Degradation assessment for machinery prognostics using Hidden Markov Models

Degradation detection and recognition of degradation pattern are crucial to the successful deployment of prognostics. A machine degradation process is known to be stochastic instead of deterministic. Recognizing the degradation pattern needs helps from stochastic and probabilistic models. Among various stochastic approaches, Hidden Markov Models (HMMs) have been proven to be very effective in modeling both dynamic and static signals [1]. In this paper, aiming to providing a guideline of how to effectively and efficiently use the HMMs to assess degradation for various machinery prognostic applications, three different approaches of applying the HMMs are reviewed and compared. It demonstrates that depending on the varieties of applications, available prior knowledge, and characteristics of degradation processes, those three implementation approaches perform differently. A full understanding of the strengths and weaknesses of each deployment approach is extremely important in order to effectively utilize this powerful tool for system degradation assessment.