The solvent recovery panel, digitized without touching the machine

The problem

The number that decides whether recovery makes money is calculated once a month.

Recovery yield — how many clean liters come out per hundred dirty liters in — is the number that decides whether the service makes money. Today it is calculated at month end from inventory, not run by run. When a run comes out with more water or a lower yield, there is no way to reconstruct the temperature, pressure and flow profile it was made with: that data was on the screen and vanished with it. The obvious fix — replacing the column control — is a several-hundred-thousand project with a unit shutdown. In an operation containing costs that project does not get approved, so the plant settles for not measuring.

• Recovery yield — clean liters out per hundred dirty liters in — is what decides whether the solvent recovery service makes money. Today it comes out of the month-end inventory, so a bad run in week one is discovered in week five.
• The still's panel shows overhead and bottom temperature, pressure, reboiler level and distillate flow every second, and none of it survives the run. The profile that explains a wet run vanished with the screen.
• The control is proprietary, off the network, and the manufacturer stopped supporting it years ago. Nobody wants to open it, and nobody is authorized to wire anything to it. So the still is the plant's oldest asset and its least documented one.
• Replacing the column control is a several-hundred-thousand project with a unit shutdown. In an operation containing costs it does not get approved, so the plant settles for not measuring.

How it fits the IRIS system

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

Connect in "photo of the analog" mode on a machine panel.

A camera pointed at a screen touches nothing: no port opened, no firmware, no modification sign-off for a piece of equipment in a classified area. That is what makes it the fastest installation in the matrix to approve, and the one that gives the recovery service a per-run history for the first time. The still keeps running exactly as before; the only thing that changes is that its screen is finally remembered.

• Fixed industrial camera aimed at the panel, with whatever protection the area classification demands.
• A reading every few seconds: text recognition with a grounded model interpreting overhead and bottom temperature, pressure, reboiler level, distillate flow and alarms.
• Each reading is structured as a time series and linked to the active run and to the batch of dirty solvent that went in.
• The history becomes searchable: profile of every run, yield obtained, and the correlation between the two.

Nothing is wired to the original control, the equipment is not modified, and there is no modification sign-off to manage.

See the full IRIS architecture →

Before and after

Today's recovery still versus the still being read

Aspect	Today	With iLEAN Connect
Recovery yield	Known at month end, from inventory	Known per run, linked to the dirty batch
Temperature, pressure and flow profile	On the screen, then gone	A time series per run
A run with water in the distillate	No way to explain it	Profile compared with the good runs
The original control	Untouchable	Untouched
Replacement capex	Blocking measurement	Deferred indefinitely
Area classification	—	Camera housing rated for the zone

from yield known at month end to yield per run · from zero correlation between distillation profile and recovered quality to root cause analysis in minutes · from replacement CAPEX "in order to measure" to zero equipment CAPEX.

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, depending on the volume of solvent recovered.
• Every percentage point of yield recovered is sellable liters, and the correlation between profile and yield is what finds those points.
• Replacement capex for the column control postponed indefinitely: the plant measures without buying a new control.
• Root cause analysis of a low-yield run goes from impossible to minutes, with the batch of dirty solvent that went in already linked and the good runs available as the reference to compare against.

every percentage point of yield recovered is sellable liters, and the control replacement is postponed indefinitely. Estimated payback of 5 to 10 months depending on recovered volume. *Estimate to validate.*

And the fair question from the production manager

“Can a camera reading digits off a panel be trusted for a yield calculation?” — this is an anchored task: fixed screen, fields in known positions and physical ranges known in advance, where the best models drop below 1.5% error [1]. And there is a second net built into the series: a reading that falls outside the possible range for that variable — a reboiler level of 150% — is discarded instead of stored, and yield is always closed against the physical liters measured at the outlet.

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