The kiln control panel, now connected

The problem

The most expensive data in the plant lives on a screen nobody records.

The kiln is the thermal bottleneck and the most expensive asset across the interconnected lines. Its panel shows everything a continuous-improvement manager would need to correlate the firing curve with defects and energy use, but today nobody does because the panel's data isn't stored in a structured way.

• The roller kiln is the thermal bottleneck and the most expensive asset across the interconnected lines. Many run for more than a decade on an isolated network, by the kiln builder's decision, and nobody wants to touch them.
• Its panel shows everything that matters: temperature by zone, roller speed, firing cycle time and energy consumption. It is the richest data source in the plant and the least used. It changes every minute and is gone the next.
• Nobody stores it in a structured way. When a batch comes out with black core, warpage or shade drift, there is no curve to put next to the defect, only the memory of whoever was on shift.
• So the continuous-improvement manager cannot answer the question that decides the scrap: was it the paste, the glaze or the firing? And without that answer, replacing the kiln starts to look like the only option.

How it fits the IRIS system

Connect in photo mode on the panel — read from outside, the kiln untouched.

Connect photo mode on the panel. Fixed industrial camera + OCR with a grounded language model interprets every frame, structures the temporal curve per batch, and cross-references it with the ERP (which batch was firing at that moment).

Nothing is installed in the kiln's control system, no port is opened and its builder's support is not affected. A camera reads what the panel already shows, and that is enough to give each batch its own firing curve. On the most expensive machine in the plant, solving it from outside is what makes the case approvable.

See the full IRIS architecture →

Before and after

The kiln panel today versus the kiln panel being read

Aspect	Today	With iLEAN Connect
Firing curve per batch	Visible on screen, then lost	Digitized and stored by batch
Roller speed and cycle time	Read by eye during rounds	Time series in central memory
Black core or warpage root cause	Argued, not proven	Curve next to the defect in minutes
Batch firing at a given minute	Guessed from the schedule	Cross-referenced with the ERP
The kiln's SCADA and network	—	Untouched, still isolated
Kiln replacement capex	On the investment plan	Postponable

Zero curve-defect correlation → root-cause analysis per batch in minutes. Kiln replacement becomes postponable indefinitely.

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.
• Scrap reduction from correlating the firing curve with defects, a correlation that simply cannot be made today.
• Avoided capex: a kiln that can be read stops being a kiln that has to be replaced, and the replacement can be postponed indefinitely.
• Root-cause analysis per batch goes from impossible to minutes, and energy use per square meter becomes visible by format.

Estimated payback 5-10 months, avoided kiln-replacement CAPEX, scrap reduction from curve-defect correlation. *Estimate to validate*.

And the fair question from the production manager

“Can a camera reading a kiln panel be trusted for root cause?” — the screen is fixed, the fields sit in known positions and each zone has a physical temperature range, which makes it an anchored task where the best models drop below 1.5% error [1]. And any frame with a value outside the kiln's possible range is discarded instead of stored, so it never contaminates the curve that the analysis rests on. Each stored value keeps the frame it was read from, so anyone can check it later.

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