Roller Bearing Fault Feature Enhancement And Detection Method

To extract the weak fault signal and achieve fault characteristics by direct detection for roller bearing diagnosis, this paper carried out research on fault feature enhancement and detection method. The bearing fault enhancement method is based on peak feature and sparse representation and detection method is built on a few compressed data for direct feature extraction. At same time, peak hold method was used for signal down-sampling and signal feature enhancement and applied to low speed bearing fault diagnosis. This paper also applied spectrum kurtosis to direct-drive wind turbine main bearing fault feature recognition. Different approaches including theoretical analysis, analogue simulation, and experimental analysis were taken for validating the effectiveness of the method. Specific content are as follows:(1) Signal transform method based on peak feature was proposed to enhance bearing vibration signal in the transform domain, which was achieved by piecewise recombination relied on peak point. Next, bearing sparse representation method was proposed by using wavelet basis and discrete cosine basis. At last, the enhancement frame based on peak feature and sparse representation was built. Compared with wavelet method and discrete cosine transform method, results Verified the effectiveness of the enhancement method.(2) Bearing fault direct detect method based on a few compressed data was proposed by applying compressed sensing frame. Fault characteristic harmonicas were taken as detect objects. Gaussian random matrix was adopted as sample matrix of compressed sensing to acquire under-sampling compressed data. Then, towards an incomplete reconstruction process, compressive sampling matching pursuit algorithm was applied to detect bearing fault harmonica at low sampling rate. Finally, the proposed method is verified through both simulation and experiments, which shows that the method can achieve a direct and reliable detection of bearings faults from compressed data. At same time, the results of compressed detection probability with different sampling rates were analyzed.(3) The low-speed bearing fault diagnosis methods researches were carried out. First of all, peak hold algorithm is applied to low-speed bearing signal processing for solving difficult problem of extracting the fault features and reducing analytical data. Compared with the normal down-sample method, the applied method could can effectively retain or enhance the fault while reducing analytical data. At the same time the method of spectral kurtosis was used to the engineering practice, a wind turbine low speed main bearing signal processing. The fault feature could not be extracted effectively by time domain analysis, frequency domain analysis and envelope spectrum analysis. For solving this problem, the method of spectral kurtosis is used to choose filter parameters. Then envelope analysis is applied to the signal after filter processing by optimized filter parameters. The final result demonstrated that the spectral kurtosis method could extract the fault characteristic frequency effectively.