Case Study

It is estimated that there are approximately 40,000 km of Asbestos Cement (AC) pipes still in service across water supply networks in Australia. Many of these pipes are starting to reach the end of their expected life, which in turn is expected to lead to increases in failure rates. In partnership with WSP, Harmonic was tasked with developing statistical predictive models to help estimate the future failure of AC water mains, and validate existing engineering approaches. These models can be used to assist the industry in developing strategies for the management of AC pipes.

Harmonic developed a data processing pipeline to cleanse and collate pipe attributes and data on repairs and replacements from eleven water utilities into a single dataset. The processed data was then used to develop a set of statistical models that predict the number of breaks and the probability of failure of AC pipes in a given year. The collated datasets enabled smaller utilities to leverage the data of others to improve predictive performance.

These statistical models associate various pipe attributes with the number of breaks in a pipe and help identify the attributes with the greatest importance in predicting pipe failure. These drivers were carefully assessed by engineering experts against conventional approaches to validate both the data-driven and engineering approaches. Development of the models helped to improve understanding on the factors of importance relating to AC pipe failure.

In addition, the models provide a way for utilities to predict future pipe failures on their network, enabling them to cater for accelerating failures on aging pipe stock. Outlier predictions were also key in detecting and correcting data quality issues.

Harmonic developed two web tools to assist the utilities in using and interpreting model outputs: a web application that allows utilities to upload their pipe asset data and generate failure predictions forecasts from the models, and an interactive dashboard that allows utilities to visualise the relative impacts of different pipe attributes on pipe deterioration.

A link to a fact sheet produced for this project is available here.

• Participating utilities have gained a deeper understanding of the factors contributing to pipe failures in their networks.
• Water utilities can leverage data-driven forecasts to develop and improve their strategy for managing an aging population of asbestos cement pipes.
• Engineering approaches to pipe-failure were validated against statistical, data-driven methods.
• Utilities can compare their performance with others, highlighting areas of success or concern.
• Developed a collated and cleansed dataset that can be used for further research.