Case study
Fusion Computing Lab: Engaging supercomputers to test the “untestable”
The UK Atomic Energy Authority (UKAEA) is advancing fusion energy research through the Fusion Computing Lab programme. Launched in partnership with the STFC Hartree Centre, Fusion Computing Lab leverages cutting-edge supercomputing, AI, and data analytics to accelerate the development of fusion power plants. Aiming to create the first digital twin of a fusion power plant, this initiative enables parallel testing and rapid optimisation, overcoming significant technical and financial barriers. The innovative digital twin technology aims to enhance the scalability and safety of commercial fusion power plants, with potential applications extending to various high-tech industries.
Key Highlights:
| Fusion devices push technological boundaries | Digital twins enable parallel testing | AI enhances fusion simulation speed |
The UK Atomic Energy Authority (UKAEA) is working to lead the delivery of sustainable fusion energy with experiments like MAST-Upgrade. However, transitioning fusion production from experiments into commercially viable power plants presents unprecedented challenges.
As fusion energy research progresses, the Fusion Computing Lab program stands at the forefront, leveraging state-of-the-art technologies to overcome complex challenges in fusion power plant development. Through innovative approaches in supercomputing, AI, and data analytics, the UK Atomic Energy Authority (UKAEA) and its partners aim to pave the way for reliable and sustainable fusion energy solutions.
Why Supercomputing?
Fusion Computing Lab was launched in 2021 as a partnership between UKAEA and the STFC Hartree Centre with the aim of integrating sophisticated computing infrastructure into fusion research. Through the Fusion Computing Lab, UKAEA is hoping to leverage the Hartree Centre’s capabilities in supercomputing, data and artificial intelligence (AI) — some of the most advanced in the UK — to promote a more cohesive, accelerated and scalable engineering design process for fusion power plants.
“There’s a real opportunity around what we call the convergence of AI and high-performance computing — the two are coming together,” says Rob Akers.
“The way I think of this is that we’ve started to use AI as being a little bit like having access to a scientist or specialist — someone who has enough experience to know the answer before you’ve even tried to simulate it. So, AI, although fallible, gives you a really good idea very often as to what the answer might be and allows us to converge very quickly with more traditional high-performance computing methods.”
Now entering its third year, Fusion Computing Lab has begun to bring additional partners into the collaboration, including STFC’s Scientific Computing Department and Digilab, an Exeter-based SME specialising in uncertainty quantification.
Digital Twin
The ultimate aim of Fusion Computing Lab is to create the first digital twin of a fusion device — essentially, this will be a model of a fusion device “in silico” (in the digital world) that allows for accelerated, parallel testing of multiple technical factors.
This digital power plant will be able to quickly and accurately adjust for changes in plasma behaviour and account for multiple variables in testing criteria. This will be particularly important for modelling how the plant will respond to age-related changes, as materials and components are expected to respond differently to strains as they get older.
“AI could replace some of the numerical simulations of plasma behaviours,” says Vassil Alexandrov, Chief Science Officer at the Hartree Centre.
“We have to simulate in intervals of 0.01 milliseconds or 0.001 milliseconds, and if you use standard computing methods, it’s too slow. So, we need to have an AI equivalent that allows for computing in this timescale to realistically model the behaviour of the plasma and be able to control it in real time.”
The creation of this first digital twin will allow for faster creation of digital models for fusion devices in the future — something which will be key to the safe, scalable rollout of commercial fusion power plants.
“Once we have both the power plant and the digital version of the power plant, we’ll be able to use the digital version to run thought experiments and to try and optimise the way we use the real plant itself and solve problems,” says Akers. “When we see discrepancies and we see behaviour that emerges that we weren’t expecting, we’ll be able to turn to the digital version to ask questions and solve problems.”
Collaboration and Future Prospects
“What is great about this collaboration is that our teams are very well integrated and are working together very efficiently. We are sharing our knowledge and skills, we are combining the Hartree Centre’s computing capabilities with UKAEA’s expertise in fusion science to help make fusion energy a reality. The future of this collaboration will support the growth of the fusion energy capabilities and community in the UK and develop a network with the right expertise to prepare the industry to adopt advanced fusion technologies.” — Vassil Alexandrov, Chief Science Officer, Hartree Centre.
“UKAEA’s unique relationship with Hartree Centre is combining UKAEA’s heritage in delivering world-class fusion science with STFC’s experience in supercomputing, big data, and Artificial Intelligence. By tapping into Hartree Centre’s experience in engaging industry, we aim to democratise the tools we are designing and the skills and knowledge we are developing across the UK’s rapidly growing fusion supply chain. Our mission is to ensure that the fusion sector is “digitally adept” in leveraging the disruptive power of extreme scale computing and AI, creating a paradigm shift in the way fusion powerplants will be certified, regulated, and operated for promoting a “simulation first” approach to engineering design.” — Rob Akers, Director of Computing Programmes & Senior Fellow, UKAEA.
Additional applications
Digital twinning is already common practice in certain industries: in the automobile industry, for example, digital twins are expected to replace traditional crash test dummies for car safety tests within the next decade. By using predictive models to determine and define automobile behaviours, digital twins are also driving innovations in the autonomous car industry and are expected to help put safe self-driving cars on the road faster.
The supercomputing methods being developed for fusion testing will go far beyond the current capabilities of digital twins as they are modelling hyper-specific, essentially untestable conditions in silico. As such, they could be used for more robust, multifactor testing to promote advancements in a range of industries, including aerospace, construction, and military.
Computing Division
Please contact the computing team if you want to learn more about this project.
