Research Of Motor Bearing Fault Diagnosis Methods Based On Vibration Signal

Motor is a kind of important drive equipment in Modern industrial process, the rolling bearing is not only one of the most basic components in motor, but also is liable to be damaged. Most motor faults are relate to rolling bearing. Thus, in order to ensure the normal operation of the motor device, it is of great practical significance to research the fault diagnosis technology foe motor bearing. The vibration signal is one of the most commonly used signal in bearing fault diagnosis, which have advantages of convenient to use and is able to detect minor fault. Thus, this dissertation mainly studies the motor bearing fault diagnosis methods based on vibration signal, three aspects of the research were fault detection, fault feature extraction and recognition. At last, this paper put forward complete motor bearing fault diagnosis method, The main research is as follows:Considering the complexity of the signal will increase when the bearing failure occurs, this paper put forward a kind of bearing fault detection method based on sample entropy and permutation entropy. sample entropy and permutation entropy are selected as the motor bearing fault detection method, in order to avoid the blindness of choosing fault detection threshold, the theory of the chebyshev inequality are used as the threshold setting method. This paper use the case western reserve university bearing data, the results proved that the permutation entropy and entropy can measure the change of the bearing vibration signal complexity and can be applied to detect motor bearing fault.In order to extract fault features from complex vibration signals, the fault feature extraction methods based on wavelet packet and ensemble empirical mode decomposition(EEMD) are studied separately. Through analyzing the effect of wavelet basis and the layer number of wavelet packet decomposition, this paper selected the optimal parameters which are fit to vibration signals from fault motor bearing and calculated the wavelet packet energy entropy of motor bearing to characterize bearing fault feature on different statuses. The basic principle and algorithm steps of EEMD are introduced, it can effectively inhibiting the modal aliasing problem of empirical mode decomposition. Feature extraction methods based on samples entropy and permutation entropy are proposed, after decomposed by ensemble empirical mode decomposition, the sample entropy value and permutation entropy value of intrinsic mode functions was calculated respectively to form value vector.In order to achieve the motor intelligent diagnosis of bearing fault. the fault identification method based on multiclass relevance vector machine(M-RVM) is studied. The technology does not need to construct multiple binary classifiers to solve the multiple classification problems, but realize the multiple classification directly. Fault feature vector extracted by wavelet packet energy entropy were used to classify the bearing fault type, the results prove M-RVM classification effectively applied to the motor bearing fault recognition, and has high fault identification accuracy.Finally, this paper puts forward two kinds of motor bearing fault diagnosis method, one kind of the fault diagnosis method is a combination of the sample entropy, wavelet packet energy entropy and M-RVM; The other fault diagnosis method is a combination of permutation entropy, EEMD and M-RVM. Case western reserve university bearing data is used to verify the effectiveness of two bearing fault diagnosis methods, the result shows that both two methods can effectively diagnosis the type and the fault diameter of motor bearing fault, and shows a good application prospect.