metadata
license: mit
tags:
- physics-ml
- graph-neural-networks
- computational-mechanics
- elastoplastic
G-PARC: Graph Physics-Aware Recurrent Convolutions
Model weights, test data, and configuration files for the G-PARC elastoplastic simulation paper.
Models
| Model | Description |
|---|---|
| G-PARCv1 | Graph Physics-Aware Recurrent Convolutions — fully learned GNN operators |
| G-PARCv2 | MLS differential operators + numerical Euler integration |
| MeshGraphKAN | Kolmogorov-Arnold Network message passing with Fourier basis |
| MeshGraphNet | Standard encode-process-decode GNN (Pfaff et al., 2021) |
Dataset
PLAID 2D Elasto-Plasto-Dynamics benchmark — high-velocity impact on steel plates.
- Variables: Displacement field (U_x, U_y)
- Normalization: Global max
- Meshes: Unstructured quad elements
Usage
from huggingface_hub import hf_hub_download
ckpt = hf_hub_download("jacktbeerman/Gparc", "checkpoints/gparcv2_best.pth")