The machine stays itself, but now it talks

The problem

The machine that cuts every enclosure is the bottleneck nobody can measure.

The punch press is a critical capacity bottleneck right as the plant scales production. Its panel shows the active cutting program, piece counter and tool-change alarms, but today no one correlates it because the information dies on screen every time the display rotates.

• The sheet-metal line that punches and bends the switchgear and UPS enclosures typically runs a 10-20 year-old proprietary panel, isolated from the network by the builder's design and never meant to share data.
• Its screen shows the active cutting program, the piece counter and the tool-change alarms. The moment the display rotates to the next page, that information is gone for good.
• Replacing the controller mid-expansion is unworkable: it stops the very machine that limits how many enclosures reach assembly each week, and the retrofit lead time alone eats a quarter.
• So the plant plans capacity at its bottleneck on estimates, and when enclosures run late nobody can say whether it was tool changes, a slow program, a jammed sheet or simply waiting for material.

How it fits the IRIS system

Connect in photo mode on the press panel — read from outside, the controller untouched.

Connect photo mode on the panel, with a fixed industrial camera. OCR plus a grounded language model reads every frame and structures pieces-per-hour and stop causes, cross-referenced with the ERP for which order was running — without touching the original PLC or panel.

Nothing is installed on the press, no port is opened and the builder's warranty is not at stake. A camera reads what the operator already sees, and that is enough to turn an opaque bottleneck into stop causes by part number. In an expansion, knowing where the press loses its hours is worth more than a new press ordered on intuition.

See the full IRIS architecture →

Before and after

The silent enclosure press versus the enclosure press being read

Aspect	Today	With iLEAN Connect
Enclosure panels per hour	Estimated at the end of the shift	Counted frame by frame
Which cutting program ran	Remembered by the operator	Read from the panel and tied to the order
Tool-change alarms	Cleared and forgotten	Logged with time and duration
Stop causes by enclosure part number	Unknown	Analyzed in minutes
The original PLC and panel	—	Untouched, no software installed
Capacity planning at the bottleneck	Rule of thumb	Based on measured rates

Zero OEE visibility on this machine → root-cause analysis of stops by part number in minutes.

Impact estimate

Punch press visibility: estimated impact, to validate with your cutting plan.

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, mostly from capacity recovered at the machine that sets the pace of the whole enclosure flow, without the capex and downtime of a retrofit.
• Better cutting-capacity planning right at the bottleneck of the expansion ramp, which decides how many cabinets assembly can start each week.
• Stop analysis by enclosure part number in minutes instead of zero OEE visibility on this machine.
• And a replacement decision made on measured data, not on the feeling that the old press is slowing everything down. Sometimes the data shows the press is fine and the loss is upstream, in sheet supply.

Estimated payback 5-10 months, better cutting-capacity planning right at the bottleneck of the expansion ramp. *Estimate to validate.*

And the fair question from the production manager

“Can a camera on a fifteen-year-old screen give numbers we can plan with?” — reading a fixed panel with fields in known positions is an anchored task, where the best models drop below 1.5% error [1]. On top of that, a counter that jumps backwards or a rate outside what the press can physically punch is discarded, not stored, so a bad frame never turns into a planning figure.

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