Neural-network-based aerodynamic modeling and fixed-time neuro-adaptive control of a novel “diamond-wing” morphing aircraft

This paper addresses the aerodynamic modeling and attitude control of a “diamond-wing” morphing aircraft subject to strong nonlinearities, configuration-dependent aerodynamics, and external disturbances. An aerodynamic modeling framework based on a Dendrite Net is developed by leveraging CFD-generated datasets to obtain an explicit polynomial representation through white-box feature extraction and least-squares identification. This formulation is particularly suited to morphing configurations, where continuously varying geometry induces high-order feature couplings and calls for a model that remains consistent and interpretable over the entire morphing envelope. The identified aerodynamics are embedded into the nonlinear longitudinal dynamics with morphing-induced forces and moments explicitly accounted for. On this basis, a fixed-time neuro-adaptive backstepping controller is constructed to handle non-affine input characteristics, model uncertainties, and disturbances. RBFNNs approximate state-dependent unmodeled dynamics, while adaptive update laws estimate disturbance-related terms, thereby reducing the learning burden and facilitating Lyapunov-based fixed-time stability analysis. Input constraints and backstepping complexity are addressed via hyperbolic-tangent saturation and command filtering, respectively. Simulations under multiple operating conditions, including parameter deviations that reflect residual CFD and identification discrepancies, demonstrate faster responses and smaller tracking errors compared with an ESO-DI benchmark, while guaranteeing fixed-time stability.