Research On Fault Diagnosis Method Of Rolling Bearing Based On Improved Empirical Wavelet Transform

Nowadays,modern machinery and equipment are developing in the direction of automation,digitization and intelligence,such as high-speed railway,intelligent machining center,aerospace equipment and large-scale equipment production line are inseparable from "industrial joint" —rolling bearing.Rolling bearings often work in tension,compression,alternating and other complex stress states and high stress conditions,which are prone to minor damage and expansion,leading to rolling bearing damage,thus affecting the normal operation of the equipment,and even causing serious accidents and huge economic losses.Therefore,it is particularly necessary to monitor the running state of rolling bearing and diagnose its fault.When the defects such as spalling appear in the rolling bearing,the periodic high-speed rolling element will stimulate two kinds of mutual coupling vibration while passing through the defects: the periodic impact vibration characterized by the relationship between structure and motion and the pulse attenuation vibration characterized by the natural frequency of the bearing element.Therefore,the accurate and rapid diagnosis of rolling bearing can be realized by extracting the periodic impact information from the vibration signal.In recent years,non-linear and non-stationary signal analysis method has become one of the research hotspots in the field of fault diagnosis,among which the signal adaptive decomposition method is the most representative.Empirical wavelet transform（EWT）stands out due to its adaptability,complete mathematical theory basis and low mode mixing effect.According to the characteristics of signal Fourier spectrum,the empirical wavelet transform adaptively divides the spectrum and reconstructs it to extract the empirical mode（EM）components with different frequency components.In this paper,we deeply study the spectrum segmentation problem that directly affects the rationality of the decomposition results in empirical wavelet transform,propose an improved empirical wavelet transform based on the mean envelope spectrum trend,discuss the influence of different interpolation methods on the spectrum trend estimation,and give the iteration stop criterion to obtain the optimal spectrum trend.In view of the shortcoming that the traditional spectral kurtosis method is easy to be disturbed by accidental impact,the paper proposes a robust method named Enfigram to extract the optimal frequency modulation band by taking the spectral kurtosis of envelope spectrum as a new statistical index—envelope Fourier index（Enfi）and the improved empirical wavelet transform.In the process of calculating Enfi,the accidental impact interference is transferred to the initial position of spectrum of envelope spectrum.The low-pass filter can easily eliminate the interference,and improve the accuracy of the central frequency and bandwidth of the resonance demodulation band.Using the method of numerical verification,the anti-interference ability of spectral kurtosis,sparse value,spectral negentropy,unbiased autocorrelation kurtosis and Enfi are compared and analyzed.Then,the characteristics of each statistical index under different signal-to-noise ratio,different bandwidth and aperiodic transient impact are summarized.Moreover,the Enfigram,Fast Kurtogram,Sparsogram,Infogram and Autogram are analyzed by using the fault signal of rolling bearing and the comparsion results indicates that the Enfigram method is more robust to aperiodic transient shock and harmonic disturbances.The aforementioned EWT has shown its superiority in the analysis of non-stationary and non-linear signals,however it is helpless for multivariate signal analysis.As to this EWT shortcoming,a quaternion empirical wavelet transform quaternion is proposed.The quaternion Fourier transform is used in QEWT instead of the traditional Fourier transform to make it suitable for multivariate signal processing,and combined with a spectrum segmentation method based on spectrum trend to optimize the filter boundary.QEWT realizes the comprehensive utilization of data in different directions in space,which greatly improves the accuracy and credibility of the diagnosis results.Simulation and experimental signals verify the above methods,and the results show that the above methods are suitable for the extraction of fault features of the inner and outer rings of rolling bearings.