Allergen cleaning, validated by camera

The problem

The procedure is good. The problem is what proves it was followed.

today, cleaning is validated with a sign-off and no objective evidence; if it starts poorly, the first batch gets contaminated.

• Going from a formula with mustard, celery or sesame to one that declares none of them is the most sensitive moment in the plant, and it happens several times a week.
• The cleaning procedure exists and is well written. What fails is the evidence: a signature on a sheet, taken under time pressure because the changeover sits on the shift's critical path.
• Powder hides where a glance does not reach: the underside of the ribbon, the corner of the load hopper, the sampling drawer, the transfer between conveyor and bagger. A fine seasoning blend gets everywhere, and nobody is going to dismantle the mixer twice a day to look.
• If the mixer restarts with residue, the first batch is contaminated and declares an allergen it does not list. And when the investigation comes, it runs into a ticked box and no proof.

How it fits the IRIS system

Edge with JIDOKA AI — the mixer does not restart without conformity and a sign-off.

Edge cameras at N critical points + JIDOKA AI — no startup until every camera confirms the correct state and the quality manager signs off.

The camera does not decide alone: it checks each critical point against its clean reference image and it is the quality manager who authorizes on that evidence. What changes is not who decides, but what they decide on — and what is left behind afterwards, which today is a signature and tomorrow is a dated set of photographs.

See the full IRIS architecture →

Before and after

Today's changeover versus the validated changeover

Aspect	Today	With iLEAN Edge
Validating the cleaning	A signature on a sheet	N reference points checked
Residue in the load hopper	Found later, or never	Detected before startup
Under the ribbon and in the drawer	Out of reach of a glance	Covered by a fixed camera
First batch after the change	Under suspicion	Authorized on evidence
Evidence for the auditor	A ticked box	Dated photographs with a sign-off
Changeover frequency	More changes, more exposure	More changes, the same check

sign-off with no evidence → objective photographic evidence with history.

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.

• No standalone payback figure is claimed here: the brief measures this case by cross-contamination risk, and that is not honestly expressed in months.
• What changes is that the cleaning stops being validated by signature and starts being validated by objective evidence, with a dated history behind every single changeover.
• That evidence is also what closes a customer audit question in minutes instead of opening a corrective action you then have to answer in writing.
• And it pairs with SMED AI: the time the changeover gains is the margin people add today precisely because there is no evidence.

reduced cross-allergen contamination risk. *Estimate to validate*.

And the fair question from the production manager

"What if the camera blocks the mixer for no reason?" — that is the right objection, because the changeover is on the critical path. Conformity is checked against the clean reference for that specific point, an anchored comparison where the best models drop below 1.5% error [1], and in case of doubt the system does not block silently: it escalates to the quality manager, who signs on the evidence.

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