Every part of the system cut in half. Move the sliders — the physics reacts. No black boxes: a coil, a diaphragm, a microphone on a pipe, a radio in a manhole, and one rule about 3 a.m.
Every district already has one: a concrete box under a cast-iron lid where the main enters and a valve can shut it. Nothing is dug. The flow meter goes in-line, the pressure sensor screws onto a tapping, the node box hangs on the wall, and the antenna sits just under the lid.
Water conducts electricity a little. Push it through a magnetic field and it behaves like a wire moving through a magnet: a voltage appears across it (Faraday). Two electrodes on the pipe wall read that voltage. Faster water, higher voltage. Nothing turns, nothing wears, nothing clogs.
A thin steel diaphragm sits between the water and a sealed chamber. Water pushes, the diaphragm bends by a few microns, and four tiny resistors printed on its back stretch and change resistance — a Wheatstone bridge turns that into a voltage. 4–20 mA or 0–5 V out, read every minute at night.
Four things in an IP68 box the size of a paperback: a microcontroller that sleeps, a LoRa radio, a lithium pack, and the connectors to the two sensors. It wakes up, reads the sensors, sends ~20 bytes, and goes back to sleep. The battery life comes from how little it is awake.
| state | current | time / day |
|---|---|---|
| deep sleep | 15 µA | 23 h 55 min |
| measure | 30 mA | 96 × 1 s |
| LoRa transmit (SF9, 14 dBm) | 120 mA | 96 × 0.3 s |
| ≈ 2.3 mAh / day → 19 Ah pack lasts > 8 years | ||
A crack or a corroded hole. Water at 4 bar is forced through a gap of a few millimetres at 20–30 m/s. Three things happen at once, and each one is a sensor:
Bigger hole: louder, more flow, bigger pressure drop. Slide it.
By day a leak is buried under thousands of taps. Between 02:00 and 04:00 a district of 1,000 homes uses almost nothing — a few toilets, a bakery. Everything else flowing in is leaking out. The system reads the minimum in that window and compares it with the same district's own history.
It is the IWA standard method, usually done by a technician once a year with a clipboard. Here it happens every night, in every district, and the alarm has a confidence score because the baseline is a distribution, not a number.
Water loses pressure as it moves through pipe — slowly if the pipe is healthy. A leak between two sensors is extra flow through that section, so the pressure drop across it is steeper than across the others. Compare the drops, and the leak segment stands out.
Move the leak. Watch which pair of sensors disagrees.
The hiss from the hole runs along the pipe wall in both directions at the speed of sound in that pipe — about 1,180 m/s in cast iron, 400 m/s in plastic. It reaches the nearer logger first. The two recordings are slid against each other until they line up; the slide needed is the delay Δt, and the delay is a distance.
Loggers are a piezo accelerometer on a magnet, stuck to a valve or hydrant. The crew moves them; they are not permanent.
A cast-iron lid and a metre of wet concrete eat 20–35 dB of signal. LoRa is built for exactly this: it spreads each bit over a long chirp so the receiver can dig it out below the noise floor. Slower chirps (higher spreading factor) go further and through more concrete, but cost airtime and battery. The thesis measures this from real chambers and picks the setting per node.
| SF | sensitivity | airtime, 20 B | open range | from chamber |
|---|---|---|---|---|
| SF7 | −123 dBm | 61 ms | 3 km | 0.8 km |
| SF9 | −129 dBm | 206 ms | 6 km | 2 km |
| SF12 | −137 dBm | 1,483 ms | 14 km | 5 km |