Neural Network Based Aerodynamic Flight Simulation

The objective of this effort is to develop a neural network-based aerodynamic prediction model of an existing weapon systems and integrate this model into a PC-based six degree-of-freedom (6DOF) simulation of the weapon. Two extensive sets of wind tunnel data have been used to demonstrate the potential of using Artificial Neural Network programs as limited aerodynamic prediction codes. The algorithm used in this preliminary work is a feed-forward, back propagation network with special techniques implemented to aid the convergence to minimum error. The wind tunnel data sets used in the present training process contained several orientation/environmental variables, several geometric variables and, of course, the normal aerodynamic coefficients as outputs. The assessment of the accuracy of the neural network algorithm shows the predicted coefficient values to be in very good agreement with actual wind tunnel data with an average mean square error for each coefficient to be less than one part in one thousand for most of the data. Comparison with Missile Datcom for some of the configurations clearly demonstrate the potential of developing a generic aerodynamic prediction code suitable for use in a flight simulation.