Adaptive flight control using dynamic inversion and neural networks

This paper presents an application of dynamic inversion control with adaptive neural networks to a tailless fighter aircraft. This technique has, in the past, typically been implemented using proportional-derivative desired dynamics, and has recently been extended to a special case of proportional-integral desired dynamics using implicit model-following. This paper presents a method to incorporate adaptive neural networks into an explicit model-following dynamic inversion controller with arbitrary PI desired dynamics. An adaptive controller designed with this method is compared with an adaptive dynamic inversion controller using proportional-derivative desired dynamics and a baseline dynamic inversion controller with PI desired dynamics and no adaptation. The controllers are evaluated by simulation on a six degree-of-freedom tailless fighter simulation during a simultaneous longitudinal and lateral maneuver. The PI dynamic inversion controller with adaptation outperforms the other two controllers and achieves the tightest tracking of commanded maneuvers both during normal flight and with a failed control surface.