Airfoil GAN: Encoding and Synthesizing Airfoils for Aerodynamic Shape Optimization

Yuyang Wang,Kenji Shimada,Amir Barati Farimani

DOI: https://doi.org/10.1093/jcde/qwad046

2023-07-06

Abstract:The current design of aerodynamic shapes, like airfoils, involves computationally intensive simulations to explore the possible design space. Usually, such design relies on the prior definition of design parameters and places restrictions on synthesizing novel shapes. In this work, we propose a data-driven shape encoding and generating method, which automatically learns representations from existing airfoils and uses the learned representations to generate new airfoils. The representations are then used in the optimization of synthesized airfoil shapes based on their aerodynamic performance. Our model is built upon VAEGAN, a neural network that combines Variational Autoencoder with Generative Adversarial Network and is trained by the gradient-based technique. Our model can (1) encode the existing airfoil into a latent vector and reconstruct the airfoil from that, (2) generate novel airfoils by randomly sampling the latent vectors and mapping the vectors to the airfoil coordinate domain, and (3) synthesize airfoils with desired aerodynamic properties by optimizing learned features via a genetic algorithm. Our experiments show that the learned features encode shape information thoroughly and comprehensively without predefined design parameters. By interpolating/extrapolating feature vectors or sampling from Gaussian noises, the model can automatically synthesize novel airfoil shapes, some of which possess competitive or even better aerodynamic properties comparing to airfoils used for model training purposes. By optimizing shapes on the learned latent domain via a genetic algorithm, synthesized airfoils can evolve to target aerodynamic properties. This demonstrates an efficient learning-based airfoil design framework, which encodes and optimizes the airfoil on the latent domain and synthesizes promising airfoil candidates for required aerodynamic performance.

Computational Engineering, Finance, and Science,Machine Learning

What problem does this paper attempt to address?

The paper attempts to address the issue of computationally intensive simulations in the current aerodynamic shape design process (such as airfoils) and the limitations of new shape synthesis relying on predefined design parameters. Specifically, traditional aerodynamic shape design methods require a large amount of computational resources to explore the possible design space and often rely on manually selected design parameters, which limits the ability to generate new shapes.

To overcome these challenges, the paper proposes a data-driven shape encoding and generation method based on deep learning. This method leverages existing airfoil data to automatically learn representations and uses these learned representations to generate new airfoils. The specific objectives include:

1. **Learning expressive representations**: Learning low-dimensional feature representations from existing airfoils through a Variational Autoencoder-Generative Adversarial Network (VAEGAN) model.
2. **Generating realistic designs**: Using the Generative Adversarial Network (GAN) part to generate high-quality new airfoils.
3. **Optimizing generated designs**: Optimizing the generated airfoils in the learned feature domain using a genetic algorithm to achieve the desired aerodynamic performance (such as lift and drag coefficients).

Through these objectives, the paper aims to provide an efficient learning framework that can encode and optimize airfoils in the latent domain and generate new airfoils that are competitive or even superior to existing ones. This will help in designing a new generation of airfoils more intelligently and efficiently, reducing the reliance on experimental experience and manually selected design parameters.