MACHINE VISION · INDUSTRIAL AUTOMATION

Every unit checked.
At line speed.

Machine vision AI that reads date codes, barcodes and labels and inspects product as it moves — on every unit rather than a sample. We build and train the AI; your line hardware and your integrator stay exactly where they are.

The problem

At line speed, nobody is really checking.

A person can verify a date code on one carton. They cannot verify every carton on a line running at production speed, for a full shift, without missing some. So most lines check a sample and hope the rest were fine — until a batch reaches a retailer with an unreadable code, a wrong label or a defect that should never have left the plant.

Dates, batches and barcodes at speed

Printed expiry and batch codes read and verified on every unit as it passes — not a sample. A smudged, missing or wrong date is caught on the line rather than at a distribution centre or, worse, on a shelf.

Barcode and label verification

Barcodes read at production speed and checked against what should be there, so a mislabelled run is stopped in minutes rather than discovered in a recall.

Visible defects and presentation

Fill level, cap and seal integrity, label placement and alignment, damage and contamination — the visible faults that determine whether a unit is saleable.

Decisions the line can act on

Detection is only useful if something happens. Models output a decision the line can use, driving reject, divert and sorting mechanisms through your existing control system.

Our scope

We do the AI. Not the hardware.

This is deliberately a narrow engagement. Kevlar Risk supplies the machine vision AI and analytics layer: the models, the training on your product and your conditions, and the integration that turns a detection into something your line can act on. Cameras, lighting, mounting, guarding and mechanical work stay with your equipment supplier or integrator, who knows your line and already supports it. That keeps the project focused on the part that is genuinely hard — making a model work reliably on your product, at your speed, in your lighting — and stops us duplicating capability you are already paying for.

How it works in practice

Trained on your product, not a demo dataset.

A model that reads date codes on one manufacturer’s cartons will not necessarily read yours. Print quality, substrate, lighting, line speed, vibration and the angle the product presents at all change the problem. So the work starts with footage from your actual line, and the honest first conversation is about what the line looks like and what currently gets missed — not about accuracy percentages that mean nothing until a model has seen your product.

Book a free site assessment.

Tell us about your site and we will come back within one business day. No obligation, no pressure.