Fusion Energy

 

digiLab sits at the forefront of machine learning (ML) applications for the fusion sector. With a catalogue of fusion energy capabilities, serving as the UK Atomic Energy Authority's strategic partner in Uncertainty Quantification, and a strong coupling with the private fusion sector, digiLab is pioneering new ML solutions across fusion’s challenges.

twinLab for Fusion

Utilising Bayesian approaches to accelerate the commercial deployment of fusion energy

twinLab is a machine learning (ML) platform that makes it simple to build ML models specialised for advanced engineering applications. Powerful cloud-based functionality “under-the-hood” is seamlessly combined with accessible interfaces via Python, Excel, and API.

With twinLab, you can build high-fidelity digital twins and emulators of complex physical models and systems. Accelerate design cycles with active learning, increase understanding with inverse methods, and reduce risk with uncertainty quantification.

Request a Demo
No code required
No code required

Developments in fusion energy have gone through an explosion over the last few years with new records for energy output and advances over a range of technologies.

As fusion looks to scale-up from scientific experiments to a clean energy provider on the industrial scale, there are many science and engineering challenges to be tackled.

digiLab’s advanced data science capabilities and cutting edge applications of ML are leading the way.

Digital Twin

digiLab are developing digital twin solutions for UKAEA's flagship spherical Tokamak, MAST-U, the precursor to the commercial Spherical Tokamak for Energy Production (STEP).

A data-driven solution used historical data and Gaussian processes through twinLab to significantly improve the predicted output of reactor control systems, given a range of inputs.

This approach has been utilised to optimise the desired control inputs for a desired output.

Request a Demo
No code required

Recent Projects

Plasma Science

Fusion requires matter to be compressed and heated to create the conditions at the centre of stars, known as plasma.


Controlling this harsh and complex physical environment requires fast and high precision diagnostics and control. digiLab are leaders in the development of digital twins of fusion reactors, using Bayesian approaches to predict and control plasmas as new physics and regimes are explored.

Disturbance prediction and mitigation

digiLab emulated cutting edge plasma gyro-kinetics simulations to enable a vast range of parameter space to be assessed and the uncertainty quantified.

 

Using twinLab's active learning capabilities, recommendations were made to optimise experimental campaigns to minimise the uncertainty in the physical regime of plasma kinetics and in turn disturbances.

Material Science

Fusion energy pushes materials to the edge of physical and engineered solutions, with many challenging environments interacting with material, including extreme temperatures, magnetic fields and radiation.

 

twinLab has been extensively tested and benchmarked for material science and engineering, driving materials development and assurance.

Tritium desorption

Tritium, a heavy form of hydrogen and core component of fusion fuel, diffuses into the surrounding materials of a fusion reactor.

 

This affects the structural integrity of materials, while causing radiological hazard in the maintenance and decommissioning of fusion energy systems.

 

Uncertainty Quantification, a core capability of twinLab was used to predict the ability to extract tritium from materials, utilising experimental data, and elicitation of expert opinion.

In Association with:

fusionCluster fusionIndustry ukaea

UKAEA Case Study

How digiLab’s unique uncertainty quantification platform is helping make fusion a 2040 reality

To design the first fusion power plant, the UKAEA has to rely on simulations using models. But those models contain epistemic uncertainty. In other words, because no one has built a nuclear fusion reactor before, there simply isn’t a set of existing data to model from.

Rob Akers, Director of Computing Programmes at UKAEA, explained, “Our models are powerful interpretive tools but they’re not actionable for engineering design purposes. We want to move to an environment where we’re using probabilistic design methods.” This is why Rob and his team are keen to wrap those models with the technology that comes from uncertainty quantification (UQ). And this is where digiLab comes in.

Read The Case Study
No code required

First of a kind solutions

digiLab is solving some of the world's biggest challenges in sustainability.

Whether it's supporting the design of the first fusion reactor, building the first electric aircrafts or reducing the chemical treatment of rivers by 50%, digiLab does not step back from the big, important challenges.

Want to find out more?