Published September 2, 2026
| Version v1
Dataset
Open
Data of: Decoding microstructure-property relationships in polycrystals: An explainable GNN surrogate for crystal plasticity
Authors/Creators
Description
This dataset contains the data and code analyzed in the research journal publication: "Decoding microstructure-property relationships in polycrystals: An explainable GNN surrogate for crystal plasticity" which was funded from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (grant agreement No 101001567).
# Decoding microstructure-property relationships in polycrystals: An explainable GNN surrogate for crystal plasticity
Code and data accompanying:
> Ali R. Safi, Pedro Campos, Frederic E. Bock, Benjamin Klusemann. "Decoding microstructure-property relationships in
> polycrystals: An explainable GNN surrogate for crystal plasticity."
> Machine Learning for Computational Science and Engineering, 2026.
> DOI: https://doi.org/10.1007/s44379-026-00092-w
## Contents
- `main.py`, `model.py`, `split.py`, `util.py`, `postproc.py`, `integrated_grad.py` -- core training pipeline
- `config/production_*.json` -- exact hyperparameters for each of the six reported stress-component models (s11, s22, s33, s12, s23, s13)
- `run_plots.py` -- dataset, loss, and prediction figures
- `run_ig_analysis.py` -- Integrated Gradients XAI figures
- `data/Data.tar.gz` -- raw DAMASK simulation data (1,000 microstructures x 3 loading conditions)
## Setup
```
conda env create -f environment.yml
conda activate gnn_xai_crystal_plasticity
pip install -r requirements.txt
```
A CUDA GPU is required for training and for the Integrated Gradients analysis.
## Data
```
tar xzf data/Data.tar.gz
```
This extracts `Data/raw/instance_<type>_..._<loading>/` (3,000 folders) and
`Data/processed/` into the repository root, which is where all scripts expect
to find it.