Fault diagnosis of CNN bearing cage based on EMD-SDP feature fusion

ZHENG Yi-zhen1, NIU Lin-kai1,2, XIONG Xiao-yan1,2, QI Hong-wei1, XIE Hong-hao1
(1.School of Mechanical and Transportation Engineering, Taiyuan University of Technology, Taiyuan 030024, China;
2.Key Laboratory of New Sensors and Intelligent Control of Ministry of Education, Taiyuan University of
Technology, Taiyuan 030024, China)

Abstract: Aiming at the problems of instability, non-impact characteristics and failure characteristics of rolling bearing cage fault vibration signals a CNN fault diagnosis method based on EMD and SDP feature information fusion was proposed. The characteristic information of the EMD inherent modal component of the fault vibration signal was fused by SDP method, and the time-frequency characteristics of the different cage fault vibration signals were demonstrated. The SDP image difference characteristics of the rolling bearing cage under different fault conditions was studied, and the CNN model was used for SDP image recognition to realize the fault diagnosis of the bearing cage, a CNN of fault diagnosis method based on the fusion of EMD and SDP feature information was designed. Finally，the simulated fault experiment was conducted through the fault test bed. The results indicate that the method can achieve a fault recognition rate of more than 99%, and further verifies that the method of adaptively extracting SDP information fusion image features through deep learning algorithms can be effectively applied to the fault diagnosis task of bearing cages.

Key words: cage fault diagnosis; empirical modal decomposition(EMD); symmetry dot pattern (SDP); convolutional neural network(CNN); feature fusion