Hidden Markov models in bearing fault diagnosis and prognosis

Hidden Markov model (HMM) have powerful capability of pattern classification. It can be used for fault diagnosis. Hierarchical Hidden Markov model (HHMM) can exactly represent the full life process of bearing. It can be used for fault prognosis. A framework for fault diagnosis based on HMM and fault prognosis based on HHMM was presented. Unfortunately, the original inference algorithm of HHMM is somewhat complicated, and takes long time. To represent HHMM as Dynamic Bayesian Network (DBN) and use a inference algorithm from as in can shorten the inference time. The proposed methods were applied to fault diagnosis and fault prognosis (Remaining useful life, RUL) of rolling bearing. The results show the validity of the methods.