Research On Vibration Signal Based Rolling Element Bearing Feature Extraction And Fault Diagnosis Method

Rolling element bearing is one of the most commonly used components in the rotating machinery, whose working condition has direct influence on the performance of the whole equipment and the safety of the entire production line. Therefore, it is extremely significant to research on the condition monitoring and fault diagnosis methods of rolling element bearings. The rolling element bearing is taken as the research object in this paper. Aiming at the key issue of the bearing fault diagnosis, viz., feature extraction, a series of research work is conducted by using advanced signal processing techniques. The main contents are listed as follows:1) The background and significance of the chosen topic are discussed and the vibration mechanism of the rolling element bearing is analyzed. The development status of the rolling bearings data acquisition methods, fault feature extraction methods and fault pattern recognition methods are reviewed. Based on the above analysis and summary, the research contents of the dissertation are presented.2) A bearing fault diagnosis approach based on IMF (intrinsic mode demcomposition) kurtosis and SVM (support vector machine) is proposed. The shortcomings of the kurtosis for bearing fault diagnosis are discussed. The bearing vibration signals are firstly decomposed by the EMD (empirical mode decomposition) method into a series of IMFs which represent different frequency bands respectively, and then the IMFs including dominant information are selected to calculate their kurtosis values. Finally, these values are treated as the fault fearture vectors and input into the trained SVM to identify different bearing conditions. The experimental results validate the effectiveness of the presented method.3) In order to better recognize the level of fault severity, a method for rolling element bearing fault diagnosis based on IMF envelope sample entropy is presented. Firstly, the notion of entropy and the definition of the sample entropy are introduced. Then, in order to improve the performance of sample entropy, the envelopes of IMFs are chosen to compute their sample entropy values by using the modulation characheristic of the bearing vibration signal. Finally, the IMF envelope sample entropy values are used to train the multiclass SVM classifier and then the trained SVM are utilized to yield diagnosis results. The experimental results indicate that the proposed method based on IMF envelope sample entropy can identify different bearing fault types as well as levels of severity effectively and is superior to the method based on IMF sample entropy.4) The hierarchical entropy is introduced into fault feature extraction of the rolling element bearings. The drawbacks of the sample entropy and multiscale entropy when assessing the complexity of time series are firstly discussed, and then the fundamental notion and algorithm of hierarchical entropy are introduced. The hierarchical entropy not only calculate the sample entropy of lower frequency components of a time series by averaging the components in the previous scales but also compute the sample entropy of the higher frequency components by taking the difference of two consecutive scales. Finally, the SVM is used to accomplish the diagnosis and a comparison with the above mentioned methods is made. The experimental results show that the proposed approach can diagnose the bearing fault types as well as fault sizes effectively and accurately.5) Based on EMD and correlation coefficient (the normalized value of the cross-correlation function at zero-lag point), an incipient fault detection and identification method for rolling element bearings is presented. The limitation of pattern recognition is that the fault pattern classification is based on the assumption that the distribution of available data between classes is relatively balanced. Hovever, it is almost impossible to obtain data for all fault types of bearings in practical applications, because data describing fault conditions is often non-existent and acquiring it wound require damaging the bearing in purpose. To overcome this limitation, the proposed method take the bearing fault detection as novelty dectetion problem, based completely on the normal data and eliminating the need for collecting failure data in advance. After the fault detection, the envelope analysis is applied to identify the location of the faults. The experimental results verify the efficiency of the proposed approach.6) The Hilbert vibration decomposition (HVD) is introduced into the fault diagnosis field of rolling element bearings. Base on the analysis of the mode mixing phenomena of EMD, the HVD method is introduced. By using the simulation signal, the decomposition performances of the HVD and EMD are compared, and then the HVD method combined with envelope analysis is applied to analyze the experimental signal. The experimental results illustrate that the proposed method based on HVD is feasible and effective.