Camera that reads the cold-press panel

The problem

The line that makes the margin shows its data for a few seconds and forgets it.

Extraction yield varies between batches, and today nobody correlates that variation with the real curve on the press or centrifugal decanter panel, because the data rotates and disappears from the screen every few seconds. Replacing the panel with a modern connected one would mean stopping the plant's most critical line and a cost few factories take on during peak season.

• The cold-press line — malaxing, decanter, polishing — is the most expensive and most differentiating asset in an avocado oil plant, and the one that turns fruit rejected for export into margin.
• Its proprietary panel shows paste temperature, flow and yield every few seconds, and the value is gone as soon as the screen refreshes.
• Yield varies from batch to batch, and nobody can say whether it was the fruit's ripeness, the paste temperature or the decanter flow, because the curve was never kept. Every batch is a lesson the plant pays for and does not learn.
• Replacing the panel with a connected one means stopping the plant's most critical line — a cost few plants accept in the middle of the season.

How it fits the IRIS system

Connect in photo mode on the extraction panel — read from outside, nothing opened.

Connect photo mode on the panel: a fixed industrial camera points at the proprietary panel. OCR with a grounded LLM interprets each frame and structures the temperature, flow and yield time curve per batch, cross-referencing it with the fruit lot that was entering the press at that moment (already available in central memory).

The camera does not guess the yield: it reads what the press already knows and nobody writes down. The value appears when that curve meets the fruit lot that was in the malaxer at that minute, with the ripeness and dry matter recorded at receiving. Over a season, that pairing tells you which fruit is worth sending to oil and at which settings.

See the full IRIS architecture →

Before and after

The cold-press line as a black box versus the line being read

Aspect	Today	With iLEAN Connect
Paste temperature per batch	Seen on screen, then lost	A stored curve per batch
Decanter flow and yield	A figure at shift end	A time series for the whole batch
Why a batch yields less	Nobody can tell	Root cause in minutes
Link to the fruit lot	None	Crossed with the lot in the malaxer
Proof of cold extraction	An operator's word	The temperature record of every batch
Replacing the panel	A capex line waiting for the off-season	Deferred

Zero curve-to-yield correlation → root-cause yield analysis per batch in minutes.

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.
• From extraction yield gained by fine-tuning paste temperature and decanter flow against real curves.
• And from avoided capex: the proprietary panel and the line stay exactly as they are, with no revalidation and no stop in the middle of the season.
• Correlating curve and yield goes from impossible to a per-batch root-cause analysis in minutes.

Estimated payback 5-10 months, from improved extraction yield through fine parameter tuning and avoided CAPEX for panel/line replacement. *Estimate to be validated.*

And the fair question from the production manager

“A camera reading a panel in a room full of oil mist and condensation?” — it is a fixed screen, with fields in known positions and physical ranges the process cannot leave, which makes it an anchored task where the best models drop below 1.5% error [1]. A fogged frame or a value outside the physical range is discarded rather than stored, so the series stays clean enough to tune the press on.

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