Deep Attention Network (DAN)

---

The repository mainly records how to use VIT to code the airfoil geometry and how to use the encoded information to reconstruct the flow field of the airfoil. 🖐

More details about this work can be found from our paper:

Fast aerodynamics prediction of laminar airfoils based on deep attention network

---

1、Airfoil Visual Transformer (VIT) geometry encoding

---

1、timm_ V2.py is implemented using the timm visual algorithm function library. The traditional implementation method can refer to VIT_author.py file.

pytorch-image-models references：

author_home_page：https://github.com/rwightman

pytorch-image-models open source code：https://github.com/rwightman/pytorch-image-models

Zhi hu：https://zhuanlan.zhihu.com/p/350837279

2、Visual Transformer Code Reference for Attention Visualization：https://github.com/zuokuijun/Transformer-Explainability

3、 VIT_ Airfoil_ Encoder is a Pycharm engineering file that uses Transformer to encode geometric parameters for UIUC airfoil database

How to use ?👉👉👉👉👉

• cd VIT_Airfoil_Encoder

• run python plot_airfoil.py generate airfoil images.

• run python get_gray_images.py generate airfoil gray images

• run python get_airfoil_map.py generate airfoil three channel airfoil heat-map images

• run python vit_explain.py get airfoil geometry encoding information

e1412 airfoil attention visualization

naca4412 airfoil attention visualization

---

2、 Airfoil flow field prediction

• cd VIT_flow_field_prediction

• run train.py file to train DAN

• run mlp_test.py to get DAN prediction results

  Tips: The test model and test data can be found in [vitAirfoilEncoder](vitAirfoilEncoder)

If you feel that our work is helpful to you, please cite our work in your article

@article{zuo2023fast,
  title={Fast aerodynamics prediction of laminar airfoils based on deep attention network},
  author={Zuo, Kuijun and Ye, Zhengyin and Zhang, Weiwei and Yuan, Xianxu and Zhu, Linyang},
  journal={Physics of Fluids},
  volume={35},
  number={3},
  year={2023},
  publisher={AIP Publishing}
}