Font Size: a A A

Identification Method Based On Higher Order Statistics Of Mechanical Failure Analysis

Posted on:2006-10-06Degree:MasterType:Thesis
Country:ChinaCandidate:Z H JiFull Text:PDF
GTID:2192360155969536Subject:Mechanical and electrical engineering
Abstract/Summary:PDF Full Text Request
As we know, feature extraction is the most important and difficult topic in the field of machinery condition monitoring and mechanical fault diagnosis. To some extent, feature extraction is a puzzling problem in the development of fault diagnosis technique. Furthermore, the interference noise brought from monitoring process has the key factor that to affect the accuracy of feature identification from fault signal. In order to reduce the noise and extract the feature from fault signal efficiently, the research of noise reduction and the method of feature extraction in fault signal based on high order statistics theory are presented on in this thesis.To learn the noise source and sorts is the necessary prerequisite to correct realization. The source of mechanical and electromagnetic noise in machinery condition monitoring signals is discussed. Random noise is classified by its statistical characters and power spectral density shapes. The distribution of noise in the signals of machinery condition monitoring and mechanical fault diagnosis is investigated. Almost all the noise signal is Gaussian colored noise. Two sorts of normality tests (graphical method and Jarque-Bera test) are introduced.For the purpose of feature extraction in vibration signals with Gaussian colored noise, the conception of time averaged three-order cumulant and its theoretical illumination are presented. The theoretical derivation indicates that time averaged three-order cumulant of signals, which are from random phase harmonic processes, is nonzero and its one-dimension slice also has a harmonic process. Theoretically, the bigger variance of noise is, the stronger harmonic component identified by the slice is.A parametric harmonic retrieval method in Gaussian colored noise is presented. Hilbert transform is used to transform the real measuring data into their complex counterpart. Then a kind of elaborately defined fourth-order cumulant of these complex measuring data is used to identify the AR parameters of the non-Gaussian noise. After prefiltering the noisy data with the identified AR polynomial, harmonics is retrieved successfully.More information of fault can be gotten from bispectrum of vibration signals than its power spectral density. At the other hand, there are obvioussecond harmonics in some rotor fault vibration signals. Consequently, the bispectral feature of the kind of fault is discussed. Experimentation indicates that bispectrum is more sensitive to these faults than power spectrum and can be used in engineering.
Keywords/Search Tags:fault diagnosis, signal processing, higher order statistics, Gaussian noise, harmonic retrieval
PDF Full Text Request
Related items