Flight control system design using neural networks and genetic algorithms

The author presents a design methodology to develop a flight control system using neural networks and genetic algorithms. The flight simulation model considered has various types of uncertain parameters, such as aerodynamics, actuator dynamics, sensor dynamics, environmental conditions, and time delay, the stochastic properties of which are defined a priori. The flight control system requirements are defined based on the flight simulation results, and neural networks with a preselected architecture are used to develop flight control laws. Training of neural networks is accomplished through genetic algorithms. In order to determine the performance of the trained neural networks, a Monte Carlo method is applied to the results of a large number of simulated flights to estimate the probability of satisfying the requirements, or the probability of total mission achievement. Simulation results for a benchmark problem of pitch attitude control of a flight vehicle are presented to illustrate the performance of the proposed methodology. The results of the proposed neural networks model are compared with the proportional-integral-derivative model.