Research On Fault Feature Extraction Method Of Rolling Bearing Based On Manifold

Rolling bearing as the key part of rotating machinery plays a very important role in the entire mechanical system. The operating status of rolling bearings usually determines the performance of the machine directly. Therefore, research on the fault feature extraction methods of rolling bearings operating status is very meaningful. In this paper, for the characteristics of early weak fault signal of rolling bearings under strong background noise, two fault feature extraction methods are studied by combing the manifold learning methods with other modern signal processing methods. The main contents of this paper are as follows:1. The fault mechanism and common fault feature extraction methods of rolling bearings are discussed. The fault mechanism is analyzed by giving the basic structure and common fault forms of rolling bearing. Two mainly frequency components of rolling bearing and the formulas are also studied. Finally, by using a set of laboratory data of rolling bearing the effectiveness of time-frequency analysis method is verified.2. A noise reduction method based on EEMD (Ensemble Empirical Mode Decomposition) and manifold learning is proposed. In order to overcome the mode mixing of EMD and the deficiency of traditional noise reduction method based on threshold, the proposed method effectively removes the noise components of original signal by combining the EEMD decomposition and LTSA(Local Tangent Space Alignment) manifold algorithm. The simulation and actual bearing signals verified the effectiveness of the new method.3. A feature extraction method based on wavelet packet and manifold sample entropy is proposed. The wavelet packet decomposition can effectively analyze the high-frequency detail components of signal and sample entropy is an important parameter to show the self-similarity of data system. By using the KPCA(Kernel Principal Component Analysis) manifold algorithm, the sub-bands of wavelet packet decomposition can be reduced into low dimension manifold sub-bands, the sample entropy of which are chosen as the characteristic parameters to distinguish the different working states of rolling bearing. Finally, the validity of the proposed method is verified by four different states data of rolling bearing.4. A vibration signal analysis system of rolling bearing based on C/S structure is developed by using the virtual instrument technology, database technology and signal processing technology. The analysis system contains five independent modules which represents the five key parts of vibration signal analysis process. The entire system is verified to be effective and reliable by actual rolling bearing data.