Roller Bearings Fault Diagnosis And Research Based On 1(1/2) Dimension Spectral Entropy And Support Vector Machine

The rolling bearing is one of the most ordinary and important mechanical parts in the rotating machinery, and is vulnerable to damage. Many faults of rotating mechanism are related to the state of rolling bearings. Its faults can result in abnormal vibration and noise in machine, equipment damage and personal casualty. Therefore, it is important to study the fault diagnosis of rolling bearing.The method of vibration analysis is the most common in bearing vibration test and fault diagnosis. Vibration signal of rolling element bearing is collected, and then its feature is extracted through signal processing. The running conditions of bearing are recognized by using these methods of pattern recognition. Feature extraction and condition identification are privotal.The research work of this paper includes two parts:The vibration signal feature extraction of rolling bearing is studied. The 1(1/2) dimension spectrum could not only entirely remove Gaussian noise in theory but also hold the phase information compared with conventional fault diagnosis technique. It can quantitatively describe the non-linear quadratic phase coupling. So it is very suitable for extracting the fault feature from the sensor signals that describe the characteristics of their working condition. The fault vibration signal of rolling bearing has many characters, such as non-Gaussian, non-linear, non-stationary. In view of these characters, In this paper, we use the method of 1(1/2) dimension spectrum calculation, combined with the theory of wavelet packet and spectral entropy, and put forward a 1(1/2) dimension spectral entropy feature extraction method in this paper.The application of support vector machine (SVM) is researched. The paper also established the pattern recognition classifier model and studied the parameters and feature vectot extraction that influence the classifier model's classification ability; on the basis of analyzing the parameter's influence on the classifier's recognition accuracy, it proposed the self adaptive optimization algorithm for the SVM classifier model using genetic algorithm. Through experiment, this paper has extracted the malfunction features by contrast of 1(1/2) dimension spectrum and wavelet package. It shows that the genetic algorithm can effectively improve the recognition rate, but the feature vector selection is the key, therefore the feature vector extracting by 1(1/2) dimension spectral entropy can more characterize the roller bearing fault information, to obtain satisfactory results.