arthurgonze / tnsfnets

Tensor Navier-Stokes Flow nets - Physics Informed Neural Networks

TNSFnets

Tensor Navier-Stokes Flow nets - Physics Informed Neural Networks

Some of the codes need this experiments to run: https://drive.google.com/drive/folders/1Kl9U2Q1BvAQAaP5W3w6HpkXs79gT7TP6?usp=sharing

############################################################################################################################################################################

Folder structures:

############################################################################################################################################################################

TNSFnets -> cube

->-> data

->->-> cube_00

->->->-> files .npy to train and test the model

->->->-> folder of the experiments slices, the exact ones ps.: to verify the model: "y_equal_15_exact", and the predict ones: "y_equal_15_pred"

->->-> cube_01

->->->-> files .npy to train and test the model

->->->-> folder of the experiments slices, the exact ones ps.: to verify the model: "y_equal_15_exact", and the predict ones: "y_equal_15_pred"

->-> models

->-> (files like cube_data, cube_plotting, cube_test, etc)

-> figures

->-> experiments folders ("beltrami_3d", "cube_00", "cube_01", etc)

->->-> folders of the experiments slices, like: "y_equal_15_exact", "z_equal_15_exact", "z_equal_10_exact"

->->-> images of the losses, and other images of the training proccess

-> models

-> sims

->-> here is the folder avaible to download in google drive, like: "cube_00" or "cube_01"

->->-> timesteps

->->->-> files of the exact positions (x, y, z) and velocities (u, v, w)

-> (files like .gitignore, .git, BeltramiFlow..., CylinderWake... etc)

############################################################################################################################################################################

To run the experiments:

############################################################################################################################################################################

1- execute the experiment_data.py file, here you can define wich slice of the experiment you're going to use in training and others parameters

2- training the model (experiment_training.py), here you can define the model name and other training parameters;

3- test the model (experiment_test.py), here the predict files will be generated in the folder you especified/created;

4- plot (experiment_plotting.py), self explanatory too.

ps.: there is the networks python files ("NSFnet_fluidborders_model.py", etc), you can customize then too and specify in the import of the training file which network you're going to use.