EngineeringNewVerified Real deployment at a named or precisely described company, with a measured result from a primary source.
AI runs a first-pass check of drawings and models before the design review, so engineers discuss real issues (INFICON, McDermott)
Added 2026-10-21 · Last updated 2026-10-21
Role
Engineering (design engineering and drafting: drawing checking and design review) · Typical users: design engineer, drafter or CAD designer, drawing checker, engineering manager, manufacturing engineer
The problem
Drawing checks are repetitive and detail-heavy: missing dimensions, conflicting tolerances, wrong hole callouts, title blocks, material notes, and the same mistakes the team has flagged before. They are easy to miss when engineers are at capacity, and a mistake found late costs a new loop through drawings, suppliers and parts. At INFICON, reviewers often saw a design for the first time in the meeting, so a one-hour review turned into several hours and the format depended on who led it; its engineering manager estimates that a significant late-stage problem can nearly double an 8 to 12 week development cycle. At McDermott, checking piping and instrumentation diagrams (P&IDs), some up to 100 pages, is done several times per project by experienced engineers and can take up to tens of hours per project.
How AI helps
Each new drawing or drawing update goes through an AI checker before a person reviews it. At INFICON, CoLab's AutoReview runs a first pass on 2D drawings and 3D parts against engineering standards and the team's own checklists, and puts its markups on the design; engineers then review those findings and add their own feedback asynchronously before the meeting. At McDermott, the in-house ReCo tool uses image recognition to find common mistakes in P&IDs (and later in piping isometrics, including third-party PDFs); the engineer decides whether each flag is a real error and how to fix it. The AI covers the baseline checks; engineers keep the judgment calls.
Step by step
- Collect the last six months of review findings and engineering changes, and sort them by type: which mistakes come back again and again.
- Turn your drafting standard and those recurring findings into a written checklist; INFICON used AI to pull its own requirements out of long industry standards into checklists and templates.
- Run the AI first pass on new drawings alongside the normal check for a few weeks and log each flag as real issue, false alarm or missed.
- Make the first pass a required step before the review meeting, so reviewers arrive with findings instead of discovering the design in the room.
- Keep sign-off with the responsible engineer and checker; the AI does not approve or release drawings.
- Track review closure time, findings per drawing, engineering changes after release and prototype loops.
Tools
- CoLab with AutoReview (AI first pass on drawings and 3D models against standards and checklists): used at INFICON
- McDermott's in-house ReCo tool (deep-learning image recognition), built on VIKTOR: used for P&IDs and piping isometrics
- Other tools to evaluate: the checking functions of your CAD or PLM (for example SOLIDWORKS Design Checker, NX Check-Mate), drawing-reading AI such as Werk24 or CADDi, and DFM checkers from your CAD vendor
- Any LLM that reads images (ChatGPT, Claude, Copilot, Gemini) with the prompt below, on one exported drawing and your checklist, for a first test
Result
INFICON (measurement and control instruments, Bad Ragaz, Switzerland; teams in Europe, the US and Asia), CoLab with AutoReview since March 2026: design-review closure time from 62 days to 7 (89% shorter); design feedback captured from about 50 to more than 400 pieces a month; one full development program reached first-time-right with no additional engineering changes, which the team estimates removed four to six weeks from that cycle. These results come from the whole move to CoLab's shared review workflow, of which the AI first pass is one part (CoLab case study, vendor-published). McDermott (engineering and construction for the energy industry, 54+ countries), ReCo tool built on the VIKTOR platform: checks that can take up to tens of hours per project are done in seconds, at 100 P&IDs in about 3 minutes; the tool also gives junior engineers a reference manual of common P&ID mistakes (VIKTOR customer interview with McDermott's AI projects manager, 2023).
As reported by the source; most figures are vendor or self-reported and not independently verified.
Human review: cautions
- The AI is a first pass, not the checker of record: the responsible engineer and checker still review and sign every drawing.
- Expect false alarms and misses at first; log them and refine your checklist before relying on the tool.
- At INFICON the results come from the whole shared review workflow, not the AI check alone.
- Figures are vendor-published (CoLab case study; McDermott interview on the VIKTOR site), and the INFICON development-time saving is the team's own estimate.
Try it yourself
You are a senior drawing checker at a manufacturer. I will paste or attach: (1) one engineering drawing (image or PDF), with the 3D model's key data if available [material, mass, critical features]; (2) our drawing checklist and drafting standard (ASME Y14.5 or ISO GPS, title block rules, our notes and finish conventions); (3) the main manufacturing process for the part (machined, sheet metal, molded, cast, welded); (4) findings from past reviews that keep coming back, if any. Do this: 1. Check the drawing against each checklist item and list findings in a table: item, location on the drawing (view, zone or balloon), what you see, why it is a problem, suggested fix, severity (must fix / should fix / question). 2. Flag missing or duplicate dimensions, conflicting tolerances, incomplete hole and thread callouts, GD&T without a clear datum scheme, missing material or finish notes, and title block gaps. 3. List manufacturability questions for the stated process (tool access, thin walls, inside radii, bend relief, draft), as questions for the engineer. 4. Say which past recurring findings appear again. 5. Only report what you can read; if a view or text is unreadable, write 'not readable' instead of guessing. The engineer and checker decide and sign off; do not approve the drawing. Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.
Want more like this? Join the newsletter
Related cases
- Automatic feasibility check of customer drawings at RFQ intake
- AI reads customer tenders and specification packages clause by clause and builds the deviation list (ROBEL, Bosch Manufacturing Solutions)
- AI drafts and checks PFMEA / DFMEA
- AI proposes material formulations so R&D runs fewer lab experiments (Resonac, formerly Showa Denko)
Sources
- https://www.colabsoftware.com/case-studies/inficon
- https://www.colabsoftware.com/product/autoreview
- https://www.viktor.ai/customer-cases/46/mcdermott-artificial-intelligence-AI-piping-instrumentation-diagram-image-recognition
Weekly: 1–2 practical AI use cases for your plant
Free, one email a week. By subscribing you agree to receive the weekly newsletter (and its sponsor message) from AI Agents for Manufacturing. Every email includes an unsubscribe link, and you can unsubscribe anytime. See our Privacy Policy.