HYELE uses a machine-learning technique to learn features and distinguish demand and maintenance activity from leak or burst signatures, and they correlate known maintenance and repair events where available to reduce false positives.

HYELE acknowledged unmapped service connections and explained that their ML stage provides coarse localization, and then their time-reversal method uses induced waves and on-site measurements to achieve finer localization near connections.

Time-reversal actively probes the system by sending waves and can identify and pinpoint existing leaks, while machine learning alone cannot identify past leaks unless provided with historical data (one to two years).

Waves are generated using a temporary side-discharge box or controlled hydrant flashing with rapid valve closure to introduce safe transient waves for a short data-capture interval.

The wave generator is temporary and used only during the condition-assessment or pinpointing exercise; it is removed after testing and the measurement window is typically seconds while setup takes longer.

HYELE uses piezo-resistive pressure sensors by default, and their hardware can accommodate other sensor types (for example, accelerometers or flow sensors) if needed.

Sensor spacing depends on the utility's required resolution and accessibility; generally HYELE recommends a couple of hundred meters between sensors.

Devices upload pressure-versus-timestamp measurements over 4G using SIM cards, and all data are synchronized with GPS.

The system requires at least two sensors per DMA as a minimum for the time-reversal application and that further sensor count depends on desired resolution and client needs.

The minimum detectable leak depends on system noise and signal-to-noise ratio; the team measures system dynamics first and then reports the smallest detectable leak for that DMA.

Pipe material and diameter do not materially affect time-reversal results if the pipe is fully pressurized, but unknown valve states or air/blockages are detected as part of the condition assessment, and accurate system data improves results.

Yes; the team builds a hydraulic/transient model and works with the client to obtain pipe material and layout information to improve accuracy.

The team proposed either simulating a leak in a healthy DMA or testing a DMA with unidentified leaks and comparing detection rates and localization accuracy against previous techniques to match client expectations.

We can share presentations and videos from Hong Kong cases, noted plans to implement in Guangzhou, and mentioned exploratory cases in Africa and Uganda; they will share those materials.