USE OF NEURAL NETWORKS IN CONTROL OF HIGH-ALPHA MANEUVERS

A method is presented by which an appropriately constructed artificial neural network can be ''trained'' to Predict the force and moment coefficients of a 70-deg sweep delta wing during a high-angle-of-attack excursion. The angle-of-attack time history is a sinusoidal motion from 0 to 90 deg and returning to 0 deg. Experimental data are used to train the network, and it is demonstrated that the network has indeed learned to model the behavior of the delta wing over a range of frequencies of this type of angle-of-attack time history. The longitudinal equations of motion for a delta wing aircraft are integrated for three sinusoidal angle-of-attack time histories using the predicted network aerodynamic data. This integration generates the longitudinal control deflection time histories required to produce these maneuvers. An exploration is then made as to whether a second artificial neural network can be trained as a neural stick gearing for such maneuvers. This is investigated by attempting to train a network to associate each required control deflection time history with a specified stick position schedule.