Optimisation of bearing diagnostic techniques using simulated and actual bearing fault signals

In this paper, bearing fault vibrations are modelled as a series of impulse responses of a single-degree-of-freedom system. The model incorporates slight random variations in the time between pulses so as to resemble actual vibration signals. Although the bearing fault harmonics in the raw spectrum are caused by the random fluctuations to smear over one another, they remain quite clear in the spectrum of the envelope. However, the envelope spectrum is still prone to masking by discrete and random noise. Therefore, the simulated bearing fault signals were used to investigate the efficient application of self-adaptive noise cancellation (SANC) in conjunction with envelope analysis in order to remove discrete frequency masking signals. Two ways of combining these techniques have been suggested, both of which require the original signal lo be band-pass filtered and frequency-shifted in order to reduce the number of samples to be processed by SANG. The subsequent envelope analysis can then be performed by using the Hilbert transform technique or band-pass rectification. Band-pass rectification is simpler but requires extra zero padding above and below the demodulation band, making the length of the signal processed by SANG twice as long as with the former method, but still only a fraction of the length of the original signal. On the other hand, the Hilbert technique requires an extra forward and inverse discrete Fourier transform operation compared with band-pass rectification. These two methods reduce the masking effects in the envelope spectrum by removing pseudo-sum frequencies or placing them outside the frequency range of interest. This is illustrated with examples of simulated and actual vibration signals. The removal of discrete frequency noise using SANG is also demonstrated for actual vibration signals. The threshold for which analysing the squared envelope or its higher powers gives an improvement in the envelope spectrum has also been defined using simulated and actual vibration signals. The treatment in the paper is qualitative and non-mathematical for purposes of clarity, but reference is made to a quantitative treatment of the effects of masking. (C) 2000 Academic Press.