Date of Award
8-2022
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Civil Engineering
Committee Chair/Advisor
Kalyan R. Piratla
Committee Member
Da Li
Committee Member
Kapil Chalil Madathil
Committee Member
Ilya Safro
Abstract
Overpopulation and climate change have direly challenged the freshwater resources, specifically potable water supplied by water distribution networks (WDNs). One aggravating issue associated with the WDNs is associated with the pipeline leakage, which accounts for almost 20% of freshwater loss in WDNs throughout the US. Leakage detection and severity measurement are of top
asset management priorities in water utilities to minimize and mitigate complicated risks attributed to background and burst leakage. Accordingly, decline in other pipe condition parameters such as effective hydraulic diameters and roughness coefficients, which are complex and uncertain in nature, abets leakage by worsening the WDN status quo over time and can subject pipes to catastrophic
breaks if not addressed in time. Novel model-based condition assessment approaches leveraging advanced metering infrastructure (AMI) and cyber-monitoring data offer promising benefits and challenge conventional, hands-on methods, which are labor-intensive, costly, and often inefficient. Moreover, employment of machine learning algorithms coupled with optimization methods in such model-based schemes constitutes the state-of-the-art research horizons regarding condition assessment practices in water utilities. Along similar lines, the major goals of this study are: (1) introduction and development of a reverse-engineered, model-based condition assessment framework using consumption and monitoring data, (2) employment of genetic algorithms and particle swarm optimization to predict condition parameters such as effective hydraulic diameters, roughness coefficients, and leak severity, (3) incorporation of optimized artificial neural networks (ANNs) to circumvent the time-consuming EPANET 2.2 simulator toolkit, (4) validation of the robustness of the model through various performance metrics by exposing the framework to different WDN benchmarks and a comprehensive series of sensitivity analyses, (5) implementation of an exhaustive optimization procedure for the location and number of cyber-monitoring sensors, and (6) stochastic prediction of condition parameters by incorporating Monte Carlo simulation into the proposed framework.
Recommended Citation
Momeni, Ahmad, "Machine Learning-Enabled Model-Based Condition Assessment of Water Pipelines by Leveraging Hydraulic Monitoring Data" (2022). All Dissertations. 3110.
https://open.clemson.edu/all_dissertations/3110
Author ORCID Identifier
0000-0002-4506-517X