Prognosis based on Multi-branch Hidden semi-Markov Models: A case study

Abstract

This paper proposes a multi-branch model to deal with Remaining Useful Life (RUL) estimation problem in the case where several deterioration modes co-exist within a single component. By basing on Hidden semi-Markov Models (HsMM), the component is supposed to pass through some discrete and unobservable health states until it fails. The rate and the manner of these state transitions, however, depend on the mode of deterioration that is actually active. We show that by taking into account the co-existence of different deterioration modes, the multibranch model can help to improve prognosis results, which is essential for the implementation of a predictive maintenance. A practical case study is investigated to evaluate the advantages of the proposed model.