A data analysis tool that uses artificial intelligence and machine learning to predict water pipe failures has been awarded the Eureka Prize for Excellence in Data Science.

Developed by the CSIRO Data61’s Smart Infrastructure Team, the tool is able to pinpoint pipes with a high risk of failure, mitigating the cost of managing the assets and the inconvenience of burst water pipes.

It has been put to the test across 30 national and international water utilities including SA Water, KWR in the Netherlands and UKWIR (in the UK) and has completed data analysis for nearly 9 million pipes.

Dr Fang Chen, Group Leader and Senior Principal Researcher at Data61’s Enterprise Analytics Group, told IPWEA’s magazine inspire that the data collected falls into four categories: attributes (like pipe manufacturer details, material, coating), environmental (such as weather, soil, traffic volumes), historical (including failures, maintenance orders) and operational.

The machine learning algorithms become more reliable as more data is ingested over time, giving a better prediction for what is a probabilistic case of failure. This allows utilities to better target their maintenance resources rather than guessing where attention is needed.

Such data science methods are becoming increasingly popular in areas where large amounts of resources are used on a widespread issue. According to the CSIRO, $1.4 billion is currently spent by water utilities on reactive repairs and maintenance; switching to preventative repairs could halve this.

Mathematical modelling uses complex collections of data to make sense of an area where there’s both huge uncertainty and many different contributors (such as predicting incidences of cancer or other illnesses based on risk factors).

“The old pipes seem to in general fail more than the younger parts, but you can’t say that one pipe made in the 1950s is definitely is going to fail ahead of 1998,” Chen told inspire.

“So, our attributes contribute to this mix, which could be the operational pressure, it could be traffic conditions, it could even be the material type.”

“The key innovation is to self-learn those relationships among the inputted factors towards the failure. After you learn those then you can use that to predict failure.”

Read the full story from the Jul-Aug edition of inspire here. 

Described as the Oscar’s of Australian science, the Eureka Prizes are presented annually by the Australian Museum in recognition of outstanding achievements in Australian science and science communication, in the fields of research and innovation, leadership, science engagement, and school science.