In view of the limitations of SVM in processing data and classification, a bearing fault diagnosis method based on LMD support vector machine is proposed. The parameter tuning of kernel function directly affects bearing fault diagnosis efficiency. Seven kernel functions are selected for parameter tuning evaluation in this paper.In this paper, the signal is decomposed into a series of PF components by the local decomposition algorithm, and six components are selected to form the eigenvector. Secondly, the experimental data were randomly extracted and combined as a training set and a test set to test the prediction accuracy of seven kernel functions under different penalty parameters. Finally, seven kernel functions are evaluated by Frideman test, and the radial basis kernel function have the best performance.