How to add document OCR to Ignition (MES/SCADA)
Short answer
Ignition has no OCR engine, so run extraction in a separate service that watches a folder or receives uploads, then writes fields and the source file into SQL tables. A Perspective review screen handles low-confidence fields before records go live.
Where Ignition is used as an MES, the plant is often partly digital. Tags, alarms and line states are captured automatically, while the paperwork around them is not: routers and travelers, first-article inspection sheets, material certificates, delivery notes, hand-written gauge logs and operator checklists. Someone retypes them into Perspective screens, or they stay in a binder.
OCR in this context is not the headline feature, it is the bridge between the paper that still exists and the structured tables Ignition already uses for traceability, quality and genealogy. The value is in linking the extracted data to the right work order, lot or asset, and in being able to show an auditor the original image next to the stored values.
The engineering decisions are practical: where files land, how extraction results are validated against existing MES data, what confidence is enough for auto-accept, and how reviewers fix errors fast. Modern vision-language models read messy forms better than classic OCR, but they still need validation against known work orders, part numbers and units.
What teams use it for
Inspection sheet digitisation
Scanned first-article or in-process inspection sheets are read into measurement rows against the lot and characteristic, ready for SPC charts.
Material certificate capture
Mill certificates and certificates of conformance are read for heat number, chemistry and standard, then linked to inbound lots.
Traveler and router import
Customer-supplied or legacy paper travelers are converted into operations and routing records so operators work from screens.
Delivery note matching
Packing slips are matched against expected receipts so material can be released to the line with the right lot references.
Handwritten log transcription
Daily check sheets and gauge logs written on paper are transcribed into tables, with doubtful handwriting flagged for review.
Inductive Automation Ignition (MES/SCADA) integration surface
- Gateway file access and scheduled scripts, or an external watcher, to detect new scans in a network folder; large file handling should happen outside Gateway threads.
- Web Dev module doPost endpoints or direct SQL inserts for the extraction service to deliver fields and file references.
- Perspective components for image display, tables and forms, used to build side-by-side review screens; exact component names and capabilities vary by version.
- Named Queries to validate extracted work order, lot and part numbers against existing MES and ERP-linked tables.
- Blob or file path storage in SQL for the original document, so a record can always show its source image.
How it works
- 1
Define document types and target tables
List the three or four document types that matter, such as inspection sheet and material cert, and the table and columns each should populate.
- 2
Set up capture
Use a scanner to a network share, a mobile upload page in Perspective, or an email attachment feed. Normalise to PDF or PNG at a fixed resolution.
- 3
Run extraction in a service
A separate service performs OCR or vision-model extraction, returning typed fields, bounding boxes and a confidence per field. Run it on a local GPU if documents are sensitive.
- 4
Validate against MES data
Check work order, lot, part and unit values with Named Queries. Reject impossible values, such as a measurement outside a physical range.
- 5
Route by confidence
High-confidence, fully validated documents post to live tables. Others go to a staging table and appear on a review screen with the image and highlighted fields.
- 6
Store and link the source
Save the original file path or blob with the record, plus the extraction version, so corrections can be audited and models improved over time.
Example flows
Mill certificate to inbound lot
Trigger: Receiving scans a mill certificate to a shared folder.
- The watcher picks up the PDF and submits it to the extraction service.
- Heat number, grade and chemistry values are returned with confidence scores.
- The service checks the heat number against the inbound lot record.
- A mismatch on carbon content versus the spec sends the item to review.
Result: The lot gets certificate data attached, with out-of-spec values flagged before use.
Handwritten check sheet to measurements
Trigger: An operator photographs a paper check sheet from the Perspective mobile page.
- The image is uploaded and enhanced for contrast.
- Numeric cells are read and compared to characteristic limits.
- Unclear digits are highlighted for the operator to confirm on screen.
- Confirmed values write to the measurement table and feed SPC.
Result: Paper readings appear in charts the same day instead of weeks later.
Keeping data on-prem
Documents in manufacturing often carry customer drawings numbers, export-controlled data or certification details, so extraction on a public cloud API may not be acceptable. Run OCR and a vision-language model on-site on a GPU server reachable by the Gateway over the plant network, with open-weight models. The Gateway never needs to run the model: it only hands over files and receives structured results through a Web Dev endpoint or SQL staging tables. For fully air-gapped plants, the extraction service, review screens and storage all stay inside the site, and updates to models are brought in through your approved transfer process.
Security checklist
- Treat uploaded files as untrusted and scan them before processing.
- Restrict the extraction service to staging tables and the file store only.
- Keep the original image with every extracted record for audit and dispute handling.
- Log field-level corrections by reviewers, with user and time.
- Apply the same access controls to document images as to the records they support.
Ways to get it done
Off-the-shelf document capture product writing to SQL
Effort: Two to six weeks including template setup.
High volume of standard forms such as invoices and delivery notes.
In-house pipeline with open OCR plus a Perspective review screen
Effort: Four to eight weeks.
Teams that have a developer, stable form layouts and modest volumes.
Partner build with vision-language models on-prem and MES validation rules
Effort: Scoped per document family; Netray can deliver with your Ignition integrator.
Mixed, messy and handwritten forms in regulated environments.
Pitfalls
- !Classic OCR struggles with handwriting and stamps, so measure accuracy on your own documents before promising automation.
- !Reading a value correctly but attaching it to the wrong lot is worse than a failed read. Validate identifiers first.
- !Unit confusion, such as inches versus millimetres, quietly corrupts SPC data; extract units explicitly.
- !Storing images as large blobs inside the main historian or MES database can bloat backups; use a file store with path references.
Frequently asked questions
Does Ignition include OCR?
No. Ignition does not ship an OCR engine, so you add an external extraction service. The Gateway or an external watcher passes files to it and receives structured fields back through SQL tables or a Web Dev endpoint.
Can OCR results go straight into live MES tables?
Only when every required field passes validation and confidence thresholds. Most plants start with a staging table and a review screen, then auto-accept the document types where the error rate is demonstrably low.
How accurate will extraction be?
It depends heavily on document quality and type. Clean typed certificates can be read very reliably, while photographed handwriting is much worse. Test on a representative sample of your own documents and set thresholds from those results.
Can the models run on-prem?
Yes. Open-weight OCR and vision-language models can run on a local GPU server. Documents never leave the site, and the Gateway only exchanges files and structured results with the service over the internal network.
Where should the scanned files be stored?
Prefer a file share or object store with path references in SQL, plus a backup policy. Large blobs in the main database make backups and Gateway queries heavier than they need to be.
Want Document OCR working on your Inductive Automation Ignition (MES/SCADA)?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment