Format change without paper on the pressing line

The problem

The press setting that explains a caliber problem sits in a folder.

The shift's work order is printed from ERP data, and any mold or pressure adjustment gets noted by hand. The format-change sheet stays on paper for 4-8 hours before digitization. If a caliber issue shows up later at the kiln, nobody can quickly correlate it to the actual press setting.

• The shift's work order prints from the ERP with format, collection and target square meters. Everything that happens afterwards at the press — die set, ejector height, specific pressure, fill density — is written by hand on the mold-change sheet.
• That sheet stays on paper for 4-8 hours, until someone collects the folder at the end of the shift and types it, if they type it at all. Over a weekend the gap is longer, and the press keeps running. During that gap the batch exists physically but not in any system anyone can query.
• When a caliber or flatness problem appears at the kiln exit or at sorting, the question is always the same: what pressure and which spray-dried powder batch was the press running at that moment?
• Answering it means going down to the line, finding the right folder and reading a handwritten figure. By the time it is done, the next two batches have already been pressed with the same setting, and the problem has moved downstream to glazing and firing. Every hour the setting stays wrong is more tiles that will be downgraded at sorting.

How it fits the IRIS system

Connect in photo mode — two photos at the press, and the batch exists.

Connect photo mode. A grounded language model extracts structured fields (format, caliber, thickness, paste batch, pressure) from the photos and stores them in iLEAN's central memory the instant they're taken. Zero habit change for the press operator.

The press operator keeps filling in the same sheet with the same pen. The only new gesture is two photos, and from that second the format change is a record you can search by batch, not a sheet you have to go and find. That is why it is usually the first case a tile plant approves: it asks the floor for almost nothing.

See the full IRIS architecture →

Before and after

Format change on paper versus format change captured at the press

Aspect	Today, on paper	With iLEAN Connect
Mold-change sheet in the system	4-8 h later, typed by hand	At the moment of the photo
Specific pressure and fill density	Handwritten, rarely transcribed	Structured fields per batch
Spray-dried powder batch on the press	Inferred afterwards	Linked to the format change
Caliber deviation found at sorting	Hunt for the folder	One lookup by batch
Reconstructing a past format change	Hours of searching	Instant web query
Press operator's routine	—	Unchanged: two photos

4-8h latency → zero latency. Manually reconstructing a past format change → instant web lookup by batch.

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 format changes per month.
• The production lead's transcription time is recovered every shift, and it stops being the bottleneck for knowing what the press did.
• From 4-8 hours of latency down to zero: the format change exists in the system before the first tile reaches the dryer.
• And reconstructing a past change goes from an afternoon in the archive to an instant lookup by batch, which is what a caliber claim actually needs. The answer to the customer stops depending on who remembers the shift.

Estimated payback 4-9 months depending on monthly format changes, recovering the production lead's transcription time. *Estimate to validate*.

And the fair question from the production manager

“What if it misreads the pressure on a handwritten sheet?” — reading fields from a document whose layout never changes is an anchored task, where the best models drop below 1.5% error [1]. And there is a second net: specific pressure and fill density have a physical range for each press and format, so a value outside it is flagged for the operator to confirm before it is stored.

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