The mold changeover start-up sheet, captured with a photo

The problem

The parameters that actually made a good part never leave the paper.

The start-up sheet prints with the textbook parameters, but on top of them the setter writes down what actually had to be changed to get a good part that day: ten degrees more in zone three, an extra second of holding time, one cavity blocked off because the ejector is marginal. That is precisely the knowledge that prevents the next bad part, and it never reaches the system. The first-off approval record, in turn, is mandatory evidence in a certified automotive environment. If a complaint arrives three weeks later, somebody has to physically go and find it in a folder to prove the check was done. Between the moment it is written and the moment somebody collects it, four to eight hours pass. If the changeover happens on nights, a full day. And in a plant that changes mold, color and part number several times a week, that is a permanent hole in the process history.

• The sheet prints from the ERP with the nominal molding parameters, and the setter writes on top what really made the door panel come out right that day: ten degrees more in barrel zone three, an extra second of holding time, a cavity blocked off because the ejector is marginal.
• The first-off record is handwritten by standard, with the dimensions checked against the control plan, the appearance check and the quality signature. In a plant certified to IATF 16949 it is mandatory evidence.
• Between the signature and the moment somebody collects the sheets, four to eight hours pass — and a full day if the changeover falls on nights.
• In a shop that changes mold, color and part number several times a week, that latency is a permanent hole in the process history of every large-tonnage press.

How it fits the IRIS system

Connect in photo mode — two photos at start-up, the routine untouched.

Connect in photo mode. The operator photographs the annotated sheet and the first-off record with a phone or an industrial tablet. A grounded language model extracts the structured fields — part number, revision, mold number, active cavities, resin and masterbatch lot, actual versus nominal parameters, signature and approval time — and ties them to the live ERP production order. The summary appears on the tablet at the press so a human can validate it in two taps (see case 5) and only then does it cross into the central system. Zero change of habit: people keep writing on the same paper.

It is the simplest case of the twelve and the one with the least friction on the shop floor, because it asks nothing of the setter: they keep writing on the same sheet and add two photos. The mold's real recipe stops living in the heads of two people and becomes something the next shift can look up.

See the full IRIS architecture →

Before and after

Today's mold start-up versus the captured start-up

Aspect	Today, on paper	With iLEAN Connect
Latency of the start-up sheet	4-8 h, a full day on nights	Second zero
Real parameters per mold	In the heads of two setters	Searchable history per mold and press
First-off approval record	Hunted for in a folder	Tied to the lot and the serial number
Resin and masterbatch lot	On the sheet, if anyone wrote it	Extracted and linked to the ERP order
Start-up on a difficult mold	Rebuilt from memory	Last known-good parameters reused
Setter's habit	—	Unchanged: two photos

from 4-8 hours of latency (or a full shift) to second zero · from "the real parameters live in the heads of two setters" to a searchable history per mold · from hunting for the first-off record in a folder to having it tied to the lot and the serial number.

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 4-9 months, depending on the number of mold changeovers per week.
• Transcription time of the production technician recovered, every changeover, every week.
• Shorter start-up on difficult molds by reusing the parameters that already worked instead of rediscovering them.
• And the first-off record stops being something you go and find: it is tied to the lot when a complaint lands three weeks later.

estimated payback 4-9 months depending on changeovers per week; recovery of the production technician's transcription time and shorter start-up on difficult molds by reusing the parameters that already worked. *Estimate to be validated.*

And the fair question from the production manager

"What if it misreads a holding pressure the setter scrawled in the margin?" — reading fields from a document with a known structure is an anchored task, where the best models drop below 1.5% error [1]. And there is a second net: the summary goes to the tablet at the press and the setter confirms or corrects it before anything crosses into the central system; a value outside the physical range of that press is flagged, not accepted.

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