Optimal Sensor Placement for Contamination Detection and Identification in Water Distribution Networks

Water Distribution Networks (WDN) is often exposed to either intentional or accidental contamination. In order to protect against such intrusions, an effective and efficient online monitoring system through sensors is needed. Detection of contaminants in WDN is challenging and it is not possible to place sensors at each and every potential point of intrusion, due to cost and maintenance reasons. Instead, as few sensors as possible, should be located optimally such that intrusions can be detected quickly. This is known as sensor network design problem for intrusion detection in WDNs. Several optimization models and algorithms have been proposed for intrusion detection in a WDN. In this study, we design sensor networks which satisfy the two important properties of observability and identifiability. Observability denotes the ability of the sensor network to detect the occurrence of the intrusion, whereas identifiability refers to the ability to unambiguously deduce the point (or source) of intrusion from the set of sensors affected. A hydraulic analysis of the network is first carried out for a given loading condition to determine the flow directions. The concept of a directed path is then used to construct a bipartite graph, and map the sensor network design problem to that of a minimum vertex set cover problem. Algorithms based on greedy heuristics are used to solve the set covering problem and obtain the corresponding sensor network. The proposed method is illustrated using a fairly large scale urban WDN.