Deep Residual Shrinkage Network Based Fault Diagnosis for Quadrotor UAV Formation System

The multi-type and multi-intensity noise of UAV formation system in actual flight could affect the performance of fault diagnosis. However, the traditional noise reduction method cannot adaptively change the threshold to cope with the influence of different noises, which leads to the problem of high misdiagnosis and missed diagnosis rate. Therefore, this paper proposes a fault diagnosis method based on deep residual shrinkage network for fault diagnosis of quadrotor UAV formation system. This method combines adaptive slope and soft thresholding, and introduces attention mechanism to automatically infer adaptive slope and realize automatic soft thresholding, so as to set different thresholds for different samples. The XGBoost classifier is used to replace the traditional softmax classifier to classify the extracted features to achieve fault diagnosis. Finally, the above method is verified by experiments on the built simulation platform. The results show that the accuracy of fault diagnosis is more than 90 % when the signal-to-noise ratio of multiple noises is greater than 0 dB, and the average diagnostic accuracy is improved by 11.94 % compared with the existing fault diagnosis model in the case of large noise intensity.