Artificial intelligence is one of the major technological advancements of the last decade, transforming industries across the board. There’s barely a sector where its impact couldn’t be felt, including nuclear fusion—a technology as disruptive as AI itself.
Hailed as a near-limitless source of clean energy, nuclear fusion is a promising energy source that can prove to be game-changing.
Among its many benefits include abundant fuel, energy efficiency, minimal greenhouse gasses, reliability, intrinsic safety, and sustainability. However, it is still in the experimental stage.
This is despite the fact that a nuclear fusion energy record was achieved this year when a team of scientists and engineers sustained 69 megajoules of fusion energy for 5 seconds using only 0.2 milligrams of fuel. While this energy is enough to power about 12,000 households for that duration, it still requires more energy as input than it generates.
So far, scientists have only been able to sustain fusion for a few seconds. However, things may be about to change thanks to AI.
Continuous breakthroughs in AI in recent years have not only made it a part of our daily lives but also made it critical for achieving scientific advancements. AI is now being actively explored to improve nuclear fusion in several ways.
From controlling fusion reactions by predicting and preventing instabilities in real-time to analyzing reactor designs to find the best ones, maintaining stable control of fusion reactions for longer periods, and even reducing the time and money of analyzing reactor designs and predicting changes in plasma—AI’s role in nuclear fusion is rapidly growing.
While generating large-scale energy from nuclear fusion may not be possible before around 2050, AI has certainly emerged as a promising tool to accelerate this process.
Steven Cowley, professor of astrophysical sciences and laboratory director of the Energy Department’s Princeton Plasma Physics Laboratory (PPPL), believes:
“Fusion might turn out to be AI’s “killer app.”
Earlier this year, scientists reported finding a way to forecast instabilities in plasma, which is highly susceptible to tearing and escaping the power magnetic fields of machines designed to keep it contained and prevent them from happening in real-time using AI.
The AI controller was able to predict potential plasma tearing up to 300 milliseconds in advance, as per the experiment run by the researchers from Princeton University and the PPPL.
Disruptions are one of the big obstructions in nuclear fusion, as one wants the reactor to be operating 24/7 for years without any problem.
“(Given that disruptions and instabilities are) very problematic… developing solutions like this increase their confidence that we can run these machines without any issues.“
– Egemen Kolemen, Study author and professor of mechanical and aerospace engineering at Princeton University
Building AI to Find New Alloys for Nuclear Fusion Facilities
A recent study by the Oak Ridge National Laboratory (ORNL) of the Department of Energy (DOE) has created an AI model to help identify new alloys that are used in nuclear fusion reactors as shielding for housing fusion applications components.
The project started many years ago under the leadership of David Womble, a former AI Initiative Director, and is a big step towards improving nuclear fusion facilities. Then, it continued as the priority area of research under the Artificial Intelligence for Scientific Discovery (AISD).
The current limitations of large-scale computational approaches, the study noted, are shedding new light on the potential benefit of integrating ML within traditional material science and quantum mechanics.
Deep learning (DL) models, in particular, have been effective in recording relevant nonlinearities that are triggered by an atomic system’s atomic configurations.
Once the DL model is trained, it can be used for inference in a fraction of the time it would take to run a full DFT calculation while producing accurate results. This considerable time-saving in forecasting alloy properties using atomic information effectively accelerates the discovery and design of new multi-component alloys.
So, the focus of the study has been on alloys, which are required to “achieve exceptional performance at very high temperatures.“ This performance needs to be in terms of resistance to high temperatures as well as the structural mechanical properties required for the alloys’ usage in complex nuclear plants. For a material to achieve high strength at a higher temperature, its decomposition temperature needs to be higher.
To make these materials, tungsten has been used as the primary element traditionally. Additional elements, meanwhile, are added as a supplement. The resulting alloy composition has been resistant to high temperatures but inconsistent in maintaining proper shielding.
Recently, researchers looked into replacing these standard technology materials with what ORNL AI data scientist Massimiliano Lupo Pasini described as “something completely new and disruptive.”
But, of course, identifying potential metallic combinations is a big challenge, given the vast number of possibilities.
So, researchers used AI to get around the test-run period and discover usable alloy candidates in a more efficient manner. This involved generating the data to create the AI model that spotted three elements for testing as potential new alloy candidates.
According to the results of the study, which are published in the journal Scientific Data, the team presented four open-source datasets. The datasets provided results of density functional theory (DFT) calculations of ground-state properties of alloys.
The alloys involved are niobium-vanadium (NbV), niobium-tantalum (NbTa), tantalum-vanadium (TaV), and ternary alloys NbTaV ordered in body-centered-cubic (BCC) structures with 128 Bravais lattice sites.
The AI model from ORNL is of great significance in the world of nuclear fusion because this promising technology can provide clean and nearly limitless energy-needs materials that can withstand extremely high temperatures, radiation, and mechanical stress. So, by finding high-performance alloys, we can ensure the longevity and reliability of fusion reactors.
Besides helping identify new alloys far more quickly and cost-effectively and removing a major hurdle in making nuclear fusion reactors practical and safe, such an AI model can enable disruptive technological advances in nuclear fusion as well.
AI-Driven Data Modeling to Expedite Scientific Discoveries
Having an AI-generated database is just one part of the project. The next part involves using the generated data for further research regarding the development, training, and deployment of ML models for the discovery and design of materials.
According to Lupo Pasini, six elements are needed to support the design of new refractory high entropy alloys.
High entropy alloys, or HEAs, are a new frontier for materials scientists, and currently, there are very few experimental results. A study from a couple of years ago surveyed 14 elements and the combinations yielding HEAs. Using high-throughput quantum mechanical calculations, they found the stability and elastic properties of over 7,000 HEAs.
That was reportedly the “largest database of the elastic properties of high-entropy alloys,“ as per study author Wei Chen, an associate professor of materials science and engineering at the Illinois Institute of Technology.
Now, for the latest study, there’s also the matter of the expensive task of running quantum mechanical calculations on existing supercomputers. So, “the data alone will not be enough.”
Yet another challenge the team needed to overcome with these massive calculations was the time. It took more than a year for the team to finally make the calculations on the Perlmutter and Summit supercomputers and generate data.
The Summit supercomputer is part of the Oak Ridge Leadership Computing Facility and is located at ORNL, while Perlmutter is housed at Lawrence Berkeley National Laboratory, with both computing systems being DOE Office of Science user facilities.
In the next step, the team will utilize the generated data to train the AI model. The model will accelerate the vast range of compounds as a result of mixing the six elements at different concentrations as alloys.
“We are trying to help the material scientists with their trial-and-error approaches in identifying the relative percentage of the different elements that need to be mixed together in order to come up with alloys that can lead to disruptive technological advances in fusion.“
– Lupo Pasini
The use of AI to gain an understanding of the complexities of materials at a deeper level has also been advancing at a rapid pace. Last year, ORNL’s Lupo Pasini also led a team of computational scientists that generated datasets of unprecedented scale that provided the ultraviolet visible spectral properties of over 10 million organic molecules.
Understanding a molecule’s interaction with light is crucial to revealing its electronic and optical properties, which have potential photoactive applications in medical imaging systems and solar cells.
The quantum chemistry calculations were run using high-performance computing resources to generate the vast datasets. These datasets will then be used to train a DL model to identify molecules with tailored optoelectronic and photoreactivity properties.
At the time, Lupo Pasini noted that determining how matter and energy interact at the subatomic level via labor-intensive experiments and existing first-principles calculations, which can easily slam supercomputing facilities, is simply unaffordable. DL models, however, “provide very promising tools to overcome these barriers.”
The team is building a HydraGNN architecture that takes in the atomic structure and converts it into a graph before trying to predict what the first-principles code would produce as an output.
Earlier this year, Lupo Pasini and the team showed HydraGNN scaling on the Perlmutter system as well as Frontier and Summit supercomputers. HydraGNN is an implementation of GNN architectures to produce fast and accurate predictions of material properties.
With the illustration, the team showed just how to scale graph neural networks (GNNs) to map connections between thousands or even millions of variables and unravel their relationships to expedite scientific discovery.
Companies Standing to Benefit from the Fusion
Both AI and nuclear fusion industries are seeing billions of dollars in investment, which goes on to show the excitement and development surrounding these two transforming technologies. So, let’s see some prominent names from these fields:
#1. General Electric (GE)
General Electric is involved in nuclear power research and development through its GE Vernova, which aims to accelerate the path to more reliable, affordable, and sustainable energy. Its efforts in the field include new sensors and imaging techniques to track used nuclear fuel materials in real-time, an inspection technique called RADMASS to make reprocessing fuel more affordable, and building boiling water nuclear reactors (BWRs).
With a market cap of $204.469 bln, the company shares are currently trading at $188.94, up 47.83% YTD. It has an EPS (TTM) of 4.28, a P/E (TTM) of 44.09, and a dividend yield of 0.59%.
For Q2 2024, GE Vernova reported $8.2 billion in total revenue and $1.3 billion in net income. Meanwhile, total orders of $11.8 billion exceeded revenue by 1.4x. Cash from operating activities at the end of the quarter came in at $1 bln, while $5.8 billion in cash balance was up from $4.2 bln upon spin-off from GE in April.
“Global electrification and decarbonization trends continue to drive demand for our products and services.”
– GE Vernova CEO Scott Strazik
#2. IBM (IBM)
While NVIDIA is a leader in AI and supercomputing technology, which has driven its stock prices to rally 134.76% this year, IBM is also among the leading ones in quantum computing and AI research. The company’s quantum computers are cloud-based quantum computing systems that can be used for research and exploration.
By running complex simulations on these quantum computers, IBM allows for critical breakthroughs in various sectors. A couple of years ago, the company wrote about optimizing nuclear power generation by unlocking the value of data.
With a market cap of $201.54 bln, the company shares are currently trading at $218.80, up 34.82% YTD. It has an EPS (TTM) of 9.53, a P/E (TTM) of 2.95, and a dividend yield of 3.05%.
For Q2 2024, IBM reported revenue of $15.8 billion, which was an increase of 2% with the biggest upside seen in its software segment. The company’s gross profit margin was 56.8% while operating (Non-GAAP) and came in at 57.8%. Net cash from operating activities was $6.2 billion for year-to-date and free cash flow was $4.5 billion.
#3. ATI Inc. (ATI)
ATI specializes in the production of specialty metals and alloys, and this expertise positions them well for fusion reactor applications. High-Performance Materials & Components (HPMC) is one of its segments, which produces high-performance materials like titanium and titanium-based alloys used in energy, medicine, and aerospace. Advanced Alloys & Solutions (AA&S) is another company that produces zirconium and related alloys that are used in the automotive, electronics, energy, and defense markets.
With a market cap of $8.20 bln, the company shares are currently trading at $66, up 44.74% YTD. It has an EPS (TTM) of 2.94 and a P/E (TTM) of 22.48.
For Q2 2024, ATI reported $1.1 billion in sales, which was up 5% from the previous quarter. Net income of the quarter was $81.9 million, or $0.58 per share, up 26% from 1Q24. Adjusted EBITDA of the company meanwhile was $182.6 million.
“Our execution and ability to capitalize on market opportunities allows us to drive increased margins and generate strong operating cash flow.”
– Kimberly A. Fields, President and CEO
Conclusion
AI mania is constantly growing, and its integration into nuclear fusion research marks a pivotal moment for both fields. By addressing the complex challenges inherent in fusion, AI has the potential to drastically shorten the timeline to achieve practical fusion energy.
From real-time plasma predictions to discovering new alloys capable of withstanding extreme conditions, AI-driven advancements are steadily pushing nuclear fusion closer to becoming a reality, offering a cleaner and more sustainable energy solution to the global energy crisis.
Click here to learn all about investing in artificial intelligence.
