Research on Fault Diagnosis of Electric Actuator Based on VMD and t-SNE

Abstract

As an important part of the control system, the fault diagnosis of actuator is very important for the stable and safe operation of the control process. In order to further the accuracy of fault diagnosis and get a handle on the potential problems arising from the weak features among different faults of electric actuator, a fault diagnosis method based on the combination of VMD and t-SNE manifold learning is proposed. The original vibration signals of the actuator are decomposed into several intrinsic mode components (IMF) by VMD, through which the original signals are grouped into high-dimensional fault features, and then t-SNE is used for secondary extraction of fault features. After that, the low-dimensional sensitive features are obtained and used as the input of the -means classifier to achieve fault type recognition. The proposed method is applicable to the fault diagnosis of actuators and compared with VMD + PCA original feature+ t-SNE method. The results show that the VMD+ t-SNE method considerably furthers the precision of fault diagnosis.