Abstract: The impulse generated by bearing damage is interfered by discrete frequencies and much noise. The interference brings troubles to bearing faults detection. A new enhanced detection method of bearing faults is proposed based on combination of pre-whitening technology using cepstrum editing procedure (CEP) and weak signal detection based on stochastic resonance (SR) theory. Signal pre-whitening could be of advantage in enhancing the impulsiveness of the bearing signals, and the eventual result is a white signal, which contains both noise and the components resulting from localized bearing defects. The vibration signal pre-whitened is enveloped without the knowledge of optimal frequency band of bearing. The bearing faults character components could be enhanced, while bring more noise interference. The input of SR model is envelope signal of bearing. By normalized scale transform, bearing characteristic frequencies are enhanced. The local spectrum kurtosis and local signal-to-noise ratio of bearing faults characteristic components are proposed as indicators for comparison. The detection results of normal and outer race fault planted bearing show that the method proposed is better than the envelope analysis based on optimum spectrum kurtosis and signal pre-whitening only.

Key words: Bearing, Fault diagnosis, Signal pre-whitening, Stochastic resonance