listen to this articleThe Princeton Plasma Physics Institute last week published a new paper presenting important results from the institute's artificial intelligence research.
In a paper published in Nature Communications, PPPL researchers describe how they used machine learning to avoid the magnetic perturbations and disruptions that make fusion plasma unstable .
“This result is particularly impressive because we were able to achieve the results in two different tokamaks using the same code,” said PPPL staff research physicist SangKyeun Kim, lead author of the paper.
A tokamak is a donut-shaped device that uses a magnetic field to hold plasma.
Egemen Coleman, associate professor in the Department of Mechanical and Aerospace Engineering, said plasma instability could cause significant damage to fusion devices.
“They cannot be installed on commercial fusion ships. Our research advances the field and suggests that artificial intelligence will play a key role in managing future fusion reactions, ensuring that plasmas are maintained while avoiding instability.” “It shows that we can generate as much fusion energy as possible,” Coleman said.
groundbreaking work
NJBIZ recently reported on the groundbreaking of a more than $100 million effort at PPPL headquarters on Princeton University's Forrestal Campus in Plainsboro. This location will serve as an international fusion research hub. PPPL is a Department of Energy (DOE) National Laboratory managed by Princeton University.
Institute director Steve Cowley said the milestone was a sign the institute was moving forward.
“We are ramping up our efforts to deliver fusion energy,” Cawley continued, “and we are also working to support other parts of the economy, particularly the microelectronics and sustainable manufacturing sectors. We're leveraging our expertise.'' To advance that vision, PPIC will need a laboratory and an office. ”
Fusion energy offers enormous potential to produce virtually unlimited clean energy through the fusion of plasmas. The complex dance in which atoms fuse and release energy is something scientists have studied and pursued for decades.
This lab uses AI machine learning to power these efforts.
The ability to learn and adapt using these techniques allows PPPL researchers to perfect the design of the vessel surrounding the ultra-hot plasma, optimize heating methods, maintain stable control of the reaction, and more. We believe that we can improve the control of fusion reactions in a variety of ways. increasingly longer periods.
Operational optimization
The latest breakthrough builds on PPPL's previous work in using AI to reduce instability.
In 2019, PPPL principal research physicist William Tan and his team demonstrated for the first time the ability to transfer a process from one tokamak to another.
“Our research is a breakthrough by using artificial intelligence and machine learning combined with powerful, modern, high-performance computing resources to process vast amounts of data in a thousandth of a second. We have developed a model to address disruptive physical phenomena long before they occur,” said Tan. “You can't effectively deal with disruption for more than a few milliseconds. It's like starting treatment for a deadly cancer that's already too advanced.”
These are among the many AI-powered initiatives and projects underway across the lab.
Other examples include:
- Using machine learning to improve the design of another type of fusion reactor, the stellarator.
- Utilizes AI to speed up the code called HEAT.
- We are perfecting a technique known as ion cyclotron radiofrequency heating to find the optimal conditions for heating ions in plasma.and
- We use experimental data from various tokamaks, plasma simulation data, and artificial intelligence to study the behavior of the plasma edge during nuclear fusion. Researchers hope this will reveal the most effective way to confine plasma in commercial-scale tokamaks.
Fatima Ebrahimi, principal research physicist at PPPL, said: “We are applying machine learning to all our data and simulations to help bridge the technology gap and integrate high-performance plasma into viable fusion power plant systems. I would like to explore how it can be used.”
Ebrahimi is the lead investor in the four-year project. D.O.E.Advanced Scientific Computing Research Program.
“Machine learning offers great potential for optimizing code,” said Alvaro Sanchez Villar, associate research physicist at PPPL. “Essentially, we can predict how the plasma will evolve and we can modify it in real time, giving us better control over the plasma.”
