kopaaParametric/ec3-steel-portal-frames

EC3 Steel Portal Frame Structural Analysis Dataset

Dataset Summary

A parametric structural analysis dataset of 995 single-span steel portal frames designed and verified to Eurocode 3 (EN 1993-1-1:2005). Each record represents a unique frame configuration solved with the Karamba3D 3.1 FEM solver inside Grasshopper / Rhinoceros 3D.

The dataset is intended for training and benchmarking AI models that predict structural performance, section selection, and code compliance in the early design phase of steel-framed buildings — industrial sheds, agricultural buildings, warehouses, and sports halls.

Dataset Structure

Two linked JSONL files:

Frames and elements are linked via run_id (format: SF-00001 → SF-01000).

Frame Schema

Each frame is a 2D single-span portal in one of two configurations:

Element identifiers: C1, C2 = columns; B1, B2 = roof beams (rafters)

Parametric Ranges

Key Statistics

Fields

frames.jsonl — 30 fields per record

elements.jsonl — 47 fields per record

Load Case Combinations (10 per frame)

All combinations follow EN 1990 Annex A1 ULS persistent/transient design situation with ψ₀ combination factors per EN 1991:

Data Generation Pipeline

Grasshopper / Rhinoceros 3D
  └── Parametric frame geometry (uniform random sampling)
  └── Karamba3D 3.1 FEM solver
        └── 10 ULS load case combinations (EN 1990 Annex A1)
        └── EC3 EN 1993-1-1:2005 cross-section + stability checks
        └── BeamView utilization component (max across all LCs)
        └── BeamDisplacements component (midspan deflection)
  └── GHPython 3 extractor (CPython 3.9, Rhino 8)
        └── frames.jsonl
        └── elements.jsonl

Solver: Karamba3D 3.1 (Clemens Preisinger, TU Vienna / Bollinger+Grohmann) Design code: EC3 EN 1993-1-1:2005 with EN 1990 load combinations Platform: Rhinoceros 8 / Grasshopper, Windows

Known Limitations

• —Column profiles limited to HEB only — HEA, UC, HD sections not included
• —Rafter profiles limited to IPE only — welded, cellular, and tapered sections not included
• —Wind pressure only — wind suction (uplift) load cases are not included; wind uplift on roof can govern rafter design in practice and is absent from this dataset
• —2D single-span frames only — no multi-bay, crane gantry, mezzanine, or 3D effects
• —Connections not modelled — fully rigid frame assumption throughout
• —No second-order (P-Δ) effects — first-order linear elastic analysis only
• —5 missing run_ids in SF-00001–SF-01000 sequence due to solver failures during batch generation

Intended Use Cases

• —Training ML regression models to predict util_max, frame_weight_kg, roof_mid_vert_defl_mm from geometric and load input parameters
• —Training classification models to predict is_ec3_compliant
• —Surrogate modelling for structural optimisation loops
• —Benchmarking graph neural networks on 2D frame topology
• —Research in AI-assisted structural design and automated EC3 checking
• —Education — demonstrating EC3 design parameter sensitivity

License

CC BY-NC 4.0 — free for research and non-commercial use with attribution. For commercial licensing enquiries contact the author.

Citation

@dataset{neiburgs_ec3_steel_frames_2025,
  title     = {EC3 Steel Portal Frame Structural Analysis Dataset},
  author    = {Neiburgs, Arturs},
  year      = {2025},
  publisher = {KOPAA},
  url       = {https://huggingface.co/datasets/kopaaParametric/ec3-steel-portal-frames},
  license   = {CC BY-NC 4.0},
  note      = {Generated with Karamba3D 3.1 / Grasshopper / Rhinoceros 8.
               995 frames, 3450 elements, EC3 EN 1993-1-1:2005.}
}

Contact

Arturs Neiburgs KOPAA ✉ arturs.neiburgs@kopaa.io 🔗 LinkedIn