Latent Variable Analysis Based on Hidden Markov Model in Rolling Element Bearing Diagnostics
Résumé
In a previous paper, the authors have explored the potential of hidden Markov models in rolling element bearing diagnostics. Based on the Gabor transform, noisy measurements are decomposed in time and frequency, respectively. Then a stochastic model is built in each frequency band. As an extension of the former one, this paper introduces a more exact model which considers a latent variable as a switch corresponding to different distributions and a more complex spectral structure. Therefore, different types of fault based on distinct structures of spectral correlation are separated and estimated. The estimated latent variable carries the information according to the spectral structure. Based on the developed model, a new fault separation scheme is proposed. And the performance is demonstrated by simulated and experimental cases. In simulated case, it reveals expected fault frequencies even in heavy background noise when noise to signal ratio is high. For experimental case, it achieves the superior results to the classical envelop analysis.
Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...