394265 - Leakage detection in water distribution systems using routine measurements under model and data uncertainty
Wednesday, June 6
10:30 AM - 12:00 PM
Location: Lakeshore C
Mohd Anwer, Greensboro – NC A&T University; May Almousa, Greensboro, NC – NC A&T University
In most modern water networks, pressure and flow measurements are routinely collected at various points in the network for operational reasons. Since leaks typically induce a signature on these measurements, they can be used in non-intrusive leak detection approaches that rely on a hydraulic model such as EPANET. This research extends previously developed iterative linear approximation methods (based on linear and mixed integer linear programming) for detecting leaks by incorporating a number of new enhancements. These include the following: (a) change in model formulation to use flow measurements in addition to pressure, (b) algorithmic changes to handle measurement and model uncertainties, and (c) a more realistic leakage representation in the hydraulic model. The performance of the linear approximation methods will be compared against a machine learning approach that uses only sensor observations (pressure, flow) to detect leaks. The methods are validated using simulated leakage scenarios for representative networks in North Carolina and South Africa.