Physics-Constrained Bayesian Neural Network

Physics-Constrained Bayesian Neural Network for Fluid Flow Reconstruction with Sparse and Noisy Data

Luning Sun, Jian-Xun Wang

PyTorch implementation of Physics-Constrained Bayesian Neural Network.

Noisy Stenotic Flow

CFD	Mean	Standard Deviation

Noisy Junction Flow

CFD	Mean	Standard Deviation

Dependencies

• python 3
• PyTorch 0.4 and above

Installation

• Install PyTorch, TensorFlow and other dependencies

• Clone this repo:

git clone https://github.com/Jianxun-Wang/Physics-constrained-Bayesian-deep-learning.git
cd Physics-constrained-Bayesian-deep-learning

Training

Noisy Stenotic Flow

Train a parametric DNN surrogate for pipe flow

cd code
python mainsolve.py

Noisy Junction Flow

To be added

Citation

If you find this repo useful for your research, please consider to cite:

@article{sun2020physics,
  title={Physics-constrained Bayesian neural network for fluid flow reconstruction with sparse and noisy data},
  author={Sun, Luning and Wang, Jian-Xun},
  journal={arXiv preprint arXiv:2001.05542},
  year={2020}
}

We also have a relational research of using PINN to build surrogate modeling without using labelled data

Acknowledgments

Thanks for all the co-authors and Dr. Yinhao Zhu for his valuable discussion.

Code is inspired by cnn-surrogate