Feeders verified before the changeover starts

The problem

One swapped feeder and the whole work order is scrap.

Today it depends on an operator visually checking feeder by feeder with no objective evidence. One swapped feeder can mount the wrong component on hundreds of boards.

• Changeover is where a multi-customer EMS plant makes or loses its margin, and feeder setup is the part of it with the worst risk profile: many positions, similar reels, time pressure.
• Today it depends on an operator checking feeder by feeder by eye, with no objective evidence that the check happened at all.
• One swapped feeder mounts the wrong component on hundreds of boards before anyone notices, and on a fine-pitch passive the substitution is invisible to the naked eye — a resistor of the wrong value looks exactly like the right one.
• What follows is worse than the scrap: rework on assembled boards, a schedule blown for the customer whose program you were running, and a containment conversation you cannot win without evidence. And because the plant runs several programs a shift, the same exposure comes back every few hours.

How it fits the IRIS system

Edge with JIDOKA AI — conformity before the first board, signed on visual evidence.

Edge + JIDOKA AI: fixed cameras over the feeder cart compare the current setup against the new program's BOM, position by position. The line won't start until every position passes and the engineering lead signs off based on visual evidence.

The cameras do not decide alone: each position is compared against the new program's BOM and the engineering lead signs off on that evidence. What changes is not who decides, but what they decide on — a checked list of positions instead of a memory of having looked.

See the full IRIS architecture →

Before and after

Today's changeover versus the validated changeover

Aspect	Today	With iLEAN Edge
Feeder verification	By eye, feeder by feeder	Position by position against the BOM
Evidence that it happened	None	Visual evidence per position
A swapped reel	Found after hundreds of boards	Blocked before the first board
Line start	Whenever the operator says go	Only when every position passes
Who signs off	Nobody, in practice	The engineering lead, on evidence
With high program mix	More changeovers, more exposure	More changeovers, the same verification

Eyeballed verification with no evidence → feeder-by-feeder visual-evidence-based verification, in seconds. Protection against scrapping an entire work order.

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 payback range on this one — the brief frames its return as protection against scrapping an entire work order, which is an avoided catastrophic event rather than recovered minutes.
• Eyeballed verification becomes feeder-by-feeder verification on visual evidence, in seconds, before the first board is placed.
• JIDOKA AI holds the line start until every position matches: the wrong component stops being mountable rather than merely stopping being likely, which is a different class of guarantee.
• And the evidence generated at changeover feeds the traceability dossier and the inventory cross-check, so the same verification pays three times over without any extra work on the floor.

Protects against scrapping a full WO · JIDOKA AI. Estimate to be validated.

And the fair question from the production manager

“What if the cameras block the line for no reason?” — that is the right objection, because changeover sits on the shift's critical path. The system stops on a verifiable mismatch between the cart and the BOM, never on model uncertainty: comparing a loaded position against a declared part number is an anchored task where the best models drop below 1.5% error [1], and any doubtful position escalates to the engineering lead instead of blocking silently.

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