Improving gear signals analysis methods for radial basis function neural network automatic recognition

Automatic gear signals recognition suffers from lower performance in noisy and modulating conditions. The cepstrum is sensitive to the changes in the gear signals environment. But the cepstrum does not work well in the automatic recognition of the gear faults. This article shows the possibilities offered by the use of the improved gear signals methods for radial function neural network automatic recognition. First, how to use the cepstrum is discussed in the vibration signals from a test gearbox, and the improved cepstrum is presented. That is, use wavelet package to decompose vibration time signals of gear to reconstruct the wavelet packet necessary coefficients, make a Hilbert transform into the wavelet packet necessary coefficients, and also, apply the cepstrum analysis to the Hilbert transform result of the wavelet packet necessary coefficients. The result confirms that the Wavelet packet-Hilbert-cepstrum does work well in the measured vibration signals from a test gearbox. Second, radial function neural network was applied to identify the gear fault patterns. The results show that the method of the Wavelet packet-Hilbert-Cepstrum -Radial function neural network can not only detect the exiting of the fault in gear, but also effectively identify the fault patterns.