Research On Fault Feature Extraction Method Of Rolling Bearing Based On Empirical Wavelet Transform

In the fault diagnosis of electromechanical equipment,the vibration signal generated in mechanical work is an important basis for detecting the health status of electromechanical equipment.As an important component of electromechanical equipment,rolling bearings often produce vibration signals with non-stationary and complex components.In recent years,there are many methods for analyzing non-stationary vibration signals,mainly represented by wavelet transform(WT),empirical mode decomposition(EMD),and fast spectral kurtosis(FK).Although the traditional signal processing method can extract the effective fault feature components in the vibration signal,it cannot guarantee the extraction of the most abundant fault feature information,which is difficult to accurately diagnose the fault type of the rolling bearing.As a novel signal decomposition method,EWT is widely used in the fault diagnosis of electromechanical equipment.The method overcomes the shortcomings of EMD,such as: mode confusion and endpoint effects,and retains the basic characteristics of wavelets.It has the characteristics of adaptive division of signal frequency domain and is concerned by more and more experts in the field of fault diagnosis.Extracting the sideband component containing the fault frequency from the vibration signal becomes the key to effectively diagnose bearing faults.As an adaptive frequency band division method,EWT provides a powerful theoretical basis for effectively extracting the optimal sideband.At the same time,EWT's frequency band division still has certain deficiencies,and there is an urgent need for improvement in the rationality of frequency band division.Therefore,this paper fully studies the characteristics of rolling bearing fault vibration signal.Using EWT as the main research method,a new method will be explored to find the best sideband,effectively extract the fault feature information,and realize the fault diagnosis of rolling bearing.The paper mainly includes the following contents:(1)The basic principle of EWT is introduced in detail.The characteristics of rolling bearing fault vibration signal are studied.The difference between EWT and EMD in the extraction of bearing fault characteristic frequency is compared.(2)Using the spectral partitioning characteristics of empirical wavelet transform,combined with the spectral kurtosis principle,the empirical scanning spectral kurtosis method(ESSK)is proposed to realize the extraction of the sideband of the rolling bearing fault feature.The method can accurately find the center frequency and bandwidth corresponding to the optimal sideband and extract the most abundant fault feature information.Simulation and experimental signals demonstrate its effectiveness.(3)Optimize the ESSK spectrum partitioning part to improve the operation speed,and propose the fine spectral negentropy method(ASNE)based on the principle of spectral negentropy.The method uses spectral negentropy as the screening index,and can successfully capture the periodic frequency components in the signal,and extract the fault characteristic frequency through the EWT filter.(4)Aiming at the problem that ESSK and ASNE are susceptible to accidental impact interference,a new method of multi-segment fusion empirical wavelet transform(MMEWT)and sparsity selection optimal segment(OSSS)is proposed(MMEWT-OSSS).Firstly,EWT is improved by multi-segment fusion method,and its frequency band division method is optimized.By combining with Hilbert change,the EWT envelope component is obtained,and the optimal segment is selected by the sparse value.The method is applied to the bearing.On the fault simulation signal and experimental signal,the fault characteristic frequency component is successfully extracted,and the fault diagnosis of the rolling bearing is realized.