Neural Inversion of Flight Guidance Dynamics

Abstract

Differential flatness, a property of some dynamic systems which has been recognized only recently, has made possible the development of new tools to control complex nonlinear dynamic systems. Many dynamic non linear systems have been proved to be differentially flat. In this paper, it is shown that the inertial position coordinates of an aircraft can be considered as differential flat outputs for its flight guidance dynamics. Since this differential flatness property is implicit, a neural network is introduced, as a numerical device, to deal with the inversion of the guidance dynamics. Then, once conveniently structured and trained, a neural network is able to generate in real time directives to conventional autopilot systems so that a reference trajectory can be tracked. Numerical results relative to the training of the neural network and to trajectory tracking are displayed and discussed.