Nvidia’s Omniverse for designing fusion reactors


With accelerating climate change and increasingly contentious fossil fuel supply, ensuring a secure supply of clean energy is a priority for many researchers, businesses and governments. Fusion power – which is carbon-free and low in radioactivity – has long been the holy grail in this quest, but has (until now) not been successfully scaled up for production, suffering from high energy input requirements and general instability of the delicate fusion reaction.

Now, as Nvidia steps up its Omniverse tool for digital twins — physically and photographically realistic simulated duplicates of real-world objects and environments — at least one research team has wondered: Could Omniverse be used for help in the design of fusion reactors?

This was the subject of a session at Nvidia’s GTC22 presented by Lee Margetts, a professor at the University of Manchester and UK Atomic Energy Authority (UKAEA) Chair of Computational Engineering for Nuclear Fusion. ).

“The basic idea is that we need to heat hydrogen to very high temperatures and then a fusion reaction occurs,” Margetts explained. “The outputs of that are helium – which is waste – and energy and neutrons. So it heats water, which turns water into steam and drives a steam turbine in exactly the same way as a gas or coal-fired power station.

The problem, of course: fusion currently requires far more energy than it produces. But research teams have recently made substantial progress on fusion energy, with a UK-based team last month doubling the world record for energy extracted from a fusion reaction at the fusion reactor much larger, ITER, which is under construction in France.

Margetts said that while there are many projects underway to support ITER, “the most relevant program for this presentation is STEP,” which stands for the Spherical Tokamak for Energy Production program. “This is an initiative that aims to deliver a prototype fusion power plant before 2040. … In order to carry out this program, [we need] a revolution in the way we do engineering.

This, he said, is where Omniverse could help.

The role of digital twins in the design of fusion reactors

“The challenges of reactor development are immense,” Margetts said. “There are many different components, and we have to consider many different areas of physics and engineering. And each of those, if we make a design change in a system, it has an effect of training on other systems. Now this may sound like science fiction, but it would be really great if the design team could work together in a real-time simulation environment considering the design of the whole of the machine rather than individual sub-components.

“This can be achieved by disruptive technologies like Nvidia Omniverse, in concert with exascale computing and artificial intelligence,” he continued. “The latter could help us build … surrogate models of different components and sub-assemblies so that we can simulate the whole machine.”

Margetts compared the process of designing the fusion reactor to producing movies and designing video games.

“There are a number of business drivers common to each of these industries,” he said. “The need for increased realism; the desire to support a large team of people working on the same project; and the pressure to deliver a product in a short period of time. And that brings us to the Omniverse, because what the marketing says is that the Omniverse offers enhanced realism; it has support for managing teams and assets; and through co-designing rendering tools…there is an opportunity for real-time collaboration.

But, of course, using Omniverse for fusion reactor design meant doing a lot of testing — and, Margetts said, “in order to evaluate that from an engineering perspective, we needed the funding to throw in a a look.” The team therefore applied for and received funding from the UKAEA’s Fusion Industry Challenges programme: £117,000 (~$154,000) for a six-month project to assess Omniverse for the computational engineering of fusion reactors.

Omniverse evaluation for fusion reactor design

The goal, Margetts explained, was to evaluate Omniverse based on four key questions: Was it easy to add extensions for scientific software? Could the team create a digital copy of an existing factory? Were there advantages for robotic maintenance? Could they extract key information from plasma images? Each of them played on different elements of Omniverse, from integrating open-source tools to GPU-driven tools aimed at photorealism.

GIANT4 in action. Image reproduced with the kind permission of the researchers.

First, they integrated GEANT4, a software package for simulating the trajectory of particles through solid matter that fusion researchers use to simulate the trajectory of neutrons in a fusion reactor. “What we’ve been able to do very lightly in a very short project is extend the Omniverse to use this simulation framework with very little coding,” Margetts said.

When it came to collaboration, he continued, the status quo was that many engineers were using many different software packages and collaborating offline. “It becomes very difficult to get a good picture of the design status of the whole reactor,” he said.

In contrast, another project researcher, Muhammad Omer, showed a demonstration of the simulated interactive process through Omniverse. “We can see three different engineers working on three different components of a reactor in three different packages from three different locations,” Omer said, explaining that Omniverse helped them achieve photorealism using Nvidia’s RTX capabilities. and that they could easily compare different design options for a component in Omniverse.

Multiple engineers using different tools collaborating to design fusion reactors through Omniverse. Image reproduced with the kind permission of the researchers.

Next: robotics. “In the reactor, it’s not a very nice place for human beings, and so robotic arms are used to do maintenance,” Margetts said. “The long-standing research activity is looking at ways to automate or allow the robot to make its own decisions. One way is to look at digital twins. It showed a juxtaposition of an actual assembly with real robots and a digital mockup of the same scene. “To begin with, the Omniverse offers many different capabilities for robotic simulation and manipulation, and all of that is already built in and available,” he said, explaining that while Lidar, cameras, and other capabilities sensors were implemented in real-life robots and Omniverse robots, the benefits could be immense.

Focusing on the cameras, Margetts said fusion reactor cameras could be used to provide researchers with a live feed from the plasma, but it was often confined to a single feed. “The question is, could we use this single stream to get information about what’s going on in the plasma as a whole?”

“Scientists at the UK Atomic Energy Authority, in collaboration with Imperial College London, have developed algorithms to create synthetic visualizations of plasma so they can train neural networks to make decisions about how to control the device or the fusion reaction,” he continued. And Omniverse, he said, could help by simulating “very precisely what a camera in the real device could see,” helping to train those neural networks.

First impressions and what’s next

Overall, Margetts and his colleagues were impressed with Omniverse’s promise for fusion reactor design. “We have completed our first phase of evaluation of the Omniverse and are very happy and excited with what we have seen. We believe it has great potential as a platform for digital engineering. This what we hope is that the UK Atomic Energy Authority will select our project for the next phase and invite us to submit a proposal.

For this second phase, researchers hope to build a prototype design platform and create digital twins of the CHIMERA facility at the Culham Center for Fusion Energy and a robotic arm, among other goals.


About Author

Comments are closed.