AI helps to make great leaps forward for nuclear fusion and the energy industry

Sustainable nuclear fusion is the energy industry’s ‘boy who cried wolf’. It’s been just around the corner for so long that people can’t believe it’s just around the corner now.

 “As a physicist, we always joke that fusion has been 50 years away for 50 years,” said Daniel Kammen, a professor of energy at the University of California, Berkley. As with many jokes, this comment once contained a kernel of truth. However, “in the last four or five years, with the effort that’s going on here, the effort that’s going on with Commonwealth Fusion in Massachusetts, you’re suddenly seeing that the old idea – that fusion is great but infinitely far away – has gone away,” commented Dr. Kammen. 

we can’t do all kinds of things until we can do much better code

At a virtual workshop on Tackling Climate Change with Machine Learning held on April 26, Kammen expressed his opinion on transformative achievements in computing: “There are now people who are projecting small-trial fusion plants that couldn’t have been done before without higher computing. We can’t adjust the magnets, we can’t do all kinds of things until we can do much better code. That is a transformative idea.” 

The notion that you hear fusion is another 20 years away, 30 years away, 50 years away – it’s not true. We’re talking commercialisation coming in the next five years for this technology

The Applied Science branch of Google Research has partnered on a machine learning project with TAE Technologies (formerly known as Tri-Alpha Energy) in order to devise modern solutions to decades-old maths problems. John Platt, head of Google’s Applied Science division, stated: “We’ve had a really great collaboration with TAE, which is a fusion company in Southern California, and we’ve helped them with optimisation and Bayesian inversion, and that’s been great for them, and they’ve achieved the physics goals that they’ve been working on for a bunch of years.” 

As a result of this impactful machine learning project, TAE CEO Michl Binderbauer said: “The notion that you hear fusion is another 20 years away, 30 years away, 50 years away – it’s not true. We’re talking commercialisation coming in the next five years for this technology.” Although Binderbauer’s announcement was met with predictable scepticism, TAE is not alone. Vancouver-based General Fusion Inc. is devoting the next five years, with support from the Canadian government, to developing a prototype of its fusion reactor. In addition, the Massachusetts Institute of Technology announced in 2018 that it expects to bring its fusion reactor to market in 10 years. It is also purported that the 35-nation ITER project (which began in 1985 as a Reagan-Gorbachev initiative) expects to complete a demonstration fusion reactor in France in 2025. 

Harvard and Princeton academics applied deep learning tools to forecast disruptions that could potentially stall reactors.

So, why is machine learning thought to be a major difference-maker in the field of nuclear fusion? In 2019, Harvard and Princeton academics applied deep learning tools to forecast disruptions that could potentially stall reactors. “When plasma in a fusion experiment becomes unstable, it can escape confinement and touch the wall of the machine, causing severe damage and sometimes even melting or vaporising components,” stated researcher Julian Kates-Harbeck. “If you could predict those escapes, or disruptions, you could mitigate their effects and build in safety protocols that would cool the plasma down gently and keep it from damaging the machine.” The Fusion Recurrent Neural Network (FRNN) deep learning code devised by the U.S. scientists was able to predict true disruptions within the 30-millisecond time frame required by the global ITER energy project. Commenting on the findings, which were published in the journal Nature, Kates-Harbeck stated: “We don’t have good strategies for completely avoiding these disruption events yet. The best thing we can do is to predict when they are going to happen so we can avoid most of their adverse effects. This might be, for example, by injecting neutral gas that cools the plasma before it smashes into the wall. But you can’t mitigate anything if you don’t know it’s coming.”

Apart from predicting destructive disruptions, the deep-learning algorithm is able to analyse and make sense of not only simple sources of data (such as the average density or current in the plasma) but also complex, high-dimensional data (like the temperature of electrons as a function of radius in the plasma). 

Nuclear fusion has always been 30 years away. However, this no longer seems to be the case. Owing to advances in machine learning, and the contributions this could make to fusion energy, Daniel Kammen believes that: “The technology has gone from wonderful and theoretical and someone else’s lifetime to a next generation project.”