The oldest machine on the floor can speak too

The problem

The most expensive machine on the floor keeps no memory of what it did.

Foaming defects — voids, sink marks, short shots, out-of-range hardness — show up downstream, at wrapping or final inspection, once labour has already been invested in the part. Root cause analysis would need to know what temperature that specific mold had on that specific cycle. With no history, root cause gets rebuilt from conversation and memory, and corrective actions are taken on hypotheses rather than data. Any attempt to connect the machine hits the same wall every time: the builder gives no access, the network is isolated by design, and touching the control could void the warranty or the validated state of the process. So the conversation always ends in the same place: the machine would have to be replaced, and that costs a seven-figure sum and months of downtime.

• The steering wheel foaming carousel is one of the oldest and costliest assets in the plant. Its panel shows the temperature of every mold, shot weight, mix ratio, demold time and alarms — and all of it dies on screen because the panel is proprietary and the network is isolated by design.
• Foaming defects — voids, sink marks, short shots, hardness out of range — show up downstream, at wrapping or final inspection, once the labor of the whole wheel has already been spent on the part.
• Root cause would need to know what temperature that specific mold had on that specific cycle. With no history, it gets rebuilt from conversation and memory, and corrective actions are taken on hypotheses.
• Every attempt to connect the machine ends the same way: the builder gives no access, touching the control could void the warranty or the validated state of the process, and the replacement costs a seven-figure sum and months of downtime.

How it fits the IRIS system

Connect in photo-on-panel mode — the machine, its network and its warranty stay untouched.

Connect in photo-on-panel mode.

It does not integrate with the carousel's control, opens no port and installs nothing on the machine: a fixed camera looks at the screen the way an engineer would, only every cycle and without forgetting. On the asset whose validated state nobody wants to disturb, that is the difference between a case quality approves and a project that never leaves the meeting room. And the plant finally gets to decide about the replacement on evidence instead of on fear.

• A fixed industrial camera points at the panel screen. Zero physical contact with the machine and zero contact with its network.
• Optical recognition with a grounded LLM interprets each capture and structures the time series by mold and by cycle.
• It is cross-referenced with the MES/ERP to know which order and which variant were running at that moment.
• When a parameter drifts out of window, the alert reaches the right role there and then, not in the monthly report.
• The same pattern replicates onto any other critical vintage machine in the plant, touching none of them.

See the full IRIS architecture →

Before and after

Today's carousel versus the carousel being read

Aspect	Today	With iLEAN Connect
Mold temperature, shot weight, demold time	Visible on screen, then gone	Time series by mold and by cycle
Root cause of a void or sink mark	Rebuilt from memory	Parameter-to-defect correlation in minutes
A parameter drifting out of window	Noticed at final inspection	Alert to the right role there and then
Which order and variant was running	Reconstructed by hand	Cross-referenced with MES/ERP automatically
Machine control and warranty	—	Untouched, network untouched
Replacement CAPEX	Presented as inevitable	A decision postponed on evidence

from zero process history to a complete time series by mold and cycle · from root cause rebuilt from memory to parameter-to-defect correlation in minutes · from a replacement CAPEX presented as inevitable to a decision you can postpone on evidence.

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.
• Avoidance or deferral of the carousel replacement CAPEX, which is the seven-figure line in the plan.
• Scrap reduction through the process-to-defect correlation that is impossible to make today.
• The same pattern replicates onto any other critical vintage machine in the plant, touching none of them.

estimated payback 5-10 months; avoidance or deferral of the asset replacement CAPEX; scrap reduction through process-to-defect correlation. *Estimate to validate.*

And the fair question from the production manager

"Can a camera reading a screen be trusted for root cause on a safety part?" — here the task is anchored: fixed screen, fields in known positions and expected physical ranges, where the best models drop below 1.5% error [1]. Any frame whose value falls outside the possible range for that mold is discarded instead of accepted, and the series is what an engineer correlates with the defect: the camera supplies history, the person supplies the judgment.

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