Detection of Patterns in Water Distribution Pipe Breakage Using Spatial Scan Statistics for Point Events in a Physical Network |
| |
Authors: | Daniel P. de Oliveira Daniel B. Neill James H. Garrett Jr. Lucio Soibelman |
| |
Affiliation: | 1Project Manager, Office of Campus Planning, The University of Texas at Austin, 5404 Grover Ave., Austin, TX; formerly, Ph.D. Candidate and Graduate Research Assistant, Dept. of Civil and Environmental Engineering, Carnegie Mellon Univ. (corresponding author). E-mail: danielpinhodeoliveira@gmail.com 2Assistant Professor of Information Systems, H.J. Heinz III College, School of Public Policy and Management, School of Information Systems and Management, Carnegie Mellon Univ., 5000 Forbes Ave., Pittsburgh, PA 15213. E-mail: neill@cs.cmu.edu 3Professor and Head, Dept. of Civil and Environmental Engineering, Carnegie Mellon Univ., 5000 Forbes Ave., Pittsburgh, PA 15213. E-mail: garrett@cmu.edu 4Professor, Dept. of Civil and Environmental Engineering, Carnegie Mellon Univ., 5000 Forbes Ave., Pittsburgh, PA 15213. E-mail: lucio@andrew.cmu.edu
|
| |
Abstract: | Infrastructure systems of many U.S. cities are in poor condition, with many assets reaching the end of their service life and requiring significant capital investments. One primary requirement to optimize the allocation of investments in such systems is an effective assessment of the physical condition of assets. This paper addresses the physical condition assessment of drinking water distribution systems by analyzing pipe breakage data as the main source of evidence about the current physical condition of water distribution pipes over space. From this spatial perspective, the primary questions are whether data sets present unexpected clustering of pipe breaks, and where those break clusters are located if they do exist. This paper presents a novel approach that aims to detect and locate clusters of break points in a water distribution network. The proposed approach extends existing spatial scan statistic approaches, which are commonly used for detection of disease outbreaks in a two-dimensional spatial framework, to data collected from networked infrastructure systems. This proposed approach is described and tested in a data set that consists of 491 breaks that occurred over six years in a 160-mi water distribution network. The results presented in this paper indicate that the adapted spatial scan statistic approach applied to points in physical networks is able to detect clusters of noncompact shapes, and that these clusters present significantly higher than expected breakage rates even after accounting for pipe age and diameter. Several possible hypotheses are explored for potential causes of these clusters. |
| |
Keywords: | Spatial analysis Statistics Water distribution systems Water pipelines Failures |
|
|