Graph neural networks for strut-based architected solids

Link to paper:

https://doi.org/10.1016/j.jmps.2024.105966

Code structure

├── lattices: submodule for lattice processing, elasticity and plotting functions
├── gnn: ML modules
    ├── ...
├── scripts
    ├── train_utils.py: utilities for training
    ├── train_fromcfg.py: training script for the main model
    ├── load_datasets.py: script that can be used to assemble train/val/test datasets

Usage

Set up environment using requirement.txt or environment.yml file.

Try the scripts in scripts folder. You can run the code using the following:

cd scripts
python train_fromcfg.py --cfg-path params-0.yml

This will use a small sample dataset of 1500 training and 500 validation lattices.

Data to assemble full datasets is available for download from https://doi.org/10.17863/CAM.115813