physics-informed CNN for HSL-TFP

This is the implementation of ‘’Physics-informed convolutional neural networks for temperature field prediction of heat source layout without labeled data“

• the Pipeline of our method.

• The visualization of the steady-state temperature field obtained by finite difference method (FDM) and the proposed method.

The case with simple heat source layout.

• The case with complex heat source layout.

Usage

Environment

Note that you need to choose the right version of pytorch-lightning.

torch=1.12.1+cu113
torchvision=0.13.1+cu113
pytorch-lightning=1.1.0
tensorboard
opencv-python

Data Preparation

• the datasets used in this paper are uploaded to here. Please download the dataset to your local hard drive, and modify the data address in the configuration file config_ul.yml, consisting of the data_root, train_list, val_list, and test_list.

data_root: /mnt/jfs/zhaoxiaoyu/data/ul/complex_component/FDM
train_list: /mnt/jfs/zhaoxiaoyu/data/ul/train.txt
val_list: /mnt/jfs/zhaoxiaoyu/data/ul/val.txt
test_list: /mnt/jfs/zhaoxiaoyu/data/ul/test.txt

Train and Test

• Train the model without labeled data.

cd example
python train_ul.py

• test the trained model. Please modify the position of trained model in the test.py, and

python test.py

Citation

If you find our codes or models useful, please consider to give us a star or cite with:

@article{zhao2023physics,
  title={Physics-informed convolutional neural networks for temperature field prediction of heat source layout without labeled data},
  author={Zhao, Xiaoyu and Gong, Zhiqiang and Zhang, Yunyang and Yao, Wen and Chen, Xiaoqian},
  journal={Engineering Applications of Artificial Intelligence},
  volume={117},
  pages={105516},
  year={2023},
  publisher={Elsevier}
}