Competing deterioration processes modeling and RUL estimation based on factorial hidden Markov models

Thanh Trung Le (1) , Florent Chatelain (1) , Christophe Bérenguer (1)

1 GIPSA-SAIGA - GIPSA - Signal et Automatique pour la surveillance, le diagnostic et la biomécanique

Deterioration modeling and Remaining Useful Life (RUL) estimation play an important role inimplementing a predictive maintenance program since it helps to predict early the failure time caused byan incipient deterioration, and thus to provide sufficient time for maintenance crew to act before the actualfailure eventually occurs. One well-identified challenge in deterioration modeling is that equipment does notalways deteriorate following only one single mechanism, instead several deterioration modes can co-exist incompetition due to the dynamics and variability of operating and environmental conditions. In this paper, westudy the phenomenon where different modes co-evolve in parallel within a single component until one of themlead the component to fail. The observations contain hence information of the states of each mode, leading theinference problem being difficult. To deal with this problem, the Factorial Hidden Markov Model (FHMM) andthe variational approximation technique are adapted. The associated RUL estimation method is also developed.To evaluate the performance of the proposed approach, a numerical study is given

Domaines: Automatique / Robotique; Statistiques [stat]

Thanh Trung Le, Florent Chatelain, Christophe Bérenguer. Competing deterioration processes modeling and RUL estimation based on factorial hidden Markov models. MIMAR 2016 - 9th IMA International Conference on Modelling in Industrial Maintenance and Reliability, Institute of Mathematics and its Applications, Jul 2016, Londres, United Kingdom. pp.108-113. ⟨hal-01411405⟩