Starting from the theory of information fusion

a algorithm of bearing friction fault feature fusion was proposed based on Kmeans singular value decomposition (KSVD) and maximum relevance minimum redundancy (mRMR) principle. First

in order to represent the nonlinear fault information

the algorithm uses KSVD to sparse the signals

and the dictionary atoms corresponding to the sparse coefficients were used as the parameters of the feature fusion. Second

in order to optimize the selection of dictionary atomic set

a criterion based on mutual information mRMR was proposed to determine the number of atoms in the optimal atomic set. Finally

the sparse coefficients were fused by maximizing the principle to extract the valid information for fault condition monitoring. The results of simulation experiment of bearing friction fault show that the proposed method can better integrate the feature information of the redundancy and complementarity. Compared with the single feature and other fusion method

the proposed method can improve fault recognition rate about 12%.