Compressed air — the plant's most expensive energy, and the worst governed.

The problem

You pay for air that escapes and for pressure you don't need.

Plant compressed air is the only form of energy manufactured in-house — which is exactly why it is the worst measured. The main compressor meter is usually there, but traceability by shift, by line, by zone almost never is. And without traceability two things happen at once with nobody seeing them: air escapes through dozens of small leaks that add up to a great deal, and the plant runs at a higher pressure than it actually needs, a historical safety margin nobody has revisited since.

Every unnecessary bar of pressure is between 6 and 8 % of extra electrical consumption every hour of every shift. And between 20 % and 40 % of the air generated is lost to undetected leaks — the compressor notices, because it keeps loading at 4 a.m. on a Saturday with the plant stopped, but the energy manager doesn't find out until the manufacturer's monthly summary.

The classic system (scheduled preventive maintenance plus an annual leak walk) finds 30 % of what is there. The rest stays put, blowing into the atmosphere and costing money.

How it fits the IRIS system

iLEAN doesn't buy new compressors — it gets everything out of the ones you already have.

The compressed air problem isn't a lack of technology: it is information living on islands (the compressor panel on one side, the pressure gauges on another, production planning on another, the electricity meter on another). iLEAN acts as the filler that closes those gaps — without replacing a single compressor.

Edge samples ultrasound and flow. Connect reads each compressor panel and the production plan. The agent cross-checks, proposes lowering the pressure and orchestrates the sequencing. A person signs off every step.

The three iLEAN pieces applied to compressed air:

• Edge — ultrasonic sensors in critical zones and flow sampling at the highest-consumption points. The ultrasonic signature of a leak is recognized by Edge locally; it reports automatically, with nobody walking around with a gun. It works with no network.
• Connect — reads each compressor panel through whatever route applies (a modern interface, or an old screen captured by photo), pulls the production plan from the ERP/MES to understand the real expected demand of each shift, and absorbs whatever comes from outside (a manufacturer alert about the model, the electricity bill from the utility).
• Agent — cross-checks detected leaks, pressure point by point, real production demand and the efficiency curve of each compressor. It proposes pressure reduction in measured steps, orchestrates which compressor starts, modulates or rests, and opens a repair work order in the CMMS for every located leak, with coordinates and priority. A person validates — the agent does not write critical setpoints; the rings are there.

See the full IRIS architecture →

Before and after

Classic compressed air management vs. optimization with iLEAN

Aspect	Classic management	With iLEAN Edge + Brain + Agent
Leak detection	Annual walk with an ultrasonic gun — finds 30 %	Edge sensors + continuous consumption pattern analysis
Working pressure	Fixed historical setpoint — nobody dares lower it	Reduction in measured steps down to the real minimum needed
Multi-compressor	Rotation by running hours — static rule in the master compressor PLC	Dynamic sequencing based on each unit's real efficiency curve
Visibility by zone/shift	Main meter only — no idea which line consumes what	Traceability by zone, shift and line
Leak repair	A spreadsheet list that gets lost before the next shutdown	CMMS work order with coordinates + priority + estimated saving
ISO 50001 reporting	Rebuilt by hand before the audit	Automatic energy report, ready for the auditor

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Standard industrial plant with 2-4 compressors totalling 100-400 kW installed, working pressure between 7 and 9 bar, no traceability by shift or zone.
• Edge pilot in the compressor room + 3-4 critical points of the network. First value expected within a few weeks — the big leaks and the surplus pressure show up very fast.
• Indicative payback between 4 and 9 months, depending on the current cost per kWh and the state of the installed base.
• Hard levers: leak detection (recovering between 20 % and 40 % of the air lost), pressure reduction (each bar = 6-8 % of consumption), multi-compressor orchestration (a further 3-7 %). Electrical consumption reduction for air in the order of 15 % to 30 % as a conservative target.

And the energy manager's reasonable doubt

"What if the AI proposes lowering the pressure and a critical application falls short?" — the drop is never made in one go. The agent proposes a measured step (0.2-0.3 bar), a person validates it, it is observed for 48-72 h with the whole production plan covered, and if all goes well the next step is approved. The AI does not decide on working pressure by itself; a person signs off every reduction. The three safety rings are there precisely so that critical setpoints are not changed without human validation. And in anchored tasks such as reading a panel and comparing it against the demand recipe, the best models brought the error below 1.5 % ^[1].

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