It is difficult to dectect the gear fault signal in early stage because of weak intensity and strong interference.To solve this problem

a method for incipient fault diagnosis of gears is proposed based on vibration signals using kurtosis

wavelet packet energy features extraction and discriminative weighted probabilistic neural networks.The method uses the advantages of the kurtosis statistics on the impact load feature extraction method in feature extraction and reserves the merit of wavelet packet decomposition in extracting energy characteristics of various frequency bands.Meanwhile

the discriminative weight probabilistic neural network（DWPNN） is introduced to solve the problem of the scene noise pollution.The experimental results show that the method achieves a good identification of incipient faults of gears and has strong robustness against noise disturbance.