NTT Uses Fiber Optics and AI to Locate Water Leaks

NTT and the Fukuoka City Waterworks Construction Cooperative successfully tested a method using fiber-optic sensing and AI to locate leaks in water pipes in Fukuoka City. The test achieved 88% accuracy within a 75-centimeter horizontal error margin. NTT said it was the world’s first successful demonstration of locating water-pipe leaks with fiber-optic sensing.

Traditional surveys relied on workers listening for sounds to estimate leak locations, a skill that took several years to learn. The demonstration exceeded the stated benchmark for surveys before leak-repair work: at least 80% accuracy within an 80-centimeter error margin. It also confirmed that even less-experienced workers could automatically identify a leak’s location in about 15 minutes, from measurement through AI analysis.

For the survey, densely arranged fiber-optic sensors were laid on the road directly above the water pipe, and faint vibrations associated with leaks were measured for 5 minutes. AI separated leak-related vibrations from surrounding noise and estimated the leak location by identifying a spot with distinctive vibrations compared with other points. The tested pipes were buried 30 to 120 centimeters underground.

NTT plans to improve the accuracy of the AI analysis and the system’s usability in the field, with the goal of introducing it to water utilities and local governments across Japan.

NTT and the Fukuoka City Waterworks Construction Cooperative successfully tested a method using fiber-optic sensing and AI to locate leaks in water pipes in Fukuoka City. The test achieved 88% accuracy within a 75-centimeter horizontal error margin. NTT said it was the world’s first successful demonstration of locating water-pipe leaks with fiber-optic sensing.

Traditional surveys relied on workers listening for sounds to estimate leak locations, a skill that took several years to learn. The demonstration exceeded the stated benchmark for surveys before leak-repair work: at least 80% accuracy within an 80-centimeter error margin. It also confirmed that even less-experienced workers could automatically identify a leak’s location in about 15 minutes, from measurement through AI analysis.

For the survey, densely arranged fiber-optic sensors were laid on the road directly above the water pipe, and faint vibrations associated with leaks were measured for 5 minutes. AI separated leak-related vibrations from surrounding noise and estimated the leak location by identifying a spot with distinctive vibrations compared with other points. The tested pipes were buried 30 to 120 centimeters underground.

NTT plans to improve the accuracy of the AI analysis and the system’s usability in the field, with the goal of introducing it to water utilities and local governments across Japan.