Melbourne Water is using IoT sensors, a unified data platform, machine learning and a digital twin of its treatment plants to predict recycled water quality two days in advance with 75 percent accuracy.
The Victorian-owned corporation wanted to provide earlier warnings to the farmers, companies and households that depend on its Class A recycled water for non-drinking purposes when quality would be less than standard.
Innovation lead Blair Smith said it was necessary to predict turbidity – the measure of water clarity – because it could negatively impact production.
“If the water is too cloudy, we can’t take it through our tertiary treatment plant and we have to advise our customers that we can’t produce water, which means they need to seek alternative water sources,” he said.
IoT sensors
Prior to using predictive analytics, Melbourne Water had already invested in IoT and an in-house virtual model to enable real-time insights into treatment plant discharges.
According to the organisation’s 2021-22 annual report at Boags Rocks, on Victoria’s Mornington Peninsula, “in May 2022 a new buoy – fitted with an automatic identification system (AIS) and GPS monitoring device – was installed.
“The buoy enables us to capture live data about the quality of water at Boags Rocks where high-quality Class A recycled water is discharged from the Eastern Treatment Plant,” the report states.
“This world-class technology is being used to calibrate a sophisticated, 3D hydrodynamic model of the treated wastewater discharge environment at Boags Rocks.”
According to AWS, Melbourne Water currently has “700 sensors and meters in water treatment plants.”
Unified data platform
Melbourne Water sought to progress from catching real-time data about recycled water quality to predicting it in advance.
This was part of a broader ‘Iintelligent network enablement’ program, which Melbourne Water contracted AWS partner Arq Group to help it realise in October last year when it awarded the NCS-owned provider a one-year, $782,760 cloud-managed services contract.
According to an Arq statement, the first phase of the ‘intelligent network enablement’ program was building an organisation-wide platform to view all Melbourne Water’s data sources from a single pane of glass.
It used a combination of Snowflake and AWS services to build the capability, including “a data lake and analytics platform called unified data store,” over six months.
“The platform employs best-of-breed tools such as AWS Glue for data ingestion into the Snowflake cloud database platform, dbt Core for data transformation, and delivers data to consumers via Snowflake to range of SQL-compliant reporting, advanced analytics and AI/ML workbenches,” Arq said.
Digital Twin and machine learning
The unified data platform enabled Arq to then build a virtual model representing Melbourne Water’s recycled water productions on AWS TwinMaker in three months.
The digital twin relies on AWS IoT SiteWise to collect and analyse real-time IoT sensor data, laboratory data, and weather data – such as wind speed, rain levels, and temperature.
The data is then fed to Amazon SageMaker to provide machine learning models that predict these factors’ impact on water quality.
Melbourne Water’s Smith said that the solution had enabled the utilioty to go from having real-time data about recycled water quality to being able to predict it two days in advance with 75 percent accuracy.
“We have an accurate three-day outlook on turbidity and other conditions that impact our recycled water production,” he said.
“We can detect water quality issues faster now, so we can see whether or not we can produce water.
“We can then address water quality issues quickly while also predicting when water conditions will support the resumption of production.”
Digital transformation roadmap
The project fits into a broader expansion of IT, OT and IoT capabilities across Melbourne Water’s business.
Arq said that the next step it would take to support Melbourne Water on its intelligent network enablement program was to build another digital twin project based on real-time video analytics for drones.
The intelligent network enablement program follows an earlier digital transformation – its automation and operations strategy – that began in mid-2017.
The strategy saw Melbourne Water use historical data to predict the most efficient routes to pump water from its reservoirs to the Victorian capital’s taps.
As part of the same strategy, Melbourne Water also used IBM’s cloud platform for computer vision to detect grates being blocked.