Research Of Rolling Bearing Fault Feature Extetraction

As one of the most important parts in mechanical equipment, the normal working state ofrolling bearing has greatly effects on the performance and safety in production. It ismeaningful to study the fault diagnosis methods for rolling bearing. After extracting thefeature which represents fault characteristics, the faults can be diagnosed effectively.Therefore, the research of feature extraction is getting more and more attention and becomesthe key to fault diagnosis.The time-frequency analysis is the main analysis method for unstable vibration signals ofrolling bearing. The wavelet analysis has compact support characteristic and implements thedecomposition and reconstruction for signal in time-frequency domain. Furthermore,harmonic wavelet analysis has good box spectrum characteristic and specific expression intime and frequency domain. But wavelet analysis and harmonic wavelet can’t be divided atany frequency with good compact support base and box spectrum features. Harmonic waveletpacket can divide signal unlimitedly at any frequency. The harmonic wavelet packet featureextraction technology and SVM are combined to extract fault features to verify the validity ofharmonic wavelet packet decomposition method compared with other methods in laboratoryenvironment. A large-scale fault features with complexity need reduction, the kernel principalcomponent analysis (KPCA) is obtained by the kernel function to solve the problem of linearinseparable data. Extract feature with harmonic wavelet packet firstly and then implementfeature extraction with KPCA secondary in this paper. The feature extraction method withKPCA is illustrated with the vibration data of rolling bearing in the case western universityelectrical engineering laboratory. This paper summarizes and analyses several kinds of featureevaluation methods and points out the applicable object of the method and applicablecharacteristics, which provide the convenience for the study of the feature evaluation.