Cyclic statistics in rolling bearing diagnosis

Mechanical signals with amplitude modulated are characterized by periodic time-varying ensemble statistics and can be considered as cyclostationary. In this paper, the second order cyclic statistics, i.e. cyclic autocorrelation and cyclic spectrum, are introduced. A method of demodulation based on cyclic autocorrelation is derived from a signal model. The modulators and carrier are exhibited, respectively, in low- and high-frequency band of cyclic frequency domain. The three-dimensional spectral correlation figure, which represents cyclic frequency, frequency and spectral correlation strength simultaneously, is developed to express the demodulation results clearly. The method is tested by simulation signal and applied to diagnose rolling bearing faults. It obtained more information than other conventional methods, such as the frequency domain and the envelop detection. Furthermore, its effect is demonstrated by comparing with the wavelet envelope demodulation.