Prediction of Aerodynamic Characteristics Using Neural Networks

This study presents a systematic neural network approach for the prediction of aerodynamic characteristics from the wind tunnel test data. The research is based on the Back propagation neural network method/radial basis function neural network method, which uses information about alpha, frictional drag coefficients and Mach number. A simple Backpropagation network of two input nodes (for the graph parameters), three hidden layers (18, 28, 10 neurons) and one output node was developed and compared with a radial basis function for the predicting power. For a training set of 136 data points and a training set with Mach number ranging from 0.6-3, the radial basis function neural network consistently out-performed the Backpropagation network regression model in time effectiveness. The results from the Backpropagation network and the radial basis function neural network are compared with the graphs taken from the database.