operational factors affecting component reliability and to assess whether these can be used to reduce the number of unscheduled occurrences (i.e. failures). To do so, a data-driven approach is adopted where historical operational and maintenance data is gathered and analysed to identify operational factors with a measurable influence on maintenance event occurrence. Both time-independent and time-dependent Proportional Hazard Models (PHMs), models which incorporate operational factors as covariates, are employed to generate reliability estimates. Results obtained from analysing historical data of a set of nine components with respect to unscheduled removals indicates that adopting new maintenance schedules, derived from the proposed reliability models, can reduce the number of unscheduled occurrences.

Original languageEnglish
Pages (from-to)23-30
Number of pages8
JournalJournal of Industrial Information Integration
Publication statusPublished - 1 Jan 2018


  • Predictive maintenance
  • Proportional hazard model
  • Unscheduled maintenance


Dive into the research topics of 'Predictive maintenance for aircraft components using proportional hazard models'. Together they form a unique fingerprint.

Cite this