Engineering Drawing Extractor

What Problem It Solves

Engineering drawings (e.g., CAD DWG/DXF or PDF) contain dense text, coordinates, symbols, and tables that are tedious and error-prone to manually interpret. For instance, extracting equipment tags from P&ID diagrams or room numbers from architectural plans requires structured data for systems like equipment ledgers or material lists.

How the Skill Works

The Drawing Extractor Assistant follows an automated discovery, extraction, and legend interpretation workflow to convert drawings into structured outputs:
- Core Extraction: For vector PDF/DXF, it directly pulls text, coordinates, and block names (e.g., named blocks like Ball Valve (Flanged)), bypassing OCR; scanned PDFs or images use OCR (requiring the user's Tencent Cloud keys) to extract text coordinates.
- Legend-Prioritized Interpretation: The skill first identifies legend tables in the project to build a "sample-to-name" mapping. When processing drawings, it prioritizes this project-specific mapping to explain symbols, falling back to standard libraries (e.g., ISA-5.1) for gaps or anomalies, and marks items with tags like needs_review.
- Output Format: By default, it generates an Excel table with a tags column, including text lists and symbol statistics; for ledger entry, it can output JSON or CSV per user-defined fields. Batch processing supports folder scanning, resume from breaks, and parallel OCR.

Applicability Boundaries and Considerations

• Accuracy Limits: Vector PDF/DX text extraction is near 100% accurate; anonymous geometry (unnamed blocks or pure lines) may require manual review.
• Dependencies: DWG conversion needs the user to install ODA File Converter locally; scanned OCR requires setting environment variables TENCENTCLOUD_SECRET_ID and TENCENTCLOUD_SECRET_KEY, with automatic fallback to local tesseract if keys are missing (lower precision).
• Security and Privacy: The skill runs entirely locally, with no built-in accounts or data uploads; it processes files read-only without modifying originals.
• Legend Handling: Conflicts between legends and tags trigger marks like suspected_legend_error to avoid silent misinterpretation; output data always includes tags for human verification of critical information.