Camera reads the stator winding machine panel

The problem

The machine that defines the stator keeps its data on a rotating screen.

- The winding machine's panel shows vital parameters (winding turns, wire tension, coil temperature) that rotate screens every few seconds and get lost if no one writes them down. - Replacing the machine means months of downtime and a seven-figure investment, so in practice it never gets replaced, and today no one correlates its parameters with the final electrical test result.

• The stator winding machine is one of the most expensive, least replaceable assets on the floor. It can be 10-20 years old, run on an isolated network, and nobody wants to open its controller.
• Its screen shows the vital parameters — winding turns, wire tension, coil temperature — rotating through pages every few seconds. If nobody writes them down, they are gone.
• Replacing the machine means months of downtime and a seven-figure investment, so in practice it never gets replaced and never gets connected either.
• The result: when a stator fails the surge or insulation resistance test after varnish impregnation, nobody can correlate that failure with how it was wound.

How it fits the IRIS system

Connect in photo mode on the winder's screen — read from outside, controller untouched.

Connect photo mode on the panel: a fixed industrial camera captures every screen, OCR with a grounded LLM interprets the frame and structures the time series per winding lot, cross-referencing it with the ERP/MES to know which unit was being wound at that moment.

The winder keeps its controller, its programs and its isolation. The camera only reads what the machine already shows its operator, and that is enough to turn a rotating screen into a time series per winding lot.

See the full IRIS architecture →

Before and after

The isolated winder today versus the winder being read

Aspect	Today	With iLEAN Connect
Turns, wire tension and coil temperature	Visible for seconds, then lost	Time series per winding lot
Which stator was on the spindle	Guessed from the shift log	Cross-referenced with the ERP/MES order
Surge or hipot reject after impregnation	No link to winding conditions	Root cause per lot in minutes
Slow drift in wire tension	Noticed when rejects already climb	Visible as a trend on the series
The original controller	—	Untouched, no ports opened
Replacement of the critical machine	Seven figures and months of downtime	Indefinitely postponable

Zero correlation between winding and electrical failure → root-cause analysis per lot in minutes, with replacement of the critical machine indefinitely postponable.

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.
• Part of it comes from avoiding the CAPEX of replacing a critical machine that still winds well but gives no data.
• The rest comes from scrap reduction through the parameter-to-defect correlation that is impossible to make today.
• From zero correlation between winding and electrical failure to root-cause analysis per lot in minutes, which is what a customer 8D asks for first.

Estimated payback 5-10 months, avoiding the CAPEX of replacing a critical machine and reducing scrap through parameter-to-defect correlation. Estimate to be validated.

And the fair question from the production manager

“Can a camera reading a screen be trusted for turns and tension?” — here the task is anchored: fixed screen, fields in known positions, values within known physical ranges, and in anchored tasks the best models stay below 1.5% [1] error. A frame whose value falls outside the physically possible range for that stator is discarded, not stored, so a glare or a half-refreshed page never pollutes the series.

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