The cut that doesn't pass inspection

The problem

At cutting pace, the defect that gets through is the one the eye cannot catch.

At a steel service center's cutting pace, a defect — burr, dent, rust — can slip past the human eye. The piece reaches the customer's facility, and that's where the out-of-tolerance cut or surface damage gets found.

• A steel warehouse cuts to size all day, in short batches, changing grade and thickness constantly. A burr along one edge, a dent from handling, a patch of rust on a sheet that sat too long: any of them can pass.
• The eye that is supposed to catch them belongs to somebody who is also loading, strapping and moving on to the next order, with a customer waiting at the counter for a job promised within the hour.
• The piece reaches the workshop, and that is where the defect is found — on the customer's bench, with their own job already late.
• And the cost is never just the piece: it is the return, the replacement cut, the transport paid twice and a metal fabricator who starts phoning somebody else.

How it fits the IRIS system

Edge — a camera over the cutting line and a model trained on your own material.

An Edge camera above the cutting line, with a CNN trained on correct and defective pieces of that specific material, inspects every piece in milliseconds and ejects it before packaging.

Inference runs locally, in milliseconds, on the machine itself: it does not depend on the warehouse network or on the cloud, and it keeps up with the pace of the line. The piece is ejected before it is packed, which is the only moment when fixing it is still cheap.

See the full IRIS architecture →

Before and after

Today's inspection versus Edge inspection

Aspect	Visual inspection	With iLEAN Edge
Coverage	Whatever the eye catches while loading	Every piece, at cutting pace
Burr along the cut edge	Found by the customer	Ejected before packing
Handling dents and rust	Argued about after delivery	Classified piece by piece
Point of detection	The workshop's bench	The cutting line
Cost of a miss	Return, recut and double transport	One piece recut on the spot
Borderline pieces	Pass, because nobody is sure	Escalated to a person

Defect discovered at the customer's facility → defect ejected before leaving the warehouse.

Impact estimate

Impact estimate — to be validated with your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• At least a 30% reduction in incidents of this type, as an estimate to validate against your own returns log.
• The cost avoided is never the piece: it is the return, the replacement cut and the transport paid twice.
• And the customer relationship, which is what a stockholder actually competes on once the price per kilo is the same everywhere.
• Plus a defect history by grade, thickness and supplier that tells you whether the problem is your shear or the material you bought.

Estimated reduction of at least 30% in this type of incident. Estimate to be validated.

And the fair question from the production manager

“What if it rejects good pieces?” — the false positive is the real risk of any vision system, which is why the model is trained on good and bad pieces of your own grades and thicknesses, under your own lighting, rather than on a generic model. Borderline pieces are not simply thrown out: they are escalated for a person to decide, and every decision feeds back into training. On the classification itself the task is anchored, where the best models drop below 1.5% error [1].

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.