DOI: 10.14489/td.2014.010.pp.052-057

Sokolova E. S., Pashkovskiy А. I.
HIDDEN MARKOV MODEL BASED VIBRATION DIAGNOSTICS METHODOLOGY
(pp. 52-57)

Abstract. This paper describes an approach to construct diagnostic systems based on Markov and semi-Markov processes. The current state of detection methodology based on trained hidden Markov and semi-Markov models which describe the technical state of the object by observed values of the parameters is proposed. Sequence of samples obtained after spectral processing of vibration signal and its filtering is used as training data. Model training algorithms with using Baum-Welch iterative procedure where logarithmic likelihood function is chosen as objective function are described. Proposed diagnostic models and algorithms are used to determine the defect by real vibration signals obtained from rolling stock wheel pairs.

Keywords: defect, hidden markov and semi-markov models, checkpoint, vibration, spectral analysis, log-likelihood function.