Water Distribution Burst Detection and Localization Using Advanced Metering Infrastructure Data Collection Systems

Abstract

Pipe bursts are one of the most common failures in water distribution networks (WDNs). To minimize their impact, numerous burst detection and localization methods have been developed that identify failures using hydraulic measurements (e.g., pipe flow and pressure) collected from supervisory control and data acquisition (SCADA) data collection systems (DCSs). However, since their monitoring networks are sparse, SCADA systems are often insufficient to identify realistic sized bursts. A clear next step is to develop detection and localization methods for smart systems that collect advanced metering infrastructure (AMI) data (i.e., AMI systems). However, no previous work has proposed tools for the AMI DCSs.The goal of this dissertation is to develop a series of burst detection and localization methods that employ AMI data. Depending on the data being measured (whether AMI flow and/or pressure), appropriate detection/localization approaches are proposed and tested under a range of realistic burst sizes. In addition, these methods are compared for different levels of AMI DCSs to identify which method works best for a given WDN. This dissertation is composed of six journal manuscripts that propose several burst detection and localization approaches for AMI DCSs. First, a new burst detection algorithm employing AMI demands is developed for AMI system where individual end-user demands and system inflow rates are measured. With that approach as a basis, the impact of missing AMI data is assessed. Then, convolutional neural network deep learning models and linear programming based burst detection and localization methods are developed for AMI DCSs where both AMI demand and pressure are measured. Finally, the proposed methods are tested for alternative WDNs and are evaluated using metrics of detection probability, false alarm rate, time to detect, and localization pipe distance.