Turning engineering drawings into a digital-twin asset database
We built a computer-vision pipeline that reads engineering drawings and spec tables, extracts the assets on them, and writes them into a structured database that feeds a digital twin — with low-confidence items routed to a human before anything is trusted.
The asset data was locked inside drawings
A digital twin is only as good as the asset data behind it — and here that data lived in stacks of engineering drawings and spec tables, not in a database. Building the twin by hand meant people reading each drawing and re-typing every asset: slow, expensive, and error-prone at exactly the scale where errors compound.
The goal was to get the structured asset database out of the drawings automatically — fast enough to be practical, accurate enough to trust as the foundation of a digital twin.
Vision does the reading. Structure and review keep it trustworthy.
Each stage has one job, and low-confidence results are escalated to a person rather than guessed.
Ingestion Drawings + tables
Takes in engineering drawings and their spec tables in the formats they already exist in — no re-drawing, no manual prep. Every source is tracked so extracted assets can be traced back to where they came from.
Detection & reading Computer vision + OCR
Computer vision detects the assets and symbols on the drawing; OCR reads the labels and spec-table values. This is the part that replaces hours of manual eyeballing — the model does the reading across the whole set.
Structuring Schema mapping
Detected elements are mapped into a defined asset schema — types, attributes, relationships — so the output is a clean, queryable database, not loose text. Each extracted field carries a confidence score.
Human review Human-in-the-loop
Anything below the confidence threshold is routed to a person to confirm or correct, instead of being trusted blindly. The model handles the volume; a human owns the edge cases — so the database earns the trust the twin depends on.
Digital-twin sync Structured output
Verified assets flow into the asset database that backs the digital twin — turning a pile of drawings into a live, queryable model of the real-world infrastructure.
Speed of AI, accuracy you can stake a twin on
What the system gives you
How we ship CV to production
Featured in Bentley's iTwin Innovation Showcase
Bentley's iTwin Innovation Showcase credits our AI with improving point-cloud analysis inside OpenTower IQ on the iTwin platform — a rare case of the platform vendor publishing the work rather than the vendor's own marketing.
That entry covers one slice of the engagement. The wider scope also read engineering drawings and spec tables, corrected OCR and table-parsing errors under human review, and kept the resulting asset database in sync with the twin as source documents changed.
Have documents or images hiding structured data?
Drawings, forms, inspection photos, spec sheets — if the data is trapped in images, we can get it out, structured and verified.
Book a discovery callAnonymized case study. Client details are confidential; described by domain and outcome.