Legacy Drawing Mining for Retrofit & Service
Your transformers will run for sixty years. Their drawings are trapped in scans nobody can search.
The most profitable business in the energy equipment world right now is not the new unit — it is everything that happens afterwards: retrofit, life extension, spare parts, service. The grid build-out is pushing operators to squeeze more decades out of installed fleets, and every one of those decades generates enquiries about machines your works built in 1985.
The knowledge to answer those enquiries exists. It sits in tens of thousands of paper and TIFF drawings — scanned at some point, filed by order number, carrying no searchable metadata whatsoever. Today, answering a retrofit request means a veteran engineer digging through the archive for days, reading title blocks one by one. And the veterans who can still interpret a 1970s drawing convention are the same people retiring in the next five years.
Drawing mining changes the economics of that archive: AI reads every scan — title block, parts list, materials, key dimensions — classifies it, and turns the pile into a structured, searchable inventory inside Teamcenter. The archive stops being a liability and becomes what it always should have been: the asset behind your service business.
// The archive problem, in numbers
Scanned is not digitized
Most works “digitized” their archive years ago — meaning someone fed the paper through a scanner. The result is a folder tree of TIFFs and PDFs organised by order number: findable only if you already know what you are looking for, readable only by people who know the old drawing conventions, and invisible to every system you run today.
The cost shows up in every service transaction. A grid operator asks for a bushing replacement on a 1988 unit — someone spends two days finding the right revision. A retrofit enquiry needs the tank dimensions of a discontinued series — the one colleague who remembers where to look is on holiday. A component supplier disappears — and nobody can say quickly which delivered units carry that part, because parts lists exist only as pixels.
Drawing mining attacks exactly this gap between scanned and usable: extract the data that is locked in the pixels, attach it to the drawing as structured metadata, and index everything where your engineers already work — in Teamcenter.
In — scanned drawing, 1987
Out — structured record in Teamcenter
From TIFF pile to searchable archive: how the mining works
Drawing mining is a batch process with a quality loop, not a magic button. It runs where your data must stay — on-premises or in an EU cloud — and it is honest about uncertainty: every extracted field carries a confidence score, and everything below the threshold goes to a human review queue instead of silently into your PLM.
Take the archive as it is
TIFFs, PDFs, microfilm scans, mixed resolutions, rotated sheets, handwritten additions — the pipeline normalizes what it can and logs what it cannot. No pre-sorting project, no re-scanning campaign as a precondition.
Sort tens of thousands of sheets automatically
Assembly drawing, part drawing, schematic, parts list, sketch — each sheet is typed and grouped by product family and component class, so the archive gets a structure before a single field is read.
Read the data locked in the pixels
Title block fields, drawing and order numbers, revisions, materials, parts lists, key dimensions — extracted with a confidence score per field. Old conventions and works-specific title block layouts are learned from your samples, not assumed.
Humans check the doubtful — Teamcenter gets the result
High-confidence records flow straight into Teamcenter as items with attached source scans; low-confidence fields queue for fast human review. The outcome: a searchable archive inside the PLM your engineers already use.
3D where it pays, not everywhere
For active retrofit and spares candidates, the mined data accelerates NX re-modeling: extracted dimensions and parts lists become the starting point for a proper 3D reconstruction — done for the parts that earn it, not the whole archive.
What a mined archive unlocks
The index is not the goal — it is the foundation. Once the archive is structured and searchable, four businesses that used to run on veteran memory start running on data:
Retrofit quoting at enquiry speed
"Which drawings define the 1992 unit at substation X?" becomes a search, not an expedition. Retrofit offers go out while the operator is still talking to you — not three weeks later.
Spare parts & re-manufacturing
Extracted parts lists and dimensions feed directly into NX re-modeling and supplier RFQs. Discontinued part? The mined archive shows every unit that carries it — and what it takes to rebuild it.
Obsolescence at a glance
When a supplier or material disappears, a metadata query answers what took weeks of manual digging: which series, which customers, which installed units are affected.
Knowledge that survives retirement
The veterans' ability to "just know where to look" is encoded into structure and search. New service engineers become productive on the archive in weeks instead of years.
When does drawing mining pay off?
The business case scales with the age of your installed base and the size of your archive. You are a candidate if most of these apply:
- Your units stay in the field for decades, and retrofit/service is a growing share of revenue.
- The archive holds tens of thousands of scanned or paper drawings with little or no searchable metadata.
- Service and retrofit enquiries regularly wait on days of archive research — or on one specific colleague.
- The engineers who can interpret the old drawing conventions are approaching retirement.
- You run Teamcenter (or a comparable PLM) and want the mined archive indexed there — not in yet another island tool.
- Your drawings must not leave your infrastructure — on-premises processing is a requirement, not a preference.
What the mining pipeline handles
Built for the messy reality of a works archive — not for lab-clean sample documents.
Before and after: the same retrofit enquiry, two archives
A grid operator asks for a life-extension quote on a thirty-year-old unit, including replacement of the conservator and two bushings. Here is how the enquiry runs against an unmined and a mined archive:
Before — veteran & folder tree
- Find the order number via old lists and memory — ½ day
- Open scans one by one, hunt the right revision — 1–2 days
- Re-type the parts list for the RFQ by hand — ½ day
- Hope the one colleague who knows the 80s conventions isn't on leave
- Quote goes out in week three — if the operator is still waiting
After — mined archive in Teamcenter
- Search by serial, order or component class — minutes
- Latest revision and linked sheets already resolved — instant
- Extracted parts list exports straight into the RFQ — minutes
- Any service engineer can run the search — no veteran required
- Quote goes out the same week — retrofit revenue you used to lose
Why inteliScience
Generic OCR vendors can read a clean title block. The hard part of drawing mining is everything around it: works-specific title block layouts that changed three times since 1975, material designations in old norms, revision chains across sheets — and above all, getting the result into Teamcenter as proper items and datasets instead of another spreadsheet nobody maintains.
That last mile is our home ground. We build Teamcenter and NX customizations for manufacturers every day, on Siemens’ supported extension points — so mined drawings become searchable PLM objects with attached source scans, ready for the retrofit workflows and NX reconstructions that follow.
And the quality loop stays honest: confidence scores on every field, human review where the AI is unsure, and accuracy measured on your scans — not on a vendor’s demo documents. AI does the reading; your people stay the authority on what is true.
Fixed price — no day rates, no surprises
Drawing Mining Pilot
- 1,000 drawings from your real archive, mined end-to-end
- Extraction tuned to your historic title block layouts & conventions
- Measured accuracy report per field type — your scans, your numbers
- Searchable demo index — Teamcenter integration concept included
- Cost & timeline projection for mining the full archive
6–8 weeks · processed on-prem or EU cloud · all mined data is yours, whatever you decide afterwards.
For comparison
Manual re-indexing
- Typing metadata for 50,000 drawings by hand, one sheet at a time
- Multi-year project that most works start once and never finish
- Meanwhile: every service enquiry still starts with a 2–3 day archive dig
Your archive already paid for itself once. Mine it, and it pays again — every service enquiry.
Send us 20 sample scans from your archive — we'll show you what the extraction gets out of your title blocks and parts lists.
Talk to a Drawing Mining EngineerSubscribe to our Newsletter
Always here to support you – reach out to us with your questions, and we’ll guide you every step of the way.
- Phone
- +40 774 463 429
- Our Location
- Str. Ploiești 9, 400157
Cluj-Napoca, Romania - inteliScience on LinkedIn
- YouTube
- inteliScience on YouTube