gtc1030 / cebgan_jmd_2021

CEBGAN and airfoil optimization code associated with our accepted JMD 2021 paper: "Inverse Design of 2D Airfoils using Conditional Generative Models and Surrogate Log-Likelihoods."

CBGAN and CEBGAN for 2D Airfoil Inverse Design

CEBGAN and airfoil optimization code associated with our accepted JMD 2021 paper: "Inverse Design of 2D Airfoils using Conditional Generative Models and Surrogate Log-Likelihoods."

License

This code is licensed under the MIT license. Feel free to use all or portions for your research or related projects so long as you provide the following citation information:

Required packages

Dataset

Our airfoil designs are optimized airfoils in SU2. The original airfoils are generated by BezierGAN. BezierGAN was trained on UIUC airfoil coordinates database.

CFD solver and airfoil optimization

We use SU2 as the CFD solver to evaluate the performance of the airfoil design.

Usage

Train/evaluate Conditional-Bézier-GAN

Results

Airfoil samples of training data:

Quantitative performance of conditional GANs:

Reduction in instantaneous optimality gap:

Reduction in cumulative optimality gap:

Reduction in cumulative optimality gap of an example airfoil:

Practicability of the Surrogate Log-Likelihood