No start until everything checks out

The problem

The restart after an alloy changeover rests on a blind signature.

Starting the first lot wrong after an alloy or format change can contaminate the entire shift's production, and today the only safeguard is a blind sign-off.

• Changing the grid alloy or the battery model means purging the melting pot, swapping book molds, adjusting the paster and changing assembly tooling. Each of those steps has its own checklist, and each one is done under pressure to restart quickly. Book mold swaps, paster gap adjustments and assembly tooling are all done by different people at the same time.
• Today the restart depends on a human sign-off with no objective visual evidence that each point is really ready. Everybody wants the line back up, which is exactly when steps get assumed rather than checked.
• If the first lot starts wrong, the wrong alloy or the wrong grid can contaminate the whole shift's production before anyone notices. Lead-calcium and lead-antimony grids mixed in a run cause problems that only show up in the field.
• And when that happens, the sign-off sheet proves only that someone signed, not what they saw. That is a weak position to defend in front of an OEM asking how the mix-up happened.

How it fits the IRIS system

Edge with JIDOKA AI — no start until every camera says ready.

Edge + JIDOKA AI. Industrial cameras with CNNs trained to recognize "ready vs. incorrect state" at every critical point (casting, pasting, assembly). Startup is blocked until every camera is OK and quality signs off based on that visual evidence.

The quality lead still signs, but now he signs on images: casting, pasting and assembly each show their state against the reference. If any camera does not confirm, the line simply does not start. The restart takes the same time when everything is right, and stops only when something is not.

See the full IRIS architecture →

Before and after

Alloy changeover restart today versus validated by camera

Aspect	Today	With iLEAN Edge
Basis for the restart sign-off	A blind signature	Visual evidence per critical point
Book mold and pot state at casting	Checked by eye, if checked	Compared with the reference image
Paster set for the new grid	Assumed ready	Confirmed by camera
Assembly tooling for the new model	Discovered at first unit	Verified before startup
Startup with an incorrect state	Possible	Blocked by JIDOKA AI
Risk to the shift's first lot	Contamination of the run	Caught before the first grid

Blind sign-off with lot-contamination risk → sign-off based on visual evidence per critical point.

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.

• No fixed payback range: we do not set one because it varies with how often you change alloy or model. The range would hinge on two numbers that only your plant has.
• The return comes from avoided startup scrap: the first lot after a changeover stops being the risky one. One contaminated shift avoided can be worth more than the cameras themselves.
• The restart sign-off moves from a blind signature to evidence per critical point. That is a stronger position in front of an OEM asking how a restart was approved.
• We estimate it with your changeover log and the scrap of your last bad restarts, before committing. So the figure you see is yours, not a generic estimate.

Payback varies with changeover frequency and avoided startup scrap. *Figure to validate*.

And the fair question from the production manager

“A camera saying ready does not mean the alloy is right.” — true, and that is not what it claims. It checks the visible state at each critical point against a reference image, which is an anchored task where the best models drop below 1.5% error [1]. The alloy itself is confirmed by the casting log and the ERP; the camera closes the gap a paper sign-off leaves open. Each restart then keeps its images and signature on record.

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