AI speeds UK fusion tokamak design by 100,000 times

digiLab has worked with the UK Atomic Energy Authority (UKAEA) on artificial intelligence models that reduce simulation times and computing workloads for fusion research reactor design studies. digiLab's Uncertainty Engine provided UKAEA researchers a step-change in the speed of plasma turbulence simulations for spherical tokamak designs, as part of UKAEA's STEP programme, which focuses on Spherical Tokamak for Energy Production.

Fusion developers face a persistent challenge in predicting turbulence in superheated plasma, which can reach temperatures above 150 million °C. Turbulence can bleed energy from the plasma and undermine sustained reactions. Researchers use large computational models to simulate these effects, consuming millions of CPU hours.

digiLab's approach centres on uncertainty quantification, modelling the "known unknowns" in complex systems. This reduces the need for repeated simulation work and clarifies where models carry more risk. The models allow UKAEA to explore reactor designs around 100,000 times faster than traditional methods for relevant workloads, reducing research cycles from months to hours in some cases, with savings of hundreds of thousands of CPU hours and a four-fold reduction in redundant simulations.

The work included turbulence simulation for spherical tokamaks, one of the least understood areas in plasma physics. Machine learning models now predict behaviours in these configurations, with quantified uncertainty attached to their outputs, enabling researchers to assess reliability even when data are sparse or incomplete. The models remain explainable.

The collaboration also covered diagnostic and sensing design. Probabilistic AI was applied to sensor placement for fusion devices, using genetic algorithms and Bayesian optimisation to identify configurations. Improved sensor placement reduces the likelihood of costly late-stage redesigns and strengthens the resilience of "multi-billion-pound fusion assets." Sensor configurations influence how operators observe plasma conditions and how control systems respond, with late changes potentially impacting engineering, procurement, and build schedules.

UKAEA framed the work within a wider push for digital tools. Dr Rob Akers, Director of Computing Programmes and Senior Fellow at UKAEA, said, "Delivering the fusion roadmap will require a big investment in digital technologies. And at the heart of those technologies are the solutions digiLab is working on."

digiLab positioned uncertainty as a primary driver of cost and delay in complex engineering programmes. Amanda Niedfeldt, Head of Business Development at digiLab, said, "In fusion research, the cost of uncertainty is high - in compute, in time, and in design decisions. With digiLab's Uncertainty Engine, we've helped UKAEA create fast, explainable models that don't just make predictions, but also quantify uncertainty making data actionable, even when it's sparse. Some examples of the powerful impacts this unlocks are drastically faster simulations, fewer redundant runs, and smarter diagnostic decisions, which accelerates design programmes and prevents costly late-stage redesigns."

UKAEA highlighted digiLab's practical tooling approach. Adam Stephen, Head of Advanced Control Unit at UKAEA, said, "digiLab brought a sharp focus on understanding our design challenges and finding ways to create practical and accessible tools. We had access to the senior technical staff and ideas from our discussions were rapidly converted into new features via their product team, maximising the benefits and impacts of the project effectively."

Commercial terms of the partnership and benchmarks for the speed improvements were not disclosed. STEP remains one of the UK's highest-profile fusion initiatives, aiming to progress from research to an operating plant concept. digiLab's work indicates a broader role for uncertainty-aware AI in energy, aerospace, and infrastructure programmes, where modelling and design decisions carry high costs.