ORCID Identifier(s)

0000-0002-2502-0433

Graduation Semester and Year

Summer 2026

Language

English

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Industrial Engineering

Department

Industrial and Manufacturing Systems Engineering

First Advisor

Jay M. Rosenberger

Second Advisor

Victoria C.P. Chen

Third Advisor

Mohsen Shahandashti

Fourth Advisor

Yuan Zhou

Abstract

We present a multi-objective risk-based optimization approach to water pipe network rehabilitation considering uncertainty. In this study, we design an experiment using Sobol’ sequence to represent rehabilitation policies and pipe impact scenarios. We use Sobol’ sequence due to its space-filling property which ensures little to no multicollinearity, as a result we represent the design space with fewer samples efficiently. We use System Serviceability Index (SSI) as a performance measure and run hydraulic simulations using EPANET 2.0 and Water Network Tool for Resilience (WNTR) to assess serviceability of water network using Pressure-Dependent Demand (PDD) analysis as the performance of a water network largely depends on the available flow, demand, and pressure at each node. To quantify network serviceability risk, we employ quantitative risk measures: Value-at-Risk (VaR), Conditional Value-at-Risk (CVaR) along with standard deviation and averages to assess serviceability across various severities of scenarios. Also, in multi-objective problems, although uncertainty in the objectives is represented in several ways, the solution comparison technique largely relies on using point estimates, stochastic dominance of various orders, and/or simple properties on intervals, such as mid-point, half-width, etc. which may result in incorrectly identifying a solution as non-dominated when considering uncertainty in the estimators. To address this limitation, we present and discuss an Interval-based Pareto Frontier approach for comparing solutions considering uncertainty using pairwise comparison and generate Pareto frontier using intervals obtained from pairwise differences between rehabilitation policies. In addition to this, currently, most of the multi-objective optimization problems are solved using Evolutionary Multi-objective Optimization (EMO) algorithms. These algorithms may suffer from computational challenges and may produce sub-optimal solutions. To address the drawbacks of EMO algorithms, we developed an alternate algorithm called Non-dominated Integer Program Algorithm (NIPA) for multi-objective optimization, which is faster, efficient, and produces better quality solutions. We construct multiple linear regression surrogates to capture the underlying mathematical relationship between rehabilitation policies and network serviceability risk measures; these surrogates act as a proxy to computationally expensive hydraulic simulation and we integrate these surrogates as constraints to an Integer Program (IP) to generate non-dominated rehabilitation policies. NIPA is solved to optimality using GUROBI. The solution quality, time, and efficiency are compared to NGSA-III. For Modena network, NIPA identified non-dominated solutions that are better in the range of 2% - 11% across all the risk measures, about 5.3-folds faster with only 1/9th of the simulation runs used by NSGA-III and for Oberlin network, NIPA identified solutions that are better in the range of 1% - 10% across all the risk measures, about 47-folds faster with only 1/50th of the simulation performed by NSGA-III.

Keywords

Water Network Rehabilitation, Uncertainty, Design of Experiment, Risk measures, Intervals, Multi-objective Optimization, Regression Surrogates

Disciplines

Other Operations Research, Systems Engineering and Industrial Engineering | Risk Analysis | Water Resources Engineering

Download

Available for download on Friday, August 18, 2028

Share

COinS