The twenty-year-old machine that finally speaks

The turbo-emulsifier where the color cream is made is the chemical heart of a hair-color plant and is usually between ten and twenty years old. Its panel shows what decides whether the batch comes out right: temperature ramp and hold, vacuum, rotor-stator speed, phase timing and pH. And it is connected to nothing. An external camera pointed at the panel is enough for the plant to hold the process curve of every batch in central memory, without touching the machine's control logic.

‹ See all cases of hair color

Fixed industrial camera pointed at the control panel of a colorant mass vacuum mixer, showing the temperature ramp and vacuum readings on screen
The problem

The data that explains the deviation dies when the screen rotates.

Replacing the machine means seven-figure capex and weeks of downtime, so the reasonable decision has always been to leave it alone. The price of that decision is that the panel data dies as soon as the screen rotates, and all that survives is the single reading somebody copied by hand at one moment of the phase. The consequence is very concrete for whoever has to close a deviation: when the lab returns an out-of-range viscosity, a phase separation or a shade that does not match the standard, correlating that result with what the machine actually did is impossible. Deviations get closed with hypotheses instead of data, and the same deviation comes back.

  • The panel shows what determines whether the batch comes out right: temperature ramp and hold, vacuum, rotor-stator speed, phase time and pH.
  • All of it disappears the moment the screen rotates. The only thing that survives is the one-off note somebody copied by hand at a particular moment.
  • When the laboratory returns an out-of-range viscosity, a phase separation or a shade that does not match the standard, correlating it with what the machine did is impossible.
  • Deviations get closed with no real root cause, and what has no cause happens again.
  • The obvious way out is blocked by its price: replacing the machine is seven-figure capex and weeks of downtime.
How it fits the IRIS system

Connect in photo mode on the panel — without touching the machine's control.

Connect in photo-on-panel mode. A fixed industrial camera points at the screen and captures every few seconds. OCR plus a grounded LLM interprets each frame, rebuilds the full time curve of the batch and cross-references the ERP to know which order and which shade were in the machine at that moment. Without touching the original control system, without revalidating the equipment, without opening its electrical cabinet. The curve becomes available per batch and can be crossed with the lab result.

The camera does not connect to the machine's control system and installs nothing on it. In a GMP plant that matters more than it sounds: there is no control change to revalidate, and the mixer's qualified status is the same as the day before.

See the full IRIS architecture →

Before and after

Replacing the mixer versus reading its panel

AspectReplacing the machineWith iLEAN Connect
InvestmentSeven-figure capexIndustrial camera and commissioning
Production downtimeWeeksNone: installed while running
Machine qualificationStarts overUntouched: control not modified
Process curve per batchYes, from zero historyYes, from the first batch captured
Closing a deviationWithout objective dataWith the batch curve in front of you
Parameters recordedThree handwritten notesA complete time series

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.
  • Fewer batches reprocessed because of a physicochemical deviation.
  • Avoiding or indefinitely deferring the replacement capex of a manufacturing vessel.
  • From zero correlation between curve and deviation, to root cause analysis per batch in minutes on objective data.

estimated payback of 5 to 10 months from the reduction in batches reworked for physico-chemical deviation, plus the avoidance or deferral of manufacturing-equipment replacement capex. *Estimate to validate*.

And the fair question from the production manager

«Is a camera watching a screen reliable enough for a quality investigation?» — it would not be if the model were asked to interpret an arbitrary image. Here it is an anchored task: fixed screen, fields in known positions and expected physical ranges, where the best models drop below 1.5% error [1]. And the system discards any frame whose value falls outside the possible range instead of accepting it.

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

Frequently asked questions

What people ask about reading the mixer panel

Does it affect the machine's qualification?

No. The camera looks at the panel from outside: it does not connect to the controller, no software is installed on it and its program is not modified. There is nothing to revalidate.

What if the panel cycles through screens?

That is designed for. Capture cadence and optics are set to the rotation, and the system identifies which screen it is looking at before extracting any value from it.

What happens if somebody stands in front of it?

The frame is discarded and that instant is marked as having no reading. The value is not interpolated or invented, which is what would invalidate the record for an investigation.

Does it work for the other old machines in the plant?

Yes: the same pattern serves the premixer, the developer reactor or the packaging tunnel. You start with the mixer because it is the chemical heart and where the most expensive deviation sits.

How much history before it is useful?

To investigate a specific deviation, having the batch in question captured is enough. To detect machine drift you need some weeks, which is how long the trend takes to draw itself.

Let's talk

Tell us which critical machine you have today with no data output.

We work on your plant's real data, not ours. Assessment with no commitment.

Request estimated ROI within 48h ‹ See all cases of hair color See cosmetics