Edge ultrasound for compressed air leaks in automotive — the ones that bleed most, first.

The problem

Everybody knows there are leaks. Nobody knows which ones cost the most.

In an automotive group with several workshops (body and paint, machining, assembly), compressed air is one of the most expensive utilities the plant produces. And the leak is the most universal and worst-managed loss of all:

1. Each plant runs an annual leak campaign with the external company that shows up to inspect on a Friday, leaves a PDF report and goes.
2. The report arrives with 47 leaks detected, with no ranking by real consumption. The crew repairs the 5 easy ones and forgets the rest.
3. The leaks left alive are precisely the ones that cost the most, because the cost is not visible on a single bill — it is visible in the 75 kW compressor that starts 18 minutes earlier every shift.
4. And between plants no comparison is possible: nobody knows whether workshop C leaks three times more per hour than A, because each one measures its own way (when it measures at all).

The corporate maintenance manager knows this, but his travelling crew decides by hearsay: first to the plant that complains loudest, not to the one that bleeds most. The classic system (annual campaign + checklist + intuition) works — and at the same time the group is paying for the equivalent of an extra compressor that nobody asked for.

How it fits the IRIS system

iLEAN Edge does not replace the leak technician — it gives him the ranked map he never had.

The compressed air problem is not a lack of technology: ultrasound has been detecting leaks for decades. What is missing is the layer that links the detected leak to its real cost, compares it across plants and proposes an order of attack. iLEAN is that filler that closes the gaps left open by the compressor SCADA, the CMMS and the spreadsheet from the last campaign.

Edge hears what the ear cannot. Connect captures tariffs and shutdown plans from every plant. The agent cross-checks and prioritises. The person signs off the crew's weekly plan.

The three iLEAN pieces applied to multi-plant automotive compressed air leaks:

• Edge — terminal with an integrated ultrasonic sensor (20-100 kHz). It walks the network with the technician or is installed fixed in critical zones (paint booths, solenoid valves, fittings). A trained CNN locates the leak, estimates the flow rate and translates it into kWh and €/hour using that plant's tariff. It works with no network: it records the map on cabinet power and syncs when the network comes back.
• Connect — captures from inside the real network pressure and compressor consumption from the SCADA, each plant's electricity tariff from the ERP, the shutdown plan from the CMMS, the historical leak campaigns. And from outside, the seal supplier's notice, the WhatsApp from the plant C shift leader ("number 7 is still hissing").
• Agent — cross-checks the Edge leak map of every plant with tariffs, repair accessibility and shutdown plan, and proposes a weekly ranking to the travelling crew: first the ones that bleed most and are fastest to repair. The corporate manager signs off the plan. Leaks not repaired in the last campaign are not forgotten — the system keeps counting them as € lost every hour.

See the full IRIS architecture →

Before and after

Annual leak campaign vs. continuous iLEAN mapping

Aspect	Annual campaign + checklist	With iLEAN Edge ultrasound
Detection frequency	Once a year, on a Friday	Continuous, in every plant
Prioritisation	By order of discovery	By real €/hour, across all plants
Leak cost	Estimated in the PDF, no real tariff	Quantified with the plant's electricity tariff
Comparison between plants	Impossible — each one measures differently	One corporate map, same criterion
Travelling crew	Goes where the shift leader complains most	Goes where it costs most — signed weekly ranking
Leak left unrepaired	Forgotten until the next campaign	Keeps counting as € lost every hour

Impact estimate

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

The block below is an estimate to be validated with the concrete data of your plants. We set it out so the committee has an order of magnitude; we refine it during the diagnostic.

• Group with 3-6 automotive workshops (body/paint/assembly), compressors between 55 and 110 kW per plant, a central travelling crew of 2-4 technicians.
• Edge pilot in one critical plant + integration with the compressor SCADA and the real electricity tariff. First value expected within a few weeks: the first ranked map and the first leaks repaired with verified €.
• Indicative payback between 4 and 9 months, depending on current accumulated leak flow and each plant's electricity tariff.
• Expected reduction of leak-related consumption ≥ 30% against baseline in the first half-year, with the crew attacking first the ones that bleed most.
• Hard lever: fewer kWh on the bill, month after month, comparable against the compressor SCADA. The pilot pays for itself with the savings of the first three months.

And the CFO's reasonable doubt

«What if the AI exaggerates the leak quantification?» — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI is limited to reading an ultrasonic sensor, computing flow rate with a known physical model and multiplying by the real tariff, the best models brought the error below 1.5% ^[1]. And even so, what is critical is not decided alone: iLEAN proposes the ranking, the person signs. And the kWh saved are checked against the compressor bill — no tricks.

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