Princeton AI restores missing fusion data to improve reactor control

An international team led by Princeton University has developed an AI system called Diag2Diag that generates synthetic sensor data inside fusion reactors to enhance plasma monitoring and control. By analyzing existing sensor measurements, the AI effectively acts as a virtual sensor, filling gaps when physical sensors fail or are too slow. This capability provides more detailed insights into plasma behavior, such as validating the theory that small magnetic fields create “magnetic islands” to suppress damaging edge-localized modes (ELMs) by flattening temperature and density profiles—effects that physical sensors alone could not fully capture. The improved diagnostic detail from Diag2Diag is crucial for the development of commercial fusion power plants, which must operate continuously without interruption, unlike current experimental reactors that can be shut down if sensors fail. The AI also offers economic and design advantages by potentially reducing the number of physical sensors needed, making future reactors more compact, simpler, and less costly to build and maintain. Beyond fusion, the team suggests this AI approach could enhance sensor data in