Research On Fault Feature Extraction And Diagnosis Of Rolling Bearing Based On Kurtogram-Based Algorithms

As the "joint" of rotating machinery,rolling bearings are one of the most widely used parts with the highest failure rates in rotating machinery.Research on condition monitoring and fault diagnosis of rolling bearings is of great significance for evaluating the running state of rotating machinery and ensuring the safe and stable operation of equipment.It also provides a reliable basis for Condition-Based Maintenance.In this paper,the rolling bearing is taken as the research object,and the vibration signal is taken as the analytical medium.The most widely used Kurtogram-based resonance demodulation preprocessing algorithm is studied.The limitations of this kind of algorithm in dealing with the vibration signal of rolling bearing are summarized,and the corresponding solutions are put forward.The main contents and innovations of this paper are as follows:The basic structure and principle of Kurtogram-based algorithm(Kurtogram,SE Infogram/SES Infogram)are summarized.The sensitivity of time-domain statistical characteristic index(spectral kurtosis,square envelope negentropy)and frequency-domain statistical characteristic index(envelope kurtosis,square envelope spectrum negentropy)to non-periodic transient characteristics,periodic cyclic impulse characteristics and noise is analyzed.The limitations of Kurtogram-based algorithm in dealing with three common problems(harmonic interference,compound fault and aperiodic transient component interference)are summarized.Aiming at the interference caused by the special harmonic components introduced by the spindle rotation frequency and gear meshing frequency to the Kurtogram-based algorithm(Protrugram,SES Infogram)defined by the statistical characteristic index in frequency domain,An adaptive narrowband notch preprocessing method is proposed.Particle swarm optimization algorithm is used to optimize the parameters of narrow-band notch,so as to reduce the influence of human factors,maximize the removal of harmonic components,and retain the periodic impact characteristics as far as possible,waveform matching variance is used to verify the notch results,which improves the efficiency and accuracy of notch and facilitates the subsequent analysis of resonance demodulation.Finally,the notch results are verified by the variance of waveform matching.The proposed method improves the notch efficiency and accuracy,and facilitates the subsequent resonance demodulation analysis.In order to solve the problem of missed diagnosis and misdiagnosis caused by the loss of fault components in the secondary frequency band result from the masking of the secondary frequency band in the process of maximizing the statistical characteristics in the time-domain and frequency-domain in the case of compound faults,the maximum correlation kurtosis deconvolution filter bank preprocessing method based on periodic fault impulse feature matching and the variational mode decomposition preprocessing method based on frequency band division are proposed.The Kurtogram algorithm is used to analyze the filtered components after deconvolution inverse filtering or the modal components of each order after variational mode decomposition,which effectively avoids the misdiagnosis and missed diagnosis caused by resonance band submergence.Influenced by operating environment and signal transmission path,strong background noise and large non-periodic transient impulse components will be introduced into rolling bearing vibration signals,while kurtosis/square envelope negentropy and other time-domain statistical indicators are sensitive to non-periodic transient components,and have poor robustness at low signal-to-noise ratio,which makes it impossible to accurately locate the frequency band of periodic cyclic impulse characteristics caused by faults.To solve this problem,a resonance f-requency band detection method based on time-domain teager energy negentropy and frequency-domain teager energy spectrum negentropy is proposed to improve the detection ability of periodic cyclic impulse characteristics,and a TEERgram is constructed by calculating the ratio of TEE(Teager Energy neg-Entropy)in fault state to that in normal state to further suppress the strong noise source.Aiming at the problem of bearing fault type and fault degree identification with small sample and unpredictable clustering number,a fault identification method based on normalized time-frequency TEE of wavelet packet subband and MS-FCM clustering algorithm is proposed.Through the wavelet packet decomposition of the vibration signal,the time-frequency TEE of the wavelet packet coefficients of each subband is calculated and the normalized discrete data processing is carried out,and the eigenvectors which can reflect the characteristics of the vibration signal in time and frequency domain are constructed.The eigenvectors can effectively represent the fault characteristics and have obvious distinctions between the components of the classes.In the fuzzy C-means clustering,the number of clusters needs to be preset,and it is difficult to classify accurately when the number of clusters is small and can not be predicted.Aiming at this problem,a Meanshift assistant pretreatment method is proposed,which can search the region with larger probability density distribution as the initial location adaptively,and the number of initial positions as the initial clustering number,and then the tuzzy C-means clustering method is used to update the membership degree and clustering center,and determine the clustering category.