Font Size: a A A

Feature Extraction And Fault Diagnosis Of Non-Stationary And Non-Gauss Mechanical Vibration Siganls

Posted on:2003-01-05Degree:DoctorType:Dissertation
Country:ChinaCandidate:H B ZhengFull Text:PDF
GTID:1118360092455046Subject:Mechanical Manufacturing and Automation
Abstract/Summary:PDF Full Text Request
As we know,feature extraction is the most important and difficult topic in the field of mechanical fault diagnosis. To some extent,feature extraction is a problem which hinder the mechanical fault diagnosis technique from getting further improvement. In this thesis the problem of feature extraction and fault diagnosis is addressed using wavelet,time-frequency distribution and bispectrum of non-stationary,non-Gauss vibration signals.The application of discrete and continuous wavelet transform to gear vibration signal feature extraction and fault diagnosis is investigated. An efficient fault diagnosis system based on discrete wavelet transform and artificial neural network is presented,which can accurately estimate gear fault advancement. A new concept of Time Averaged Wavelet Spectrum (TAWS) based on Morlet continuous wavelet transform is proposed. Two fault diagnosis methods named Spectrum Comparison Method (SCM) and Feature Energy Method (FEM) based on TAWS are established. The results show that TAWS is fairly sensitive to gear condition and the two fault diagnosis methods are effective.An automated de-noising algorithm based on the energy of wavelet packet not Donoho's threshold algorithm is established. A gear feature vibration signal extraction method using the wavelet packet energy is proposed,which can separate gear meshing vibration,noise vibration and gear fault vibration signal from the original gearbox case vibration signal. Basis Pursuit (BP) algorithm is introduced into the filed of mechanical signal de-noising and feature extraction for the first time. The effectiveness,sensitivities and limitations of the application of BP algorithm are analyzed in detail.The Reassigned Smoothed Pseudo Wigner-Ville Distribution (RSPWVD) is introduced,which has less interference and higher resolution than other time frequency distributions like Wigner-Ville,Choi-Williams etc. It is used to detect the multi-knocking of engine such as cylinder,piston pin,valve,crank bearing,connecting rod bearing and jib knocking. A new fault diagnosis technique based on the RSPWVD for the engine multi-knocking problem is proposed.By comparing the bispectrum and bicoherence of the gear vibration signals,it is demonstrated that bispectrum has an advantage over bicoherence for feature extraction of such signals. A method based on bispectrum and neural network for feature extraction and fault diagnosis of gearbox vibration signal is developed.To apply the techniques proposed in this thesis to the real industrial application,a virtual instrument named Gear De-noising and Diagnosis Virtual Instrument (GDDVI)is designed in the end.
Keywords/Search Tags:feature extraction, fault diagnosis, wavelet transform, time frequency distribution, bispectrum, de-noising, gearbox gear, engine knocking, neural network, virtual instrument
PDF Full Text Request
Related items