The twenty-year-old pelletiser starts to talk

The problem

The four numbers that explain a good batch are born and die on a screen.

And yet that screen holds exactly the four numbers that explain why a batch turns out well or badly: the zone temperature profile, the melt pressure, the screw amperage and the screen-changer differential pressure. Rising screen-changer pressure is the early warning that incoming material carries more contamination than expected. Screw amperage is the early warning that the blend has shifted. Both are visible on the panel, and both die on the panel.

• The temperature profile by zone, the melt pressure, the screw amperage and the screen-changer differential pressure.
• The screen-changer pressure rise is the early warning that incoming material carries more contamination than expected.
• The amperage is the early warning that the blend has changed.
• Both are visible on the panel and lost there: not historised, not tied to the batch, not comparable between shifts.

How it fits the IRIS system

A camera pointed at the panel. No PLC, no touching the original control, no warranty risk.

Connect solves this from the outside. A fixed industrial camera points at the panel and captures every few seconds. iLEAN reads each frame with an anchored model, structures the time curve and ties it to the active batch. No PLC work, no touching the original control, no warranty exposure and nothing to ask of the machine builder. The machine stays exactly as it is, but now the plant owns its curve. With the curve tied to the batch, the plant can finally cross process conditions against lab results. Root-cause analysis on a batch goes from being a conversation based on memory to being a query that takes minutes. And filter saturation gets caught while it is still climbing, rather than once it has already turned into out-of-spec material.

With the curve tied to the batch, the plant can finally cross process conditions against lab results. Root-cause analysis stops being a conversation and becomes a query.

See the full IRIS architecture →

Before and after

Extruder control, before and after

Aspect	Today	With iLEAN Connect
The process curve	Seen on the screen and lost	Time series tied to the batch
Screen-changer pressure	Seen by whoever walks past	Continuous trend, with an alert
Screw amperage	Same	Cross-referenced with the batch blend
Root cause of a bad batch	A conversation between shifts	A query over the curve
Intervention on the machine	Touch the PLC and risk the warranty	None: external camera
A fleet of different ages	One project per machine	The same method for all

Impact estimate

Impact estimate — to validate against 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.
• A process curve per batch without touching the machine or its warranty.
• Control-retrofit CAPEX deferred, which on a pelletiser is not a minor line.
• And the early warning of incoming contamination, which today surfaces once the filter is already blinded.

Payback 5-10 months · a process curve per batch without touching the machine or its warranty. Estimated payback is five to ten months. Two things usually left off the sheet should be added to that: the replacement CAPEX for the pelletiser is deferred indefinitely, and yield loss from undetected filter saturation goes down. An estimate to validate against the plant's downtime and rework history. Conservative ranges, an estimate to validate against the plant's real data.

And the fair question from the production manager

"What if the camera misreads a digit?" — reading a fixed-layout panel is about as anchored a task as there is [1]. And nothing is decided on a single reading: the alert comes from the trend, which one bad reading does not move. If somebody nudges the camera, the system flags it instead of carrying on reading wrong.

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