Application of neural network to adaptive control theory for super-augmented aircraft

Abstract

The neural network structures developed in this thesis demonstrate the ability of parallel distributed processing in solving adaptive control problems. Adaptive control theory implies a combination of a control method and a model estimation. The control method investigated is the Lyapunov Model Reference Adaptive Control or MRAC and the model estimation investigated is the linear least square estimator. The neural network theory is introduced with emphasis on the back-propagation algorithm. The implementation of the neural network adaptive control structure is demonstrated on the longitudinal dynamics of the X-29 fighter aircraft. Three configurations are proposed to train the neural network adaptive control structures to provide the appropriate inputs to the unstable X-29 plant so that desired responses could be obtained. These configurations are presented in eight cases, which emulates stable systems like the X-29 closed-loop plant or the optimal and the limited X-29 controllers, and unstable systems like the X-29 plant or its inverse.