Nonlinear flight control system with neural networks

Abstract

In this study an adaptive critic based neural network controller is developed to obtain near optimal control laws for a nonlinear automatic flight control system. The adaptive critic approach consists of two neural networks. The first network, called the critic, captures the mapping between the states of a dynamical system and the co-states that arise in an optimal control problem. The second network, called the action network, maps the states of a system to the control. This study uses nonlinear aircraft models in the stall regions from a paper (Garrad and Jordan2 to develop optimal neural controllers for an aircraft; we then compare the results with singular perturbation based nonlinear controllers developed in the literature. The results show that with the neural controllers the aircraft can operate in a broader region of angles of attack beyond stall as compared to other linear and nonlinear controllers.