Smellevision, hoverboards and colonies in space – plenty of futurists’ predictions have failed to take off. But when the Australian Water Association (AWA) published AI and the Water Industry a decade ago, it was right on the money.

“Imagine an operations model that gradually adapts itself to the ageing of pipes, changing weather predictions and … incidents (planned and unplanned),” wrote Peter Radcliffe, an optimisation specialist then working with world-leading AI company, Optimatics.

“[Imagine] an intelligent model that can recommend when assets need to be replaced, when, where and how much water to pump, and also predict when an emergency event is likely.”

Tick, tick, tick and tick – decades worth of research are starting to pay off. In fact, CSIRO’s Data61 is currently undertaking AI and machine learning projects with more than a dozen Australian water utilities across no fewer than eight focus areas.

Research Group Leader Dr Fang Chen said Data61’s collaborations were yielding results in water quality monitoring, chemical dosing, prioritising active leakage detection areas, water demand analysis, intelligent network optimisation, and the prediction of water pipe failure, sewer corrosion and sewer chokes.

“In these projects, the use of AI and machine learning techniques has demonstrated significant outcomes for improving productivity, efficiency, safety and customer satisfaction for the industry,” said Chen, the 2016 AWA NSW Water Professional of the Year.

Latest feats

Melbourne Water’s cutting-edge software platform, based on Python, is expected to reduce treatment plant pump-station energy costs by a fifth.

The system uses artificial intelligence (AI) and machine learning to determine the daily required flow rate and the most efficient combination of pump usage and speed settings to achieve it.

Melbourne Water Automation Delivery Service Team Leader Russell Riding said the utility is now looking to apply the technology solution to other pump stations and treatment processes across the business.

“The benefits would be a reduction in energy use, operational expenditure savings, less reliance on individuals to determine the most efficient operational regime (knowledge retention), and the freeing up of operational resources to optimise the system further,” he said.

https://www.youtube.com/watch?v=9Cm1jvkZh8A

At Sydney Water, the ‘Sewer Scout’ was developed to reduce risks involved in Avoid Fail Sewer traverse inspections.

It’s delivered much more – it’s now 98% accurate in locating maintenance chambers and holes, and is learning to locate and identify sewer system defects.

Sydney Water Asset Infrastructure Research Co-ordinator Dammika Vitanage said the utility had also just started operationalising Data61’s critical water main failure prediction model after six years’ development.

“Within the first 20% of the predictions we are hitting 80% of the failures,” Vitanage said.

“That has now provided quite good confidence to our operators who are planning to use it for the next condition assessment investigation.”

Meanwhile, Queensland’s Unitywater is, among other things, working with GHD to develop a wet weather overflow prediction model, said Ivan Beirne, Manager Asset Knowledge and Performance.

“This project is using weather radar images, past performance of our network in wet weather events and Google’s TensorFlow,” said Beirne. 

Unitywater’s primary machine learning project is a model that predicts the likelihood of water main failure.

“We’ve developed it with Data Science consultants and Data61, and it has enabled us to predict the likelihood of failure of every water main asset in our network over the next 15 years.”

Changing minds

Despite the increasing adoption of AI and machine learning technology across the water industry, CSIRO’s Chen said: “we haven’t reached a paradigm shift as yet”.

“From a mindset perspective, there’s a strong willingness to change, however, there’s also a strong sense of fear associated with potential changes,” she said.

Associate Professor Sarath Kodagoda, from the UTS School of Mechanical and Mechatronic Engineering, said acceptance was growing as smart technologies became more common in day-to-day life.

“We’re starting to see more driverless cars and things like Google Home coming onto the market, even Amazon is based on huge AI backend processing power,” Kodagoda said.

“People are starting to believe in those technologies and use them without so much fear, so that is a very good sign in influencing cultural change for our sector.”

To more proactively address the mindset issue, Chen recommended quantifying both the benefits and the risks of using cutting-edge technology, and establishing objective measures for assessing the successful adoption of AI.

“Start with baby steps and progress gradually,” she added.

“Don’t be afraid of failures with controlled risks, and keep learning and improving.”

Determining data

Another big challenge has been data quality and quantity.

Kodagoda noted: “Before, the availability of data was a bit sparse. Implementing machine learning and robotic devices with sparse data is not impossible, but having more relevant data is very important”.

“Now we’re in a position where we have more data – more relevant data and more informative data – so machine learning and AI can be better utilised for making decisions.”

For Unitywater, data quality and quantity presented a challenge when it was implementing its water mains failure prediction model, Beirne said.

“In the process of developing the model, we found we were unable to achieve reliable and highly granular data on individual assets. This was because the quality of the maintenance history database was unreliable and only contained four years of data,” he said.

“We used in-house skills to develop algorithms to correct these database inconsistencies based on maintenance log notes.”

Beirne encouraged more utilities to share their data and learnings, successes and setbacks.

Beirne encouraged more utilities to share their data and learnings, successes and setbacks.

“Sharing code would also help – it’s like sharing best practice in any industry. If someone has already developed and implemented something, we should be open to sharing it.

"Machine learning models improve as the volume of good data increases, therefore if we all shared basic information about our networks, for example our asset and maintenance databases, we should be able to improve the quality of everyone’s analysis.”

Chen added that data issues could be collaboratively solved if the industry addressed siloing, both within organisations and across the industry.

She added: “Facilitate data driven solutions or decision support based on a holistic view from operations for example, asset attributes, environment data, customer data, policies or regulations, payment and operations data.

"Evaluate the effectiveness of the current data collection and commission new data collection with an end goal in mind.”

Collaboration

Water sector collaboration on next-gen technology is set to reach a new high with the recently launched Smart Linings for Pipe and Infrastructure project sponsored by the Commonwealth through the Cooperative Research Centre for Projects.

The $20 million project involves 10 Australian water utilities, three Australian universities, the Water Services Association of Australia Limited, UKWIR, and two US water research foundations.

Vitanage is the water industry lead for the three-year project, which plans to use robots, smart linings and innovative materials to extend the lifespan of ageing pipes. Vitanage said the project had started with three manufacturers and applicators onboard and many more were inquiring.

“We are driving innovation in the private sector through this project – they’re saying ‘we are happy to reformulate our linings to ensure that we have a better product than before’,” he said.

“UTS is also conducting a feasibility study to see if we can mix some sensors into the lining products.”

Vitanage stressed this kind of private sector engagement was essential for the water industry to achieve maximum benefits from AI and machine learning technologies.

CSIRO’s Chen added that collaboration across industries would also be important.

“The transport industry, energy and a lot of other industries are using smart sensing faster and more real-time than water utilities,” she said.

“We can build a business case through those learnings.”

Tech and our future

Talk of smart linings, robots, sensors and zettabytes of data doesn’t sound very customer-centric, but that’s exactly what we should be achieving through the application of AI and machine learning, Chen reminded us.

“With an open mind, together with strategic and collaborative effort from the industry, the water industry will look very different,” she said.

“Efficiencies in operations will strengthen the resilience of infrastructure and fundamentally move the focus to more customer-centric services providing a healthier, more liveable society.”

Human judgement will also remain vital to operations, Vitanage said.

“The vision that we have is talking pipes. If pipes can talk – as well as the infrastructure associated with the pipes and even the treatment plants – then we will be really intelligent going into the future,” he said.

“But we shouldn’t take all the interpretation away from people who are trained in the system and experienced in the system – there are some judgements you have to make.” 

First published as 'Fast forward' in Current magazine October 2018.