Skip to content

WP-01 Drawing Tolerance Extraction

Field Value
Priority High
Status Framing
Problem Space Drawing intelligence
Linked Capability geometry reasoning, tolerancing, standards interpretation
Best Collaboration Shape thesis + benchmark + prototype
Owner unassigned

What This Work Package Is

Build a structured extraction workflow for GD&T, dimensions, symbols, views, and related drawing semantics from real engineering drawings.

Why It Matters To RapidDraft

  • Supports review automation directly
  • Creates reusable structured data for DFM and compliance logic
  • Offers a clean, product-visible thesis or prototype outcome

Expected Deliverables

  • benchmark definition and labeled subset
  • first extraction pipeline
  • error taxonomy
  • evaluation summary on realistic industry drawings

Candidate Lanes

Lane People or Group Why They Matter Best First Ask Priority
TUM CCBE AI4CADCAM Dr. Stavros Nousias, André Borrmann, CCBE drawing subgroup strongest local signal for drawing digitisation, semantic extraction, and CAD-to-downstream reasoning co-supervised thesis or HiWi feasibility project using SOMIC-style drawings High
FAU design-methods lane Sandro Wartzack strong drawing and tolerancing methods lens with KBE depth advisory call on evaluation scope and publishable framing High
Global benchmark lane Seung Ki Moon strongest direct public signal on engineering drawing extraction as a benchmark node narrow technical exchange after local lane is framed Medium
Munich execution lane Vahid Salehi Douzloo not the strongest drawing-specific signal, but a practical local execution route for a thesis or prototype wrapper applied thesis with drawing-to-CAD/CAE focus Medium

Ideal Partner Profile

  • strong in drawing understanding or engineering document analysis
  • comfortable combining learning and deterministic post-processing
  • able to work on benchmark quality, not just model novelty

Candidate Evidence Types

  • prior work on engineering drawing extraction
  • CAD-to-semantics pipelines
  • standards or compliance interpretation
  1. Pull all drawing-specific findings from the report set into a candidate table.
  2. Use TUM CCBE as the first local technical lane and FAU as the methods/advisory counterweight.
  3. Define the first benchmark slice around dimensions, views, symbols, and GD&T rather than generic full-document understanding.
  4. Keep the first ask thesis-shaped, not grant-shaped.

Open Questions

  • Should title-block and metadata extraction remain part of this package?
  • What minimum benchmark size would make the result useful internally?

Sources

  • TextCAD/04_Marketing and Outreach/13_Universities/deep-research-report.md
  • TextCAD/04_Marketing and Outreach/13_Universities/deep-research-report Monster.md
  • TextCAD/04_Marketing and Outreach/13_Universities/RapidDraft Academic & Applied Research Partner Mapping (Monster Brief, v1).md