Distributed Network Fusion for Water Quality

To protect drinking water systems, a contamination warning system can use in-line sensors to detect accidental and deliberate contamination. Currently, detection of an incident occurs when data from a single station detects an anomaly. This paper considers the possibility of combining data from multiple locations to reduce false alarms and help determine the contaminant's injection source and time. If we consider the location and time of individual detections as points resulting from a random space-time point process, we can use Kulldorff's scan test to find statistically significant clusters of detections. Using EPANET, we simulate a contaminant moving through a water network and detect significant clusters of events. We show these significant clusters can distinguish true events from random false alarms and the clusters help identify the time and source of the contaminant. Fusion results show reduced errors with only 25% more sensors needed over a nonfusion approach.