The Application of Spectral Kurtosis to Bearing Diagnostics

The choice of demodulation band for envelope analysis of faulty bearings is often made by spectrum comparison with a healthy bearing, to choose resonance frequencies where the largest change occurred as a result of the fault. It has recently been established that the so-called “spectral kurtosis” gives a very similar indication of the band to be demodulated without requiring historical data. The kurtosis is a statistical parameter based on the fourth moment of a signal, which is close to zero for gaussian noise and other stationary signals, but large for impulsive signals containing series of short transients, such as a bearing fault signal. The spectral kurtosis (SK) is obtained by calculating the kurtosis for each frequency line in a timefrequency diagram. It has also been found that the spectral kurtosis can be used to form a filter to select out that part of the signal that is most impulsive, considerably reducing the background noise and improving the diagnostic capability. The initial definition of the SK used the short time Fourier transform (STFT) for the time-frequency analysis, but this does give some artifacts and anomalies in the results, and the paper discusses the potential use of other time/frequency analyses such as Wigner-Ville and wavelets. The paper illustrates the use of the SK for bearing diagnostics, including a number of extensions and improvements in the basic technique and the choice of optimum analysis parameters. The results are illustrated using simulated and actual bearing fault signals.