Neural Network Estimation of Low Airspeed for the V-22 Aircraft in Steady Flight

The objective of this study is to utilize a neural network based approach for the estimation of airspeed and sideslip angle in the low airspeed flight regime for a tilt-rotor aircraft. Steady flight data from the flight test of a V-22 aircraft covering a wide range of azimuth angles was used to develop the neural network models. Two neural network paradigms were evaluated, the Back-Propagation and the Radial Basis Function paradigms. A sensitivity study was conducted to understand the significance of the input parameters in the overall performance of the neural network models. Results indicate that the Radial Basis Function paradigm is the better predictor achieving a Root Mean Square error of 3.22 knots for airspeed estimation. Similarly, a Root Mean Square error of 3.54 and 3.59 knots was achieved for the estimation of the longitudinal and lateral components of the velocity respectively. Sideslip angle is derived from the estimated velocity components resulted in an error of less than 15 degrees for 88 percent of the data utilized. While results for steady flight are encouraging, further study is needed to examine such effects as aircraft gross weight, ground effects and rate of climb or descent. This can then lead to the development of a comprehensive network that can provide accurate low airspeed indication over the entire low airspeed flight regime.