Digital twins in the water sector

[Article by Kirsten Kelly, Editor: Water and Sanitation Africa]

What is a digital twin? And how can it be used in the water sector? Professor Annie Bekker – research chair at Rand Water and professor in the Department of Mechanical and Mechatronic Engineering at Stellenbosch University – elaborates on the standard definition and its potential use.

It took me a very long time to understand what a digital twin actually is and to many people, a digital twin is still a foreign concept. It is commonly defined as a digital representation of the state and behaviour of a real asset within its operational context towards decision support. But the best way to explain the meaning of a digital twin is to use examples,” says Bekker.

A digital twin example: pumps

A digital twin can function on various levels. Looking at a pumping station of a Rand Water distribution network, a digital twin can be:

  • a single component – an impeller or a seal of a pump
  • a system – an assembly of components like the pump itself
  • a system of systems – a pump station with multiple pumps and several pumping stations connected to supported pipelines or an entire water distribution network.

“When designing a pump, an engineer will evaluate different elements such as how changing the number of impeller blades or blade angle will affect the pump performance (e.g. flow and discharge pressure). A model is created; the pump is then manufactured and sold to a customer and that model is never used again. With a digital twin, one can look at the model and the real outcomes (such as the flow rate of water), so you are entangling a model with the real-life operation of an asset. Engineers can unhinge the benefits of a model beyond the design phase by using it in the operational phase as well,” adds Bekker.

Data-driven digital twins can also allow engineers to take shortcuts where the geometrical detail is no longer modelled. “Going back to the example, an engineer may only be interested in the input-output relationship of water pressure in the pump. A model can be generated from data that is measured while the pump is in operation in different conditions. Therefore, an engineer would not have to go back to the design of the pump and use specialised software to make detailed engineering representations. They can use mathematical models to simulate the performance of the pump and create a black box model to create input-output relationships through techniques such as machine learning,” explains Bekker.

Data-driven modelling is especially advantageous to assist in decision support in applications with low risk, where a wrong prediction would not result in a catastrophic result such as loss of life or ethical ramifications. It calculates quickly, is cost-effective and does not require domain-specific knowledge.

Detecting anomalies

Another advantage is that digital twins can be used to detect anomalies. This can be done by looking at differences in the behaviour of the asset predicted by the model, as opposed to its actual performance. A model can be used to generate a hypothetical ‘virtual sensor’ feed for normal or expected behaviour. An anomaly is detected if sensor feeds from the real asset deviate from the expected response and an inspection is triggered. Additionally, certain standard failures of a machine can be modelled and used to create a catalogue of possible signal attributes under such conditions. These patterns in measurements can then be used as an early recognition system from a catalogue of possible errors. A digital trend in the water sector is the use of existing hydrological models of a pipeline network in complement with sensor feeds on the real network to measure information at key points. Anomalies are found by comparing what the model reveals is happening in the network to what is revealed by the sensor feeds. One such anomaly can be leak detection. A model can be developed to determine the daily water demand; it can be a data-driven expectation where demand is measured over time. Then, as a function of certain variables, it can create a daily cycle on how demand might progress throughout a certain day on a network. The digital model will take environmental conditions and the latest state of the distribution network as an input and create an estimate of the typical demand. Then by measuring the actual demand, and looking for the difference between the two, potential leaks could be detected if there is an unexplained increase in demand.

Digital twins can also be used to prepare distribution networks for future scenarios such as disaster management, population growth, and climate change, including:

  • What will happen if demand increases by 20%?
  • What if we have a drought?

Requisite interdisciplinary knowledge

Bekker maintains that while a digital twin is not a new concept, it is difficult to implement because it requires a vast amount in interdisciplinary knowledge and complexity. “For example, if one looks at a potential system of systems digital twin like the Rand-Vaal water network, multiple factors will need to be considered such as dam levels, demand pattern of users, rainfall, pumps, corrosion of pipelines, and leaks.” However, a powerful advantage of digital twins is that it can become an aggregator. It requires many diverse fields of specialisation to put together models and sensors, thereby creating cross-domain types of benefits.

What prevents the use of digital twins?

In addition to the interdisciplinary nature of digital twins, Bekker says that there is a concern around security. “Digital twins require a certain amount of openness and sharing. Understandably, manufacturers have concerns around their intellectual property rights.” Furthermore, the cost of digital twins is an inhibiting factor. “However, it must be remembered that a lot of money can be wasted by the installation of redundant sensors and redundant modelling efforts,” adds Bekker.

Decision support

When using digital twins, there is a theory determining what to invest in depending on the time period about which you would like information:

  • the past – invest in analytics of data
  • the present – invest in sensors
  • the future – invest in modelling of a predictive nature.

“Ultimately, a digital twin’s value is linked to a need and value associated with a decision support service. But when using a digital twin, the real challenge is to understand the user. What does the user need? And what is the user willing to pay for? Does the user want to be upskilled and trained? How willing is the user to adopt this technology?” expands Bekker.

The Digital Twin Pump Laboratory

Rand Water is a potential user that is interested in digital twins. It have partnered with Stellenbosch University to investigate the use of digital twins within the water network. Stellenbosch University is establishing a small simulation platform called The Digital Twin Pump Laboratory. This is equipped

with pumps, pipes, pressure sensors and valves to circulate water – similar to the components found in a water distribution network. Stellenbosch University will experiment with different digital twins of the laboratory in a stratified environment by using existing models for the system components (pumps and pipes). This facility can be used to evaluate ideas, demonstrate digital twin concepts, and to train students from academia and industry. The controlled environment will allow for the systematic testing of different ideas. Performance monitoring and prediction could be applied on different pump and pipe configurations, faults can be induced, and the model and sensor responses can be tracked to evaluate concepts such as fault detection and diagnosis through previously catalogued errors. A particular focus will be on leak monitoring and slurry pumps, where

Rand Water believes that cavitation is causing early failure. “The idea is to make a scaled-down prototype in an environment that we can control, manipulate and understand, and then – through Rand Water – we will start to apply the knowledge we have gained in the field,” adds Bekker.

“Digital twins require collaboration, openness and a willingness to learn. Specialists from different fields must collaborate. It takes courage to get going and I think that Rand Water has progressed well in this respect. My experience in working with this company is that it is hungry to grow in this era of digitalisation, as clearly demonstrated by its endeavours to build its own Innovation Hub,” concludes Bekker.

Read the online version here.


Insert, top right: Professor Annie Bekker – research chair at Rand Water and professor in the Department of Mechanical and Mechatronic Engineering at Stellenbosch University.

Insert, bottom right: Stellenbosch University has set up The Digital Twin Pump Laboratory, which is used to evaluate ideas, demonstrate digital twin concepts, and to train students from academia and industry.