Based On Non-stationary Time Series Analysis Of The Rolling Bearing Fault Feature Extraction Method Is Studied

The rolling bearing is one of the most widely used and damageable parts in rotating machinery with its working state directly impacting on the performance of the rotating machine, and the fault diagnosis of rolling bearing has a very important practical application significance. Because the vibration signals of the running bearing are typical non-stationary random signals, from the vibration signal can be extracted accurately reflect the running status of bearing fault feature is the key to the fault diagnosis.Auto-Regressive(AR) parameter model is the most basic and widely used time series model in time series analysis method, but the AR model analysis signal based on the random stationary hypothesis, cannot accurate analysis non-stationary random signals of rolling bearing. Therefore, the paper puts forward a method based on empirical mode decomposition(EMD) and AR model combining feature extraction method of rolling bearing fault. The method using EMD decomposition of the rolling bearing vibration signal is decomposed into a number of intrinsic mode function(IMF), using correlation analysis to extract the first five IMF components and establish AR model, then extract model parameters and the variance of error of the singular values as a reflection of running state of rolling bearings feature vector. The experimental results show that the method extracts features high precision, good effect.In order to overcome the EMD decomposition signal precision is not high, the paper use wavelet packet decomposition(WPD) has good time-frequency localization characteristics and multi resolution feature, based on the non-stationary vibration signals into stationary signals, which can make the vibration signal was decomposed into each frequency band and make the AR model can effectively track the signal, presents based on wavelet packet autoregressive(WPD_AR) model and wavelet packet time varying autoregressive(WPD_TVAR) model two kinds of rolling bearing fault feature extraction method. First to the rolling bearing vibration signal wavelet packet decomposition, then respectively establish the decomposition of the nodal coefficient to the AR model and TVAR model, and respectively extract WPD_AR model and WPD_TVAR model parameter singular value as reflection of running state of rolling bearings feature vectors. The experimental results show that, the WPD_AR model is better than EMD_AR model fault feature extraction is more effective, faster; the WPD_TVAR model is better than WPD_AR model fault feature extraction results in more accurate, high precision.The fault characteristics by using the three feature extraction methods in this article, which is sended into the SVM classifier to make faults classing and diagnosis. Experiments show the proposed methods can effectively and accurately identify roller bearing fault condition, validated the proposed based on non stationary time series analysis of rolling bearing fault feature extraction method is effective.