Adaptive Dynamic Surface Control for a Hypersonic Aircraft Using Neural Networks

A hypersonic aircraft dynamic model is highly nonlinear because its flight conditions are usually determined at high altitude and Mach number. Therefore, there always exist differences between the dynamical model and real system, and uncertainties during the flight, thus, leading significant degradation of control performance. To solve the performance degradation problem, this paper proposes neural networks-based adaptive velocity and altitude tracking controllers. In order for that, the hypersonic aircraft model is transformed into an uncertain feedback system, which has both matched and unmatched uncertainties, by differentiating the velocity and altitude with respect to time. Then, the overall tracking control system is designed systematically by introducing virtual control inputs and dynamic surface control. During the design process, an inverse of an input gain matrix is directly trained and adapted to remove the matched uncertainty and controller singularity problem simultaneously. In addition, several adaptive elements with saturation functions are added to handle all the matched and unmatched uncertainties. The proposed controller guarantees the uniformly ultimate boundedness of the tracking error by utilizing deadzoned errors. Finally, numerical simulations with the uncertain hypersonic aircraft are performed to demonstrate the effectiveness of the proposed approach.