Comparison of Pipeline Failure Prediction Models for Water Distribution Networks with Uncertain and Limited Data

Abstract

This paper addresses the problem of specifying, estimating, and validating Weibull hazard rate models for water distribution networks under incomplete data. The most notable advantages of the proposed model are the ability to incorporate expert opinion and the spatial analysis of break histories in order to improve the predictive performance of the model with respect to identifying the highest risk cohorts of pipe. The methodology was demonstrated on a large utility in the southeast United States. The expert opinion of utility professionals was elicited to fill in data gaps associated with pipe material; categorical variables were introduced to account for the bias associated with such data imputing. Additionally, a kriging model was used to estimate the spatial distribution of pipe break rates on subsets of the network, and the resulting break rate parameter was added to the Weibull model. Validation metrics showed that including both the parameter estimates and break rate parameter increased the model’s capability of identifying pipe groups with the highest risk of failure. These results show that utilities with limited known parameters describing pipe networks can elicit additional data through expert opinion and spatial analysis to develop prioritization models for repair, replacement, and rehabilitation.