Machine-readable Building Information Models (BIM) are of great benefit for the building operation phase. Losses through data exchange or issues in software interoperability can significantly impede their availability. Incorrect and imprecise semantics in the exchange format IFC are frequent and complicate knowledge extraction. To support an automated IFC object correction, we use a Geometric Deep Learning (GDL) approach to perform classification based solely on the 3D shape. A Graph Convolutional Network (GCN) uses the native triangle-mesh and automatically creates meaningful local features for subsequent classification. The method reaches an accuracy of up to 85% on our self-assembled, partially industry dataset.
| Authors | Collins, F. C.; Braun, A.; Borrmann, A. |
|---|---|
| Year | 2021 |
| Venue | EC3 2021 (European Conference on Computing in Construction) |
| DOI | 10.35490/ec3.2021.168 |
| Source Database | Backtracking |
| Bridge-to-GNN | Category A |
| GNN Architecture | GCN (heuristic — verify) |
| Graph Encoding | BIM Element Topology Graph (inferred from title) |
| AEC Task | Not yet extracted from PDF |
| Cohort | Early Infrastructure (2020–2021) |
| Implementation Framework | PyTorch Geometric (PyG), DGL, PyTorch, Revit / Dynamo (heuristic — verify) |
|---|---|
| Key Hyperparameters | 250 epochs; batch size 30; lr 0.001 (heuristic — verify) |
| Primary Metric | Accuracy (heuristic — verify) |
| Primary Value | 85% (heuristic — verify) |
| Key Finding | the workflow leading from IFC models to two different shape encodings, laying the base for GDL (heuristic — verify) |
| Quality Assessment | 4/6 · rigor: Medium · Code/Data: ✓Baseline: ✓Ablation: ✓Reproducible: ✓ (heuristic — verify) |
| GNN architectures detected | PointNet / PointNet++ · Knowledge Graph Embedding (TransE/RotatE/DistMult) · GCN |
|---|---|
| Frameworks / libraries | PyTorch Geometric (PyG) · Deep Graph Library (DGL) · PyTorch · Revit / Dynamo · ArchiCAD · IFC OpenShell / IfcOpenShell |
| Code repositories found in text | https://github.com/fclairec/geometric-ifc |
| Hyperparameters (regex-detected) | epochs=250 · batch_size=30 · learning_rate=0.001 |
| Dataset stats found | 1024 points |
| Reported metrics + values | Accuracy: 85% |
| Quality (heuristic, 6-flag) | 4/6 · rigor: Medium Code/Data: ✓Benchmark: —Baseline: ✓CV/Split: —Ablation: ✓Reproducible: ✓ |
| Seed of .bib export → verified neighbours | This paper was queried as a Connected Papers seed (41 entries scanned). In-corpus neighbours verified by the bibliometric ranking: [46] 2026 · Abouelaziz · DRAGON: a dynamic risk-aware graph optimization network f… [100] 2016 · Nejur · Ivy: Bringing a Weighted-Mesh Representations to Bear on… [109] 2022 · Fuchkina · Space Matcher An interactive toolbox for assisting in spa… |
|---|
Part of the GML/GNN in AEC Systematic Review (PRISMA 2020) — 112 papers, 2020–2026