Fault Diagnosis Method for Flight Control System with small samples Based on Triplet Network

In view of the characteristics of modern flight control system with few fault data and many fault types, it is difficult to rely on deep learning for accurate fault diagnosis. In this article, a fault diagnosis method of flight control system based on Triplet Network was proposed and applied to the bearing data set of Case Western Reserve University and the Pre-flight Built-in-test(PBIT) fault data of a certain type of aircraft. For the bearing data set, the signals are firstly transferred into the frequency domain by wavelet packet decomposition. Then, build the Triplet Network model, and the network training is carried out in the form of triplet input. The model compare the similarity of characteristics of similar faults and different faults, and optimize the parameters to achieve the optimal classification effect. The PBIT fault data of flight control system is processed and network training is carried out to realize the fault diagnosis function. Aiming at small samples, I divided experimental data sets with different number of training samples to observe the diagnostic effect of the model, and compared with traditional deep learning models respectively. The experiment verified the unique advantages of Triplet Network model in fault diagnosis under small sample conditions, and the accuracy was significantly better than the traditional deep learning method.