Format change verified

The problem

After a format change, the first minutes are baked on trust.

If a format changeover isn't perfectly set, the first minutes of production come out off-weight or off-shape. Today it depends on an operator eyeballing it and signing off "ready", with no objective evidence.

• Switching from large sliced pan bread to burger buns means recalibrating the divider, the molder and the cutter: piece weight, dough shape and cut all change at once.
• If one of them is not perfectly set, the first minutes of production come out off-weight or off-shape and are already proofing before anyone notices. By the time they reach the oven exit, the dough and the oven time are both spent.
• Today the decision depends on an operator eyeballing it and signing off as ready, with no objective evidence behind the signature. It is not carelessness: the changeover sits on the critical path and the shift clock is running.
• And multi-SKU lines change format several times a week, so the exposure repeats with every switch, on every shift that runs a changeover.

How it fits the IRIS system

Edge with JIDOKA AI — no full-speed startup until every critical point checks out.

Edge + JIDOKA AI. Industrial cameras with vision models trained to recognize "correct format vs. weight or shape deviation" at each critical point. Startup blocked until visual OK + supervisor sign-off.

The camera does not replace the operator's judgment: it gives it evidence. Each critical point is compared with the new format's reference, and the supervisor signs on what the cameras saw instead of on a feeling. What changes is not who decides, but what the decision rests on.

See the full IRIS architecture →

Before and after

Blind format startup versus format startup verified by camera

Aspect	Today	With iLEAN Edge
Divider piece weight	Checked on a few pieces	Confirmed against target and tolerance
Molded dough shape	Judged by eye	Compared with the format reference
Cutter setting	Found wrong at the slicer	Verified before full speed
Ready-to-run signature	A tick with nothing behind it	Visual OK plus supervisor sign-off
Minutes of off-spec product	Baked and then scrapped	Held before they reach the oven
Evidence of the changeover	None	Images stored per change

Impact estimate

Impact estimate — to be validated against your changeover log.

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.

• Payback depends on changeover frequency: we do not fix a month range because a line that switches once a month and one that switches daily are different cases.
• On multi-SKU lines that change format several times a week, startup scrap after each change is where the return accumulates: dough that was divided, molded and proofed wrong is lost along with the oven time it used.
• The blind startup with off-spec minutes becomes a startup verified with objective visual evidence at every critical point, signed by the supervisor on what the cameras saw.
• And every changeover leaves its own images, which the audit dossier (case 11) uses without anyone rebuilding it. The evidence that authorized the start is the same evidence the auditor later reads.

Reduced startup scrap after format changes; payback depends on changeover frequency (multi-SKU lines change several times a week). Estimate to validate.

And the fair question from the production manager

“What if the cameras hold the line for nothing while the shift clock runs?” — that is the right objection, because a changeover sits on the critical path. Comparing a piece's weight and shape with a known format reference is an anchored task, where the best models drop below 1.5% error [1]. When the doubt is the model's, it does not block silently: it escalates to the supervisor, who decides on the images.

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