Scientists are implementing a new artificial intelligence framework to streamline experimental campaigns at the DIII-D National Fusion Facility, which is operated by General Atomics. The initiative, a collaboration with Argonne National Laboratory, aims to reduce the downtime between experimental plasma shots by automating complex data analysis. In typical tokamak operations, diagnostic data from brief, high-power plasma discharges must be analyzed to inform parameter adjustments for subsequent shots. This process has historically been a manual bottleneck, limiting the number of experiments that can be conducted within a given operational window. The new workflow automates this analysis, enabling a higher experimental tempo. Source: HPCwire
The operational cycle of the DIII-D tokamak involves plasma shots lasting only a few seconds, followed by a necessary cooling period for the device's magnetic coils and other systems. The new computational approach leverages this inter-shot period to run sophisticated analysis on high-performance computing resources. By offloading the analysis to supercomputers and using AI to interpret the results, researchers can more rapidly validate physics models and plan the next experiment. This near-real-time feedback loop is designed to maximize the scientific output of the facility's limited operational time and accelerate the overall pace of fusion energy research. Source: HPCwire
The new computational approach leverages this inter-shot period to run sophisticated analysis on high-performance computing resources.
This integration of AI and supercomputing into the experimental loop represents a significant operational shift for major fusion facilities. The goal is to move from a sequential, human-in-the-loop process to a more autonomous system that can explore plasma parameter space more efficiently. Such a system can identify optimal conditions for plasma stability and confinement, which are critical for achieving net energy gain. The techniques being developed at the DIII-D National Fusion Facility could serve as a model for future fusion devices, including burning plasma experiments like ITER, where maximizing research efficiency will be paramount due to high operational costs and complex physics. Source: HPCwire
The successful deployment of this automated workflow will be measured by its ability to increase the number of scientifically valuable plasma shots per day. While the initial announcement does not quantify the expected increase in experimental throughput, the project's focus is on reducing the critical path time dominated by post-shot data analysis. Future work will likely involve refining the AI models to handle a wider range of plasma phenomena and integrating them more deeply into the tokamak's control systems. This would enable the system to not only analyze results but also to propose and even autonomously execute optimized experimental plans, further advancing the application of machine learning in fusion science. Source: HPCwire