Research Of Rolling Bearing Fault Diagnosis Based On Blind Source Separation

Rolling bearing is one of the important component of train wheelset, its quality state is usually directly affect the safety operation of the train. Especially in the process of train running, once the train rolling bearing failure, is likely to lead to major accidents such as the train derailment, overturning. In this paper, taking the fault diagnosis of the train rolling bearing as research goal and vibration signal as the research object based on blind source separation(BSS) methods, aiming at the field condition restrictions for fault vibration signal acquisition of train rolling bearing, mainly focus on the apply of single-channel blind source separation for train rolling bearing fault diagnosis. Since the vibration signals acquisition by sensors inevitably will be affected by other vibration source and environmental noise interference, a noise reduction method based on improved singular value decomposition(SVD) is proposed. At the same time, for the extremely under-determined problem of single-channel blind source separation, the adaptive empirical mode decomposition(EMD) combined with blind source separation is applied to single-channel train rolling bearing fault diagnosis from the perspective of increasing signal dimension, numerical simulation and experimental results demonstrate the effectiveness of the proposed method, provide a new thought for rolling bearing fault diagnosis.The main research contents and the key conclusions are shown as follows:(1) The noise reduction method of traditional SVD is improved in the second chapter. On the basis of studying the shortcomings of noise reduction method of traditional SVD, the temporal constraint estimate is introduced, improved the noise reduction method of traditional SVD, for the influencing factors of noise reduction order and Lagrange multiplier respectively proposed a new selection principle and adaptive solution method to avoid the human factors influence of experimental principles. Numerical simulation experiment results show that the signal to noise ratio of the de-noised signal by improved SVD method is higher by about 2.7d B and 1.6dB compared to the traditional SVD method used maximum spectral peak, respectively, unilateral maximum value as the order of noise reduction, exhibit the better noise reduction performance of improved SVD method.(2) The third chapter concentrates on the fundamental and commonly used algorithm of BSS. On the basis of introduction in detail of the fundamental of BSS and related basic concepts of probability and statistics, and information theory, the Fast ICA algorithm based on negative entropy, the JADE algorithm based on fourth-order cumulant matrix joint approximate diagonalization and the Robust ICA algorithm based on kurtosis are studied. Numerical simulation experiment results verify the effectiveness of separating a plurality of source signals from the mixed signals by the three blind source separation algorithms, do foreshadowing for the follow-up train rolling bearing fault diagnosis.(3) A new method of the combination of improved SVD, EMD and BSS for single-channel train rolling bearing fault diagnosis is studied in the fourth chapter. Firstly, the improved SVD method is applied to the noise reduction pretreatment of the single-channel vibration signal of bearing fault, and then EMD is carried on the de-noised signal, and by using SVD, combining Bayesian information criterion to estimate the source number, according to the estimated source number, BSS is applied to the reorganized multi-dimensional signals. In the numerical simulation experiment, EMD and wavelet decomposition(WD), respectively, are carried on the de-noised signal, then use the above three BSS methods to estimate the source signals, the comparison and analysis results show that the correlation coefficient between the separated signal achieved by Robust ICA combined with EMD and its corresponding source signal is closer to 1, illustrates the relatively good separation performance of RobustICA combined with EMD. Therefore, the improved SVD and EMD combined with RobustICA is applied to the single-channel train rolling bearing fault diagnosis, the analysis results of experimental data show that the proposed method is validity for the single-channel train rolling bearing fault diagnosis.