A preprocessing method for the vibration signal of gear based on MUDW and CK

In allusion to the problem that the gear fault signal appears unsteadily, strong nonlinearity, and feature information is effected by noises, a method for vibration signal preprocessing based on weighted morphological un-decimated wavelet (MUDW) decomposition and correlated kurtosis is proposed. Firstly, the MUDW method is presented under the scheme of morphological un-decimated wavelet decomposition. On this basement, in order to increase feature information content, it combines approximate signals in various layers according to their feature contribution weights, which is measured by correlated kurtosis factor. Secondly, the initial indexes of MUDW are analyzed and a systematic method for optimal selection is presented. Finally, the validity of the method is testified by the data of measured gear fault vibration signal.