Remaining Useful Life Estimation Using Explicit Duration Hidden Markov Models

Remaining Useful Life Estimation Using Explicit Duration Hidden Markov Models

van Santvoort, Sander (TU Delft Electrical Engineering, Mathematics and Computer Science)

Verwer, S.E. (mentor)
de Oude, Patrick (graduation committee)

Delft University of Technology

Delft University of Technology

2018-06-28

Hidden Markov models are suited for modelling processes whose state is not
directly visible, but the outputs of which depend on its state. As such they can be used to model degradation of machines. Hidden semi Markov models extend hidden Markov models by removing the inherent geometric constraint imposed on hidden Markov models.
Using hidden semi Markov models the duration of each state can be modelled by
any parametric density. In this thesis modelling the state duration using dierent parametric densities are compared for a (simulated) turbofan dataset. It is shown that for the simulated dataset several parametric densities perform similarly well. Besides modelling degradation, hidden (semi) Markov models allow for prediction of remaining useful life (RUL). In this thesis a novel RUL estimation method is proposed that incorporates three components to accurately predict RUL: The uncertainty regarding the current health state, the estimated time spent in each health state and the shape of the state duration distribution. Using the turbofan dataset the novel method is shown to improve the accuracy of RUL estimations, especially towards the end of the turbofan's lifetime (at which the accuracy is most crucial).

hidden Markov model
hidden Semi Markov model
remaining useful life
condition based maintenance
predictive maintenance

http://resolver.tudelft.nl/uuid:b703abaf-74da-4efe-bbc8-5da4b2db8aa3

2023-06-28

Student theses

master thesis

© 2018 Sander van Santvoort