Abstract: Filtering noise component of mechanical vibration signal effectively can observe the characteristics of vibration signal more clearly. So based on the wavelet threshold de-noising method, a new improved threshold function was proposed. The coefficient of wavelet transform mechanical vibration signals were estimated by the threshold selection method based on kurtosis value. Not only new threshold function conforms to the distribution characteristics of the vibration signal, but also new threshold function is continuous in the threshold point. And the new threshold function overcomes the inherent deviation which traditional threshold function brings. The research of noise reduction on the simulation signal and measured mechanical vibration signal shows that wavelet de-noising method based on the improved threshold function protects the useful information of vibration signal well, and provides the method is a kind of effective de-noising method in vibration signal.

Key words: wavelet denoising, threshold function, mechanical vibration, fault diagnosis