Human technology

Machine learning facilitates ‘turbulence tracking’ in fusion reactors | MIT News

Fusion, which promises virtually limitless, carbon-free energy using the same processes that power the sun, is at the heart of a global research effort that could help mitigate climate change.

A multidisciplinary team of researchers is now bringing tools and insights from machine learning to support this effort. Scientists at MIT and elsewhere have used computer vision models to identify and track turbulent structures that appear under the conditions necessary to facilitate fusion reactions.

Monitoring the formation and movements of these structures, called filaments or “blobs”, is important for understanding the heat and particle fluxes exiting the reacting fuel, which ultimately determines the engineering requirements for the walls to of the reactor respond to these fluxes. However, scientists typically study blobs using averaging techniques, which trade the details of individual structures in favor of aggregate statistics. Individual blob information must be tracked by manually marking it in the video data.

The researchers created a synthetic video dataset of plasma turbulence to make this process more effective and efficient. They used it to train four computer vision models, each of which identifies and tracks blobs. They trained the models to identify the blobs the same way humans would.

When the researchers tested the trained models using real video clips, the models were able to identify the blobs with high accuracy – over 80% in some cases. The models were also able to effectively estimate the size of the drops and the speed at which they were moving.

Since millions of video frames are captured during a single fusion experiment, using machine learning models to track blobs could provide scientists with much more detailed information.

“Before, we could get a macroscopic picture of what these structures are doing on average. Now we have a microscope and the computing power to analyze one event at a time. If we step back, what this reveals is the power available to these machine learning techniques and ways to use these computational resources to advance,” says Theodore Golfinopoulos, a researcher at MIT Plasma Science and Fusion Center and co. -author of an article detailing these approaches.

His fellow co-authors include lead author Woonghee “Harry” Han, PhD candidate in physics; lead author Iddo Drori, visiting professor at the Computer Science and Artificial Intelligence Laboratory (CSAIL), associate professor at Boston University and adjunct at Columbia University; along with others from the MIT Plasma Science and Fusion Center, the MIT Department of Civil and Environmental Engineering, and the Swiss Federal Institute of Technology in Lausanne, Switzerland. Research appears today in Nature Science Reports.

heat things up

For more than 70 years, scientists have sought to use controlled thermonuclear fusion reactions to develop a source of energy. To achieve the conditions necessary for a fusion reaction, the fuel must be heated to temperatures above 100 million degrees Celsius. (The sun’s core is about 15 million degrees Celsius.)

A common method of containing this super hot fuel, called plasma, is using a tokamak. These devices use extremely strong magnetic fields to hold the plasma in place and control the interaction between the plasma exhaust heat and the reactor walls.

However, the drops appear as filaments falling from the plasma at the very edge, between the plasma and the walls of the reactor. These random and turbulent structures affect the way energy flows between the plasma and the reactor.

“Knowing what the blobs are doing severely limits the engineering performance your tokamak power plant needs at the edge,” adds Golfinopoulos.

The researchers use a unique imaging technique to capture video of the turbulent plasma edge during the experiments. An experimental campaign can last for months; a typical day will produce about 30 seconds of data, corresponding to about 60 million video frames, with thousands of blobs appearing every second. This makes it impossible to manually track all blobs, so researchers rely on average sampling techniques that only provide general characteristics of blob size, speed, and frequency.

“On the other hand, machine learning provides a solution to this by tracking blob by blob for every frame, not just average quantities. This gives us a lot more insight into what is happening at the plasma boundary,” says Han.

He and his co-authors took four well-established computer vision models, which are commonly used for applications such as autonomous driving, and trained them to solve this problem.

Simulation of drops

To train these models, they created a large dataset of synthetic video clips that captured the random and unpredictable nature of the blobs.

“Sometimes they change direction or speed, sometimes multiple drops merge or separate. These types of events were previously not accounted for with traditional approaches, but we could freely simulate these behaviors in the synthetic data,” says Han.

Creating synthetic data also allowed them to label each blob, which made the training process more efficient, Drori adds.

Using this synthetic data, they trained the models to draw boundaries around the blobs, teaching them to closely mimic what a human scientist would draw.

Then they tested the models using real video data from experiments. First, they measured how closely the boundaries drawn by the models matched the actual contours of the drops.

But they also wanted to see if the models predicted objects that humans would identify. They asked three human experts to identify the centers of the blobs in the video images and checked whether the models predicted blobs in the same locations.

The models were able to plot precise spot boundaries, overlapping brightness contours that are considered ground truths, about 80% of the time. Their assessments were similar to those of human experts and successfully predicted the regime defined by the blob theory, which is consistent with the results of a traditional method.

Now that they have shown success in using synthetic data and computer vision models for blob tracking, the researchers plan to apply these techniques to other fusion research problems, such as the estimation of particle transport at the boundary of a plasma, explains Han.

They’ve also made the dataset and models publicly available, and look forward to seeing how other research groups apply these tools to study blob dynamics, Drori says.

“Before that, there was a barrier to entry that most of the only people working on this problem were plasma physicists, who had the datasets and used their methods. There’s a huge machine learning community and One of the goals of this work is to encourage participation in fusion research from the broader machine learning community with the broader goal of helping solve the critical problem of climate change. “, he adds.

This research is supported, in part, by the US Department of Energy and the Swiss National Science Foundation.