New AI Program Helps Identify Elusive Space Plasmoids

Scientists now have a new tool that may give them an edge in the ongoing game of hide-and-seek in space: physicists at the U.S. Department of Energy's (DOE) Princeton Plasma Physics Laboratory (PPPL) have developed a computer program incorporating machine learning that could help identify clumps of plasma called plasmoids in space. What's novel is that the program was trained using simulated data.

The program will sift through reams of data collected by spacecraft within the magnetosphere — a region of space strongly influenced by Earth's magnetic field — to point out telltale signs of the elusive clump. Using the technique, scientists hope to learn more about the processes that govern magnetic reconnection, a process that occurs within the magnetosphere and throughout space and can cause damage to communications satellites and power grids.

Scientists believe that machine learning can improve our ability to spot plasmoids, advance our fundamental understanding of magnetic reconnection, and help researchers better prepare for the aftermath of disturbances caused by reconnection.

“To our knowledge, this is the first time that artificial intelligence trained on simulated data has been used to look for plasmoids,” said Kendra Bergstedt, a graduate student in Princeton University's Plasma Physics Program who is based at PPPL and is first author on the paper reporting the results. Earth and Space SciencesThe research combines the lab's growing expertise in computational science with a long history of studying magnetic reconnection.

Looking for a link

Scientists want to find a reliable and accurate way to detect plasmoids, so they can determine whether they affect magnetic reconnection, a process in which magnetic field lines separate and then violently reconnect, releasing enormous amounts of energy. When reconnection occurs near Earth, it can set off a chain reaction of charged particles that fall into the atmosphere, disrupting satellites, cellphones, and power grids.

“Some researchers believe that plasmoids help large plasmas reconnect quickly,” said Hantao Ji, a professor of astrophysics at Princeton University and a distinguished investigator at PPPL, “but these hypotheses remain unproven.”

The researchers want to know if plasmoids can change the rate at which recombination occurs, and also to measure how much energy recombination imparts to plasma particles.

“But to figure out the relationship between plasmoids and recombination, we need to know where the plasmoids are,” Bergstedt says. “Machine learning can help us do that.”

The scientists used computer-generated training data to make sure the program could recognise a variety of plasma signatures. Typically, plasmoids created by computer models are idealised versions based on mathematical formulas and have shapes that are not often seen in nature, such as perfect circles.

If the program had been trained only to recognize these perfect versions, it might have missed other versions of the shape. To prevent this, Bergsted and Ji decided to use artificial, deliberately incomplete data so that the program would have an accurate baseline for future studies.

“Compared to mathematical models, the real world is complicated,” Bergstedt says.

“So we decided to train the program using fluctuating data from real-world observations. For example, rather than starting the simulation with a perfectly flat current sheet, we gave the sheet some wobble. We hope that this machine learning approach will enable us to capture finer nuances than a strict mathematical model can.”

The research builds on previous efforts by Bergstedt and Gee to create a computer program that incorporated a more idealized model of a plasmoid.

Scientists say the use of machine learning will become increasingly common in astrophysics research. “It can be especially useful when making inferences from a small number of measurements, as we sometimes do when studying recombination,” Ji said. “The best way to learn how to use a new tool is to actually use it. You don't want to sit on the sidelines and miss out.”

Bergstedt and Ji will use the Plasmoid Detection Program to examine data being collected by NASA's Magnetospheric Multiscale (MMS) mission. Launched in 2015 to study reconnection, MMS consists of four spacecraft flying in formation through the plasma of the magnetotail, a region of space away from the Sun that is controlled by Earth's magnetic field.

The magnetotail's accessibility and size make it an ideal place to study reconnection.

“When we observe the Sun to study recombination, we can only measure it from a distance,” Bergstedt says, “If we were to observe recombination in the laboratory, we could put instruments directly into the plasma, but the size of the plasma would be smaller than what we typically see in space.”

Studying magnetotail reconnection is an ideal middle ground: “It's a large-scale, naturally occurring plasma that we can measure directly with spacecraft passing through it,” Bergstedt said.

Bergsted and Ji hope to take two key steps in improving their plasmoid-detection program. The first is to perform a procedure called domain adaptation, which helps the program analyze data sets it has never encountered before. The second step is to use the program to analyze data from the MMS spacecraft.

“The methodology we've demonstrated is largely a proof-of-concept because we haven't actively optimized it yet,” Bergstedt said. “We'd like to make the model work even better than it does now, apply it to real data, and go from there.”