The line cannot start wrong

The problem

The changeover is the riskiest moment of the day, and it happens many times per shift.

If the line starts with the previous variant's fixture, the wrong kit on the cart or the wrong torque program, the first units come out non-conforming. And in a just-in-time regime those first units are already committed to a specific vehicle on the customer's line. Today the guarantee that the changeover was done right is a human signature with no objective evidence, from someone under time pressure, because every minute of changeover is lost cadence. This is precisely the scenario where human error is expected, not exceptional.

• If the line starts with the previous variant's fixture, the wrong kit on the sequencing cart or the wrong torque program, the first seats come out non-conforming, and they are discovered later, if at all.
• In just-in-sequence those first seats are already committed to a specific vehicle on the customer's line. There is no buffer to absorb them.
• Today the guarantee that the changeover was done right is a human signature with no objective evidence, from someone under time pressure, because every minute of changeover is lost cadence on a line already behind the broadcast.
• This is precisely the scenario where human error is expected, not exceptional, and the plant runs five lines with different variants at once, each with its own changeovers through the shift.

How it fits the IRIS system

Edge with JIDOKA AI — the line does not start until every point confirms and a person signs.

Edge iLEAN places fixed cameras on the critical points of the changeover: the fixture nest, the component kit on the sequencing cart, the loaded foam and cover references, the tool and its program. The manufacturing system says which variant is due; the cameras confirm the line is in that state. And if any point fails to confirm, the line does not start. That is JIDOKA AI: the error is not announced, it is prevented. The system does not let it through. The AI provides the evidence point by point, the responsible person signs on that evidence, and can override and start if their judgment justifies it, leaving a reasoned record. The decision is never taken away from the person: only the blindness is. There is a side effect people usually like more than the main one. With the history of how long each point takes to confirm, the changeover itself gets shorter. The evidence we collect to avoid starting wrong becomes the raw material for SMED.

The manufacturing system says which variant is due; the cameras confirm the line is in that state. If a point fails, the line does not start: the error is prevented, not announced. And the responsible person can override with a reasoned record if their judgment justifies it — only the blindness is taken away, never the decision. The evidence collected to avoid starting wrong becomes the raw material for SMED.

See the full IRIS architecture →

Before and after

Today's changeover versus the validated changeover

Aspect	Today	With iLEAN Edge
Fixture nest	Assumed correct	Confirmed against the variant due
Kit on the sequencing cart	Checked by eye, in a hurry	Confirmed point by point
Foam and cover references loaded	A signature	Image evidence with a timestamp
Torque program	Whatever is on the controller	Confirmed for the variant
If a point fails	The line starts anyway	The line does not start
Changeover duration	Unknown per point	History per point, raw material for SMED

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, to validate against the number of changeovers per shift and the cost of a non-conforming start-up.
• The line cannot start wrong: no first seats committed to a vehicle with the previous variant's fixture, the wrong kit or the wrong torque program.
• A reasoned override record whenever a person decides to start anyway, so the decision is never taken away, only the blindness.
• And the side effect people like most: with the history of how long each point takes to confirm, the changeover itself gets shorter.

Estimated payback 5-10 months · the line cannot start wrong, and changeover gets shorter Estimated payback of five to ten months, an estimate to validate against the number of variant changeovers per shift and the average cost of a non-conforming start-up.

And the fair question from the production manager

“What if the camera blocks the line for no reason?” — that is the right objection, because every minute of changeover is lost cadence. Each point is checked against the reference state of that specific variant, which is an anchored task where the best models drop below 1.5% error [1]; and in case of doubt the system does not block silently: it escalates to the responsible person, who signs on the evidence or overrides with a reasoned record.

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