The foam carousel that finally talks

The problem

The carousel knows why the cushion collapsed, and forgets it every few seconds.

Replacing it costs a seven-figure sum and weeks of downtime. In a just-in-sequence operation those weeks do not exist. So the machine stays, and the problem stays with it. Foam is where a good share of the comfort defects the customer finds later are born: hardness out of range, collapse, voids, poor demoulding. The panel knows why it happened, low temperature at station seven, curing time cut short to recover cadence. But that data dies on the screen every few seconds.

• The foam carousel is ten to twenty years old and lives on an isolated network by the manufacturer's decision. Replacing it is a seven-figure sum and weeks of downtime that a just-in-sequence plant does not have, so the machine stays and the problem stays with it.
• Its panel shows everything that matters: mold temperature per station, curing time, mix ratio, pressure and active alarms. None of it survives the next refresh of the screen.
• Hardness out of range, collapse, voids, poor demolding: a good share of the comfort defects the customer finds later are born right here, and by the time they show at trim the cushion has left the mold and the cause has left the screen.
• Low temperature at station seven, curing time cut short to recover cadence: the cause was on the panel, but nobody can correlate it with the defect found at trim or with the claim that comes back weeks later.

How it fits the IRIS system

Connect in photo mode on the panel — read from outside, the machine stays exactly as it is.

Connect in photo-on-panel mode. A fixed industrial camera with a stable frame, a grounded reading model interpreting every capture, and the reading structured as a time series per station and per mold. We do not touch the controller, we do not touch the machine network, we do not open an integration project with the manufacturer, we do not void the warranty. That is the elegant part: the carousel remains exactly the same machine. It simply talks now. And what it says gets cross-referenced with the manufacturing system to know which reference was running at each moment. From there you can do what you cannot do today: correlate foaming parameters with defects found at trim, and with the claim that comes back from the customer. Root-cause analysis per mold in minutes, not weeks.

No controller touched, no machine network touched, no integration project with the manufacturer, no warranty voided. The carousel stays the same machine; it simply talks now, and what it says is cross-referenced with the reference that was running on each mold at each moment. The correlation between a cold station seven and the collapsed cushions found two hours later at trim becomes a query, not a theory.

See the full IRIS architecture →

Before and after

Today's carousel versus the carousel being read

Aspect	Today	With iLEAN Connect
Mold temperature and curing time	On the screen, then gone	Time series per station and per mold
Comfort defect found at trim	No link to the foaming curve	Correlated with the mold that produced it
Root cause on a customer claim	Weeks, if at all	Minutes
Curing cut short to recover cadence	Invisible	A deviation with a timestamp
Carousel controller and network	—	Untouched, warranty intact
Replacement CAPEX	In the plan	Deferred

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 5-10 months.
• Replacement CAPEX for the carousel deferred indefinitely, because the problem was never the machine.
• Comfort scrap reduced from the parameter-to-defect correlation that is impossible to make today: hardness, collapse and voids traced back to the station and the cure that caused them.
• And a root-cause analysis per mold in minutes instead of weeks, which is what the customer's quality team measures you on.

Estimated payback 5-10 months · defers the CAPEX of replacing the carousel Estimated payback of five to ten months, an estimate to validate, plus the replacement CAPEX deferred indefinitely. Because the real problem was never the machine: it was that it did not talk.

And the fair question from the production manager

“Does a camera reading a panel hold up as process evidence?” — the task is anchored: a fixed screen, fields in known positions and expected physical ranges for temperature and curing time, where the best models drop below 1.5% error [1]. And any frame whose value falls outside the possible range for that mold is discarded instead of accepted, which is what would invalidate the series in front of the customer.

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