The model changeover, without one part of the previous variant

The problem

Under schedule pressure, the line clearance signature becomes a formality.

Changeover is frequent — several a day on some lines — and pushes in two opposite directions: it has to be fast, because a stopped line costs money, and it has to be perfect, because a variant mix-up is a major non-conformity. A single surviving part from the previous variant built into the new one is a mixed unit. And since there is no way to know which one carried it, it forces you to contain everything produced between two signatures. Over time and under schedule pressure, the line clearance signature becomes a formality. Not through carelessness: because saying no costs line minutes and whoever signs knows it. And when a mixed unit turns up at the customer, the auditor's question is not whether it was signed. It is what evidence there is that the line was clear.

• A line that makes one variant in the morning and another in the afternoon has to be completely clear in between: not one part of the previous variant in a feeder, not one mislabeled bin, not one loose component on the prep table. Today that depends on a person walking the line and signing "clear", with no objective evidence.
• Changeover pushes in two opposite directions: it has to be fast, because a stopped line costs money, and it has to be perfect, because a single surviving part from the previous variant built into the new one is a mixed unit — a major non-conformity.
• Since there is no way to know which unit carried it, a mix-up forces you to contain everything produced between two signatures. Over time and under pressure the signature becomes a formality, not through carelessness but because saying no costs line minutes and whoever signs knows it.
• When a mixed unit turns up at the carmaker, the auditor's question is not whether the clearance was signed. It is what evidence there is that the line was clear.

How it fits the IRIS system

Edge with JIDOKA AI — the block belongs to the system, not to the person who has to say no.

Edge + JIDOKA AI, with SMED AI in parallel.

Fixed Edge cameras on the critical points — feeders, component bins, prep tables, assembly station, outfeed — compare the current state against the reference state "line cleared and set up for the incoming variant". The quality manager signs over images rather than over trust, and if one point does not read clear, JIDOKA AI prevents the start-up. SMED AI runs in parallel, guiding the sequence and preparing the points ahead of time, so the verification does not cost line minutes.

• Fixed Edge cameras on the N critical points: feeders, component bins, preparation tables, assembly station, outfeed.
• Each camera compares the current state against the reference state "line cleared and set up for the incoming variant".
• The per-point result is presented to the quality manager, who signs over images rather than over trust.
• If one point does not read clear, JIDOKA AI prevents the start-up: the block belongs to the system, not to a person who has to say no under pressure.
• SMED AI guides the changeover sequence and prepares the points ahead of time, so the verification does not cost line minutes.

See the full IRIS architecture →

Before and after

Today's changeover versus the verified changeover

Aspect	Today	With iLEAN Edge
Line clearance signature	Over trust, under pressure	Over an image at every critical point
A part of the previous variant left in a feeder	Found in a mixed unit at the customer	Detected before start-up
Who says no	A person, costing line minutes	The system: the start-up is blocked
Scope after a mix-up	Everything between two signatures	Bounded precisely, point by point
Changeover time	Verification competes with speed	SMED AI prepares the points ahead
Evidence for the auditor	A signed sheet	Before-and-after images per point

from a signature with no evidence to a signature backed by an image at every critical point · from variant mix-up risk to a physical block on start-up · from containing everything produced between two signatures to bounding it precisely.

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 6-12 months, depending on changeovers per day.
• Protection against a variant mix-up incident, whose typical cost runs from tens to hundreds of thousands across containment, rework and customer penalty.
• From containing everything produced between two signatures to bounding it precisely.
• A reduction in changeover time by guiding it, which is what turns the objection from the floor into support.

protection against a variant mix-up incident — whose typical cost runs from tens to hundreds of thousands across containment, rework and customer penalty — plus a reduction in changeover time by guiding it. Estimated payback 6-12 months depending on changeovers per day. *Estimate to validate.*

And the fair question from the production manager

"What if a camera blocks the start-up for a shadow?" — that is the right objection, because the changeover sits on the shift's critical path. Each point is compared against the reference state for that specific variant, an anchored comparison where the best models drop below 1.5% error [1]. When the doubt is the model's, it does not block silently: it shows the quality manager the image and they decide. It stops on a verifiable discrepancy — a part where there should be none — not on model uncertainty.

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