JIDOKA AI applied to mold, color and material changeovers

The problem

The five changeover mistakes all produce parts that look fine.

The five classic changeover mistakes are always the same: previous material not fully purged, wrong masterbatch, a cavity left blocked off from the last shift, the robot end-of-arm tool from the other mold, and the previous packaging container with its old label still on. Any one of the five produces parts that look fine. The pre-start check exists — there is a checklist — but it gets signed from memory, because whoever signs it has forty minutes of downtime weighing on them and the pressure to start. It is a problem of time, not of willingness. And the consequence does not show at the changeover: it shows in the first hundred parts if you are lucky, and at the customer if you are not. Material contamination and starting up on the previous setup are among the most expensive and least documented causes of non-quality in this sub-sector.

• Previous resin not fully purged, wrong masterbatch, a cavity still blocked off from the last shift, the robot end-of-arm tool from the other mold, and the previous packaging container with its old label still on.
• The pre-start checklist exists. It gets signed from memory, because whoever signs it has forty minutes of press downtime weighing on them and the pressure to start.
• The consequence does not show at the changeover: it shows in the first hundred parts if you are lucky, and at the OEM if you are not.
• Material contamination and starting up on the previous setup are among the most expensive and least documented causes of non-quality in interiors molding.

How it fits the IRIS system

Edge plus JIDOKA AI — the line does not start until every point reads conforming and the lead signs.

Edge plus JIDOKA AI. Fixed industrial cameras at each critical changeover point, with models trained to recognize "ready versus not ready" at each one: resin color at the feed throat, octabin label, visible mold number, free cavities, correct robot tool, packaging container and label. Start-up stays blocked until every camera reads conforming and the lead signs on visual evidence. And SMED AI sequences and overlaps the changeover tasks so verification falls into dead time: the goal is a changeover just as short or shorter, but with evidence.

The camera does not decide alone: it compares each point against its reference state for that mold and color, and it is the lead who signs on what they can see. SMED AI sequences and overlaps the tasks so the verification falls into dead time: the goal is a changeover just as short or shorter, but with evidence. The signature at the end is the same one as today; what changes is what stands behind it.

See the full IRIS architecture →

Before and after

Today's changeover versus the verified changeover

Aspect	Today	With iLEAN Edge
Resin color at the feed throat	Assumed purged	Read against the reference
Octabin and masterbatch label	Glanced at	Verified before start-up
Blocked cavity from last shift	Found in the first parts	Detected before the first shot
Robot end-of-arm tool	Right one, usually	Confirmed by camera
Packaging container and label	Old label still on	Verified before the first part is packed
Changeover time	Best part of an hour	Just as short or shorter, with SMED AI

from a blind signature at the end of a long changeover to a signature backed by point-by-point evidence · from finding contamination in the first parts or at the customer to being unable to start with it · from a changeover lengthened by verification to one just as short or shorter, thanks to SMED AI.

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, depending on changeover frequency.
• The first contaminated lot after a changeover, eliminated.
• Machine minutes recovered per changeover on a large-tonnage press, multiplied by the number of mold and color changeovers per week.
• And a signature backed by point-by-point evidence instead of a blind signature at the end of a long changeover.

elimination of the first contaminated lot after a changeover, plus machine minutes recovered per changeover multiplied by the number of changeovers per week. Estimated payback 5-10 months depending on changeover frequency. *Estimate to be validated.*

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. Each point is checked against the reference state of that specific mold and color, and in doubt the system does not block silently: it escalates to the lead, who signs on the evidence. Reading a mold number or an octabin label is an anchored task where the best models drop below 1.5% error [1]; the press stops on a verifiable discrepancy, not on model uncertainty.

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