The vintage shear starts talking

The problem

The machine that decides cut quality leaves no record of how it cut.

The cutting shear is over 15 years old and runs a proprietary controller with no data output to any system. Its panel shows pressure, blade angle and feed speed, but that data dies on screen as soon as the next cut starts. Replacing it costs hundreds of thousands of euros and weeks of downtime.

• The shear is over fifteen years old and runs a proprietary controller with no data output to any system. It cuts perfectly well, which is exactly why nobody is going to replace it.
• Its panel shows hydraulic pressure, blade angle and feed speed. Those values die on the screen the moment the next cut starts, and the next order is already on the table.
• When a workshop sends a batch back for burr or for a cut that is out of square, there is nothing to correlate the complaint with: no pressure, no angle, no clearance recorded for that batch.
• Replacing the machine costs hundreds of thousands of euros and weeks with the cutting service stopped, which for a stockholder means sending its own customers to the competitor down the road.

How it fits the IRIS system

Connect in photo mode on the panel — read from outside, with nothing wired in.

A fixed external camera points at the panel. OCR with an anchored LLM reads each value every few seconds and structures it by batch, cross-referencing it with the order being cut at that moment. The machine stays the same, but now it talks.

Nothing is connected to the controller, no cabinet is opened and no port is enabled. What changes is that the panel's readings now exist outside the screen, stamped with the order, the grade and the heat that were on the table at that moment. Keeping an old shear stops meaning going blind about it, and the decision to replace it can finally be taken on tonnage and condition rather than on the absence of data.

See the full IRIS architecture →

Before and after

The silent panel versus the panel being read

Aspect	Today	With iLEAN Connect
Pressure, blade angle and speed	Visible, volatile, unrecorded	A reading every few seconds, per batch
A burr or out-of-square claim	Nothing to correlate it with	Root cause per order in minutes
Which order was on the table	Reconstructed from paper	Stamped on every reading
The machine's controller	—	Untouched, nothing installed
Replacing the shear	The only way to get data	Deferrable: capex avoided
Blade clearance drifting	Noticed when pieces come back	Visible as a trend

Zero correlation between cutting parameters and measurement claims → root-cause analysis per order in minutes.

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, counting the replacement capex you do not spend.
• Hundreds of thousands of euros of machine capex deferred, plus the weeks of stopped cutting service that would come with it.
• Root-cause analysis per order in minutes when a batch comes back for burr or squareness, instead of a conversation nobody can win.
• And blade wear stops being discovered through returns: a clearance drifting batch after batch shows as a trend before a customer finds it.

Estimated payback of 5 to 10 months, avoiding machine replacement CAPEX. Estimate to be validated.

And the fair question from the production manager

“Does a camera reading a panel hold up in front of a customer?” — the task here is anchored: a fixed screen, fields in known positions and physically expected ranges, where the best models drop below 1.5% error [1]. On top of that, any frame with a value outside what that shear can physically do — zero pressure mid-cut, a speed the machine does not reach — is discarded rather than stored, which is precisely what would otherwise ruin the series in a claim.

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