Rolling bearing fault diagnosis based on Self-Weight and t-SNE

In order to solve the problem that the fault signal of rolling bearing is nonlinear and the fault features are various, and it is difficult to classify accurately, a method combining Self-Weight feature selection with t-SNE algorithm is proposed. Firstly, WPT is used to process the original fault signal and extract the features. Then Self-Weight is used to evaluate the sensitivity of each feature to obtain the optimal feature. Then, these optimal features are visualized by t-SNE to obtain low dimensional sensitive features, which are used as the input of affine propagation clustering (AP) to achieve 100% accuracy of fault type identification. The results are verified by the bearing data of the MFS-MG, Compared with t-SNE without feature selection and Self-Weight + PCA, the results show the advantages of the proposed method.