Research On Feature Extraction Method Of Rolling Bearing Vibration Signal

Because of the small structure and large bearing capacity of rolling bearing,it is one of the most vulnerable parts,which is often used as an important monitoring object for mechanical fault diagnosis.In the case of measured signal contains noise signal source,using a single channel blind source separation algorithm for purification of bearing signal,and for the virtual channel signal can not be accurately selected in this algorithm,proposing a sliding entropy correlation coefficient method to screen bearing fault signal source,verify the effectiveness of the blind source separation algorithm improved from the method of simulation and experiment.Taking into the accuracy of feature extraction is influenced by the modulation characteristics of the signal,this paper proposes a signal entropy features extracted method based on wavelet envelope analysis and Relief F algorithm,this method resolves bearing the original signal into different frequency bands of the envelope signal by wavelet packet decomposition algorithm combined with Hilbert transform,and the entropy features of above envelope signal that without bearing fault information be screened and removed by Relief F algorithm,so as to improve the extraction method of bearing fault features accuracy.Through experimental analysis,it is proved these features is good for distinguishing different types of bearing signals.In view of the problem that the multi domain feature extraction of bearing is not conducive to the accuracy of pattern recognition,the multi domain feature set of bearing signals is reduced to dimension.Comparing the differences of each dimensionality reduction method,the LLE algorithm is selected as the dimensionality reduction method of bearing signal features.And for the problem that the algorithm uses the Euclidean distance for the nearest neighbor samples,distorts the mapping structure of low dimensional data,a new distance formula is proposed.Through the simulation and experimental data analysis,it is proved that the improved algorithm improves the accuracy of the feature description of the bearing signal.Based on the above research methods,the fault diagnosis system of rolling bearing is developed by using MFC and MATLAB software.The actual bearing signal is acquired through experimental platform as an example,and it is verified that the system can successfully judge the bearing fault type through the feature extraction method described in this paper.The feature extraction method of bearing vibration signal is studied.Considering the feature extraction of bearing signals from multiple angles,the accuracy of bearing fault description can be improved,so that it can give a comprehensive overview of the running state of bearings.