A new artificial intelligence system called Diag2Diag is transforming how scientists monitor and control plasma in fusion energy experiments. Developed by researchers at Princeton University, the U.S. Department of Energy’s Princeton Plasma Physics Laboratory (PPPL), and international partners including Chung-Ang University, Columbia University, and Seoul National University, the AI fills in missing data from fusion diagnostics using synthetic data that matches real-world measurements with greater detail. The system works by taking data from one type of sensor and generating a high-resolution, synthetic version of what a different sensor would have measured. This allows scientists to gain deeper insights into plasma behavior without needing additional hardware. The research, published in Nature Communications, was based on data collected from experiments at the DIII-D National Fusion Facility, a DOE user facility. One of the key challenges in fusion research is capturing fast-evolving plasma instabilities—sudden changes that can disrupt energy production. While many diagnostics measure plasma characteristics like temperature and density, some do so too slowly to detect rapid changes. For example, Thomson scattering, a widely used diagnostic in tokamak fusion devices, measures electron temperature and density but not frequently enough to capture fine details at the plasma edge, known as the pedestal. This region is critical for achieving high-performance fusion, yet difficult to monitor. Diag2Diag enhances the resolution and timing of these measurements, effectively giving existing sensors a performance boost without the cost of new equipment. “It’s like giving your diagnostics a superpower,” said Egemen Kolemen, principal investigator of the project and a joint appointee at PPPL and Princeton’s Andlinger Center for Energy and the Environment. The AI also provides new evidence supporting a leading theory on how edge-localized modes (ELMs)—powerful bursts of energy that can damage reactor walls—can be suppressed. By applying resonant magnetic perturbations (RMPs), scientists can create small magnetic islands at the plasma edge, which flatten temperature and density profiles. These changes help stabilize the plasma. However, traditional diagnostics cannot clearly observe this flattening. Diag2Diag’s synthetic data revealed this phenomenon in greater detail, strengthening the magnetic island theory. This insight is vital for designing future commercial fusion reactors, which must be both reliable and cost-effective. As SangKyeun Kim, a PPPL staff research scientist on the team, noted, future reactors will likely have fewer diagnostics to reduce size, complexity, and maintenance costs. Diag2Diag enables this shift by making fewer sensors more powerful and informative. Beyond fusion, the AI has potential applications in spacecraft, robotic surgery, and other safety-critical systems where sensor data may be incomplete or degraded. The research team is already exploring broader uses, with multiple institutions expressing interest in adopting the technology. With its ability to enhance data, improve control, and reduce hardware needs, Diag2Diag represents a significant step toward making fusion energy a practical and sustainable power source.