Neural-Network-based Control Method Research To Inhibit Uncertainties Of Hypersonic Aircraft

The dynamic model of hypersonic aircraft shows highly nonlinearity, strong coupling, elastic deformation and various uncertainty perturbations, especially in the maneuver at high angle of attack and high angle of sideslip, which will enhance hypersonic vehicle’s coupling and uncertainty. This increases the difficulty of designing the controller. Previous researches always suppose hypersonic vehicle as a rigid body and ignore the complex coupling between elastic mode and rigid-body mode, which will lead to the loss of accuracy of the model and the degradation of performance of controller. Designing the controller for hypersonic vehicle under the complex coupling is a difficult problem.This paper aims at hypersonic vehicle with elastic mode and various uncertainties, focuses on three areas:stability analysis with complex coupling, nonlinear decoupling control and uncertainty inhibition. A controller with high robustness is proposed to manage the flight at high angle of attack and high angle of sideslip. The main context and innovation of this paper are shown in below:(1) Based on lots of papers and NASA technical reports, a six degree-of-freedom full states Winged-Cone model of hypersonic aircraft is built, including kinetic and motion equations, aerodynamic force and moment model, scramjet engine model and elastic mode. This model can describe the features of hypersonic vehicle more precise and comprehensive, include nonlinearity, coupling and elastic deformation.(2) Based on the complex coupling model of hypersonic aircraft, a new closed-loop stability criterion is derived. It is more accurate and widely-used than classic stability criteria. It can offer ideas to solve the coupled stability problem of hypersonic vehicle.(3) Applay Trajectory Linearization Control theory to linear and decouple the model of hypersonic vehicle. Furthermore, introduce control allocation algorithm to decouple the control laws from control surfaces. This controller can perliminary solve the stability problems caused by aerodynamic coupling, kinematic coupling, inertia coupling and control coupling, etc.(4) utilizing the Radial Basis Function Neural Network, a controller is designed to suppress a variety of uncertainty perturbations. Genetic algorithm is used to optimize the center of radial basis functions and spread. Then combine RBFNN with TLC controller to improve its precision and real-time capability.Lots of simulations and contrast experiments with fuzzy control law demonstrate that the proposed control law has a good performance and high robustness under various uncertainty disturbances.