Cleaning between products stops being signed for and starts being proven

The problem

Cross-contamination is caught in the lab or at the customer, never at start-up.

Cross-contamination is not caught at start-up. It is caught when the lab analyzes the first container, or when the customer finds it. By then there is a whole batch to rework or destroy, and if the contaminated product was going to a food-grade or high-purity customer, there is also a conversation that can cost the account. The current barrier is one person's declaration: somebody signs that the line is clean, with no objective evidence for each point. It is a real barrier, but a fragile one, and it leaves no trace an auditor can use. And there is a silent cost the plant pays every day: every changeover ties the line up for one to three hours, and that time stretches out of caution. Without objective evidence, everyone would rather over-rinse.

• Between an aromatic solvent and a food-grade or high-purity blend, the only thing separating a good batch from a contaminated one is that the purge and the cleaning were done properly. Today that is signed, not shown.
• The signature is one person's declaration with no objective evidence per point. It is a real barrier, but a fragile one, and it leaves nothing an auditor or a customer can use.
• When the contamination is found — by the lab on the first container, or by the customer — there is a whole batch to rework or destroy and, if the customer was food-grade, an account at risk.
• And a silent cost every day: each changeover ties the line up for one to three hours, and the time stretches out of caution. Without evidence, everyone prefers to over-rinse.

How it fits the IRIS system

Multi-point Edge with JIDOKA AI and SMED AI — the line protects itself, then speeds up.

multi-point Edge + JIDOKA AI + SMED AI.

Each camera compares its critical point with the purged-and-clean reference for the incoming product family: visible residue, remaining color, deposits, the right filling head. While any camera withholds approval, the line does not start. When all pass, the quality manager signs on evidence instead of on judgment, and the over-rinsing that caution was adding becomes unnecessary. The result is a changeover that is both safer and shorter, which is the combination nobody believed possible while it was signed by hand.

• Edge cameras at the critical changeover points: blend tank opening, pump, hose section, filling head and drip tray.
• Each camera compares the current state with its "purged and clean" reference: visible residue, remaining color, deposits, and the correct head for the new product family.
• While any camera withholds approval, the line does not start. That is JIDOKA AI: the line protects itself.
• When they all pass, the quality manager signs on the evidence rather than on judgment, and that evidence files itself.
• With cleaning demonstrated, over-rinsing becomes unnecessary: SMED AI cuts changeover time without lowering the guarantee.

See the full IRIS architecture →

Before and after

Today's changeover versus the proven changeover

Aspect	Today	With iLEAN Edge
What the signature rests on	A declaration	Photographic evidence of every point
Contaminated batch	Found in the lab or at the customer	Start-up blocked before the first liter
Blend tank, pump, hose, head, tray	Checked by eye, if at all	Compared with the reference state
Wrong head for the product family	Discovered on the first drum	Detected before start-up
Changeover time	1-3 h, inflated by caution	Tightened, with a guarantee
Trace for the auditor	A ticked checklist	Evidence filed by changeover

from a signature without evidence to a signature on photographic evidence of every point · from contaminated batches caught in the lab or at the customer to start-ups blocked before the first liter · from changeover times inflated by caution to times tightened with a guarantee.

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 6-12 months, depending on changeovers per day.
• At least 30 % less changeover time, because over-rinsing stops being the only way to feel safe. On a line that changes product twice a day, that is capacity.
• Virtual elimination of batches reworked or destroyed for cross-contamination, which is the cost that never fits in a spreadsheet until it happens.
• And the first liter of the new product is no longer born under suspicion: it is released against evidence, not against a signature.

at least 30% less changeover time and the virtual elimination of batches reworked for cross-contamination. Estimated payback of 6 to 12 months depending on changeovers per day. *Estimate to validate.*

And the fair question from the production manager

“What if a camera blocks the line for no reason?” — that is the right objection, because the changeover sits on the shift's critical path. Comparing a critical point with the reference state for that specific product family 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 escalates the image to the quality manager, who signs on what they see. It stops on a verifiable residue, not on model uncertainty.

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