SFEM Dataset: 3D Finite Element Stress Fields under Stochastic Point Loads for Graph Neural Network Surrogates
- Provider
- Jessica Ezemba, Carnegie Mellon University Design Research Collective
- License
- MIT (source)
- Size
- 39,902 files
- Stored at
- Hugging Face (https://huggingface.co/datasets/cmudrc/SFEM)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Physics
- Source
- https://github.com/cmudrc/SFEM
About this dataset
Three-dimensional finite element meshes with stress and displacement fields under stochastic point loading, built at the Carnegie Mellon University Design Research Collective for neural network surrogates of the stochastic finite element method (SFEM). About 16,000 CAD geometries generated with BrepGen were meshed with Gmsh (Frontal-Delaunay tetrahedra) and solved with FEniCSx/DOLFINx as linear elastic parts (E = 2.303 GPa, nu = 0.4002) under point loads of 200 N, 2,000 N and 20,000 N, with 50 realizations per geometry and load. Per sample fields include vertices, von Mises stress, displacement, the full stress tensor, load class and a fixed-facet boundary mask. Step_Files/ holds the original STEP geometries (about 39,900 files, 3 GB); Pytorch_Geometric_Files/ holds processed PyTorch Geometric train and validation sets (about 7 GB and 1.2 GB) and raw tar.gz archives, split 85/15 by geometry. Total about 18.6 GB. Curated by Jessica Ezemba; licensed MIT as stated in the card; cite Neural Network Surrogate Modeling for Stochastic FEM using 3D Graph Representations (2025).
How to load this dataset in Python
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="cmudrc/SFEM", filename="Pytorch_Geometric_Files/processed/pre_filter.pt", repo_type="dataset")
print(path)Data structure
Files
Example file: Pytorch_Geometric_Files/processed/pre_filter.pt
Analyze this dataset without downloading it
On Scigantic, SFEM Dataset: 3D Finite Element Stress Fields under Stochastic Point Loads for Graph Neural Network Surrogates opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Physics datasets
- SPH-Simulated Laser Powder Bed Fusion Melt-Pool Dataset (Ti-6Al-4V, 241 Single-Track Simulations)
- CMS Level-1 Trigger Anomaly Detection Dataset (2025 Zero-Bias Data and Simulations)
- ColliderML Drifting Beamspot: Single-Muon Tracking Samples with Smeared Digitisation
- Top Quark Tagging Reference Dataset (Parquet)
- Rotor37: 3D Compressor Blade RANS Dataset under Geometric Variations (PLAID)
- The Well: Euler Multi-Quadrant Riemann Problems (Open Boundaries)
- Pythia 8 Quark and Gluon Jets (Komiske, Metodiev, Thaler), Parquet with Particle Transformer Features
- The Well: Rayleigh-Taylor Instability Turbulence (3D)