Starting injection with the wrong tooling after a reference change — and not finding out until the first quality check.

The problem

The reference change mixes three data points that almost never arrive together.

Verifying that the injection mold is correctly mounted is the crossing of three realities that, in most plants, each live on an island:

1. Which reference is due to be produced — the MES order. The production manager knows it; the night-shift setter not always.
2. What is actually mounted on the press — the master mold's frame is usually the same across similar references; what changes are the inserts and cavity blocks, which is exactly what a setter in a hurry can confuse.
3. Which configuration the mold's data sheet demands — which insert, which cavity block, which active cavities, which closing parameters. It usually lives in the ERP, in a vertical MES, or in a spreadsheet the tooling manager keeps updated on their own.

The setter knows this, but cannot hold all three data points in their head at three in the morning. The press closes and starts all the same if the clamps are fine — it does not know what it has inside. The first warning usually comes from the quality check an hour later, with a whole bin of out-of-drawing parts. If the block does not quite fit, on top of that, you get mold damage.

How it fits into the IRIS system

iLEAN does not add another system — it seals the crack between the order, the data sheet and the mounted mold.

The tooling problem is not a lack of information: it is information living in islands that, at the critical moment (the reference change at three in the morning), does not reach the person deciding in time. iLEAN acts as the filler that closes that gap, without asking you to change the MES, the ERP or the mold management system you already have.

Edge sees the mounted mold. Connect reads the order and the data sheet, wherever they come from. The agent cross-checks, and if something does not add up, holds the start-up. The person signs off — never the other way round.

The three iLEAN pieces applied to tooling verification in plastic injection:

• Edge — a terminal with machine vision (CNN) facing the open press. It reads the engravings, the cavity block's differential geometries, the insert's visible codes and the cavities' layout. It compares cavity by cavity. If something does not add up, it triggers the actuator (dry contact with the press's closing, light signal, panel notice). It works with no network. If the plant loses its WiFi, Edge keeps confirming and blocking start-ups, because what is critical cannot depend on connectivity.
• Connect — captures the MES order, the mold's data sheet from the management system (the ERP, a vertical MES, or the tooling manager's forgotten spreadsheet) and the notices arriving from outside (the customer's email with a revision change on the part). All at second zero, with no forwarding.
• Agent — cross-checks the order, the data sheet and what Vision sees at the press. If there is a deviation, it does not send an email at 10pm: it holds the start-up, alerts the setter through their channel and leaves a dossier with a photo of the mounted mold and the expected reference. The person validates and signs; the press never starts on its own.

See the full IRIS architecture →

Before and after

Reference change with a checklist vs. cross-checked change with iLEAN Vision

Aspect	Manual change + checklist	With iLEAN Edge + Connect + Agent
Source of the expected reference	The setter looks it up in the MES or on paper	Order + data sheet + vision cross-checked at second zero
Insert/cavity-block verification	The setter's eye, against the clock	The CNN compares engravings and geometries cavity by cavity
Multi-cavity change	Assumption: "if one is fine, all are fine"	Cavity-by-cavity confirmation before start-up
Error detection	First quality check, a whole shift later	Before the first closing, in line, in milliseconds
Operation with no network	n/a	Edge keeps operating on cabinet power
File for the audit (IATF 16949)	Reconstructed by hand, days	Dossier per order, automatic, with a photo of the mold

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific figures of your plant. We set it out so the committee has an order of magnitude; we refine it during the diagnostic.

• An automotive plastic injection plant with 8-15 presses, multi-reference, master-mold systems with interchangeable inserts and cavity blocks.
• Edge pilot on one critical press (camera over the mold area + integration with the MES order and the tooling data sheet). First expected value within a few weeks.
• Indicative payback between 4 and 9 months, depending on the documented frequency of crossed-tooling incidents and the average cost of a shift of scrap or a mold failure at your plant.
• Expected reduction in scrap from start-ups with the wrong tooling: ≥ 30% versus the baseline, with ample margin once the system covers the presses with the highest rotation. A single avoided mold failure pays for the pilot.

And the tooling manager's reasonable doubt

"What if the AI confuses two similar inserts and lets a bad start-up through?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI simply compares what it sees with a specific mold data sheet, the best models brought the error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN holds the start-up and the person signs off. The three safety rings are there for exactly this.

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