Research On The Detection And Feature Extraction Of Weak Impact Signal Showing Rolling Element Bearing Fault

Condition monitoring and fault diagnosis of rolling element bearings can be realized by extracting and analyzing the characteristic impact signals related to the bearing faults.However,in some operation conditions,the fault-related impact signals are relatively weak or even submerged due to the existence of heavy background noise and other source of vibration in the rotating system,which greatly affects the performance of condition monitoring and fault diagnosis of rolling element bearings.To settle this problem,the detection and feature extraction algorithm of weak bearing impact signal under heavy background noise and large disturbance is systematically studied.In the dissertation,several methods are proposed respectively for the detection of weak impact signal,the extraction of fault signal and the feature extraction of the characteristic signal.The main contents are as follows:In order to detect the fault-related weak impact signals and determine the real characteristic frequencies of rolling element bearings,a method based on improved singular value ratio spectrum is proposed.In the improved singular value ratio spectrum,cubic spline interpolating function is used to change the sampling frequency of the vibration signal,so the ideal number of sampling points in each row of the matrix will be an integer,which can eliminate the row phase error accumulation in matrix construction.Then the improved singular value ratio spectrum is introduced to detect the impact signal of bearing fault.The detection and real characteristic frequency determination of fault impact signal can be realized by checking the large prominent components in the improved singular value ratio spectra of the measured vibration signals.In order to extract the weak impact signal of rolling element bearing fault,an improved time averaging method is proposed.The proposed method is an improvement of the time synchronous averaging method,which considers the slippages between rolling elements and inner/outer race and the fluctuating load of bearings during operation.In the proposed method,correlation coefficients are introduced to evaluate the similarity of short-time patches.The ideal patches are searched around the designated time to eliminate the impact of rolling element slippages.Meanwhile,weighted coefficients are introduced to the averaging process to minimize the impact of load fluctuations.The improved method performs well in suppressing the interferences and extracting the characteristic impulsive components of bearing fault.Aiming at the feature extraction of characteristic impact signal of rolling element bearing fault with the impact of independent frequency interferences and in-band noises,a feature extraction method based on narrowband envelope spectra fusion is proposed.The method suppresses in-band noises and enhances fault features by averaging the envelope spectra at neighborhood narrow bands.Meanwhile,some minor improvements are made to conventional narrowband envelope spectrum calculation method to enhance the similarity of the narrowband envelope spectra containing fault features,and finally optimize the fusion process.Then sparsity values of these denoised envelope spectra are utilized to determine the optimal band and select the optimal envelope spectrum.By narrowband envelope spectrum fusion and introducing the sparsity values,the influence of noise and independent frequency components can be reduced,and the feature extraction of bearing fault impact signal can be more accurate and effective.Aiming at the feature extraction of characteristic impact signal of rolling element bearing fault with multiple interferences,a method that combines narrowband envelope spectra with blind source separation is proposed.The method transforms the extraction of characteristic envelope spectra into blind source separation problem,which is implemented by fusing envelope spectra at multiple narrow frequency bands.These informative frequency bands are determined by Jarque-Bera statistic of envelope spectrum amplitudes,and the fast independent component analysis is introduced to extract the fault-related source spectra in the selected bands.The proposed method does not need to determine the optimal frequency band of the signal,which can effectively avoid the uncertainty and errors in band selection.Since all repetitive transient features are preserved,the feature extraction of bearing impact signal can be more robust and accurate.