Airfoil Optimization Using Deep Learning Models and Evolutionary Algorithms for the Case Large-Endurance UAVs Design

Abstract

This article presents the development of the AZTLI-NN network and the evaluation of this network as a set of evolutionary algorithms in airfoil optimization tasks. AZTLI-NN has the characteristic of predicting the aerodynamic coefficients of the airfoils in the form of images (graphs of the aerodynamic coefficients as a function of the angle of attack) from parameter vectors corresponding to the parameterization method CST. This feature allows the network to achieve good performance when generalizing the predictions of the aerodynamic coefficients, being on par with neural networks that have the aerodynamic coefficients encoded in the form of structured data, and has the ability to handle a wide range of usage airfoils in general aviation. In addition, a case of how AZTLI-NN together with an adaptive evolutionary algorithm and population size reduction methods achieve good performance in finding the airfoil that provides the highest possible endurance value is shown, so this work is considered as an option in the early stages of the design for the selection of airfoils in the design of large-endurance UAVs.

Keywords

• airfoil optimization
• OpenFOAM
• neural networks
• evolutionary algorithms
• method CST
• large-endurance UAV