Dynamic Diagnosis Approach of Multi-state Degradation System Using Hidden Markov Model

A methodology for developing dynamic diagnosis of multi-state degradation system was proposed in this paper. Wavelet packet energy entropy was employed to characterize the uncertainty and complexity of the signal. Current state evaluation and multi-state recognition had been implemented by hidden Markov model. The recognition performance was verified by a bearing vibration experiment, and the effects of decomposition levels and wavelet mother functions on the recognition performance were taken into account. Compared with classifiers of K-means, BP neural networks (BP-NN) and support vector machine (SVM), hidden Markov model (HMM) achieved a better recognition performance for multi-state degradation system and provided theoretical explanation of the system failure evolution.