Two problems, not one
A PDF of a drawing and a STEP file of a part hold different kinds of information. The PDF is a page: lines, arcs, text and symbols placed on a sheet, which a person reads as views, dimensions and notes. The STEP file is a model: surfaces, edges and their topology in three dimensions. Nothing in the PDF says that two circles in two views are one hole. That link is made by interpretation.
So the path from one to the other has two stages. Reading turns the page into requirements: this is a Ø12 through hole, located 35 mm from datum B, with this tolerance. Building turns the requirements into geometry. The two stages fail differently. Reading fails by misreading or by filling a gap. Building, given complete requirements, can be done deterministically and checked. Most of the question “can this be automated” is about reading.
What is inside a drawing PDF
Native vector PDF
A PDF exported from a CAD or drafting system usually stores the drawing as vector paths (lines, arcs and curves with exact coordinates) and the text as text objects (characters in a font, at a position). Software can extract a callout such as “6X Ø5.5 THRU” exactly, together with its location on the sheet, and relate it to nearby geometry. That is the best case for automated reading. It is still a drawing: the paths are projected edges with line types and weights, not the part’s model.
Raster or scanned PDF
A scanned drawing is an image wrapped in a PDF. There are no lines or characters in it, only pixels. Every symbol has to be recognized from the image, and the characters that matter most on a drawing are the ones most easily confused: Ø and 0, 1 and 7, a decimal point and a speck, ± and +. A scan can also be skewed or slightly stretched, so even the geometry is approximate. Automated reading of raster drawings is possible, but its errors depend on the scan, and each reading needs a person to confirm it against the image.
The text layer, and how it can mislead
Whether a PDF has usable text is a separate question from whether it is vector: a vector export can turn its text into outlines, and a scan can carry an OCR layer that is a recognizer’s reading rather than the drafter’s text. For a reader, only characters the drafting system wrote count as evidence. The vector versus raster guide covers the mixed cases and how to tell them apart.
| What software can extract | Outlook for automated reading | |
|---|---|---|
| Native vector, with text | Exact paths, exact characters and positions | Best case: stated callouts can be read exactly; interpretation still needed |
| Native vector, text as outlines | Exact paths, no characters | Characters must be recognized from shapes; each reading needs review |
| Scan with an OCR text layer | Pixels, plus another program’s reading of them | Treat as raster: the text layer is not evidence |
| Raster scan | Pixels only | Recognition throughout, quality-dependent; review every value |
| Multi-sheet or multi-part | Any of the above, across pages | Views and notes must be associated across sheets, and parts kept apart |
STEP in brief
STEP is ISO 10303, the international standard for exchanging product data. A .stp or .step file is usually in the clear-text exchange format of ISO 10303-21. Which data a file can carry is set by its application protocol (AP). For mechanical parts, three are common:
| Protocol | Scope | In practice |
|---|---|---|
| AP203 | Configuration-controlled 3D design of mechanical parts and assemblies | Long-established and widely supported for exact solid geometry |
| AP214 | Core data for automotive mechanical design processes | Used well beyond automotive; also carries data such as colors and layers |
| AP242 | Managed model-based 3D engineering | Brings together the scope of AP203 and AP214, and extends it with fuller support for 3D annotations (tolerances, datums and notes, in semantic and graphic form) and tessellated geometry |
For exchanging the nominal geometry of a single part, any of the three can carry an exact solid. The difference matters when tolerances and annotations travel with the model: AP242 can carry tolerances, datums and notes on the model in a form software can read. A STEP file built from a drawing carries such annotations only if someone, or something, authored them, and most such files carry geometry alone.
B-rep versus mesh
- B-rep (boundary representation)
- A solid described by its bounding faces, each lying on an exact surface (plane, cylinder, cone, torus, spline), trimmed by edges and vertices, with the topology that says which faces meet where. A Ø12 bore is a cylindrical face of radius 6 mm (some systems split it into two half-faces).
- Mesh
- A surface approximated by triangles, as in STL. The same bore becomes a ring of flat facets. Its apparent diameter depends on how finely it was tessellated, and nothing in the file says it is a hole.
Machining, inspection and further design need the B-rep: CAM selects a cylindrical face to drill, and an inspection plan measures a diameter. A mesh is fine for 3D printing and visualization. A STEP file can contain either. A mesh converted into a B-rep with one planar face per triangle is a valid STEP file that is still a mesh, so check the face types, not just the file extension.
What automation can and cannot infer
| Item | Can be automated when | Needs a person when |
|---|---|---|
| Callout text (Ø12 ±0.05 THRU) | It is real text in a vector PDF | It is raster, outlined, or in a symbol font that does not extract cleanly |
| Title block (material, units, general tolerance) | Fields are labeled and the text extracts | A field is blank, handwritten, or contradicted by a note |
| Pattern count and spacing (6X) | Count and spacing are stated in the callout | The count is only drawn, or the pattern’s reference is not dimensioned |
| Overall sizes | They are stated as dimensions | They are only implied by the views |
| Positions | They are dimensioned from a datum | They are shown only by a centerline or symmetry |
| Hidden and internal features | A section or a note states them | They must be inferred from hidden lines across views |
| Values the drawing does not state | Never | Always: ask whoever owns the drawing |
| Tolerance meaning (fit, datum order) | Recording the stated tolerance as data | Deciding what it means for manufacture or inspection |
| Material choice and manufacturability | Never from the drawing alone | Always: this is engineering judgment |
The building stage is different in kind. Given complete, approved requirements, a solid can be built by a deterministic process and measured against those requirements: envelope, feature counts, positions and diameters. That check shows the solid matches what was approved. It cannot show that what was approved matches what the designer meant, which is why the reading stage needs review. The verification guide covers the checks.
Why a STEP file is not manufacturing approval
A STEP file is a geometry exchange file, usually at nominal size. Release for manufacture is a decision taken in your configuration and change process: the right revision, the tolerances, material, finish and inspection requirements, and a responsible person’s sign-off. A shop that receives only a STEP file still needs the drawing, or machine-readable tolerances and notes on the model under a model-based definition practice such as ASME Y14.41, to know what is toleranced and how. Unless your organization has released the model as the authority, the drawing remains the controlling document and the STEP file is a representation of it.
What no conversion can supply
Software can read what the drawing states and build what was approved. It cannot supply a value the drawing does not state, decide which of two conflicting dimensions is right, or approve a part for manufacture. A complete-looking STEP file built from a guessed value is more dangerous than an incomplete one, because nothing in the file shows the guess.
A workable PDF-to-STEP workflow
01 Classify the PDF
Vector or raster, real text or outlines or OCR, one sheet or several, one part or several. This decides how much of the reading you can trust without review.
02 Read the requirements
Extract each callout and title-block field as a requirement, with the place on the sheet it came from.
03 Find the gaps
Check every feature for size and location. A missing value becomes a question, not a default. The missing-dimensions guide has a checklist.
04 Approve the specification
A person reviews every requirement, including the answered questions, and approves one revision of it.
05 Build and measure
Build the solid from the approved requirements only, then measure it against them.
06 Export, and keep the drawing with it
Export STEP (and a mesh only where a mesh is wanted). Send the drawing, or the requirements with their tolerances, alongside it.
Where TruePart sits today
TruePart reads a native vector PDF with a text layer: one sheet, one part, millimeters, up to 10 MB. Scanned, raster or multi-sheet drawings stop with a reason. Currently supporting select part families: today one family, rectangular plates with a centered bore and a hole pattern, read by a deterministic drawing reader. AI-assisted interpretation: in evaluation, not enabled in this preview. Its output is the exact B-rep solid as STEP, and a mesh of it as STL, exported once the solid is measured against the approved specification. The checks measure the envelope, volume, hole count, and each hole at planned position and diameter; they do not check tolerance bands, material or manufacturability. Tolerance bands are carried as data and the CAD is built at nominal. Output is for review by a qualified person, not engineering advice. See PDF to STEP.
Questions
Can any PDF drawing be converted to STEP automatically?
No. A vector PDF with real text gives software exact callouts to read, but the drawing still has to be interpreted into features, and any value it does not state has to be supplied by a person. A raster scan adds recognition errors on top. A tool that never asks a question, even about a drawing with a missing dimension, is supplying that value itself.
Which STEP protocol should I ask for?
For exact part geometry, AP203, AP214 and AP242 all work, and all three are widely supported. Ask for AP242 when you want tolerances and notes to travel with the model and the receiving software supports it. Confirm with the receiving shop, which knows what its software imports cleanly.
Is an STL file good enough instead of STEP?
For 3D printing and viewing, usually yes. For machining or inspection, no: an STL has no cylinders, planes or holes, only triangles, so features must be reconstructed from the mesh before they can be programmed or measured.
If the STEP file passed its checks, can it go straight to the shop?
Checks show that the solid matches the requirements it was built from. Release to manufacture also needs the right revision, tolerances, material, finish and a responsible sign-off, which a geometry file does not carry. Send the drawing with it unless your organization releases annotated models as the authority.
Does a text layer mean the PDF is vector?
Not necessarily. A scanned drawing can carry an OCR text layer that makes it searchable. Zoom into the linework: if it pixelates, the page is raster, and the text layer is a recognizer’s reading rather than the original text.
Sources
- ISO 10303, Industrial automation systems and integration: Product data representation and exchange (STEP) · ISOThe STEP standard family.
- ISO 10303-21, Implementation methods: Clear text encoding of the exchange structure · ISOThe file format of .stp and .step files.
- ISO 10303-203, Application protocol: Configuration controlled 3D design of mechanical parts and assemblies · ISOAP203 scope, at a general level.
- ISO 10303-214, Application protocol: Core data for automotive mechanical design processes · ISOAP214 scope, at a general level.
- ISO 10303-242, Application protocol: Managed model-based 3D engineering · ISOAP242 scope, 3D annotations and tessellated geometry, at a general level.
- ISO 32000-2, Document management: Portable document format, Part 2: PDF 2.0 · ISOPDF page content: vector paths, text objects and images.
- ASME Y14.41, Digital Product Definition Data Practices · ASMEWhen an annotated model, rather than the drawing, is the released authority.
What TruePart does today
- TodayTruePart reads a native vector PDF with a text layer: one sheet, one part, millimeters, up to 10 MB.
- LimitScanned, raster or multi-sheet drawings stop with a reason.
- LimitCurrently supporting select part families: today one family, rectangular plates with a centered bore and a hole pattern, read by a deterministic drawing reader.
- In evaluationAI-assisted interpretation: in evaluation, not enabled in this preview.
- TodayIts output is the exact B-rep solid as STEP, and a mesh of it as STL, exported once the solid is measured against the approved specification.
- LimitThe checks measure the envelope, volume, hole count, and each hole at planned position and diameter; they do not check tolerance bands, material or manufacturability.
- LimitTolerance bands are carried as data and the CAD is built at nominal.
- LimitOutput is for review by a qualified person, not engineering advice.