Long-Term Water Pipe Condition Assessment: A Semiparametric Model Using Gaussian Process and Survival Analysis

Abstract

The maintenance and renewal of water mains demand substantial financial investments, and direct inspection of all water mains in a distribution system is extremely expensive. Therefore, a cost effective break mitigation technique such as a failure forecasting model that allows one to predict the water mains failure likelihood, would reduce the negative social impact and the cost to serve. We introduce a semiparametric Bayesian model for pipeline failure forecasting. The model is centred on a nonparametric Gaussian Process Regression (GPR), and uses a parametric survival model to capture the long-term survival probability using domain knowledge. The parametric element in our model allows the inclusion of survival probability, while the nonparametric part allows us to handle covariates and to employ incomplete prior knowledge about pipe failures. We apply our model to the proactive maintenance problem using a real dataset from a water utility in Australia. The results demonstrate that, our model performs better than competing models such as Support Vector Regression, Poisson regression, Weibull, Gradient Boosting, and GPR, leading to substantial savings on reactive repairs and maintenance. Our water pipeline failure prediction models have been deployed in three states across Australia, and are being monitored by each water authority.

Toward Sustainable Water Infrastructure: The State‐Of‐The‐Art for Modeling the Failure Probability of Water Pipes

“…Although only 15 articles have been published in 2021 and 2022, it is expected that the number will exceed that of the previous decade by the end of 2030. This is due to the fact that many utilities are taking proactive measures to construct prediction models for their WDNs (Weeraddana et al., 2020). Furthermore, the emergence of ML‐based models may promote further growth in this domain's publishing.…”

“…According to the American Water Work Association (AWWA), the USA needs to invest around $1 trillion in replacing and repairing the deteriorating components of their WDNs (Fan et al, 2022). In Australia, the estimated cost of repairing and maintaining WDNs is about AUD 1.4 billion (Weeraddana et al, 2020). In South Korea, 52.5% of the water pipes will require rehabilitation by 2024, as indicated in a study by Seo et al (2015).…”

Long-Term Pipeline Failure Prediction Using Nonparametric Survival Analysis

“…Most of the research found in the literature predict short-term failure forecasting, which spans 1-3 years into the future [12]. However, water utilities require long-term estimations for the structural deterioration of water mains to effectively plan the renewal of water distribution systems and to develop a risk based investment decisions for capital interventions [24]. Therefore, the main objective of this work is to investigate pipe failure factors and develop a long-term prediction model.…”