Overview 

A multi-agent deep reinforcement learning (DRL)-based model is presented in this study to reconstruct flow fields from noisy data. A combination of the reinforcement learning with pixel-wise rewards (PixelRL) method and physical constraints represented by the momentum equation and pressure Poisson equation as well as the boundary conditions is utilised to build a physics-guided deep reinforcement learning (PCDRL) model that can be trained without the target training data. In the PCDRL model, each agent corresponds to a point in the flow fields, and it learns the optimum strategy of choosing pre-defined actions. 

Dependencies 

Python: 3.6-3.8 

Pytorch: 1.4.0 (cuDNN=7.6, CUDA=10.1 for torch) 

Numpy: <1.19 

Data preparation 

1.Implement Add_noise.py to obtain the noisy DNS data with three levels of noise (1/SNR=0.01, 0.1, 1.0) using the dns-bluff_uv.npy data. 

2.According to a certain length of time, pack the noisy DNS data to get random-noisy data by using Random.py. 

Training 

Use the random-noisy data in various levels of noise and PCDRL_Denoising_bluffbody_train.py to train PCDRL model. 

Prediction 

After finishing the training of PCDRL model, use PCDRL_Denoising_bluffbody_predict.py to generate the reconstruction of the random-noisy data in various levels of noise. 

BibTex citation:

@article{yousif_zhang_yu_yang_zhou_lim_2023,

title={Physics-constrained deep reinforcement learning for flow field denoising},

volume={973}, DOI={10.1017/jfm.2023.775},

journal={Journal of Fluid Mechanics},

publisher={Cambridge University Press},

author={Yousif, Mustafa Z. and Zhang, Meng and Yu, Linqi and Yang, Yifan and Zhou, Haifeng and Lim, HeeChang},

year={2023},

pages={A12}

}