Use case

From scattered PDFs to structured PMI — in seconds.

Werk24 extracts dimensions, tolerances, GD&T, notes, hole tables, surface finishes, and cross-references from technical drawings — and returns clean, unit-aware JSON ready for ERP, MES, and QA.

Why manual PMI re-entry does not scale

PMI is scattered across title blocks, views, balloons, and free-text notes. Manual re-entry into ERP/MES/QA is slow and error-prone — duplicates, missing tolerances, and inconsistent symbols slip through. As drawing volume grows, headcount-dependent transcription becomes a structural bottleneck.

What it extracts — a complete, normalized PMI graph

  • Dimensions — linear, diameter, radial, and angular with upper/lower tolerances.
  • GD&T frames — feature control frames, datum references, and composite tolerances.
  • Notes & tables — free-text notes, hole tables, weld symbols, and surface finish callouts.
  • Cross-references — resolve "see detail A", "per note 7", view-to-feature links.
  • Title-block metadata — part number, revision, material, standard, and customer references.

Every item includes provenance and confidence. Units are normalized. Unresolved references are flagged for review.

PMI extraction pipeline
From drawing to structured PMIParse drawings, detect features, normalize units, cross-link references, export JSON.Ingest drawingPDF, scan, or DXFDetect & normalizefeatures & referencesExport structuredPMI + overlaysGeometry, text layers,symbols, and tables• Unit normalization• Confidence per item• JSON for ERP/MES/QA• PDF with highlights

How it works

Five stages from drawing to structured, cross-linked PMI.

  1. Ingest Werk24

    Upload PDF, scan, or DXF. We parse geometry, views, tables, text layers, and symbols.

  2. Detect Werk24

    Identify dimension leaders, GD&T frames, note anchors, hole tables, and surface finish callouts.

  3. Normalize Werk24

    Unify units, tolerances, and symbols into a single PMI schema. Original symbols are preserved in metadata.

  4. Cross-link Werk24

    Connect notes to features, features to views, and views to details. Unresolved references are flagged.

  5. Export

    JSON for ERP/MES/QA, PDF overlays with highlighted features, and a change log for audit.

Sample PMI

Sample PMI (JSON excerpt)

{
  "dimensions":[{"type":"diameter","value":18.0,"tol":{"+":0.02,"-":0.02},"feat":"H1","unit":"mm"}],
  "gdt":[{"feature":"A","frame":[{"sym":"⌀","val":0.1},{"sym":"⟂","val":0.02,"datum":"B"}]}],
  "notes":[{"id":7,"text":"Deburr all edges 0.2×45°"}],
  "surfaces":[{"Ra":1.6,"area":"S1"}],
  "refs":[{"from":"VIEW-LEFT","to":"DETAIL A"}]
}

Topline QA

Dimensions parsed
124
GD&T frames
18
Notes linked
27
Unresolved refs
2

Export a PDF overlay to review links before release.

Outcomes

Structured PMI that downstream systems can consume — without manual transcription.

  • Seconds per page, not hours

    Automated extraction replaces manual re-entry. Engineers focus on exceptions, not transcription.

  • Unit-aware, normalized output

    No more misread symbols or wrong units. Every item is parsed, validated, and confidence-scored.

  • Cross-linked references

    Notes, features, views, and details are connected — not just extracted in isolation.

  • Ready for ERP/MES/QA

    Clean JSON drops into downstream systems. PDF overlays let reviewers verify before release.

Metrics that matter

Track extraction speed, coverage, and downstream integration quality.

  • Extraction speed: Time per page (P50/P90).
  • Feature coverage: % of dimensions, GD&T, and notes correctly extracted.
  • Confidence distribution: Share of items above/below review threshold.
  • Downstream adoption: % of PMI consumed by ERP/MES/QA without manual correction.

FAQ

Which PMI types are supported?

Dimensions (linear, diameter, radial, angular), tolerances, GD&T frames, hole tables, surface finish, weld symbols, notes, and cross-references.

How are units and symbols handled?

Units are normalized to your target system (e.g., mm). Original symbols (⌀, ⌖, ±) are preserved in metadata for traceability.

What about low-confidence items?

Every item includes provenance and confidence. Unresolved references and low-confidence items are flagged for human review.

Can we integrate via e-mail instead of API?

Yes. Forward drawings to a dedicated inbox and receive JSON + annotated PDF back. API and e-mail workflows can run in parallel.