Fault Diagnosis Method Of Bearing Based On Order Statistic Filter And Empirical Wavelet Transform

The mechanical equipment in the modern industrial system,such as high-speed rail,intelligent robots,robotic arms and large-scale equipment production lines,are inseparable from rotating machinery and bearings.Bearings are subject to high-speed and heavy-duty loads,and also face complex conditions such as high and low temperature differences.Bearings that work in this environment for a long time are prone to minor damage and extend the bearing damage and affect the operation of the equipment.A healthy bearing is the basis for the normal operation of the equipment,and the bearing that is easily damaged and causes the equipment to fail is one of the biggest hidden dangers of equipment failure.Therefore,it is very necessary to monitor the condition of the bearing and diagnose whether it has failed.The signal processing method based on fault feature extraction is one of the key technologies in the field of fault diagnosis.Since the bearing is running periodically,a periodic pulse occurs in the signal collected by the acceleration sensor when a certain part of the bearing fails.Extracting the component of the shock information in the extracted signal can improve the speed and accuracy of the fault diagnosis.After studying many signal processing methods,it is found that the empirical wavelet transform can extract components of different frequency components by dividing the spectrum,filtering and reconstruction when processing the stationary signal.It has an adaptive and reliable mathematical derivation process.In the empirical wavelet transform,the spectral segmentation problem directly related to the number of empirical modes and the final decomposition effect is deeply studied.A fast empirical wavelet transform is proposed to optimize the filtering boundary.Then the relationship between key functions and periodic pulses in fast empirical wavelet transform is discussed.An adaptive fault extraction method based on trend spectrum and spectral negentropy is proposed.After studying the envelope characteristics of the order statistics filters and the filtering characteristics of the empirical wavelet transform,it is found that the order statistics filters can be used to quickly estimate the upper or lower envelope of the signal,which can be extended and applied in many algorithms.Based on this,the fourth chapter proposed the Adaptive Kurtogram method.By increasing the window width of the order statistics filters to change the number of envelope extremum points to obtain different filter banks,the best distribution according to the signal frequency components can be obtained.The method of obtaining the boundaries based on the spectral features of the signal replaces Fast Kurtogram,and the segmented modulation information has integrity.Aiming at the shortcomings and defects of the local mean decomposition method,the order statistics filters and the empirical wavelet transform are used to optimize the extraction process of the product function.Envelope rules based on order statistics filters were developed.The calculation effect and time-consuming of different interpolation times verify that the order statistics filter method is superior to the interpolation method.By simulating the signal and adding noise with different signal-to-noise ratios,it is verified that the spectral negentropy has stronger anti-noise ability,and an adaptive stopping criterion based on spectral negentropy is proposed.The simulated signal and experimental signal verify that the above methods are suitable for fault feature extraction of rolling bearing.