Neural network dynamic inversion with application to reentry process of a hypersonic vehicle

This paper studied an intelligent adaptive flight control method. The classic dynamic inversion control provides automatic adaptation at the flight point, which is particularly suited to aerospace vehicles (aircraft, pitchers or entry vehicles). However, the inversion process is sensitive to modeling errors. A possible improvement method is to compensate these errors. In this paper, neural networks have been applied to solve this problem. A reentry hypersonic vehicle has been taken as an example for application. The kinematic equations of this system found an unstable, multivariable, and nonlinear model which contains several uncertain parameters. The main idea is to firstly divide the variables into two groups according to their rates of change, and build two close loops of dynamic inversion separately for each group; then a compensation controller is designed using neural networks. Finally the simulation demonstrates the effectiveness of this technique.