External camera on a legacy press panel

The problem

The veteran press shows everything on screen and records nothing.

Not every press in a closures plant is latest-generation. The veteran ones are still productive, but their panel is proprietary and network-isolated. Swapping a press costs hundreds of thousands of euros and weeks of downtime. But its screen shows pressure, temperature and cycle time — vital data that dies on the screen today.

• Most closure plants run a mixed fleet: latest-generation presses next to veteran machines that still mold good caps at a competitive cycle. Their panel is proprietary and isolated from any network by design.
• Replacing one costs hundreds of thousands of euros and weeks without that capacity, so the business case to retire a press that still works rarely closes.
• Meanwhile the screen shows injection pressure, barrel and hot-runner temperatures, holding time and cycle time. That data dies on the screen: at best a technician copies one reading onto the startup sheet. The figures that would explain a defect are visible for a few seconds and then overwritten.
• When a batch shows short shots or out-of-range removal torque three weeks later, nobody can say which parameters it ran with. Root cause turns into an argument between process and quality instead of a query.

How it fits the IRIS system

Connect in photo mode on the panel — read from outside, nothing installed in the press.

With iLEAN Connect photo mode, a fixed industrial camera points at the panel. OCR with an anchored model reads each frame and structures the time series per batch. Cross-referenced with the ERP, the system knows which order ran with which parameters. If a defect shows up 3 weeks later, root-cause analysis takes minutes, not days.

The camera does not connect to the press controller, does not open ports and does not touch any warranty. It reads what the screen already shows, frame by frame, and turns it into a time series per batch. That is the difference between a project maintenance blocks and one it signs off. And the press keeps molding while it is being read.

See the full IRIS architecture →

Before and after

Veteran press today versus veteran press being read

Aspect	Today	With iLEAN Connect
Pressure, temperatures, cycle time	Visible on screen, then gone	Time series per batch
Link to the active order	None	Crossed with the ERP order
Short shot found weeks later	Days of guesswork	Root cause in minutes
Hot-runner temperature drift	Noticed when rejects rise	Visible as a trend
Press controller	—	Untouched, no ports opened
Capex to replace a veteran press	Pushed by lack of data	Deferred with evidence

Zero parameter-defect correlation → root-cause analysis per batch in minutes, with no machine replacement.

Impact estimate

Impact estimate — to be validated against your veteran press fleet.

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, before counting any deferred press replacement.
• Avoided capex: a press that still molds good closures does not need to be replaced just to get its process data out of a closed panel. Replacing it becomes a choice backed by data, not a way of getting data.
• Less scrap, because parameter-defect correlation becomes possible: short shots, flash or weight drift can finally be tied to the settings that produced them.
• Root-cause analysis on a defect found weeks later goes from days of reconstruction to minutes per batch, with the series of that press on the table. Process and quality look at the same curve instead of defending two memories.

Payback 5-10 months · avoids machine-replacement CAPEX. Payback of 5 to 10 months, avoided press-replacement CAPEX, and reduced scrap through parameter-defect correlation. Estimate to validate.

And the fair question from the production manager

“Can a camera reading an old screen really count as process evidence?” — yes, because the task is anchored: the panel never moves, each value sits in a known position and has a physically possible range, and there the best models drop below 1.5% error [1]. A frame whose reading falls outside the possible range is discarded instead of stored, so a reflection or a hand in front of the screen never becomes a fake barrel temperature.

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