Fault Diagnosis of Antenna Actuator Based on Multi-sensor Information Fusion

The Actuator, as the sole adjustment device for the antenna active surface, is the key part to ensure the precision of antenna reflector. Therefore, monitoring the health status of the actuator is crucial. Aiming at the limitations of single sensor diagnosis due to limited data dimensions and scarce engineering data, a fault diagnosis method based on continuous wavelet transform and group normalization parallel convolutional neural network (CWT-GPCNN) is proposed in this paper. Firstly, an actuator fault diagnosis model integrating CWT-GPCNN is established and the group normalization technology is adopted to quicken net convergence and improve diagnosis accuracy. Then, the optimal model is determined by analyzing the impact of network hyperparameters on model performance. Finally, the proposed method is validated using a dataset from an actuator transmission system experiment. The result demonstrates that this model has good generalization capabilities and superiority of multi-sensor fusion. Comparison between multi-sensor fusion and single sensor diagnosis demonstrates the superior performance of the former. The CWT-GPCNN model is compared with three other information fusion models. The CWT-GPCNN model achieves an accuracy rate of up to 93%, which indicates its excellent diagnosis performance.