MD-CNN-AE

Mode decomposition with autoencoders.

This repo is linked to the paper under review Decoder Decomposition for the Analysis of the Latent Space of Nonlinear Autoencoders With Wind-Tunnel Experimental Data. For lecture material please go to branch lecture-material-may2024.

Content:

1. Models for mode decomposing autoencoder (MD-CNN-AE) (Murata, Fukami & Fukagata, 2020), 
   hierarchical autoencoder (Fukami, Nakamura & fukagata, 2020), 
   and standard autoencoder. All as keras model subclass.
2. Methods for POD and DMD (Brunton and Kutz, 2019).
3. Ranking methods for MD-CNN-AE -- cross entropy, signal energy and contribution.
4. Two data files from the same wake experiment, downsampled to different sizes. Original data from George Rigas.

Python versions:

- python 3.9.7 
- tensorflow 2.7.0 
- numpy 1.20.3 
- matplotlib 3.4.3 
- jax 0.4.9

Setting up:

Create a file _system.ini that contains results saving location. Example config files are available in examples/config_files/

[system_info]
save_location=/home/results_location # required, training results will be saved at this location
alternate_location=/home/experiment_results # optional, this is needed for running repeated experiments with multiprocessing, can be the same as the previous

For running experiments with multiprocessing script

Create a config file using the template examples/config_files/_train_md_ae.ini. Create a file _train_mp.ini and save it in experiments-mp. See examples/config_files/_train_mp.ini. Specify the config file and python file to run as shown in the example. Run start_mp_training.py

References:

Brunton, S. L. & Kutz, J. N. (2019) Data-Driven Science and Engineering: Machine Learning, Dynamical Systems, and Control. 1st edition.
    Cambridge, UK, Cambridge Univeristy Press. Chapter 7.
Murata T.,Fukami K. & Fukagata K., "Nonlinear mode decomposition with convolutional neural networks for fluid dynamics," J. Fluid Mech.
    Vol. 882, A13 (2020). https://doi.org/10.1017/jfm.2019.822
Fukami, K., Nakamura, T. & Fukagata, K.
    (2020) Convolutional neural network based hierarchical autoencoder for nonlinear mode decomposition of fluid field data. Physics of Fluids.
    32 (9), 095110. 10.1063/5.0020721.

MagriLab / MD-CNN-AE