Neural networks based airfoil generation for a given CpCp using Bezier–PARSEC parameterization

Determining the airfoil geometry from a given CpCp-distribution is an inverse problem of paramount importance specially in the context of variable geometry aerodynamic platforms. This work describes the implementation of artificial neural nets for the airfoil geometry determination. Instead of using full coordinates of the airfoil, Bezier–PARSEC 3434 parameters have been used to describe an airfoil. Some of these parameters have been determined using a Genetic Algorithm. In the second stage CpCp-distribution in terms of clcl, cdcd and cmcm for 10 angles of attack has been input into three different neural nets for learning and then estimating the corresponding BP3434 parameters. Feed-forward backpropagation, Generalized regression and Radial basis neural nets have been trained and then compared in terms of performance and regression statistics. The work establishes the superiority of feed-forward backpropagation neural nets. The result is partly due to good function approximation properties of the neural architecture and partly due to the use of Bezier–PARSEC 3434 parameterization scheme.