Abstract：Aiming at the traditional mine hoist rocker arm bearing fault diagnosis technology which usually

only utilizes a certain specific geometric structure of the vibration dataset and fails to obtain the optimal similarity

graph structure，an intelligent diagnosis method for bearing faults based on adaptive graph embedding was

proposed. Firstly，the local and global geometric structure of the bearing data were mined using Laplacian

eigenmaps and sparse self-representation. Then the two structural information were fused by coefficient fusion，and

an ι2,1-regularization constraint with row sparsity was introduced to select the essential features that can accurately

characterize the operating state of the bearing. Finally，the selected intrinsic features were input into a support

vector machine for training and testing to construct an intelligent diagnostic model for bearing faults. To verify the

validity of the method，bearing monitoring data were obtained by simulating the working conditions of the rocker

arm drive system of the underground mine hoist using the bearing data from Case Western Reserve University and

the QPZZ-Ⅱ drive system platform provided by the R&D department. The simulation experiment results show that

the proposed intelligent diagnostic model can obtain the optimal fault diagnosis accuracy，and the average accuracy can reach 97.5%.

      Key words：rocker bearings；fault diagnosis；feature selection；Laplacian eigenmaps（LE）；sparse selfrepresentation；support vector machine（SVM）