The die change that validates itself

The problem

The checklist is well designed. What fails is how it gets signed.

The 'it's set up right, go ahead' responsibility rests on a human sign-off with no visual evidence, and a wrong setup can mean a whole out-of-spec batch.

• Every part-number change means mounting the correct die, setting it, and validating the first piece before the batch is released.
• If the press starts with the previous revision's die insert, with a shim that was not refitted or with a sensor left bypassed from the last intervention, the whole batch is born out of spec.
• The checklist exists and is well designed. What fails is when and how it is signed: under pressure, because the changeover sits on the shift's critical path, and with no objective evidence behind the box.
• When something goes wrong afterwards, the investigation runs into a ticked box and no proof. It is not bad faith and nobody in the plant thinks it is: a signature is simply not a measurement, and a setter under pressure signs what they are confident about rather than what they verified point by point.

How it fits the IRIS system

Edge with JIDOKA AI — the batch is not released without conformity and a sign-off.

Edge + JIDOKA AI: cameras with CNNs trained to recognize the correct die/tooling and validate first-piece geometry. Batch release is blocked until OK + sign-off.

The camera does not decide alone: it checks each critical point against its reference image and the supervisor signs on that evidence. What changes is not who decides, but what they decide on — and what remains on file afterwards, which is a set of images of the actual setup instead of a name and a date next to a tick.

See the full IRIS architecture →

Before and after

Today's changeover versus the validated changeover

Aspect	Today	With iLEAN Edge
Signing the setup checklist	A ticked box	Reference images checked point by point
Wrong or obsolete die insert	Found after the batch	Detected before release
First-piece geometry	Signed off by eye	Validated against the reference
Batch release	On the setter's word	Blocked until OK plus sign-off
The investigation afterwards	A ticked box and no proof	The evidence of every critical point
More changeovers per week	More accumulated exposure	The same verification every time

From blind sign-off with mis-setup risk to sign-off based on visual evidence at every 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.

• Estimated payback tied to your changeover frequency, not to a fixed figure: the brief deliberately leaves it open because it depends on how many die changes you run.
• What it protects against is concrete: a whole batch of out-of-spec structural parts from a setup nobody could verify.
• Start by counting the die changes per month and the cost of the last batch that had to be sorted.
• And the blind sign-off becomes a sign-off on visual evidence at every critical point.

Payback depends on changeover frequency; protects against whole out-of-spec batches. Estimate to validate.

And the fair question from the production manager

“What if the camera blocks the press for no reason?” — that is the right objection, because the changeover sits on the shift's critical path. Conformity is checked against the reference state for that specific die and, in case of doubt, the system does not block silently: it escalates to the supervisor, who signs on the evidence. Recognizing a die against its own reference image is an anchored task, where the best models drop below 1.5% error [1], and even then the system stops on a verifiable discrepancy, never on model uncertainty.

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