B-Rep Feature Recognition Backlog¶
Status: Active research backlog (April 2026 ingestion pass)
Purpose: Capture external model and dataset references that can improve CAD-feature extraction quality for DFM and downstream automation.
Why This Page Exists¶
Two newly ingested sources are directly relevant to geometry-first intelligence:
- a B-Rep-native machining feature recognition paper (Hierarchical CADNet)
- the MFCAD++ dataset repository surface
They are not production-ready drop-ins, but they are high-value references for training-data and model-experiment planning.
Source Triage¶
| Source | Type | What it contributes | Recommended use |
|---|---|---|---|
| Hierarchical CADNet (ScienceDirect) | Research paper page | B-Rep graph representation + neural architecture framing for feature recognition | Use as architecture reference when scoping face/edge graph encoders and evaluation metrics |
| MFCAD++ Dataset (GitLab) | Dataset/repo page | Public pointer to machining-feature dataset structure and access path | Use as candidate dataset source for benchmark experiments and label-schema alignment |
How To Use These in TextCAD Work¶
- Keep deterministic Part Facts and rules as the production baseline.
- Use CADNet and MFCAD++ as an R&D track for improved feature candidate proposals.
- Evaluate models against the same rule-relevant outputs the product needs (holes, pockets, ribs, bosses, undercuts), not only paper-level class accuracy.
- Gate any model-assisted extraction behind confidence and evidence-location quality thresholds.
Open Questions¶
- What portion of MFCAD++ labels map cleanly to RapidDraft/DFM rule families without relabeling?
- Which feature classes produce the largest practical reduction in rule false negatives?
- Should model-assisted extraction run only in deep scan, or also in light scan with strict confidence thresholds?
Sources¶
C:\Users\adeel\OneDrive\100_Knowledge\203_TextCAD\01_Product_Project_Management\TextCAD_Wiki\docs\06_Infrastructure_Research\_sources\hierarchical_cadnet_machining_feature_recognition.mhtmlC:\Users\adeel\OneDrive\100_Knowledge\203_TextCAD\01_Product_Project_Management\TextCAD_Wiki\docs\06_Infrastructure_Research\_sources\mfcadpp_dataset_gitlab.mhtml