For the blindness and one-sidedness of selection and fusion of mechanical fault features without priori knowledge

a novel method of gear fault feature extraction and classification based on feature evaluation and kernel principal component analysis is presented

where the original signals are decomposed with wavelet pocket decomposition（WPD）

and the features in time domain are extracted from the original signals and each decomposed signal to compose the combined features.Furthermore

the threshold value for stability and the filtering scale factor for sensitivity are confirmed to evaluate the features by the combined method with stability and sensitivity

and the nonlinear features are extracted from the residual features by using the method of kernel principal component analysis（KPCA）to realize the classification of different fault conditions.The experimental results of gearbox demonstrate that the method integrating WPD

combined feature evaluation method and KPCA

could better extract the feature information of gear fault

remove the unstable and insensitive ones from a large number of features

and obviously improve the result of nonlinear feature extraction of gear fault for KPCA.