No line startup without visual evidence

The problem

A gauge changeover closes on a signature with nothing behind it.

a gauge or program changeover stops the line for 1-3 hours and today closes on a human signature with no structured visual evidence — a direct risk under an IATF 16949 control plan.

• On a precision tube line, switching tube gauge or OEM program means readjusting forming stands, weld head position, sizing stands and the cut-to-length setting. The line stops for 1-3 hours, and everyone on the floor feels the pressure to restart.
• Today the changeover closes with a human signature and no structured visual evidence of the state of each station. The checklist is well designed; what fails is how it gets signed under time pressure.
• Under an IATF 16949 control plan that is a direct risk: a stand left at the previous setting can produce a whole out-of-tolerance lot, and a weld head a few tenths off the seam may pass the first-off piece and fail later.
• And when that happens, the investigation finds a signature, not a record of what the line actually looked like at startup.

How it fits the IRIS system

Edge with JIDOKA AI — no green light, no startup.

Edge + JIDOKA AI. Industrial cameras with CNNs trained to recognize "ready vs. not ready" at each station. Startup blocked until OK + sign-off.

Each camera compares its station against the reference state for the new program. The line starts when all of them read OK and quality signs on that evidence: the decision stays human, the proof is new. And the more changeovers a line runs, the more the evidence is worth. High-mix lines, with several gauges a week, are where it pays first.

See the full IRIS architecture →

Before and after

The signed changeover versus the evidenced changeover

Aspect	Today	With iLEAN Edge
Forming stands after the change	Checked by eye	Compared with the new program's reference
Weld head and seam position	Assumed right	Confirmed by camera
Cut-to-length setting	Found at first inspection	Verified before startup
Quality sign-off	On trust	On images of every critical point
Out-of-tolerance first lot	A real possibility	Blocked at the source
Changeover record	A signature	Images, timestamps and signer

a blind sign-off with risk of an out-of-spec lot → sign-off backed by visual evidence at every critical point.

Impact estimate

Estimated impact — to validate with your own 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.

• Payback set by your changeover frequency: we do not fix a month range, because the return is counted per gauge change, not per calendar month.
• Each blocked startup protects a whole lot that would otherwise run out of tolerance, with its scrap, sorting and possible containment.
• The sign-off stops being blind: it rests on visual evidence at every critical point.
• And the changeover record becomes evidence your control plan and your OEM can inspect, instead of a ticked box and no proof.

protection against a whole out-of-tolerance lot, payback depending on changeover frequency. *Estimate to validate*.

And the fair question from the production manager

“What if a camera holds the line for nothing, with the changeover clock already running?” — comparing a station against its reference state is an anchored task, where the best models drop below 1.5% error [1]. When the model is unsure, it does not block silently: it shows the image to the setter and quality decides. The line stops on a verifiable mismatch, not on a model's doubt. Every exception and every override is recorded with its author.

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