Conditional Generative Adversarial Network Framework for Airfoil Inverse Design
Emre Yılmaz, Brian German
Abstract
This paper describes the application of generative adversarial networks (GANs) to airfoil inverse design. Specifically, this work focuses on creating new airfoil shapes via conditional GANs (CGAN) based on a vector set of conditional data indicating desired performance metrics. The networks within this GAN framework are constructed based on deep convolutional neural networks (ConvNets) because of their success in learning both the inverse and the forward relationships between the airfoil shape and the resulting flow as shown in our previous papers [1, 2]. In this setting, the deep ConvNet structures are intended to avoid the problems arising due to shape parametrization in classical methods and to detect and make use of the patterns in the data at a lower level of abstraction. In order to demonstrate the framework, generator and discriminator networks are trained on a database of airfoil shapes and conditional information based on stall condition and airfoil drag polars. After training, the generator of the CGAN framework can be used to create desired airfoil shapes based on specified stall conditions or drag polar information that are input into the framework as conditional data.