Neural network development foundations for low airspeed prediction

The object of this work was to establish the framework necessary for the future development of a neural network based system capable of accurately determining airspeed (magnitude and direction) for speeds under 40 knots which comprise what is known as the low airspeed regime. Ordinary airspeed measuring devices are unable to function properly within this range while advanced mechanical systems are so expensive that only the most sophisticated aircraft can afford them. Neural network predictions comparable in accuracy to these systems were successfully achieved. A self-organizing map was utilized to insure the selection of a compact, yet representative training set, while several paradigms were then applied to perform the predictions. From these tests, a radial basis function emerged as the best predictor. In addition, a study of network response to signal frequency content was performed which uncovered ample potential for further accuracy improvements