Research On Fault Diagnosis Method Of Rolling Bearing Based On Segment Clustering

Rolling bearing is one of the most easily damaged components in rotating machinery. According to statistics, about30%of mechanical failures are caused by bearing damage. Therefore, the rolling bearing condition monitoring and fault diagnosis is important. Vibration signals of defected bearing is a pulse waveform which lasts a very short time. In this study, shock pulses have been cut out from the background noise, and the frequency of impacts are calculated to locate the source of defect, thus bearing working condition can be identified. The content of the thesis is as follows:(1) The characteristics of vibration signals of different bearing fault types have been discussed. The discussion presents the theoretical basis of bearing condition monitoring and fault diagnosis.(2) The effectiveness of wavelet analysis and Fourier transform on pulse detection has been compared. Wavelet analysis can be used to detect mutations points in the bearing fault signal. Then, the pulses can be cut out based on the mutation points.(3) Different feature extraction method for bearing pulses has been studied, including time domain features, frequency domain features, and wavelet package features. As there are many types of features, the principle component analysis is used to reduce the dimension of feature space, and the characteristics of bearing pulse can be represented by a few features.(4) The algorithm of clustering bearing pulses has been studied. Detailed definition of clustering algorithm. Classification of clustering algorithms, distinguish differences between clusters have been discussed. As fuzzy C-means method has been used to cluster the bearing fault signal, some parameters of this method have been inspected. Finally, the effectiveness of this method was verified by two experiments. One of the experiments was performed on a bearing with outer ring fault, and the other is on a bearing with both outer ring fault and rolling element fault. The results shows that the method proposed is feasible, simple, and reliable.