Rolling Bearing Fault Identification Method Based On Wavelet Package And Support Vector Machine

In the operation of mechanical systems, 30% of the rotating machinery failures were caused by the rolling bearing. If the bearing fault types were discovered and predicted earlier, the workers could take appropriate measures to eliminate faults in the bud, which can keep the equipments well and protect persons. Therefore, the identification method of rolling bearing fault has strong practical significance.First, this paper has analyzed the vibration mechanism and signal characteristics of rolling bearing. On the faults diagnosis laboratory bench in the laboratory, we have simulated four kinds of rolling bearing works, which are normal, with damage in outer track, with damage on ball and with damage in inner track, and collected the corresponding vibration signals.At present wavelet package method is a time-frequency analysis method of orthogonal decomposition based on multi-resolution analysis. They can decompose the low frequency parts and high frequency parts, adaptively determining the signal resolution at different frequency bands, and also can class any signal including sinusoidal signal in the appropriate frequency bands which have a certain amount of energy. Therefore, this article will take the signal energy in each frequency band as the feature vector to characterize the operation state of rolling bearing.Among the study of signal recognition methods which are used to identify machinery failures, SVM is a new intelligent fault detection theory created on the machine learning problem with a small sample. SVM can solve the practical issues with small sample of data, high dimensions and nonlinearity and so on better.On the basis of multi-fault classifier SVM's fault automatic identification technology, combining with decision tree theory and voting theory, this subject constructs a kind of multi-fault classifier algorithm, and finds a method which not only can add reuse of the known small sample, but also can keep each sub-classifier's symmetry of positive and negative training samples, and can improve the classification accuracy significantly. This paper has got good results after using extracted energy feature vectors of mock sample to identify and classify the rolling bearing faults.