Digitizing the concentrate evaporator panel without touching the machine

The problem

The most expensive asset of the season keeps its data on a screen that cycles.

The orange season is intensive and seasonal: a few weeks a year in which the plant runs at its limit and in which every point of yield is multiplied by thousands of metric tons. When concentrate yield drops, the question is always the same: was it the fruit, the evaporator setting or the extraction? Nobody can answer it with data, because the panel data dies every time the screen cycles. The correlation between the fruit that came in that morning, the process applied and the gallons that came out never gets made. Decisions run on experience, which in these plants is enormous, but experience leaves no record you can study next season or hand over to a successor. And the obvious alternative — replacing the evaporator with a modern one with an open SCADA — is not an alternative: it is a major investment project with a shutdown the season will not tolerate.

• The evaporator is fifteen to twenty-five years old, isolated from the network by design, and its panel shows outlet degrees Brix, temperature per effect, vacuum, feed flow and alarms — for a few minutes, until the next screen takes over.
• When gallons of concentrate per metric ton of fruit drop, the question is always the same: was it the fruit, the evaporator setting or the extraction? Nobody can answer it with data, because the panel data no longer exists.
• The season is a few weeks a year in which every point of yield is multiplied by thousands of metric tons. The decision runs on experience, which is enormous, but which leaves no record to study next season or hand over to a successor.
• And the obvious alternative — a modern evaporator with an open SCADA — is a seven-figure CAPEX and a shutdown the season cannot absorb.

How it fits the IRIS system

Connect in photo mode on the panel — solved from outside, not inside.

Connect in "photo of the analog" mode applied to machine panels. The flow:

It does not connect to the evaporator, does not touch its control logic and does not compromise its warranty. On the machine the whole season depends on, that is the difference between a case the plant manager approves this week and a project nobody dares to schedule between two campaigns. The machine is still the same machine; the plant simply keeps what it was already saying — and next season starts with a curve to compare against instead of a memory.

• A fixed industrial camera points at the panel screen. Nothing is connected to the equipment, its control logic is untouched and its warranty is not compromised.
• OCR with a grounded language model reads each frame and recognizes the panel's fields, even as the screen cycles between different views.
• The full time series is rebuilt: outlet degrees Brix, temperature per effect, vacuum level, feed flow and alarms.
• The series is cross-referenced with the fruit batch in process (from the ERP or the intake weighbridge) and with the lab's Brix reading.
• The concentration curve becomes queryable by batch, season after season.

The evaporator is still the same evaporator. The only change is that the plant now keeps what the machine was already saying.

See the full IRIS architecture →

Before and after

Today's evaporator versus the evaporator being read

Aspect	Today	With iLEAN Connect
Brix, temperature per effect, vacuum	On screen for minutes, then gone	Full time series, batch by batch
Correlation with the fruit batch	Never made	Cross-referenced with the weighbridge and the lab
Yield drop mid-season	Argued from experience	Root cause in minutes
Concentration curve from last season	Does not exist	Queryable, season after season
The evaporator's control logic	—	Untouched, no network exposure
Replacement CAPEX	Back in the plan every year	Deferrable, with the machine under control

from zero correlation between evaporation curve and yield, to root-cause analysis by fruit batch in minutes. From an equipment replacement that appears in the investment plan every year, to a sustainable deferral with the machine under control.

Impact estimate

Estimated impact — to validate with your own 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.
• The larger component is not operational: it is deferring a replacement CAPEX of a far higher order, sustainably, because the machine is finally under control.
• Plus percentage points of improvement in gallons of concentrate per metric ton of fruit, through tuning based on data instead of memory.
• And a concentration history that survives the season and the people, which is what a successor inherits today as nothing at all.

estimated payback 5 to 10 months. The larger component is not operational: it is deferring a replacement CAPEX of a far higher order, plus percentage points of improvement in gallons of concentrate per metric ton of fruit through data-based tuning. *Estimate to validate.*

And the fair question from the production manager

«Can a camera reading a cycling screen produce a series you would trust?» — here the task is anchored: a fixed panel, fields in known positions and physical ranges known in advance, where the best models drop below 1.5% error [1]. And the system discards any frame whose value falls outside the possible range — Brix does not jump twenty points between two frames — instead of accepting it, which is what would corrupt the curve.

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