The resin dryer and the high pressure compressor start talking

The problem

Two assets govern everything else, and neither is connected.

If resin reaches the screw above its target moisture, intrinsic viscosity drops during processing and the preform comes out brittle. You do not see that on the line: it surfaces three days later in a lab test or — worse — in a complaint from the bottler, when the bottle fails on their filler. The dryer panel shows dew point, hopper temperature, air flow and desiccant wheel hours. The compressor panel shows pressure and consumption. All the information you need is right there, and all of it dies on the panel. Nobody today can correlate drying conditions with breakage, or blow pressure with kilowatts per thousand units. The classic alternative — replacing the equipment with a connected one — is a CAPEX nobody signs off purely to obtain data, and it means line downtime on top. So the plant lives with the blind spot.

• The dryer-crystallizer governs quality. The high pressure compressor governs the electricity bill. Both show on their panels everything you would need to know.
• If resin reaches the screw above its target moisture, intrinsic viscosity drops during processing and the preform comes out brittle. That is not visible on the line.
• It surfaces three days later in a laboratory test or — worse — in a bottler's complaint, when the bottle bursts in their filler.
• Dew point, hopper temperature, air flow, desiccant wheel hours, pressure and consumption: all the information needed is there, and all of it dies on the panel.

How it fits the IRIS system

Connect in photo mode on the panel — no PLC, no open ports, no lost warranty.

Connect, photo-on-panel mode. A fixed industrial camera pointed at the screen, without touching the equipment. Step by step:

Nothing on the machine gets touched, and that answers the objection that blocks most of these projects: no port is opened, nothing is installed on the control system and the manufacturer's warranty stands.

• Capture. The camera images the panel every few seconds, including screens that cycle.
• Reading. OCR with a grounded language model interprets each frame and structures the values: dew point, hopper temperature, air flow, desiccant wheel hours; pressure and consumption on the compressor.
• Time series. The values become a continuous series in central memory, with no PLC, no open ports and no compromise to the manufacturer's warranty.
• Cross-reference. Matched against the ERP or MES to know which batch and which mold were running at each moment.
• Correlation. What cannot be seen today appears: drying conditions against intrinsic viscosity and breakage; pressure and consumption against the reference being produced.

The same pattern transfers directly to any other isolated asset in the plant: chiller, cooling tower, regrind mill or the panel of the oldest injection machine.

See the full IRIS architecture →

Before and after

Today's panel versus the panel being read

Aspect	Today	With iLEAN Connect
Panel data	Dies when the screen rotates	Continuous, queryable time series
Root cause of a bad batch	Done from memory	Resolved in minutes with the curve
Drying ↔ viscosity ↔ burst correlation	Impossible	Direct, batch by batch
Compressor consumption	Shows up on the bill	Attributable to batch and mold
PLC and ports	—	Untouched
Manufacturer's warranty	—	Intact

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, plus the replacement capex that gets deferred.
• Less scrap from brittle preforms, through the drying ↔ viscosity ↔ burst correlation that cannot be made today.
• Attributable energy savings on the plant's largest consumer.
• Root cause analysis of a bad batch in minutes, with the curve in front of you, instead of from memory.

Estimated payback 5-10 months, plus the replacement CAPEX you defer. Two sources of return: less scrap from brittle preforms (drying ↔ viscosity ↔ breakage correlation) and attributable energy savings on the hall's largest consumer. In a continuous process plant, the energy saving alone usually carries the case. *Estimate to be validated.*

And the fair question from the production manager

«Does a camera watching a panel hold up as technical evidence?» — here the task is anchored: fixed screen, fields in known positions and expected physical ranges, where the best models drop below 1.5% error [1]. On top of that, the system discards any frame whose value falls outside the physically possible range instead of accepting it, which is what would invalidate the series.

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