OEE in a multi-PDO cheese plant — the changeover between European formats stops eating hours.

The problem

Every PDO change is a linear chain of tasks — and OEE picks up the bill.

A multi-PDO cheese plant lives with three constraints that drag OEE down on every recipe change:

1. The "just in case" CIP — cleaning between two different PDOs is done with a generous margin because nobody wants to risk cross-contamination. If it is not measured, it is compensated with time.
2. Adjustments run in series — molds, presses, brining room, ripening room. Each one waits for the previous one to finish even though they could move in parallel. The linear chain doubles the time actually needed.
3. Start-up at half speed — the first run after the change does not come out at nominal speed because the operator plays it safe until he can see that "everything is fine". Without data, that "everything is fine" arrives late.

The classic setup (shift lead + checklist + Excel sheet) works and keeps the plant hygienic. The problem is not the what — it is the how much: a linear changeover takes hours of OEE away from the plant, hours that are sitting right there, waiting for someone to orchestrate them in parallel.

How it fits the IRIS system

iLEAN Brain does not remove hygiene steps — it overlaps the ones that can be overlapped.

The problem with the multi-PDO changeover is not excess caution — it is a lack of orchestration. The caution is necessary, but it can now be spread in parallel across the vat, the press, the brining and the ripening room without losing the hygiene sign-off. iLEAN acts as the putty between the production order, the real state of the equipment and each operator's plan, without touching the ERP, the MES or the protected hygiene procedures.

Brain (agents in Central) cross-references the real state with the next PDO. Connect takes the sub-task to each operator's earpiece. Edge validates the cleaning with data. The person signs off the start-up.

The iLEAN pieces applied to the multi-PDO changeover:

• Brain (agents in Central) — cross-references the production order for the next PDO with the current state of every piece of equipment on the line and generates a plan of parallel sub-tasks, each with an owner and a window. The vat's CIP cleaning starts at the same time as the mold adjustment and the press calibration — not after. The agents follow progress and replan if a sub-task drifts.
• Connect — delivers each sub-task to the earpiece of the operator who has to carry it out, full duplex. The operator confirms by voice; the quality manager receives the electronic signature. It also captures the data wherever it comes from — the MES, the CIP record, the plant notebook.
• Edge — verifies the critical cleaning points (vat, bottom valve, cutting harp) with machine vision and cross-references the photo with the CIP curve. If everything lines up, it releases the start-up; if not, it holds the line. It works with no network.

See the full IRIS architecture →

Before and after

Linear changeover vs. changeover orchestrated with iLEAN Brain

Aspect	Classic recipe change	With iLEAN Brain + Connect + Edge
Task sequence	Linear — each machine waits for the previous one	In parallel wherever possible, with a plan per operator
"Just in case" CIP	Generous time for safety	Measured time, validated with real data
Hygiene validation	Checklist signed at the end	Triple cross-check: CIP + Edge photo + operator signature
First start-up after the change	Half speed "just in case"	Nominal speed as soon as the data confirms the state
Changeover traceability	Rebuilt by hand for the audit	Automatic dossier for every recipe change
Visible OEE	Excel sheet at the end of the shift	Continuous availability / performance / quality curve

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• European cheese plant with several PDOs under the same roof, several recipe changes a week, a known current OEE and an identified bottleneck in the changeover.
• Brain pilot on one line (orchestration agents + Connect on the earpiece + Edge at the critical cleaning points). First value expected within a few weeks: an OEE baseline per PDO that the plant did not have.
• Indicative payback between 4 and 9 months, depending on the number of recipe changes per year and the hourly cost of a plant stopped for changeover.
• Recovery of OEE hours per PDO change of ≥ 30% as a conservative floor. The hard lever is the overlapping of tasks that are run in series today with no need, and nominal speed recovered from the very first start-up.

And the plant manager's reasonable doubt

"What if the agents orchestrate something wrong and we skip a hygiene step?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the agents simply cross-reference the production order with the real state of the equipment and propose a plan, the best models brought error below 1.5% ^[1]. And even so, the changeover never runs itself: iLEAN proposes and orchestrates, the operators confirm every sub-task and the manager signs the start-up. The three safety rings exist precisely for this.

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