Va., Aug. 14, 2026 — At a fusion science facility in balmy Southern California, the ideal conditions for experiments are in fact quite harsh. There, the primary research device contains a plasma that is hotter than the sun’s core, squeezed with extreme pressure and violently churning amid a powerful magnetic field
That hostile environment is essential to fusing lightweight atoms in an effort to harvest energy from the controlled reactions. But it can take a toll on the machine and cause some crucial components to shift. Those subtle distortions of the device’s physical structure can have outsize effects on the physics happening inside.
That led data scientists from the U.S. Department of Energy’s Thomas Jefferson National Accelerator Facility (Jefferson Lab) and partner institutions to develop a deployable machine learning (ML) framework that can adaptively predict changes in a fusion research machine’s hardware. The innovative method, recently published in the journal Machine Learning with Applications, promises to help make the operation of fusion devices more stable and accelerate humanity’s pursuit of a superabundant energy source.
“The technique that we have developed would help fusion researchers find issues before they actually happen on the physical machine,” said Kishan Rajput, a data scientist at Jefferson Lab and the lead author on the study. “This takes us one step closer to putting AI in operations for fusion diagnostics.”
The work focuses on the DIII-D National Fusion Facility in San Diego, a DOE Office of Science user facility hosted by General Atomics, with support from DOE’s Office of Fusion Energy Sciences. DIII-D is a tokamak device, which uses a strong magnetic field to confine plasma within its toroidal, or donut-shaped, volume. In fact, DIII-D is so named because its cross section is shaped like the letter D.
Outlining that D shape is a ring of large magnets called toroidal field (TF) coils. Though engineered to strict specifications and bolted in tight, the coils can shift ever-so-slightly as plasma stability changes from one fusion experiment, or “shot,” to the next.
Because shots happen roughly every 10 minutes at DIII-D and small issues can grow into bigger ones down the road if unaddressed, researchers identified a need to predict the coils’ movement for the next planned shot during the short downtime in between. This would allow machine operators to diagnose problems before they happen and adjust their plasma parameters or perform timely maintenance if needed.
“Unexpected movement in the physical machine would not be desired,” Rajput said. “You want to predict the movement of these coils during experiments to get a sense of how stable a particular shot would be and whether a problem may arise.”
To make those predictions, Rajput teamed with researchers from General Atomics, the University of Houston and DOE’s Pacific Northwest National Laboratory to build a deployment-ready digital twin of DIII-D’s TF coil system. The digital twin uses a novel ML technique with several key innovations.
“We are creating a virtual replica, more or less,” Rajput said. “You can feed all sorts of different parameters to this virtual replica that you may want to run on the physical machine.”
The framework employs deep neural networks (DNNs) traineda stream of data to stay up to date with recent data trends. It’s the first such application on non-stationary, or drifting, data streams in fusion science
“There’s a lot of drift in the TF coil data shot to shot because the behavior of the plasma is always changing,” Rajput said. “If you only train a model on historical data and try to use it without any updates, it would likely not be reliable.”
The method groups together multiple models that are trained on slightly different time series, or horizons, forming an online ensemble. Some models are trained to accommodate abrupt drifts in the data, some detect more gradual drifts, and others look at the horizons in between.
The models incorporate reliable uncertainty quantification for every prediction, supporting downstream decision-making by indicating the confidence for each prediction.
“We combine the predictions such that we inversely weight them based on the uncertainty quantification, or the width of the error envelope,” Rajput said. “We weight the models with the narrower envelopes higher, and that’s where the guidance comes in for the ensemble.”
This online learning approach was shown to reduce prediction error by 80% compared with static ML models. The use of an uncertainty-guided ensemble further reduces error by approximately 10% compared with standard single-model online learning. The system also provides uncertainty estimates that support decision-making by DIII-D’s operators.
“When you have these highly complex processes and different types of drifts, this ensemble with uncertainty guidance is really novel,” Rajput said. “That’s where the method we used shines.”
Rajput said the work isn’t finished. He said he’d like to run the ensemble on years’ worth of data to capture the behavior of less common events and improve uncertainty quantification.
“Though we’re able to adapt each shot, it’s important to show explicitly what the trend is from the ML model perspective,” he said. “We want to show how the model is evolving as opposed to how the data is evolving and what part of the models are capturing that trend.”
Capturing those trends goes a long way toward explainability, which is a hot topic in artificial intelligence and machine learning.
“People want to know what’s going on inside an AI model as opposed to it being a so-called ‘black box,’ ” Rajput said. “That’s a top priority because it builds trust.”
In the meantime, the system is ready for deployment on DIII-D and can be adapted for other fusion devices.
“From all the aspects we consider — the uncertainty quantification, the adaptive mechanism to make sure the drifts are accommodated, the constraint with respect to time — this all indicates that it’s usable in actual operations,” Rajput said. “That’s new for the field of fusion.”
Sectors:
Academia, Government, Science
Tags:
Jefferson Lab, tokamak
The Big Data Inside Amazon’s New Fire Phone
The new Fire phone that Amazon launched this week looks like your ordinary black smartphone,…
Three Reasons to be Scared of the Internet of Things
We know the Internet of Things forecasts: 50 billion connected devices by 2020. Apparently, there’s…
Wanted: Intelligent Middleware That Simplifies Big Data Analytics
We’ve seen tremendous technological innovation in the data analytics space over the past 10 years….
GPUs Tackle Massive Data of the Hive Mind
LIVE from GTC12 — The flock of birds that weaves seamlessly through the sky, propelled…
Why Hadoop on IBM Power
In the quest to achieve data-driven insight, Hadoop running on Intel X86-based processors has emerged…
Datanami’s Leverage Big Data Summit Wraps Up
Dialog and networking were on the Datanami agenda this week as we kicked off our…
DeepSeek Open-Sources the Missing Layer Between AI Models and Agents
AI models are getting better at a rapid pace. They are now able to reason,…
AI Turns Genomic Data Into 16 New Bacteria-Killing Viruses
Scientists are using GenAI and massive stores of genomic data to design new biological systems…
NVIDIA’s $500 Billion AI Bet: Jensen Huang Brings Wall Street Into the Race
Jensen Huang believes NVIDIA’s chips are becoming much more than expensive pieces of hardware. As…
Your Data is Not Ready: Solving the First Mile Gap for Enterprise AI
There’s trouble brewing in the enterprise AI world and IT leaders are faced with a…
Samsung Attacks the AI Memory Wall with 3D Packaging
The AI memory wall has emerged as the biggest bottleneck in executing AI inference workloads….
Microchip and Micron Bet AI’s Next Infrastructure Battle Will Be Storage
Earlier this week, Microchip and Micron unveiled a new PCIe Gen6 AI storage architecture that…
