Flow data based leak detectionfor underdetermined water networks

Abstract: Water leaks are a common and wasteful occurrence in all Water Distribution Networks (WDNs). There has been extensive research for detecting and localizing leaks. However, an overwhelming number of these techniques use both pressure and flow of the WDN. Such approaches, while being successful, are often solving the overdetermined case (fewer unknown measurements than known measurements), which is not applicable to many WDNs due to lack of measurement infrastructure. Therefore, there is a need for a detection model that works on underdetermined problem (fewer measurements than unknown measurements) with only flow data, to detect and localize leaks in a network. This exploratory study introduces a leak detection model that works on limited flow data to detect and localize leaks. General equations that describe every flow in a network as a sum of its downward flows are proposed. Using the equations and known measurements, the model detected areas of the network that contain leaks. To localize leaks down to a specific pipe, physical characteristics of the network pipes are taken into account to propose a probability leak model. This probability leak model is applied on the detected areas to localize leaks to a pipe. In absence of real-life data, this model has been test on a variety of simulated data. The notable results was that the leak detection model was better at detecting the correct location of a leak, than at detecting the quantity of leak present at that location."