GANTL

This is the code and data for GANTL: Toward Practical and Real-Time Topology Optimization With Conditional Generative Adversarial Networks and Transfer Learning paper published in the Journal of Mechanical Design.

Code Usage

Dependencies

• Tensorflow == 2.3.0
• H5Py == 2.10.0
• Matplotlib == 3.5.1
• Numpy == 1.21.6

Commands

• --res : The resolution of the domain.
  • Options: 4080 (default), 80160 , 120160, 120240, 160320, 200400
• --traintest:
  • Options:
    • traintest which train the model and test that
      • Note: You have to train on 4080 before you try to train on other resolutions.
    • test which test the saved model
• --batch_size: The batch size for the model training (default = 16)
• --n_epoch: The number of epochs for training (default = 500)
• --save_step : Save models every x epoch (default = 50)
• --lr : Learning rate (default = 2e-4)
• --TOloss : Whether to test the TO model or not (default = False)
  • Note 1: it should be used only on the test mode
  • Note 2: the resolution must be 200400

Data and Models

The data can be downloaded from this link. You have to extract that to the data folder ( it would be like data/'res') . To download the pretrained models, you can use this link. You have to extract the models under GANTL folder (it would be like GANTL/'res').

Using Example

To train a model you can use the following command:

python main.py --res 4080 --traintest traintest

The code will train the model and save the models in GANTL/'res'/model folder. It also save the images obtained during the training in GANTL/'res'/images. The final predictions will be saved in GANTL/'res'/prediction.

To test the model, one can run the following command:

python main.py --res 4080 --traintest test

The result will be saved in GANTL/'res'/prediction.

Note: You can run these codes for any resolutions.

To test the model on TO loss, you have to run the following command:

python main.py --res 200400 --traintest test --TOloss True

Citation

@article{behzadi2022gantl,
  title={GANTL: Toward Practical and Real-Time Topology Optimization With Conditional Generative Adversarial Networks and Transfer Learning},
  author={Behzadi, Mohammad Mahdi and Ilie{\c{s}}, Horea T},
  journal={Journal of Mechanical Design},
  volume={144},
  number={2},
  year={2022},
  publisher={American Society of Mechanical Engineers Digital Collection}
}