Latest / Internet of Things with Fexingo: Connected Devices, Sensors, and Industrial IoT / How IoT Sensors Are Detecting Water Leaks in Ancient Aqueducts
Transcript
- Lucas: So imagine a two-thousand-year-old aqueduct that still carries water into a major European city. Now imagine trying to find a hairline crack in its masonry before that crack becomes a collapse. Luna: And you're going to tell me IoT sensors are doing that right now. Lucas: Exactly. There's a pilot project that started last year in Rome, on the Aqua Virgo. That's the aqueduct built in 19 BC that still feeds the Trevi Fountain. The water authority there deployed two hundred low-power vibration and acoustic sensors along a twelve-kilometer stretch of underground tunnel. Luna: Two hundred sensors in a two-thousand-year-old structure. How do you even attach them without damaging the brickwork? Lucas: That was one of the big engineering challenges. They used non-invasive clamps and adhesive mounts designed for heritage surfaces — nothing drilled into the original fabric. The sensors themselves are about the size of a deck of cards, and they communicate via a low-power wide-area network, so no trenching for cables either. Luna: And what are they actually listening for? Lucas: Two main signals. One is vibration frequency shifts — when water flows through a crack, the turbulence creates a specific signature in the sub-sonic range. The other is acoustic emissions from the masonry itself: tiny popping sounds as crystals in the mortar fracture under stress. The system's machine learning model was trained on about six months of baseline data to distinguish those sounds from background noise like traffic rumble or nearby construction. Luna: So it's basically a continuous listening device for the structure. How accurate has it been? Lucas: The early results are pretty impressive. After the first year, the system flagged forty-three potential leaks. Traditional inspection — which involves sending a crew into the tunnel to visually check — had only identified eleven of those. And of the thirty-two the sensors caught first, seven were described by the engineers as 'imminent failures' that could have flooded parts of the tunnel within weeks. Luna: Seventy percent more detections. And presumably way cheaper than emergency repairs after a collapse. Lucas: Exactly. The water authority estimated that emergency repairs on a collapsed aqueduct section can run upwards of two million euros, between excavation, heritage restoration, and water service disruption. The entire sensor pilot cost under four hundred thousand. So even if it prevents one major incident every few years, it pays for itself. Luna: But the Aqua Virgo is just one aqueduct. How scalable is this approach? I imagine every tunnel has different masonry, different water chemistry, different environmental conditions. Lucas: That's the research frontier right now. The consortium behind the pilot — it's a joint project between the University of Rome, the water utility ACEA, and a Spanish IoT firm called Sensaqua — they're working on a transferable model. The idea is that you train the algorithm on the first six months of data from any new site, and it learns the site-specific acoustic signature. They're already planning deployments in the Roman aqueduct at Segovia and, more surprisingly, in the Channel Tunnel drainage system. Luna: The Channel Tunnel? That's not exactly two thousand years old, but the stakes are enormous. One leak in the wrong place and you've got a very expensive problem. Lucas: Right. The Channel Tunnel has a complex network of drainage pipes and sumps to handle groundwater seepage. Currently they do visual inspections every six months, but micro-cracks can grow fast. The Sensaqua team is adapting their acoustic sensors for the higher humidity and saltier environment there. It's a good test of whether the technology can move from heritage structures to modern critical infrastructure. Luna: One thing I'm wondering about: sensors in a damp tunnel, running on batteries — how long do they actually last? Lucas: The Aqua Virgo sensors are rated for five years on a single battery pack, because they only transmit data once an hour unless they detect an anomaly, in which case they wake up and send a continuous stream for thirty minutes. They also have energy harvesting from the water flow vibration — a tiny piezoelectric element that generates a trickle charge. Early field data suggests they might actually exceed the five-year estimate. Luna: That's clever. So the vibration you're monitoring also powers the sensor that monitors it. There's something elegant about that. Lucas: It is elegant. And it makes the system truly low-maintenance, which is critical for any deployment in hard to reach locations. The Segovia aqueduct runs partly above ground, so sensor access is easier, but parts of it are inside a hillside. And the Channel Tunnel sensors will be in areas that require rail traffic shutdowns to reach. Luna: Let me ask about the data side. Two hundred sensors, each sending hourly readings plus anomaly bursts — that's not a trivial data stream. How do they handle false positives? Because I imagine a train passing or a maintenance vehicle could trigger the sensors. Lucas: They had a lot of false positives in the first three months. The model initially flagged anything with a strong vibration signature — including garbage trucks on the street above. So they built a multi-layer filter. First, the sensor itself does some onboard processing: it compares the signal to a stored template of known background noise. If it's a match, it discards it. Then the cloud model does a second pass, looking at correlated signals across multiple sensors to triangulate whether the anomaly is localized or widespread. By month four, false positives had dropped by eighty-five percent. Luna: That's a solid improvement. And presumably the model keeps learning as more data comes in. Lucas: Yes, and that's actually one of the more interesting outcomes. The model has started to identify patterns that the human engineers hadn't noticed — for example, a subtle vibration shift that occurs about three days before a visible crack appears. They think it's the result of micro-stresses as the masonry adjusts to temperature changes. So now they're using that as a predictive signal to schedule proactive reinforcement. Luna: So the sensors aren't just detecting active leaks — they're predicting where leaks will form. That shifts the whole maintenance paradigm from reactive to truly preventive. Lucas: Exactly. And that's the holy grail for infrastructure managers: knowing where to reinforce before there's any sign of water. The Rome water authority is now planning a phase two that would expand the sensor network to cover the entire Aqua Virgo, about twenty kilometers of tunnels, plus the other major aqueducts in the city. They're talking about a city-wide acoustic monitoring mesh. Luna: And that kind of project — protecting infrastructure that's literally irreplaceable, using sensors that listen and learn — it's exactly the kind of practical innovation that makes this show worth doing. If today's episode gave you something useful, the way these stay ad-free is listener support. Lucas: Yeah, it's a good reminder. If you value deep dives like this one, you can help keep them coming at buy me a coffee dot com slash fexingo. No pressure, just a way to support the kind of reporting that digs into real engineering. Luna: Exactly. And now back to those Roman tunnels — Lucas, what about the preservation ethics? Some archaeologists argue that installing any modern equipment inside an ancient structure is a form of alteration. Lucas: That's a fair debate. The Sensaqua team worked closely with the Italian Ministry of Cultural Heritage to ensure all mounts were reversible and left no permanent marks. They also color-matched the sensor housings to the travertine stone so they're visually unobtrusive. But some purists still object. The counterargument, of course, is that a catastrophic collapse would be a far greater loss than a few carefully placed clamps. Luna: And the data from the sensors could actually inform better preservation techniques for other historic structures. Lucas: Exactly. The vibration signatures they're collecting are helping materials scientists understand how ancient mortar ages under different humidity and load conditions. That knowledge could influence how we restore other Roman structures, and even how we design modern masonry for long-term durability. So the sensors aren't just protecting the aqueduct — they're teaching us something about the material itself. Luna: I love that. The past informing the future, and the future helping preserve the past. What's next for this project? Lucas: Phase two expansion is scheduled to start this September. And there's talk of adapting the system for the Pont du Gard in France, which is a UNESCO World Heritage site. That one is above ground, so the acoustic environment is completely different — more wind, more tourist noise — but the core approach should transfer. And the Channel Tunnel pilot is set to begin sensor installation in October. Luna: We'll have to do a follow-up once those are live. For now, it's a reminder that sometimes the most cutting-edge tech is being used to keep ancient wonders from crumbling. Lucas: Absolutely. And that's the story from the Aqua Virgo — IoT sensors listening for whispers in the stone.