Integrated approach for pipe failure prediction and condition scoring in water infrastructure systems

Abstract

Pipe failures in water distribution infrastructure have significant economic, environmental, and public health impacts. To alleviate these impacts, pipe deterioration modeling has been increasingly implemented to characterize and predict pipe failure patterns with the aim of prioritizing repair and replacement decisions. Logistic regression has been recognized in recent literature as a strong candidate for failure prediction modeling. However, previous studies have often been limited to demonstrating the application of logistic regression for estimating failure probabilities. This study builds on previous efforts by proposing an approach for implementing logistic regression into a holistic framework for asset management decision-making. This framework incorporates logistic regression modeling, with a flexible time-interval, into a practical condition scoring methodology that accounts for the attitude of water utilities towards risk. The developed framework is demonstrated and tested on a 20-year pipe failure dataset of a large metropolitan US city. The logistic regression model displayed high accuracy in estimating the probability of failure within different time intervals, and the scoring method showed a reasonable ability to predict the criticality of repair decisions for pipes based on their condition.