Location:Home > Engineering science >  Control Theory and Control Engineering > Features Extraction and Selection in Rolling Bearing Fault Diagnosis

Features Extraction and Selection in Rolling Bearing Fault Diagnosis

Downloads: []
Tutor: PengTao
School: Hunan University
Course: Control Theory and Control Engineering
Keywords: Feature extraction,Feature selection,Mixed-domain Features Set,Fault,Rolling bea
CLC: TH165.3
Type: Master's thesis
Year:  2011
Facebook Google+ Email Gmail Evernote LinkedIn Twitter Addthis

not access Image Error Other errors

The essence of bearing fault diagnosis is pattern recognition,including signal acquisition, feature extraction and selection, and state recognition, feature extraction and selection of which are particularly important.they have a direct impact on the success of the pattern recognition.The vibration signals generated from running bearings is a complex non-stationary and nonlinear signal.How to extract the fault characteristics from the vibration signal which could fully and accurately reflect the state of bearing health is essential. For the effectiveness of various non-stationary statistical features with significant differences from time domain, frequency domain and time-frequency domain, a mixed-domain feature extraction approach was proposed. The mixed-domain features set can reflect the failure characteristics more comprehensive and accurately than a single feature or single-domain features.A large number of original features lead to high input space dimension and severe association between characteristics.According to this problem, Respectively, Principal Component Analysis and Kernel Principal Component Analysis are utilized for the second feature extraction.Compared with Principal Component Analysis,KPCA can fully extract the nonlinear components in the fault information while in dimension reduction and denoising,improve the separability of failure mode.Kernel Principal Component Analysis for feature extraction dose not fully consider the class information, According to this problem, Kernel Fisher discriminant analysis is used to extract useful information from the original feature set,then improve separability for pattern recognition.In order to weaken the effects of which the blindness of traditional threshold determination in feature selection methods on the result of feature selection,two fault feature selection methods are proposed in this paper,one is based on distance feature evaluation and Support Vector Machine ,the other is based on F-score and Support Vector Machine.On the basis of the support vector machine classification accuracy,intelligent threshold determination and effective features selection are both achieved. Number-less sensitive features which contain significant category differences information are directly selected from bearing fault mixed-domain features set.Simulation results show the proposed method can accurately and effectively identify the different operating conditions, different types and degrees of bearing fault conditions.
Related Dissertations
Last updated
Sponsored Links
Home |About Us| Contact Us| Feedback| Privacy | copyright | Back to top