Reactor-Relevant Control Strategies in Diagnostically Challenged Environments

2025 Research Campaign, Diagnostics and Actuators

Purpose of Experiment

Future fusion reactors aim to replicate the sun’s power on Earth, but they will operate in extremely harsh environments that can damage or limit traditional diagnostic tools used to monitor and control the plasma—the ultra-hot, charged gas where fusion occurs. To address this challenge, our team developed a machine learning algorithm called RTCAKENN, which accurately predicts crucial plasma properties even when key diagnostic measurements are missing or of low quality. By training RTCAKENN on extensive data from thousands of experiments, it can infer the necessary information to maintain plasma stability and performance. We also created a predictive control system that uses RTCAKENN’s outputs to make real-time decisions on adjusting inputs like heating power and fuel injection, ensuring the plasma stays in the desired state. This advancement not only prepares us for future reactors with limited diagnostics but also helps current fusion experiments overcome occasional data gaps. Our successful experiments will have demonstrated effective control of plasma properties such as rotation and density without relying on certain diagnostics, paving the way for more robust and economically viable fusion power plants.