From bridges and buildings to trucks and turbines, every new piece of physical infrastructure has to be analysed to make sure that it is safe, efficient and functional. But the mathematics and measurements that we’ve traditionally relied upon can only tell us so much. Imagine if, instead, each real-world object had a virtual counterpart – a ‘digital twin’ – that could be endlessly prodded and probed to improve the object’s functioning, perhaps before it was even built.
This is the ultimate vision of the burgeoning field of digital twins. At The Alan Turing Institute, researchers with the AI for science and government (ASG) programme are building digital twins of bridges, farms, wind turbines, trucks, ships and more. Far from being mere 3D visualisations on a screen, these advanced models are allowing engineers to monitor the stresses that a railway bridge withstands when a train passes over it, and helping farmers to optimise growing conditions for their crops. Digital twins are offering up data and insights that can change the way their real-world versions operate for the better.
“A lot of different sectors have identified that digital twinning is a potentially disruptive technology,” says Andrew Duncan, Group Leader in the Turing’s data-centric engineering (DCE) programme. But, he adds, the technology isn’t that accessible right now: creating a virtual sidekick requires a lot of know-how, and it’s expensive. The Turing’s approach is to lower the barriers to access through strong foundational research in digital twins, which is now translating into impact across multiple sectors.
So what is a digital twin? At the simplest level, it can be a virtual sibling for a single structure. Engineers routinely use computational models to design bridges based on assumptions about load, strain and so on. But turning a physics-based model of a bridge into a digital twin requires something else: data from the structure itself. A digital twin that Turing researchers are creating for the world’s first 3D printed steel bridge in Amsterdam incorporates sensors that stream live data on load and strain, as well as tilt, vibration and other variables, providing information about how the material is performing in the real world. Gijs van der Velden, CEO of the Dutch company MX3D, which printed the 12-metre-long footbridge layer by layer, says the Turing’s work “will allow us to build better, lightweight bridges in the near future, without compromising on safety”.
The project, which is also supported by the Turing’s partner in DCE, the Lloyd’s Register Foundation, is the result of efforts to formally marry together data- and physics-based approaches in engineering. As the Turing’s Chief Scientist (and former Programme Director for DCE) Mark Girolami explains, it all started with trying to help engineers predict the risks posed by powerful ocean waves called solitons to offshore structures such as wind turbines and natural gas facilities. In engineering, a mathematical modelling approach called the Finite Element Method (FEM) can be used to predict how structures such as these will respond to their external environment. However, the engineers were increasingly getting information about the ocean, the weather and the status of their offshore structures from sensors: information that the FEM approach wasn’t exploiting. So, drawing on his own statistical expertise and teaming up with FEM specialist Fehmi Cirak from the University of Cambridge, Girolami set about upgrading the method to make use of these new data sources. “We basically reinvented it so that it could incorporate data from the real thing,” he says.
The new statistical FEM (‘statFEM’) that they developed is a generalisable framework envisaged as a mathematical basis for digital twins, and Girolami and Cirak quickly put it into practice by developing a digital twin for a railway bridge in Staffordshire, UK. As a train passes over, the bridge’s network of more than 100 sensors provides the digital twin with accurate measurements of the strain felt by the bridge. Meanwhile, the twin’s physics-based modelling fills in gaps in the data by providing predictions for strain in locations where there are no sensors. The twin is already helping Network Rail keep tabs on the structural health of its bridge in real time. Now Cirak and Research Associate Sharana Shivanand, both ASG-funded, plan to adopt the statFEM framework for the Amsterdam bridge. As Cirak notes, it was a “large step” translating the initial foundational work to a bridge. “But now to a 3D printed bridge, it’s a smaller step,” he says, “and we are making significant progress.”
A digital twin can also be a virtual partner for an entire system. In Clapham, London, for instance, growers at an unusual, underground farm are using a digital twin built at the Turing to help them manage conditions as they produce salad greens for local restaurants and shops. The hydroponic farm, run by Zero Carbon Farms as the ‘Growing Underground’ project, occupies over 1,000m2 of tunnels originally designed for use as a World War II air raid shelter.
The farm’s digital twin (dubbed ‘CROP’) stems from before the statFEM work, but it too merges real data with physics. Sensors dotted around the farm bring in information about temperature, humidity, water quality and other variables to automatically calibrate CROP’s physics-based model of conditions in the tunnels. The system, initially developed by ASG-funded researchers Ruchi Choudhary and Rebecca Ward, along with PhD student Melanie Jans-Singh, all from the University of Cambridge, evolved into a fully-fledged digital twin when software engineers within the Turing’s Research Engineering Group (REG) got involved. “When I joined the Turing and I understood the existence of the REG, I saw a huge opportunity not only to make the whole system more efficient but also to create a software architecture that we will continue to be able to adapt,” Choudhary says, noting that she originally explored using commercially available platforms, but they couldn’t incorporate the physics-based model as REG’s bespoke solution does.
The CROP platform works as a remote dashboard, enabling farm managers to assess and optimise conditions on the farm (by adjusting the ventilation or lighting) and track how different conditions affect the volumes of salad they produce. With the aid of CROP’s physics-based model, which predicts conditions three days ahead, growers can also, for example, take proactive steps to stop their salads wilting in a heatwave. “This collaboration has provided us with arguably the most advanced tool possible to manage and improve our operation in terms of yields and efficient resource use,” says Chief Information Officer for Zero Carbon Farms, Jakob Thomas.
The Turing team is thinking bigger than just the one farm, though. Choudhary hopes to turn the open-source code developed by the REG into a software package for growers in other indoor farming spaces. Indoor farms could provide one way to grow more crops on a planet where the food supply is increasingly vulnerable to the effects of environmental degradation and climate change.
What about if we want to better understand a system made up of lots of similar, interacting objects? Digital twins can help there, too. Duncan, Girolami and ASG-funded Research Associate Lawrence Bull are developing ways to create digital twins of whole fleets. Their approach relies on sharing data between machines – this might be valuable, for example, in predicting when a certain machine might fail if it has never failed before, but other machines in the network have. In partnership with vehicle manufacturer Scania, Bull is testing multi-level models for predicting more accurately when components on Scania’s heavy-duty trucks will fail. Groups of similar trucks have their own models at one level, whilst also sharing data with a ‘global’ truck model at another level. “You have all the models depend on this shared global model,” says Bull. “In that way, if the data associated with one of the models is particularly sparse, that model will revert back to the average global population estimate.”
Proof-of-concept work suggests that this approach can more accurately predict when trucks will need to come in for maintenance. The eventual aim is to provide Scania with a means to plan their truck maintenance schedules so they can avoid downtime wherever possible. As Scania Data Scientist Olof Steinert points out, demands on reliability are increasing, especially as autonomous trucks take to the roads. Understanding vehicle health status, he says, “is an area where data-driven methods and machine learning have strong industrial potential.”
An answer for everything
Whilst the promise of digital twins is undeniable, practical questions remain. For instance, how do we democratise these complex models so that they’re accessible not just to specialists but to the many different types of organisations that might not have the in-house expertise to develop them.
This is the question that Quaisr, a spin-out from work at the Turing and Imperial College London, is trying to answer. As Co-Founder Indranil Pan explains, Quaisr takes the mathematical equations and data comprising a digital twin and turns them into software that a business can pick up and run with. “We are providing a platform that can take in any of these models, scale them up and operationalise them for multinationals,” he says. According to Duncan, Quaisr “drastically lowers” the barriers to accessing digital twins across different industries. It isn’t limited to models of structures or machines; in fact, the concept emerged from ongoing work with Procter & Gamble on digitising production processes for its bottle-filled goods, to increase efficiency and reduce waste.
Quaisr also collaborated with health tech company Multiwave Technologies on a digital tool to help them develop new materials for the positron emission tomography (PET) scanners used in cancer diagnosis, with the aim of improving image quality. The tool reduces the time and computing power it takes to design materials by automating the complex simulations involved in understanding material behaviour.
Another big question is: how do we trust a virtual sidekick to provide the right information, especially in industries where safety is critical? In the maritime industry, for instance, a digital twin could be a way to remotely monitor a ship’s electrical components, meaning all the data from, say, its radar and navigation systems would be continuously accessible to onshore engineers. This would reduce the need for on-board maintenance and provide access to up-to-the-minute data in the event of an incident at sea. However, for the sake of the crew’s safety, the reliability of such a system would have to be guaranteed.
Maritime communications company Furuno Hellas has already built an on-board voyage data recorder (VDR) – like a black box, but for a ship – capable of supplying this type of data in real time. In an industry-first, Duncan, Research Associate Domenic DiFrancesco and Data-Centric Engineering Group Leader Pranay Seshadri worked with Lloyd’s Register Foundation to independently verify and certify the VDR’s digital twin (marketed as ‘HermAce’) before it is used in practice, conducting an exhaustive list of software and hardware tests on the technology. DiFrancesco notes that this is akin to the verification process any physical engineering system would go through. “I think it’s important to have similar workflows for digital systems as well,” he says.
According to R&D Manager at Furuno Hellas, Nick Stavrou, the company is “so proud” of the outcome of the assessment, which Stavrou says will enable service inspections for vessels to be carried out remotely. Duncan adds that it could lay the groundwork for assuring digital twins for other vehicles, such as self-driving cars.
So, aided by the Turing’s wide-ranging work within the ASG programme, we can start to envision a future where our world’s physical infrastructure benefits from insights supplied by its virtual counterparts. The next phase of the Turing’s work in digital twins will see increased emphasis on translating its research into industrial applications and commercialising the tools we need to use these technologies, all whilst establishing the standards to do so safely. In March 2023, the Turing launches a new research hub – the Turing Research and Innovation Cluster in Digital Twins (TRIC-DT) – with a core aim of democratising digital twin technologies. The TRIC-DT will focus on solving significant societal challenges and generating tangible benefits across environment and sustainability, infrastructure, and health.
Header illustration: Jonny Lighthands