AI-Driven diagnostics: a new view into fusion plasma
A cutting-edge artificial intelligence (AI) approach is changing how scientists monitor and control fusion plasma. The Diag2Diag system analyzes data from multiple sensors in a fusion device and generates a synthetic dataset for a different type of diagnostic. In short, AI fills in missing data by learning from existing measurements, producing a view of the plasma that can be richer and faster than what hardware alone could deliver. In the words of the research team, this method aligns synthetic data with real measurements and can provide more detail than a single physical sensor. The result is a more robust control framework for fusion devices and the potential to simplify future reactors while maintaining performance.
The concept, developed by a group led by Azarakhsh Jalalvand of Princeton University, is called Diag2Diag. It takes the data from various plasma sensors, learns their relationships, and then reconstructs a high-fidelity picture of the plasma state for diagnostics that may be slower or less comprehensive. By enabling faster, more precise insight without adding hardware, Diag2Diag could help fusion systems run more smoothly and with lower costs — a crucial step for commercial fusion energy.
Collaborative science across borders and facilities
The work is an international collaboration among Princeton University, the U.S. Department of Energy’s Princeton Plasma Physics Laboratory (PPPL), Chung-Ang University, Columbia University, and Seoul National University. All sensor data used to train and test Diag2Diag came from experiments at the DIII-D National Fusion Facility, a DOE user facility. The collaboration demonstrates how cross-institution expertise can accelerate the adoption of AI in high-stakes physics experiments while leveraging existing experimental data to improve future machines.
Researchers emphasize that Diag2Diag offers a practical path forward for fusion energy: “Today’s tokamaks are experimental laboratories, and a single sensor failure can lead to costly downtime. For future fusion power plants that must operate continuously, reliable diagnostics are essential — and Diag2Diag helps achieve that reliability with fewer hardware components.”
Toward compact, economical fusion systems
A key advantage cited by the team is the potential to dramatically reduce the number of diagnostic components needed inside a fusion reactor. Fewer diagnostics mean more usable space for energy-producing hardware, easier maintenance, and lower costs. This aligns with a broader goal: making fusion devices more compact and robust so they can operate around the clock if built as commercial reactors. Such improvements are vital for the long-term viability of fusion as a reliable electricity source.
In addition to boosting overall reliability, Diag2Diag offers new insight into how plasma behaves at the edge of the device — the pedestal — where temperature and density gradients are most difficult to measure. Since edge measurements are crucial for maintaining stability, the enhanced data can guide operators toward more stable, high-performance plasma regimes and reduce the risk of costly disruptions.
ELMs, RMPs, and the edge: a data-driven theory test
Edge-localized modes (ELMs) are bursts of energy that can damage reactor walls, and researchers are exploring methods to suppress them using resonant magnetic perturbations (RMPs). The AI-generated data helped scientists see a clearer picture of how RMPs influence the pedestal. The team reports evidence supporting the idea that magnetic islands created by RMPs can flatten temperature and density at the plasma edge, a condition conducive to stabilizing the plasma and mitigating ELMs. This enhanced understanding is a valuable boost to both theoretical modeling and practical control strategies for future fusion devices.
Looking ahead: wider use and new frontiers
The researchers are already planning to expand Diag2Diag to additional diagnostics and contexts. Beyond fusion energy, the team notes that the approach could be applied to other systems where diagnostic data are missing or degraded, including spacecraft monitoring and robotic surgery. In short, the method could improve reliability in critical environments by recovering important information from imperfect sensor data.
Support for the project came from multiple DOE awards and international partners, with PPPL and Princeton leading the efforts. As the fusion field moves toward commercial energy, Diag2Diag stands out as a practical, cost-conscious tool that reinforces reliability and performance in next-generation reactors.
PPPL’s work in AI-assisted plasma stabilization, together with ongoing research into ELM suppression, positions the laboratory at the forefront of a new era where intelligent diagnostics help make fusion a dependable, scalable power source. The integration of AI with traditional diagnostics marks a significant step toward achieving continuous, economical fusion energy.
