Unified Strategy for UAV Blade Fault Diagnosis Using Khan Observer and CNN Network

This paper presents a fault diagnosis algorithm for Unmanned Aerial Vehicle (UAV) systems characterized by nonlinear terms and disturbances. The proposed method integrates the Khan intermediate observer and a Convolutional Neural Network (CNN), enabling the estimation of both the UAV blade state and the simultaneous estimation of sensor faults and disturbances. To induce fault conditions, two plastic strips were strategically applied to the upper surface of the blade, significantly altering its aerodynamics and thereby expanding the scope of the original system. The algorithm employs an intermediate variable to design an observer for the augmented system, providing a comprehensive view for the tester. This allows for the estimation of system states, faults, and perturbations. The Convolutional Neural Network serves as the residual generator, generating virtual residual signals. Furthermore, a proportional-integral action is introduced to estimate UAV blade faults. Simulation results on a quadrotor UAV system validate the efficacy of the proposed method in accurately estimating blade faults and disturbances.