Information-rich BIM models are rarely usable off-the-shelf for operations tasks. Change decisions made on the construction site can lead to significant differences between the as-designed and as-built state of buildings. The responsibility for keeping the digital representation in sync with its physical twin is not defined and will likely only fully be assigned when automatic methods facilitate the geometric update process. To this end, previous research succeeded in (1) identifying if an element was erected at the time and position it was initially designed, and (2) updating the parametric design geometry to fit its LiDAR-measured as-built state under a set of assumptions and threshold values. The research presented in this paper aims at updating the as-designed model in case of significant pose differences between the as-designed and as-built state.
| Authors | Collins, F. |
|---|---|
| Year | 2020 |
| Venue | ETH Zürich / TUM |
| DOI | Not in source spreadsheet (and not found via CrossRef/OpenAlex) |
| Source Database | Backtracking |
| Bridge-to-GNN | Category A |
| GNN Architecture | GCN (heuristic — verify) |
| Graph Encoding | Not yet extracted from PDF |
| AEC Task | Not yet extracted from PDF |
| Cohort | Early Infrastructure (2020–2021) |
| Dataset Size | 1300 nodes (heuristic — verify) |
|---|---|
| Primary Metric | Accuracy (heuristic — verify) |
| Primary Value | 85% (heuristic — verify) |
| Key Finding | pose deviations to the column elements (heuristic — verify) |
| Quality Assessment | 1/6 · rigor: Low · Baseline: ✓ (heuristic — verify) |
| GNN architectures detected | Knowledge Graph Embedding (TransE/RotatE/DistMult) · GCN |
|---|---|
| Dataset stats found | 1300 nodes |
| Quality (heuristic, 6-flag) | 0/6 · rigor: Low Code/Data: —Benchmark: —Baseline: —CV/Split: —Ablation: —Reproducible: — |
Part of the GML/GNN in AEC Systematic Review (PRISMA 2020) — 112 papers, 2020–2026