Digital Twins: Reduce Operating Costs and the Risks of Unplanned Incidents

A digital twin combines:

1. Geospatial and digital data about assets – what they are, where, and how they're connected
2. Observations, typically from sensors – such as how much it rains, and how fast the tank fills in response
3. Performance data – whether static (data snapshots) or dynamic (continuous records, ideally over years)
4. Analytics – the numerical engines used in models
5. Visualization – graphics that aid decision-making, from the ops room to budget-holders to office-holders

Read more about the Five Things >

A combination of enough computing power; enough sensors with good remote comms; and rapid changes in population, land use, and climate, make now the right time to use digital twins.

In the past, water engineers used models for one-off simulations – for example, to design the right size pipes for a new subdivision/development. Ten years ago models were updated every two to five years or even longer, and in between utilities relied on operator skill and knowledge to compensate for a lack of veracity in the model.

Now, with parallel processing, fast computers and cloud facilities, digital models are truly representative. They can represent more of the real world, including assets and minor pumping stations, and can be updated continuously to take account of changes in natural and human activity. They can take in live network-monitoring data from sources such as SCADA. And they can use the increasing information available about customer demand and forecast weather.

Information that was held in silos such as asset registries, water supply, wastewater, and flood control repositories can now be integrated and used.

The term 'digital twin' has traditionally been used for plant rather than networks, with many sensors generating real-time data about the assets in the plant.

Water and wastewater networks are a bit different. The sensors tend to be widely spaced (because of their cost and the challenges of retrieving data from remote locations). Some of the data is very far from real-time – pipe condition, for example, may be stable for many years. And historical data, not just real-time, is critically important to understanding and interpreting network behavior.

But the scale and complexity of both the data and the physical structures it represents can be comparable, and the benefits are similar: maximizing network performance, customer service, cost effectiveness, and risk control.

The digital twin can combine three types of model.

The static infrastructure model is mainly for asset management. It captures:

• What you have (pipes, sewers, drains, pumps, valves, etc)
• Where it is
• How it is connected
• How it was built
• Inspection and survey history
• What condition it is in

The dynamic network model is needed for design, analysis of events, and planning. It:

• Provides a hydraulic model of the flows, pressures and levels in the network
• Is driven by demands on the network
• Includes the response of control structures (pumps and valves)
• Is calibrated and validated using monitored data (if possible)

The real-time performance model is for everyday operational management. It:

• Provides a dynamic model representing the performance of the network as it is now
• Is validated against real-time monitoring
• Predicts network performance in the hours and days to come
• Gives warnings of potential service-level failures
• Optimizes control actions

The key to success through a digital twin is identifying its purpose first: framing the questions that you want to answer, so you end up with the right combination of tools, analysis, and visualization. Examples are capital improvement planning, risk modeling, and designing a water collection system with adequate pipe volumes for a new development.

It's important that all stakeholders – not just modelers and engineers – trust the digital twin, and can rely on it giving the best picture of network performance:

• What has happened
• What is happening now
• What will happen in the future

So that the digital twin remains realistic, the model, data streams, simulations, analytic processes must be maintained, calibrated and validated.

And visualizations must be presented appropriately for each audience, whether that's operations, management, IT, regulators, or officials. Fast, powerful graphics in 2D and 3D makes this far easier. Some tools are designed for non-specialists to use, including InfoAsset Online which gives the option of tailoring its browser views for particular roles.

It is possible to create a model or digital twin that's based entirely on data – instead of physics-based simulation models – but you must understand the shortfalls.

Pure data-driven models, built using techniques such as neural networks, can be faster, but:

• You may not know when their conclusions are determined from poor data
• They give 'black box' answers that can't be interrogated – you can't check how they arrived at the answers
• They need to have a lot of clean historical data to learn from
• Meaningless pattern associations can emerge with data
• There can be algorithmic bias based on the data

Tried-and-tested physics-based simulation models excel. They offer:

• Robustness
• Accuracy
• Results throughout the network
• Stability
• Predictive capabilities
• Defensibility

Machine learning and artificial intelligence (AI) show huge potential in combination with the simulation modeling process. Together they can improve on:

• Calibrating and validating models
• Interpreting data and model results
• Validating machine learning results
• Creating and interpreting warnings
• Optimizing control actions