Skip to main content

Critical Water Main Failure Prediction

Posted by: data61

February 24, 2015

“Data Analytics for Failure Prediction, Corrosion Estimation, & Operation Optimisation”

Data61 is using machine learning based data analytics techniques to improve pipe failure prediction and reduce maintenance and operational costs. The “Intelligent Pipes” project aims to help water utilities prioritise capital spend and minimise disruption to water supplies and the community.

Water Pipe Failure Prediction


Australian water utilities currently spend $1.4 billion per annum on reactive repairs and maintenance, including the consequence cost of social and economic impact. Focusing the asset maintenance efforts on preventative repairs has the potential to save the water industry $700 million on reactive repairs and maintenance.

Condition Assessment of water pipes is an expensive and disruptive process. Working with data from 23 water utilities around the world, the team has developed a data driven predictive analytics approach for asset failure prediction and risk management. The approach is expected to predict double the number of actual failures within the limited inspection constraint, leading to significantly improved risk analysis and economic savings.


Data61 has used advanced machine learning methods, particularly Bayesian nonparametric methods, to improve on the failure prediction accuracy of existing parametric methods. These techniques use all available data and do not rely on underlying modelling assumptions. This project focuses on the following key developments:  (i) an efficient and effective method of pipe failure prediction based on Bayesian nonparametric modelling; and (ii) a risk-aversion strategy to select water pipes for condition assessment, considering both failure probabilities and consequence costs. 




A comprehensive study carried out on a water pipe network comparing Data61’s nonparametric approach to a parametric model provided very promising results: (i) Optimise capital allocation by predicting twice the amount of high risk breaks with the same condition assessment budget; and, (ii) Significantly outperform the existing method even with various constraints, including consequence costs of pipe failures and multi-level risk analysis.


Follow-up Water Projects

Followed by Data61’s success on water pipe failure prediction, a number of water related projects are undertaken in the sequence:

  • Leakage Detection
  • Sewer Corrosion
  • Intelligent Network System


Contact: Fang Chen