Bearing Fault Feature Information Extraction And Intelligent Recognition Of SVM

Signal detection, fault feature extraction, state recognition, prediction are the main steps of fault diagnosis, and the fault feature extraction directly relates to the state recognition. Rolling bearings are common and consumable mechanical parts, they directly affect the equipment, Therefore, it is important to carry out rolling bearing fault diagnosis.The vibration signal of fault bearing is non-stationary, nonlinear with a lot of noise, it is imperfect to deal with such signal by traditional frequency domain method; The wavelet transform has a good time-frequency,but it is not adaptive; The empirical mode decomposition(EMD) is imperfect with mode mixing, illusive component and end effect. Aiming at the above situation, we summary the rolling bearing fault diagnosis technology, using ensemble empirical mode decomposition(EEMD), sample entropy and support vector machine(SVM) based on analysising the vibration mechanism and the characteristic frequency of rolling bearing, extract intrinsic mode components which include the bearing fault feature information, and realize the intrinsic mode components complexity measurement and the rolling bearing intelligent recognition of SVM. The main research contents are as follows:This thesis focuses on rolling bearings. The cause of rolling bearing’s vibration, vibration mode, characteristic frequency are analyzed, and especially, the vibration and its characteristic frequency which caused by the pitting of inner race, outer race and ball are detailed analyzed.The research suppresses the end effect of EEMD using mirror closure continuation and filters illusive components using correlation coefficientthe. Simulation signal verification shows that the mirror closure continuation can effectively suppress end effect and the correlation coefficient can exactly filter illusive components.The research calculate the complexity of intrinsic mode components using sample entropy, and the experimental result shows that the change trend increase-decrease-increase of sample entropy can reflect rolling bearing’s vibration signal varies with the degree of fault.In the experiment, we assume that the bearing is regular and inner race, outer race, ball are pitting fault, and using automaticly find out sqtwolog thresholding wavelet noise reduction after signal acquisition. Then, we extract fault feature information using ensemble mpirical mode decomposition and sample entropy. The ensemble empirical mode decomposition and sample entropy are used as a feature vector of support vector machine. The research construct Mexican hat wavelet kernel SVM, compare it with the radial basis kernel function SVM. The result of comparison shows that the accuracy of wavelet kernel SVM is better than radial basis kernel function SVM in recognizing the regular bearing and the inner race fault, outer race fault, ball fault.