Adaptive Neurocontroller for a Nonlinear Combat Aircraft Model

Abstract

This paper introduces an adaptive con- troller, based on the use of neural networks, for a non- linear 6 D.O.F. combat aircraft model. This controller is based on the determination of the inverse dynamics of the considered aircraft through a state feed-back, taking advantage of the on-line learning ability of a neural net- work, dealing with any changes of the aircraft dynamics during the flight. In particular the controller is based on a Predictor- corrector scheme: namely a network, trained to emulate the direct dynamics of a plant, placed in a cascade with another network, that acts as the actual controller. The latter is trained by back-propagat ing the error of the overall system, through a plant emulator copy (method of supervised learning). The behaviour of this controller is compared with the one of a conventional linear SCAS (normal acceleration limiter), implemented following the requirements of the military handling qualities as well as the requirements of high maneuverability. In dealing with the nonlinear neural controller, two versions of a SISO solution are evaluated and compared. They differ from each other in the performance index formulation: one is a classical PD, while the other is a modified version of the PD which presents adaptive features. The reported numerical simulations show that both of them give satisfactory performances.