Sampling Design for Leak Detection in Water Distribution Networks

Abstract

The leak detection, as well as other tasks related to control and management of water distribution systems (WDS), depends on representative data on the actual state of the system, ie, reliable values for state variables acquired by sensors. Given that the number of such sensors, as well as its location in the WDS, interferes with the usability of the acquired data, the careful design of the sampling points is necessary. Although many researchers have already addressed this issue, only some of them have focused on the specific purpose of leak detection, so that there is still no known efficient solution to this problem and improvements can be achieved by exploring different criteria and solving methods.The aim of the study here described is develop a sampling design (SD) method for localization and quantification of pressure sensors in WDS, aiming leak detection. According to the proposed method, the search for the proper SD is driven by four criteria: maximization of total leak sensitivity and sensitivity consistence, and minimization of information redundancy and sensors number. The sensitivity analysis is developed using a hydraulic simulation model of the network and incorporating artificial node leaks as pressure-driven demands. Entropy is used to estimate the consistence of sensitivity to all the considered leak events. Redundancy is evaluated by the correlation between simulated responses, in terms of pressure at potential nodes for sensors installation. Finally, reducing the number of sensors is targeted, admitting that generally their availability is limited. The optimization procedure uses the multiobjective genetic algorithm NSGAII approach to search for the complete set of nodes for sensors placement.In addition to the method explanation, its application for an existing water supply district is presented, producing a consistent near-optimal set of nine well distributed places for sensor installation. The SD method proposed here could be applied to any WDS and assist advances for data-driven detection of leaks, and even for intelligent systems development for WDS.