Fault Diagnosis for Flight Control System Based on Contrastive Learning

Flight control system is responsible for stabilizing flight attitude, maintaining flight trajectory and improving flight performance. It is an important system of aircraft. This paper proposed a method based on contrastive learning to solve low fault diagnosis accuracy and poor feature-extraction ability of fault diagnosis model due to few quantity of fault data in aircraft flight control system. Inspired by contrastive learning, multi-scale expansion based on white noise is carried out for original data firstly. Secondly, contrastive sample matching based on fault information is constructed by using the expanded data set. Then, this paper constructs a fault diagnosis model based on Siamese neural net taking sample pairs as input and the contrastive constraint between sample pairs is beneficial for improving feature extraction ability of the model. Finally, simulation data of flight control system based on AMEsim/Simulink co-simulation model are used to compare the fault diagnosis method based on contrastive learning with common fault diagnosis methods. The result proves the feasibility and effectiveness of the proposed method.