Scientists have developed a new artificial intelligence (AI) tool called Diag2Diag to improve fusion energy research. Plasma, a hot, charged gas, is the fuel for fusion. Diag2Diag helps fill in missing information about plasma by analyzing data from diagnostic sensors in fusion experiments. These sensors measure things like temperature and density in the plasma.
Diag2Diag takes data from one type of sensor and creates a detailed version of what another sensor might measure. This synthetic data is more precise than what actual sensors can provide, making it easier to control plasma and reduce costs for future fusion systems. The AI was developed through a collaboration involving researchers from Princeton University, the U.S. Department of Energy’s Princeton Plasma Physics Laboratory, and other universities. They used data from experiments at the DIII-D National Fusion Facility to train the AI.
The scientists have described the methods and results of this study in a paper published in Nature Communications.
AI could lead to compact, economical fusion systems
Diag2Diag focuses on diagnostics, which are methods to study plasma using sensors. In fusion devices called tokamaks, a diagnostic called Thomson scattering measures the temperature and density of electrons. However, it doesn’t measure often enough to catch fast changes in the plasma, known as instabilities, which can disrupt energy production. Diag2Diag improves these measurements without needing expensive new hardware. This is especially important for the plasma’s edge, which is critical for producing energy efficiently but hard to monitor.
By enhancing data, Diag2Diag helps scientists keep the plasma stable, supporting the goal of making fusion a reliable, 24/7 energy source. Fewer diagnostics also mean simpler, more compact fusion reactors with lower maintenance costs. The AI also supported a theory about controlling energy bursts called edge-localized modes (ELMs) using magnetic fields. These fields create “magnetic islands” that stabilize the plasma’s edge, improving performance. Diag2Diag’s detailed data confirmed this effect, advancing fusion research.