OLD hands at some water companies still on occasion whip out a pair of dousing rods or find a Y-shaped twig to search for a leak in an underground pipe. Dousing, or water witching as it is known in America, has no basis in scientific fact. A somewhat more reliable method involves using acoustic equipment called geophones to listen for escaping water. The trouble is it takes an experienced ear to distinguish the sound of a leak from the normal gurgle of water passing though pipes, let alone to predict from that sound where any trouble might be found. As this is a problem of pattern recognition, which is something that artificial intelligence can be good at, a Brazilian startup has used AI to develop an acoustic device that makes tracing leaks a lot easier.

Almost 37% of the water pumped into Brazil’s supply network disappears, because of leaks, faulty meters returning an inaccurate reading or theft. That amounts to 6.5bn cubic metres in wasted water a year, worth some 8bn Brazilian real ($2.3bn) in forgone revenue, according to some estimates. Stattus4, based in Sorocaba, São Paulo state, reckons its portable leak-detection tool can not only indicate if water is going missing but also identify the different sorts of losses and where they might be.

The company’s system, called Fluid, consists of a long hand-held probe on the end of which is a sensor. An operator places the sensor against an accessible area of pipework in, say, a culvert or under a manhole cover, for about 15 seconds to record vibrations in the water flowing through the local network. This sample is relayed to a smartphone app, which analyses the recording using a database stored on a cloud computer. This is a bit like how a smartphone music-recognition app, such as Shazam, works, says Marília Lara, a co-founder of the company. (The snatch of water music that Fluid listens to, however, also includes a range of frequencies which are inaudible to the human ear.)

With enough samples, the company says it is possible to match sounds captured in one place to those of a leak a bit farther along the pipework, say. An illegal connection might be detected from the vibrations caused by an outflow where one is not supposed to be.

A pilot version of the system is in use at ten water companies, which are providing 800 new sound samples a day into a database that numbers over 40,000 already. The idea is that Fluid can be used to quickly narrow down the location of any water losses, allowing operators using geophones to pinpoint the exact site. In one case, in Votorantim, a city in São Paulo state, it would have taken the local water company two years to survey its network of pipes using two operators armed with geophones. With three Fluid devices looking for potential loss points it was possible to reduce that time to four months, claims Stattus4. And unlike dousing, there was no need for a splash of good fortune.