Neural network based control system design of an advanced fighter aircraft

Abstract

The author demonstrates the application of neural network technology to optimal scheduling of control gains in real-time flight control systems commonly used in the real-time control of an advanced fighter aircraft and high-performance aerospace vehicles where a priori target outputs are not known, and must be generated in real-time. A learning algorithm and an appropriate performance model have been used to synthesize a nonlinear functional relationship between varying plant parameters and control gains. A performance model is used to exemplify the desired responses and force the plant/controller dynamics via a neural network to imitate the model. The performance model contains the proper dynamics to supply desired responses to given test inputs. An arbitrary cost function is used to indicate the quality of plant/controller performance according to which the adjustments to the weights within the neural network are made by the learning algorithm. The process is repeated until the neural network produces an optimal set of gains for each point in the plant parameter space.<>