The lactose-free changeover, verified by evidence rather than judgment

The problem

The CIP is logged. The physical state of the points it does not reach equally is not.

The CIP runs correctly and is logged in the process control system. That part is solved. What is not logged is the observable physical state of the points the CIP does not reach equally: the filler nozzle, the dosing head, the reel change area, the immediate surroundings of the machine. The release signature rests on a visual inspection made in a hurry, at the end of a stoppage the plant wants to shorten for perfectly legitimate productivity reasons. It is a structural conflict between two correct incentives. And it is worth saying plainly: the "lactose-free" declaration on the pack is a declaration to the consumer. It does not tolerate an "I think it was clean". If the startup goes wrong, the first batch of the new production comes out contaminated with that declaration printed on it.

• The CIP runs correctly and is logged in the process control system. That part is solved. What is not logged is the observable state of the filler nozzle, the dosing head, the reel change area and the immediate surroundings of the machine.
• The release signature rests on a visual inspection made in a hurry, at the end of a stoppage the plant wants to shorten for perfectly legitimate reasons. It is a structural conflict between two correct incentives.
• The lactose-free declaration on the pack is a declaration to the consumer. It does not tolerate an I think it was clean. If the startup goes wrong, the first batch comes out contaminated with that declaration printed on it.
• The same applies to the changeover between dairy and juice on a shared line: the valves make it possible in hardware, but the decision to start depends on someone's judgment alone.

How it fits the IRIS system

Edge with JIDOKA AI — the line stays locked until every point is green and quality signs.

Edge with JIDOKA AI. The flow:

The camera does not decide alone: each point returns green, amber or red with its image alongside, compared against the learned clean-and-ready state for the incoming product, and it is the quality manager who signs on that evidence. The human signature is not replaced; it is backed. The machine does not decide — it prevents a blind startup.

• Fixed Edge cameras at each critical changeover point, defined with the quality manager during implementation.
• Each camera compares the current state against the learned reference state of "clean and ready" for the incoming product.
• Each point returns green, amber or red, with its image alongside so the decision is inspectable rather than a black box.
• The line stays locked until every point is green and the quality manager signs.
• The signature, the images and the timestamp are tied to the incoming batch and feed the audit dossier directly.

The human signature is not replaced: it is backed by evidence. The machine does not decide; it prevents a blind startup.

See the full IRIS architecture →

Before and after

Today's changeover versus the verified changeover

Aspect	Today	With iLEAN Edge
What the release signature rests on	A visual check in a hurry	N timestamped images against reference
Filler nozzle, dosing head, reel area	Looked at, if there is time	Each point green, amber or red
Can the line start before quality signs?	Yes	No: locked until every point is green
Changeover duration	As long as the manual inspection	15-30 % shorter: verification is instant
Cross-contamination risk	Diffuse	Reduced by an order of magnitude
Evidence for the audit dossier	None	Signature, images and timestamp per batch

from a signature without evidence, to a signature with N timestamped images compared against reference. From a critical changeover that lasts as long as the manual inspection does, to a changeover 15 % to 30 % shorter because verification is instant. From diffuse cross-contamination risk, to risk reduced by an order of magnitude.

Impact estimate

Estimated impact — to validate with your own 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 6-12 months, combining shorter changeovers and reduced cross-contamination risk between variants.
• Changeover time 15 % to 30 % shorter, because the verification is instant and the margin added for lack of evidence disappears — this is where SMED AI comes in.
• The first batch of the lactose-free run stops being born under suspicion: it is validated before the line unlocks.
• And the signature, the images and the timestamp feed the audit dossier directly, so the changeover evidence exists before anyone asks for it.

estimated payback 6 to 12 months, combining reduced changeover time (where SMED AI comes in) and reduced cross-contamination risk between variants. *Estimate to validate.*

And the fair question from the production manager

«What if the camera locks the line for no reason?» — that is the right objection, because the changeover sits on the shift's critical path. Each point is compared against the reference state for that specific product, and in case of doubt the system does not block silently: it shows the amber point with its image and escalates to the quality manager, who decides on the evidence. It stops on a verifiable discrepancy, not on model uncertainty — and comparing a fixed view against a learned reference is an anchored task, 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.