Leak detection analytics that sends crews to real leaks, not false alarms
Send crews to real leaks, not false alarms.
See one case, screen by screen ↓A water network now listens to itself. Acoustic loggers on valves and hydrants report noise every night, satellite passes return leak polygons, district meters report each zone's minimum night flow and smart meters flag continuous flow at homes. On Lumina Utilities' East Network — 212 km of mains, 1,140 pipe segments, six pressure zones and 48,600 customers — a normal morning brings 1,286 alerts. Most are not leaks: a pump hums, a valve vibrates, a park is irrigated, a customer fills a pool.
Each source is read on its own — the logger portal, the satellite report as a PDF and a shapefile, the night-flow spreadsheet — by different people on different days. So a crew is sent on one alarm and digs a dry hole, or a non-surfacing leak that was heard but not trusted runs until it surfaces or bursts. From January to June, 66 of 212 digs found no leak, and a leak ran on average 34 days from the first alert to the repair.
Estimated split for one non-surfacing leak on a distribution main, in calendar days, by hand (34 days on average, as on the sample network) and with the solution (the target is under 10).
Six specialist agents listen, match, watch and score every pipe; the leakage engineer sends the crew.
Monday 13 October 2026. On Lumina Utilities' East Network, Millbrook Low's night flow has been up since the night of 4 October, and nobody has dug yet. Here is what happens next, screen by screen, in the working solution.
The morning run starts with the acoustic loggers: 1,145 alerts from 4,210 loggers, on 262 pipes. The logger listener keeps 196 persistent rises and sets aside 949, each with its reason — pumps and valves within 30 m, loggers noisy since install, noise not repeated the next night.
On the map: “Most are pumps, valves, hydrants, or quiet again the next night. On its own, acoustic cannot tell a leak from a hum.”
The 9 October pass returned 41 polygons. The matcher lays each one over the network, keeps 13 that sit on pipes and sets aside 28 over parks, gardens, the brook and fresh repairs. A polygon shows treated water in the ground — it says where to look, not where to dig.
“Each is about 100 m across — an area, not a point.”
Minimum night flow between 02:00 and 04:00, against each zone's normal night use. Millbrook Low has stepped from 18.2 to 32.6 m³/h since the night of 4 October, 14.0 m³/h above normal. Of 96 smart-meter continuous-flow flags, 71 are customer-side leaks — taken out of the zone figures and drafted as notices for customer care — and 25 are normal use.
What the night flow watcher found: “Millbrook Low +14.0 m³/h · all on MB-0412 Orchard Way.”
The fuser scores all 1,140 pipe segments once, from the three sources plus the pipe's age, material and repairs. 27 segments become leak cases; on 4 of them all three sources agree. If every alert were a job, the crews would make 307 visits.
“307 visits if every alert were a job — the old way, where about 3 digs in 10 found no leak.”
Hana decides which hotspots get a crew today. With the rule set to all three agree, four pipes are lit. She approves Larkspur Road, Ferry Road and Fenwick Close between 07:41 and 07:52. Orchard Way, MB-0412 in Millbrook Low, waits for her: acoustic 0.91, satellite 0.88, night flow 0.97, fused score 0.94, estimated leak 14.0 m³/h — 336 m³ a day.
On the map: “All three agree · estimated leak 14.0 m³/h · 336 m³ a day.”
Logger L-2217 on valve V-MB-118 went from 29 dB to 41 dB with a narrow 4 dB spread, every night for nine nights; correlated with L-2219, the noise sits 38 m from V-MB-118. Polygon SAT-0931 covers Orchard Way, its centre 46 m from the main. The pipe is 150 mm cast iron, 214 m, with repairs in 2019, 2022 and 2024. The score adds up on screen: 0.364 + 0.220 + 0.339, plus 0.017 for pipe risk.
“The satellite sees treated water in the ground, not the pipe — it says where to look, the loggers say where along it.”
Work order WO-26-18842, pinpoint and repair, 38 m from V-MB-118. Sam Whitlock pinpoints at 09:30 with a correlator and ground microphone; Gang 2, Delta Civils, repairs at 13:00 under lane closure LP-26-3381. The job closes only when night flow drops by 11.2 m³/h or more — 80 % of the estimate.
“You are approving as Hana Okafor, leakage engineer. The job writer drafted this; the dispatcher books the crews; the case closes only after the night-flow check.”
Orchard Way moves to To pinpoint, beside Larkspur Road, Ferry Road and Fenwick Close. Station Road and Wharf Road, where only two sources agree, are proposed for a technician to listen first. Bramble Lane is pinpointed 12 m from hydrant H-AV-31 and waits for its repair; Kingsmead Avenue has its road permit for 15 October.
“Every job starts from a leak case and ends only when night flow shows the water has stopped.”
Quayside was repaired on 11 October. At area meter SM-RL-07 night flow is down 5.1 m³/h against 5.6 estimated (91 %), and the logger is back to its normal level, so the repair verifier proposes closing. On Hazel Walk night flow fell only 0.6 of 2.4 m³/h and the logger still hears noise 60 m further along: it proposes reopening instead.
Hazel Walk: “A second leak, or the repair has not held.” The verifier proposes; Hana closes.
171 of 186 digs since go-live found a leak (92 %), against 146 of 212 from January to June. East Network leakage is 6.2 Ml/d against 7.9 on 1 July, already under the 6.32 Ml/d target for July 2027, and verified repairs recover 2,960 m³ a day, net of new leaks. Gwen exports the weekly leakage report.
“Leakage = night flow of all six zones less normal night use, as reported to the regulator each week.”
The same leak on Orchard Way, seen by the five people who find it, book it, dig it, report it and talk to customers — what their week looked like, and what it looks like now.
Six specialist agents listen, match, watch and score every pipe; the leakage engineer sends the crew.
Reads each night's logger noise and spread against the logger's own 30-night normal, keeps a rise of 8 dB or more that repeats for 3 nights with a spread of 6 dB or less, and correlates neighbouring loggers to a point on the pipe.
Lays each satellite polygon from the latest pass over the pipes, works out the share of the polygon on each, and screens out parks, gardens, sports pitches, streams and pipes repaired in the last 14 days.
Computes each zone's minimum night flow from the district meter against normal night use — smart-meter night use plus new connections — flags a step rise of 2 m³/h or more, and shares the excess over the zone's pipes by age, material and repairs.
Gives every pipe segment one score from the weighted acoustic, satellite and night-flow values, adjusted by up to ±0.10 for pipe risk, opens a case where at least one source agrees, ranks the cases and says which source drives each score.
Drafts a pinpoint-and-repair job where all three agree and a listen-first job where two agree, with the street, the distance from the nearest fitting, the pipe, access, permits, traffic management, the estimated flow and the closing rule.
Compares the two nights after a repair with the nights before, on the zone's night flow and the logger's noise, and proposes closing when the drop is 80 % or more of the estimate and the logger is back to normal — otherwise reopening, with where the noise is.
Sends the crew. The agents propose; a named person decides.
Ask about any pipe, zone or job — or what to send a crew to next. Every answer points to the readings behind it.
Why is Orchard Way deep red?
All three sources agree on one pipe, MB-0412. Logger L-2217 has heard a steady 41 dB noise for nine nights, and it correlates 38 m from valve V-MB-118. The 9 Oct satellite pass puts polygon SAT-0931 on the same street. Millbrook Low's night flow has been 14.0 m³/h above normal since 4 Oct. It is a 1962 cast-iron main with three repairs. Score 0.94, estimated 14 m³/h.
Which hotspots don't have a crew yet?
One pipe where all three agree is waiting for you: MB-0412 Orchard Way, est. 14.0 m³/h. Two more are proposed for a technician to listen first: AV-0109 Station Road (acoustic and satellite, 1.6 m³/h) and RL-0051 Wharf Road (acoustic and night flow, 0.7 m³/h).
What did we set aside this morning, and why?
1,002 of 1,286 alerts. The biggest groups: 488 not repeated the next night, 164 within 30 m of a pump or valve, 147 the same noise already in a case, 97 from loggers noisy since install. 23 satellite polygons sit over parks, gardens and Mill Brook. Each one can be put back.
Are we on track for the leakage target?
Yes. East Network leakage is 6.2 Ml/d against 7.9 on 1 July, already under the 6.32 Ml/d target for July 2027. 171 of 186 digs since go-live found a leak (92 %), against 69 % from January to June. Verified repairs recover 2,960 m³ a day.
13 screens from the working solution, on its sample data. Pick one to see it large.
How the three sources are weighed, when a crew is sent and who approves are settings, chosen by the network manager.
| Setting | Default | Choose from |
|---|---|---|
| Weight of each source | Acoustic 0.40 · satellite 0.25 · night flow 0.35 | Set with − and + for each source; the weights add to 1.00 |
| Pipe risk adjustment | ±0.10 | Age, material and repairs move a score by up to ±0.10 |
| Acoustic agrees when | A rise of 8 dB or more for 3 nights, spread 6 dB or less | A source agrees at a score of 0.60 |
| Satellite and night flow agree when | 30 % or more of the polygon lies on the pipe · the zone is 2 m³/h or more above normal | A source agrees at a score of 0.60 |
| A crew is sent when | All three agree | All three agree · Two agree and a technician listens first |
| Approves every job | Leakage engineer | Hana Okafor approves jobs and closes cases; Gwen Talbot changes rules and signs the weekly report |
| Alerts set aside automatically | On for six rules | Loggers within 30 m of a pump or valve · loggers noisy since install · alert not repeated the next night · polygons over parks, gardens, sports pitches and streams · pipes repaired in the last 14 days · customer-side leaks |
| Smart-meter flow under 0.01 m³/h | Off | On · Off — set aside as dripping taps, not leaks |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample network data · “estimated” = our arithmetic, inputs shown ($480K = 600 digs a year × 23 fewer dry digs in 100 × $3,500 a dig, the sample utility's figures; the days split is our estimate) · “typical” = published figures (Liemberger & Wyatt, 2019: 126 billion m³ of non-revenue water a year worldwide; a UK water company news release, Aug 2026: more than 800 leaks, an estimated 8.7 Ml/d, found in a 19-week satellite trial). Lumina Utilities and the people, streets, crews, contractors and records named on this page are fictional — characters and sample data in the working solution.
Leak detection analytics puts every leak signal a water network produces — acoustic logger noise, satellite leak polygons, zone night flow and smart-meter flags — on one map and scores each pipe once. Leak Detection Analytics does this every morning at 06:00 with six agents, so the leakage engineer sends a crew only where the sources agree.
Each agent reads one source: the logger listener finds persistent noise rises and correlates logger pairs, the satellite polygon matcher lays polygons over the pipes, and the night flow watcher compares each zone's minimum night flow with normal night use. The evidence fuser snaps every alert to a pipe segment and gives it one weighted score, adjusted for pipe age, material and repairs.
It sets aside alerts that are not leaks, with the reason — pump and valve noise, loggers noisy since install, noise not repeated the next night, polygons over parks and streams, fresh repairs. A crew is sent by default only where all three sources agree. In the working solution, 92 % of digs since go-live found a leak, against 69 % before.
Both, together with night flow. It works with the loggers, satellite surveys, district meters and smart meters a utility already has; it does not replace them. Satellite says where to look, the loggers say where along the pipe, and night flow says how much water is going.
After a repair, the repair verifier compares the nights before and after on the zone's night flow and the logger's noise. It proposes closing only when the drop is at least 80 % of the estimated leak and the logger is back to normal; otherwise it proposes reopening, with where the noise is.
Continuous flow inside a property becomes a drafted customer notice for customer care to review and send, never a crew job, and it is taken out of the zone's leakage figure. On the sample morning, 71 of 96 smart-meter flags were customer-side leaks.
Yes. The agents score, draft and propose; they never dispatch a crew, close a case or change a rule. The leakage engineer approves every job and closes every case, the dispatcher books the crews, and every change to the weights or set-aside rules needs the network manager and becomes a new rule version.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your pressure zones, pipe records, logger and satellite feeds and dispatch standards — and it goes live once every quality gate has passed. We will walk you through it on a sample of your own alerts first.
Leak crews sent where the loss is certain, not where a meter is in doubt.
1-min demoWater Quality Compliance MonitoringExceedances caught the hour they happen, and low chlorine seen two days ahead.
1-min demoWater Quality Complaint InvestigationBrown-water calls traced to their cause in minutes, the flush and every reply drafted.
1-min demoWe'll run Leak Detection Analytics on a sample of your own logger alerts, satellite polygons and night flows.