Engineering Drawing PDF Dimension Extractor

Extract numeric dimension candidates from the text layer of engineering-drawing PDFs for document indexing and review. Get a table with each candidate's text, numeric value, page position and surrounding text, plus JSON and CSV files for each parsed PDF.

DrawX is an extraction aid. It can miss dimensions and include drawing numbers, labels or other unrelated numbers. Check candidates against the original drawing.

What you get

• Dimension candidates with page number, position, explicit units, diameter or radius prefixes, and symmetric ± tolerance where recognized.
• Surrounding text and distance to nearby drawn lines to help review candidates.
• A document summary with processing status, candidate counts, page dimensions and recognized sheet-size matches.
• JSON and CSV files in the run's key-value store.

The high and low confidence labels describe the text syntax, not measured accuracy: high means a recognized prefix, tolerance or unit is present; low means a bare numeric candidate. Neither label confirms that a number is a dimension or that its value is correct.

Input

Use public PDF URLs, PDFs in your own private Apify key-value store, or both. Supply at least one document. PDFs need a readable text layer; image-only scans and text converted to outlines are unsupported. DrawX does not perform OCR.

For private PDFs:

1. Upload the PDFs as records in a private key-value store in your Apify account.
2. Select that store in inputStoreId. The input requests read access to the selected store.
3. Enter the exact record keys in inputKeys, including any filename extension. DrawX reads only the keys you list; it does not process every record in the store.

The actor input has no direct file-upload field. Keep private files in the selected store rather than pasting private Apify file URLs into pdfUrls.

Field	Meaning
pdfUrls	Optional list of public HTTP or HTTPS PDF URLs.
inputStoreId	Optional private key-value store selected for read access. Use with inputKeys.
inputKeys	Exact PDF record keys to read from the selected store. Use with inputStoreId.
minConfidence	low includes both labels; high returns only candidates with recognized dimension syntax in the dataset.
maxPages	Optional limit on pages processed per PDF.
maxFileSizeMb	File size limit per PDF, default 50 MB.

For example, after uploading a PDF under the key DRAWING.pdf:

{
  "inputStoreId": "YOUR_PRIVATE_STORE_ID",
  "inputKeys": ["DRAWING.pdf"],
  "minConfidence": "low"
}

Output

The dataset contains two row types:

• dimension: candidate text and value, page, position, prefix, tolerance, unit, confidence label and lineContext (surrounding text).
• document_summary: processing status, counts and per-page sheet-size information.

The Dimensions view includes lineContext; use it to check the original text around a candidate, especially when a symbol is absent from the structured prefix. Use the Document summaries view to check each file's processing status. Use recordType to distinguish the row types when exporting or processing the dataset. Positions are text centers in millimeters from the bottom-left of the PDF page, not dimensions of the physical part.

The key-value store contains a JSON report and CSV candidate table for each successfully parsed PDF, plus an OUTPUT run summary. JSON and CSV contain all detected candidates from the processed pages; the dataset's confidence filter and billing cap do not filter these files. The dataset and file exports use different field names: for example, dataset lineContext is CSV line_context. For private-store inputs, sourceUrl identifies the store and record as apify://<store-id>/<URL-encoded-record-key>; it is a source identifier, not a download link.

If a PDF cannot be read from its source or parsed, its summary records the problem and processing continues with the next PDF. If every attempted document has an input or parsing error, the run ends with FAILED after saving the summaries and OUTPUT. An unsupported PDF with no readable text receives a no_text_* summary; that status does not fail the run. An empty document list fails the run.

Limits

• Numeric labels can appear as candidates even when they are far from drawn geometry.
• Bare four-digit integers are skipped, including genuine dimensions such as 1200.
• Bare single digits are retained only when they stand alone near a drawn line. Title-block numbers can still appear.
• Some diameter glyphs, including lowercase ø, are not recognized as prefixes. Check lineContext; high-only filtering can omit these values.
• Signed values, fractions and complex tolerance layouts may be incomplete or misread.
• Units are reported only when recognized beside a value. Drawing units and scale are not inferred.
• Candidates are not linked to the features they measure. DrawX does not interpret GD&T frames.
• Sheet-size matching compares PDF page dimensions with a built-in size table. It does not assess drawing-standard compliance.
• A PDF can contain readable labels while its dimension text remains unreadable.

Use the results for document indexing and review. Do not use them to authorize manufacturing, inspection, quoting, compliance or safety decisions.

Pricing

US$5 per 1,000 dimension candidates saved to the dataset (US$0.005 each), plus US$0.00005 per run start per GB of memory, with a minimum of one start event. Document summaries and error rows are free. For example, 100 candidate rows at 1 GB cost US$0.50005. No separate actor compute charge is added to this event price; other Apify services or storage can incur platform charges.

Set a maximum charge per run. If it leaves no allowance for a dimension, the actor stops before processing more PDFs. If a document has more candidates than the remaining allowance, its dataset rows are truncated, its summary marks the truncation, and OUTPUT lists unprocessed PDFs. Full JSON/CSV files can include candidates beyond the charged dataset rows.

Privacy

PDF contents are processed inside the actor without an OCR or model API. The actor downloads public files from the URLs you supply, reads the listed PDF records from your selected private store, and writes results to your Apify storage. The input field requests read access to the selected store, not write access. DrawX does not write to that input store. Input PDFs, run logs (including source URLs or private store and record identifiers) and outputs follow your Apify storage and retention settings.

Support

Use the Issues tab for questions or extraction problems. Share only PDFs you have permission to disclose.

Built and maintained by AI agents for Madrasco. Support replies may be AI-assisted; a human owner is reachable on request.