Thames Water has developed a novel solution for understanding pressure transients in clean water networks. By applying machine-learning techniques on high-resolution pressure data from Syrinix PipeMinder-S (PM-S) devices, the Innovation team at Thames Water have been able to classify a majority of transients as normal network operations, and highlight unusual pressure events worthy of further investigation as and when Thames Water receive data.

Pressure transients occur in a pipe network when the flow of water changes suddenly, introducing huge strains on pipes and joins in the form of pressure waves that travel through the network. Some of these are unavoidable, for instance the daily activation of a booster each morning, or the switching of a pressure-reducing valve (PRV). Others suggest that a burst may have occurred, or that a big industrial user has been pumping large volumes of water out of the network.

Using over 10,500 transients recorded since January 2016 across 17 Syrinix PipeMinder-S devices in Reading, Thames Water discovered over 700 distinct groups of transients. Though this number is large, over 80% of the transients belong to just 3% of these clusters that mostly describe booster and PRV activities. The information contained in each of these transients is useful as it can help to locate the sources of leakage or excessive stress on the network, but assessing each by eye is a hopeless task. The developed algorithm is highly scalable and further research will combine with other Thames Water initiatives to help identify and locate points of interest.