A method of fault diagnosis for flight control system based on unknown input observer and one-dimensional convolutional neural network

The flight control system is crucial for ensuring the safety and reliability of aircraft. This paper proposes a novel hybrid fault diagnosis framework that integrates an unknown input observer with a one-dimensional convolutional neural network. The unknown input observer is designed to decouple system states from unknown disturbances and generate residuals that are sensitive to faults while remaining robust to disturbances. The residual sequences are then fed into a one-dimensional convolutional neural network, which automatically extracts discriminative fault features and performs accurate fault classification. The performance of the proposed method is validated through simulations on a linearized longitudinal F-16 aircraft model under normal and fault scenarios. Experimental results demonstrate that the proposed unknown input observer-one-dimensional convolutional neural network method outperforms one-dimensional convolutional neural network, Kalman filter-one-dimensional convolutional neural network, unknown input observer-long short-term memory, and unknown input observer-transformer. The unknown input observer-one-dimensional convolutional neural network model achieves a diagnostic accuracy of 99.69%, highlighting the superiority of the proposed hybrid framework.