Risk-based Renewal Prioritization Models (RPM) for Potable Water Pipeline Infrastructure Systems

Abstract

Water pipelines are critical infrastructure assets buried across the United States, responsible for delivering safe drinking water at adequate pressures from source to customers. A majority of these pipelines were installed in the mid-twentieth century without adequate financial planning for future renewal, creating a growing renewal backlog under tight budget and operational constraints. Decades of utility data and practice-based knowledge, combined with advances in Artificial Intelligence (AI) and computational resources, now make it possible to revisit how renewal decisions are made. A review of current water pipeline renewal methods reveals major gaps, including weak integration of risk with decision criteria, ad hoc selection of modeling algorithms without strategic foresight, and limited, often internal-only, real-world validation. This dissertation addresses these gaps by developing and testing an AI-enabled framework for risk-based renewal prioritization of water pipelines. The work has four main goals: (1) developing an AI model to predict the performance and Likelihood of Failure (LOF) of any water pipeline segment on a 0–5 scale, (2) creating an AI model to predict the Consequence of Failure (COF) of any segment on a 0–5 scale, spanning economic, en-vironmental, and social/service impacts, (3) building a multi-criteria optimization model to generate prioritized renewal portfolios that incorporate risk, cost, equity, and delivery con-straints within budget limits, and (4) establishing experimental protocols to evaluate, veri-fy, and validate model results against field inspections, retrospective failures, and expert judgement across multiple utilities. Applied to several U.S. utilities, the integrated LOF, COF, and portfolio models outperform age-based and heuristic baselines on predictive accuracy, calibration, and risk-reduction-per-dollar, while producing more spatially coherent and operationally feasible renewal programs in retrospective tests. Finally, this research evaluates whether the additional effort required for data collection, model interpretation, and governance is justified relative to current utility practices, with tradeoffs assessed in terms of reduced emergency failures and costs, enhanced transparency and accountability in decision-making, and improved public trust. In the short term, the proposed framework supports more cost-effective and defensible capital improvement planning; in the long term, it provides a template for shifting water utilities from reactive, break-driven repairs to proactive, data-informed management of buried pipeline infrastructure using explainable AI models with characterized uncertainties.