Rolling Bearing Faults Diagnosis Based On Hilbert-Huang Transform And Decision Tree

The life of the bearings is always unpredictable, therefore, it is very essential to examine the faultier diagnosis and monitor the device to keep it safely running. The process of faultier diagnosis includes acquisition of signal, feature extraction, failure mode identification etc, Faultier diagnosis is mainly based on Mechanical Signal Processing, Sensor Technology etc. In this regard the roller bearing 6205 and normal bearing was studded; the main failure mode was the inner ring failure, the outer ring failure and rolling element failure. The research was focused on the method of extracting the signal failure the bearing vibration feature, the method of reducing the dimensionality of feature, based on the application of multiple faultier diagnose algorithm the MATLAB GUI software is used to develop the faultier diagnosis.Feature extraction is the most important step in faultier diagnosis; the feature can be accurate and sensitive to record the bearing status. In this study the feature within time domain is statistically in the orders of center matrix and original moment. For frequency domain the feature is extracted using parametric method and nonparametric method. The feature is applied by HHT both in time domain and frequency domain.Accordingly the research was carried out as follow:The bearing vibration signal is resolved by HHT into Hilbert spectrum and partial Hilbert spectrum; the feature of the faultier diagnosis was then extracted in order to reduce the overlapping frequency points in EMD and obtain the real frequency, the relationship between the original signal and its derivative was analyzed. A new method was applied in this research; i.e., extreme point of the second derivative in the signal will supersede extreme point the signal in EMD, using cubic spline interpolation. Compared with the previous, the advantage was that the amplitude of the component from filtering the signal includes the high frequency signal, has practical value.To investigate the algorithm in the multi-faultier diagnosis problem, we choose the performance of the multi-class support vector machine and decision tree method was involved. In the multi-class support vector machine method, the seven features were extracted in the time and frequency domain, then the principal component analysis method was used to reduce dimension and get four features, including 98.8% of original signal information. In order to apply the algorithm in MATLAB, the program was compiled to train the samples and predict them. The results showed that it is possible to diagnose the typical bearing failure and accordingly the recognizable rate reached 97.3%.The second algorithm was to combine the HHT and decision tree method. The operation was to filter the original signal using EMD, the intrinsic mode functions was operated in noise autocorrelation method; the main frequency of the feature was extracted forming the C4.5 decision tree of the sample features accordingly. in order to reduce the complexity of the algorithm, we reduce to calculate the attribution and extract the feature in predicting process, the operation the node attribute information of the decision tree is feedback of the algorithm is applied. The recognizable rate of algorithm is up to 96%.