Rolling Bearing Fault Diagnosis Under Variable Rotational Speed Based On Generalized Demodulation

In machinery,transportation,aerospace,electric power and other industries,rolling bearing is a crucial part of rotating machinery and equipment.It operates under variable speed conditions in most cases.Once it fails,the loss is huge.Thus,rolling bearing fault diagnosis is particularly important for engineering application,and protection for personal and property safety.Generalized demodulation is an effective time-frequency analysis method to process non-stationary signals and highlight their local characteristics.To develop the potential of generalized demodulation,this paper studies the following three problems:(1)The phase function of generalized demodulation is estimated in the time-frequency diagram by peak search algorithm usually.However,when the peak search is used to analyze the bearing signal,the estimated result has a certain error at the end point,which makes the energy concentration point deviate from the actual point and interferes with the fault diagnosis.Therefore,how to enhance the estimation accuracy of the phase function and ensure the accuracy of the generalized demodulation is an important topic.(2)How to solve the problem of spectrum aliasing caused by the selection of parameters in the process of iterative generalized demodulation analysis.(3)How to realize the adaptive analysis ability of the bearing vibration signal based on the generalized demodulation method to make it more suitable for engineering applications.To handle the question that the phase function has some error in estimation.An estimation algorithm based on mirror image continuation method is raised.Firstly,the resonance zone,extracted from vibration signals is contained many fault characteristics of impact signal frequency band,then the mirror image continuation is used to preprocess the resonance band signal,and the extended signal is processed by shorttime Fourier transform for the time-frequency spectrum.Finally,the peak search algorithm is used to estimate the instantaneous frequency and the extended frequency of signal segment is cleared,and the residual instantaneous frequency is the required phase function.The results reveal that the method has high accuracy in the phase function estimation.Facing the question of fault feature extraction error caused of iterative generalized demodulation,and considering the generalized demodulation algorithm,a fault feature extraction method is put forward.Firstly,the signal is analyzed by the spectral kurtosis algorithm,and the resonance band containing fault impact characteristics is extracted.Then the amplitude envelope of the resonance band signal is made.Secondly,the envelope signal is decomposed by Variational Nonlinear Chirp Mode Decomposition and instantaneous frequency is estimated.It is used as the phase function to demodulate the signal until the signal is completely decomposed.Finally,all the demodulated signals are accumulated,and the results are processed by fast Fourier transform to extract the characteristic frequency.Simulation and experiments indicate that the method avoids the spectrum aliasing phenomenon and effectively improves the fault recognition rate.Facing the question of lack of adaptive feature extraction in generalized demodulation,an adaptive generalized demodulation method is proposed.The core strategy of this method is to combine ACMD with generalized demodulation effectively to realize the adaptive process in the selection of phase function and parameters.Firstly,the envelope signal is obtained by using fast spectral kurtosis and Hilbert transform.Secondly,the envelope signal is analyzed by adaptive generalized demodulation.Finally,the fault characteristic frequency spectrum of the demodulation signal is obtained,from which the fault type is diagnosed.Simulation and experiments show that the method has advantages in improving the adaptive ability,which has a certain reference significance for practical engineering applications.