The most expensive moment of your day depends today on one person's eyesight

The problem

The most expensive moment of the day is signed off with no evidence behind it.

The arbor has to be stripped and the knife and spacer set matching the new width pattern has to be mounted. The cost of getting it wrong is not the first out-of-tolerance strip that gets scrapped: it is doing the whole setup again. And there is a second check, just as critical and far quieter: leaving the rolls clean. A single steel particle stuck to a roll marks every meter of the prepainted coil that follows. It is the classic task done properly 95% of the time, and that remaining 5% is a downgraded coil. Both depend on the operator's skill and eyesight, and on a signature with no objective evidence behind it.

• The arbor is stripped and the knife and spacer set matching the new width pattern is mounted. The cost of getting it wrong is not the first out-of-tolerance strip: it is doing the whole setup again.
• And a second check, just as critical and far quieter: leaving the rolls clean. A single steel particle stuck to a roll marks every meter of the prepainted coil that follows.
• It is the classic task done properly 95% of the time, and the remaining 5% is a downgraded coil.
• Both checks depend on the operator's skill and eyesight, and on a signature with no objective evidence behind it. When a coil comes back marked, the investigation runs into a ticked box and nothing else.

How it fits the IRIS system

Edge plus JIDOKA AI — the line does not start until every camera clears and the lead signs.

Edge plus JIDOKA AI. Fixed cameras over the knife arbor and over the rolls. Trained networks recognize the mounted set against the active run's reference, and tell a clean roll from a roll with residue. The line does not start until every camera clears and the lead signs off on that evidence. SMED AI uses the same record to sequence the changeover itself: which set to prepare before stopping and in what order to mount.

The cameras do not decide alone: they compare the mounted set against the reference of the active run and tell a clean roll from a roll with residue, and it is the lead who signs on that evidence. SMED AI uses the same record to sequence the changeover: which set to prepare before stopping and in what order to mount it.

See the full IRIS architecture →

Before and after

Today's changeover versus the validated changeover

Aspect	Today	With iLEAN Edge
Verifying the mounted set	One person's eyesight	Compared against the run reference
Roll cleanliness	A glance, 95% of the time	Residue detected before start
Signature	A ticked box	On visual evidence, point by point
A repeated setup	Discovered on the first strip	Does not happen
Changeover time	Unmeasured, varies by shift	Measured and comparable
Sequencing the changeover	Each operator's routine	SMED AI: what to prepare, in what order

from signing blind to signing on visual evidence point by point; from repeated setups due to incorrect mounting to none. And changeover time becomes measured and comparable across shifts and operators, which is the precondition for reducing it.

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.

• Estimated payback 5-10 months, counted on setup minutes recovered and on prepainted coils not downgraded.
• In a plant with spare capacity, setup minutes convert directly into sellable tonnage with no CAPEX.
• From repeated setups due to incorrect mounting to none, and from signing blind to signing on evidence.
• And changeover time becomes measured and comparable across shifts and operators, which is the precondition for reducing it.

estimated payback 5-10 months, counted on setup minutes recovered — which in a plant with spare capacity convert directly into sellable tonnage with no CAPEX — and on prepainted coils not downgraded. *An estimate to be validated*.

And the fair question from the production manager

“What if the camera blocks the line for no reason?” — that is the right objection, because the changeover sits on the shift's critical path. Comparing a mounted set against the reference image of that specific run, at a fixed camera position, is an anchored task where the best models drop below 1.5% error [1]. And in case of doubt the system does not block silently: it shows the lead the frame and the reference side by side, and the lead signs or corrects. It stops on a verifiable mismatch, not on model uncertainty.

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