Investigation On Fault Features Extraction For Rolling Bearing And Application

Rolling bearing is one of the most widely used components in almost all kinds of machinery. Its working condition has direct effect on product quality and working safety in industry. Thus, it is very important to investigate on bearing condition monitoring and fault diagnosis techniques in order to avoid heavy accident and change the maintenance system.Amplitude modulation mode of fault features of rolling bearing is investigated in this research according to the approximate symmetry of working parts of a bearing and the resonance phenomenon excited by the periodical pulses in the destroyed bearing. Since vibration signal of a destroyed bearing generally exhibits strong periodicity, a cyclostationary based method of feature extraction is studied in detail in this paper. However, as to vibraton signal with low SNR(signal to noise ratio), the demodulation performance of both cyclostationary and traditional envelope methods limits its application when processing the original signal directly. The key problem is to filter out the best frequency band which contains rich fault features for precise demodulation.In order to locate the frequency band which contains rich fault features, this paper take advantage of Spectral Kurtosis(SK). The spectral kurtosis (SK) is a statistical tool which can indicate the presence of periodical pulses and their locations in the frequency domain. It can direct the parameter setting of band-pass filter and helpfully supplements the demodulation analysis of cyclostationary and envelopment. This paper proposes a wavelet packet filtering based method of Sk, and analyze its advantage by linking theoretical concepts with practical applications.Finally the SK is combined with cyclostationary analysis and envelopment analysis of wavelet packet coefficients, which has been applied to simulated signals and bearing fault signals. It can be concluded that this research is effective on the improvement of rolling bearing condition classification. It can also decrease the computational time, improve the practicability of fault diagnosis and promote the development of pattern recognition.