Research On Rolling Element Bearing Vibration Signal Processing And Feature Extraction Method

Rolling element bearing is one of the most widely used parts in a varity of rotating machinery, whose running state often directly affects the performance of the whole machine, so the condition monitoring and fault diagnosis of rolling bearing has important practical significance and economic value. In this paper, rolling bearing is the research object. Starting with the characteristic of faulty bearing signal, aiming at the key technologies in bearing fault diagnosis, i.e., demodulation, denoising and feature extraction, and applying modern signal processing techniques, a series of studies on the condition monitoring and fault diagnosis of rolling bearing are carried out. The main works of this dissertation are listed as follows:1. Based on the analysis of rolling bearing vibration mechanism, the vibration signal characteristics of the bearing with local damage are summarized as cyclical shock and amplitude modulation. Then the key issues of rolling bearing fault diagnosis are summarized as demodulation, denoising and feature parameter extraction. Only these three issues are well solved, rolling bearing fault can be accurately diagnosed.2. A bearing fault diagnosis method based on EMD (empirical mode decomposition) denoising and spectral kurtosis is proposed. EMD denoising is summarized as two methods, i.e., threshold-based processing and filter-based processing. In threshold-based method, wavelet threshold-denoising is referenced; in filter-based method, two criterions are put forward according to the characteristic of faulty bearing vibration signal. Taking a simulated signal of faulty bearing as an example, the performance of the two methods is compared. As a result, filter-based EMD denoising method is more suitable to be a preprocessing means for bearing signal. In view of the difficulty to determine the band-pass filter parameters of resonance demodulation, the theory of spectral kurtosis is introduced. The frequency band with largest spectral kurtosis value is confirmed as the resonance frequency band. Combining EMD denoising and spectral kurtosis theory, a rolling bearing fault diagnosis method is presented, which is verified by practical engineering signal.3. Based on optimal Morlet wavelet filter and autocorrelation enhancement algorithm, an automatic diagnosis method of rolling bearing fault is proposed. To eliminate the frequency associated with interferential vibrations, the bearing signal is filtered with a band-pass filter determined by a Morlet wavelet whose parameters are optimized by genetic algorithm. The objective function and constraints are also discussed. Using the optimal Morlet wavelet filter, the SNR is improved significantly. To further reduce the residual in-band noise and highlight the periodic impact characteristic, an autocorrelation enhancement algorithm is proposed, which is then used to the filtered signal. In the enhanced autocorrelation envelope power spectrum, only several single spectrum lines would be left, which is corresponding to bearing fault frequency for a defective bearing or to shaft rotational frequency for normal bearing. It is very simple for operator to identify the bearing fault type. The proposed method can be conducted in an almost automatic way. The results from simulated and practical experiments prove the method is very effective for bearing faults diagnosis.4. A denoise method based on hidden Markov tree (HMT) model in dual tree complex wavelet domain is proposed. In view of the drawback of discrete wavelet transform with shift sensitivity and complex wavelet transform without perfect reconstruction, dual tree complex wavelet transform (DTCWT) is adopted to process the bearing signal. In view of the unfavorable effect of traditional wavelet denoise without considering the correlation between wavelet coefficients and the non-Gaussian nature of these coefficients, a denoise ideal of wavelet domain HMT model is introduced. Combining DTCWT and wavelet domain HMT model, a more advantage method is raised. According to the principle that the real and imaginary parts of DTCWT coefficients are considered simultaneously or separately, two kinds of denoise method based on dual tree complex wavelet domain hidden Markov tree model are proposed. Their performance are compared with traditional methods and indicated the simultaneously considered one is better. This method is then used in practical rolling bearing fault diagnosis to verify its efficiency.5. A rolling bearing fault diagnosis and prognosis method based on wavelet packet sample entropy is proposed. Sample entropy (SE) in information theory is introduced into the field of mechanical fault diagnosis. Its performance and the choice of calculation parameters are discussed, which indicated that SE has better performance than approximate entropy in term of anti-noise, sample length requirement, or reflecting the system nature. Since the greater the sub-band energy of wavelet packet decomposition is, the more obvious the fault information is, wavelet packet sample entropy (WPSE) is proposed, which can distinguish well bearing fault type. Then WPSE is applied to bearing fault trend prediction. Calculating the WPSE of life cycle bearing test rig data and using EMD to extract tendency, the bearing running states can be well predicted. WPSE can forecast earlier fault development fault than RMS and kurtosis, which indicate WPSE is a good bearing monitoring and forecasting tool.6. Based on LabView and Matlab software, an analysis system of rolling bearing vibration signal is developed. The system development environment and overall structure design are introduced. Then the implement method of each module is discussed. Through the function demo for actual signal, the convience and efficience of the system is verified.