Case study

The case studies below were recorded using a fully functioning MVP app paired with a wildlife ultrasonic microphone connected to an iPhone. The production version (Alto 1) will be built specifically for leak detection and AR based precision. This will significantly enhance detection range, accuracy, and reliability in realworld environments.

Why nitrogen gas?
‍During HVAC servicing, F-GAS certified engineers often inject nitrogen gas into systems at double the standard pressure. This helps identify pressure drops over time, a common indicator of leaks.

Because nitrogen is inert and safe to use, it's ideal for this kind of diagnostic testing. During leak detection, the nitrogen gas remains flowing through the pipes while engineers search for the leak using handheld tools.

‍My app can detect both refrigerant and nitrogen gas leaks. However, nitrogen presents a greater challenge due to its higher frequency leak signature, often exceeding 55kHz. Unlike refrigerants, which produce strong harmonic signals that make them easier to identify, nitrogen leaks emit weaker harmonics.

This increases the risk of false positives from nearby machinery or ambient noise that may share similar high-frequency patterns.

This is where Alto 1's incredible detection capability (up to 100kHz per mic) and AI-driven frequency signature analysis come into play!

Offering a detection edge even high-end sensors often lack.

The testing environment - loud & outdoors

Thanks to an HVAC maintenance partner, I got access to a rough, outdoor rooftop with four active chillers.

It's an environment so loud it pushes any leak detector to its limits. This is where I trained the model and stress-tested the app.

Nitrogen hiss noise - typical leak

Leaks are almost inaudible amid HVAC racket.

To demonstrate, I manually adjusted the nitrogen hiss to a medium size leak so you can hear it clearly at approximately 10 cm from the valve. This gives us a sense of how faint the real leak sound is in the field.

Leak detection - AR pinpoints nitrogen tank

NOVELTY! Sito app detecting a pressurised nitrogen leak in a loud HVAC environment. The red AR bubble appears only when nitrogen flows. No false positives, no guessing.
‍
Most ultrasonic detectors will need to be inches close in this environment.
‍
‍Note: With my wildlife mic I can’t do full 3D spatial audio, so the app uses a “Signal Mapper” to estimate direction, hence the “PEAK” bubble showing a rough location of the leak.

Nitrogen valve CLOSED - No false positives

Zero false positives when the tank is off.

Our three-condition system (frequency match, harmonics, SNR & spectrum check) prevents noise from registering as leaks.

If only some conditions pass, the app shows a “possible” bubble and prompts the engineer to move closer.

Measuring the distance

This clip shows leak detection at about 1 meter away in a loud setting.

In quieter environments it can hit 2–3 meters. Alto 1’s triple-mic 3D array will push that range even further.

By comparison, most ultrasonic detectors need to be inches from a leak to register anything.

Walking around the site with the app

A longer walk through to check for false positives.

spoiler: none.

Toward the end I get unusually close to the tank because the app’s audio buffer is full of prior data, making detection momentarily harder. That’s purely an MVP issue; the production version will clear its cache in real time so it stays sharp throughout your search.

Do you have any questions?

I’m happy to dive deeper into any technical detail of the app.

Contact me