Home / Machine Vision
AI quality inspection for manufacturing

Machine Vision Solutions Built on Your Parts

Custom machine vision and AI quality control that inspects every part on the line — catching defects, missing components and foreign objects that a human eye misses at speed. Models trained on your parts and your defects, deployed at the edge. You own the code.

Last updated: 11 June 2026

Summarize with AI Prompt copied — paste it into the chat
Share
In short

A machine vision project trains AI to inspect your parts automatically — spotting defects, verifying assembly, detecting foreign objects and reading labels. Crux Digits builds it custom on your data with fixed prices: a €2,500 audit, a €20,000 proof-of-concept, and production systems from €50,000. EU AI Act-ready, and you own the code.

What machine vision inspection covers

Machine vision software turns cameras into an inspector that never blinks. We build systems that grade parts against your own quality standard — not a generic library — so the AI flags exactly the defects your line cares about. Typical inspection tasks we automate:

How we build a machine vision system

This is engineering, not a plug-in. We start at the physics: camera choice, resolution, optics and lighting geometry — because most vision projects fail on lighting, not on the model. Then we collect and label images of your real parts (good units and every defect class), train models on your data, and validate against a held-out set until precision and recall clear your acceptance criteria. Finally we deploy to the line — on an edge device or industrial PC for low-latency, on-premise inspection with no dependency on the cloud. It pairs naturally with predictive maintenance, and builds on the same core as our computer vision work.

Inspection task, what the AI does, and the outcome

Each inspection task maps to a specific model behaviour and a measurable result on the line:

Inspection taskWhat the AI doesOutcome
Surface & defect checkClassifies and localises flaws pixel-by-pixel against your defect classesBad parts rejected automatically; scrap and escapes drop
Assembly verificationDetects presence, position and orientation of each componentNo incomplete or mis-built units reach the customer
Label & code (OCR)Reads and checks codes, dates and barcodes for legibility and matchMislabelled or unreadable product caught before dispatch

Fixed-price bands

No open-ended day rates. Every engagement runs on fixed prices so you can budget with confidence: a €2,500 audit to assess feasibility, imaging and ROI; a €20,000 proof-of-concept to prove detection accuracy on your parts; and production systems from €50,000 for line-integrated inspection. See the full breakdown, or start with a free AI-readiness scan.

How we ship: audit → PoC → production

We de-risk in stages. The audit confirms whether vision is the right tool and how to image your parts. The proof-of-concept trains models on collected samples and proves accuracy against your acceptance criteria — you see the numbers before committing to build. Production hardens the system, integrates it with PLCs and the line, and hands over documentation. Everything is EU AI Act-ready and the code is yours — no lock-in, no per-inspection licence.

FAQ

Frequently asked questions

What does a machine vision project cost?

Fixed prices: €2,500 feasibility audit, €20,000 proof of concept, production inspection systems from €50,000. The proof of concept proves detection accuracy on your own parts. The audit tells you which band your project needs before you commit to anything. All prices exclude VAT.

  • Timelines: audit 1–2 weeks, proof of concept 4–6 weeks, production from 8–12 weeks.
  • Cameras, lighting and mounting hardware sit outside these prices — the audit specifies what your line needs.
  • Lighting is the most underestimated cost. Getting it wrong is the most common reason a vision pilot fails.

Do you use off-the-shelf models or train on our parts?

We train on your parts and your defect classes. Off-the-shelf models rarely match a specific production line, so we collect and label images of your real good and bad units, then validate against a held-out set until accuracy clears the criteria you set.

  • You set the acceptance threshold before we build, not after — usually as a false-reject and false-accept rate.
  • Rare defects are the hard case: if a fault appears once a month, expect a collection period before training.
  • Labelling is a joint effort; your QC operators know the defect classes better than any external team.

What is the difference between 2D and 3D machine vision?

2D inspects flat features; 3D measures shape and depth. A 2D camera image handles surface defects, print, labels and presence checks. 3D vision measures dimensions, warpage, fill level and assembly geometry. 2D is cheaper and faster, so use it wherever it suffices.

  • Choose 3D when the defect is geometric — a dent, a gap, a height difference a flat image cannot show.
  • Choose 2D for print quality, contamination, scratches, orientation and presence or absence.
  • The audit recommends one for your specific part rather than defaulting to the more expensive option.

Can it run on our line without the cloud?

Yes. We put it on a small computer at the line itself, so it inspects fast and keeps going if the internet drops. Your images never leave the building, which makes the privacy conversation a lot shorter too. No cloud account needed.

  • Edge deployment is the norm for inspection: cloud round-trips are too slow for line speed.
  • Running on-premise means no images of your products or processes sit on a third-party platform.
  • You own the code, so the system keeps running regardless of any supplier relationship.

Is machine vision affected by the EU AI Act?

Yes, but industrial quality inspection is generally lower-risk. Those tiers still expect documented data, traceable decisions and clear human oversight, all of which we build in. Obligations get heavier if the system is a safety component of the machine itself.

  • Regulation (EU) 2024/1689 sets the risk tiers; Article 4 requires AI literacy for the operators using the system.
  • Vision used for machine safety rather than quality can fall under the Machinery Regulation as well — classify before you build.
  • Check your case in four questions with the free EU AI Act risk checker.

See if machine vision fits your line

Start with a fixed-price €2,500 audit: we assess your parts, imaging and ROI, and tell you exactly what a working system would take. No open-ended commitment.

Book your machine vision audit