Edge validates the changeover before startup

The problem

The first lot after a changeover carries the whole risk of the changeover.

- Visually confirming the correct feeder, stencil, or mold is mounted depends entirely today on human attention under changeover time pressure. - If the confirmation fails, the entire first lot of the new reference ships nonconforming.

• Every reference changeover on the SMT line or the winding machine stops production for 1 to 3 hours, and the pressure to restart grows with every minute.
• Visually confirming that the correct feeder, stencil or mold — or the right wire gauge and winding head — is mounted depends entirely today on human attention under changeover time pressure.
• The check leaves no objective evidence: a signature on a checklist says someone looked, not what they saw.
• If the confirmation fails, the entire first lot of the new reference ships nonconforming, and the error is found at test or, worse, by the customer.

How it fits the IRIS system

Edge with JIDOKA AI — N cameras confirm, the quality lead signs, then the line starts.

Edge cameras at N critical changeover points compare the current state against the correct reference state for that part. The line doesn't start until N cameras confirm OK plus the quality lead's sign-off — JIDOKA AI.

It is JIDOKA applied to the changeover: the stop is built into the process instead of relying on someone noticing. The cameras check what is mounted, the quality lead decides, and every changeover leaves a photo of the state the line restarted in.

See the full IRIS architecture →

Before and after

Changeover confirmation today versus Edge confirmation

Aspect	Today	With iLEAN Edge
Stencil for the new ECU board	Confirmed by reading its label	Matched to the reference image for that part
Wire gauge and winding head on the winder	Checked against the traveler, if time allows	Confirmed from the image
Restart condition	The operator decides it looks right	N cameras OK plus quality lead sign-off
Evidence of the setup	A tick on a checklist	Photo of every critical point
First lot of the new reference	Carries the changeover's risk	Starts from a confirmed state
Well-executed changeover time	Stretched by repeated rechecks	Improved: one objective check replaces several

Human visual confirmation with no objective evidence → Edge confirmation plus human sign-off, with photographic evidence of every change.

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 4-9 months.
• Fewer nonconforming lots at startup, because the first lot of the new reference starts from a confirmed setup.
• Better well-executed changeover time: one objective confirmation replaces repeated visual rechecks under pressure.
• From human visual confirmation with no objective evidence to Edge confirmation plus human sign-off, with a photo of every change.

Reduces nonconforming lots at startup, improves well-executed changeover time, estimated payback 4-9 months. Estimate to be validated.

And the fair question from the production manager

“And if the camera blocks a correct setup and we lose even more time?” — comparing a mounted setup with its reference image is an anchored task, where the best models stay below 1.5% [1] error. When a camera flags a difference, it shows which point and the reference next to the current state, so the quality lead resolves it in seconds instead of rechecking the whole line.

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