Can digital twinning find transformative solutions for our problems?

A research team led by Daniel Klocke from the Max Planck Institute for Meteorology has just put together a new digital model of Earth. They call it an Icosahedral Nonhydrostatic (ICON) model. It does weather forecasting and climate modelling.

 

ICON divides Earth’s surface into a grid of 336 million cells, with each cell representing 1.25 square km.  Another 336 million cells represent the atmosphere. Each cell runs interconnected models reflecting dynamic weather and climate systems.

 

Weather is calculated by looking at “fast” systems like natural energy and water cycles, which induce short-term changes. “Slow” processes like the carbon cycle, long-term changes in the biosphere, ocean temperatures, and chemistry are also modelled to judge climate trends.

 

Combining fast and slow, ICON spits out weather forecasts and climate predictions.  This is computationally intense. ICON incorporates 1 trillion “degrees of freedom”, which equates to calculating about the same number of values. It improves drastically on earlier models which could, at best, handle only digital twins where each cell represents 40 square km.

 

Every day, ICON runs calculations for the next 145 days (145.7 days to be exact). Apart from clever maths and software tricks, it needs humongous computing power. It uses two supercomputers, which run on arrays of over 20,000 GH200 Grace Hopper chips. Each chip combines a Hopper graphics processing unit (GPU) from Nvidia and a Grace CPU from ARM.

 

As a dynamic digital model of a (very large) physical object, ICON is a “digital twin” of Earth, even a “virtual twin”. It predicts the behaviour and future outcomes of its real-life physical counterpart.  Although digital and virtual are used almost interchangeably in casual parlance, every digital twin is not a virtual twin. Virtual twins do calculations using real-time data or near-real-time data. The concept of digital twinning is intuitively easy to grasp: Put together a detailed, scaled digital model of something, plug in all the rules and numbers that you know about it, and see how things turn out when everything interacts.

 

Most physical processes are very complex, however, and this sort of modelling can require billions of calculation. It has only become meaningfully possible in the recent past, given advances in raw computational power, and improved understanding of algorithmic processes. Digital twins are now becoming an increasingly powerful tool for monitoring and studying all sorts of things.

 

One common application of digital-twin technology we use on a daily basis is online route maps.  Google Maps, for example, offer digital twins of transportation systems, showing all possible routes and travel modes between two points and updating in (near) real-time by using data from traffic, weather, etc.  

 

Twins can be built for all sorts of objects and tasks, and for components of systems, as well as entire systems. For example, engineers may build a digital twin of the blades of a turbine in a jet engine, or they may build a digital twin of the entire engine, or a twin of an entire plane. The automotive industry has enthusiastically embraced the concept. Architects and civil engineers now twin buildings, and twin many projects allowing them to easily experiment with different materials, designs, and technologies. When it comes to education, digital twins could be a massive force multiplier with applications ranging from kindergarten to post-doc.  

 

City authorities “twin” metro systems being planned and also existing metros during operations. This can smooth and optimise entire processes, and flag possible flaws in designs. Oil companies routinely set up twins of oil rigs, pipelines, and refineries, and continuously update the twins with real-time data. Many other industries use them as well.

 

Health care is another sector where twins have proved very useful. Global usage of digital twins in health care is expected to hit $21 billion by 2028, up from $1.5 billion in 2023. Hospitals are twinned to optimise work-flows for bed management, pathology tests, human resources allocation, etc. Surgeons twin virtual organs to practise opersations, and nurses and paramedics practise on virtual bodies. Creating digital twins of individual health profiles is another use case that’s catching on.

 

As Internet of Things (IoT) penetration increases, more twins would become virtual, courtesy real-time data. Efficiency gains are already visible, and as twins get better, these artefacts could have a transformative effect on multiple industries.