--Overview--

We develop a 2D3DGAN to reconstruct the three-dimensional velocity fields from flow data represented by a cross-plane of unpaired two-dimensional velocity observations.

--Dependencies--

1. Python 3.6-3.8

2. tensorflow >=2.2.0<2.4.0 (cuDNN=7.6, CUDA=10.1 for tensorflow-gpu)

3. Numpy < 1.19

--Data preparation--

200 snapshots (channel flow at Reτ = 180) of training data and 100 snapshots of testing data are provided (two different sections have been matched already).

--Training--

Use 2D3DGAN.py to train the deep learning model. It will save architecture and weights files automatically.

--Prediction--

1. After finishing the training process, use predict_test.py (within Predict file) to generate the prediction of the reconstructed 3D data (normalized) based on architecture and weights files gotten from training process.

2. Use the denormalization.py (within Predict file) to get the final reconstructed 3D data.

--Cite this article--

Yousif, M.Z., Yu, L., Hoyas, S. et al. A deep-learning approach for reconstructing 3D turbulent flows from 2D observation data. Sci Rep 13, 2529 (2023). https://doi.org/10.1038/s41598-023-29525-9

--Note--

There is a typo in eq(4) and (5). The sigmoid function should be applied for the whole right side of the equation.