Researchers at the DIII-D National Fusion Facility have successfully deployed a new control platform, the Plasma-facing Actuator Control using Machine learning (PACMAN), to predict and mitigate a critical plasma instability. In recent experiments, the system identified the onset conditions for a neoclassical tearing mode approximately 200 milliseconds before it would have formed. The PACMAN system then initiated a corrective action, which executed in just 20 milliseconds, successfully preventing the instability from developing and disrupting the plasma discharge. This demonstration marks a significant step in developing real-time, autonomous control systems necessary for sustained, high-performance operation in future fusion power plants. Source: Interesting Engineering
Neoclassical tearing modes are a persistent challenge in high-performance tokamak operation. These magnetohydrodynamic (MHD) instabilities manifest as helical magnetic islands that degrade plasma confinement by allowing heat and particles to escape more easily across magnetic field lines. If left unchecked, their growth can cool the plasma, reduce fusion power output, and ultimately lead to a major disruption, an event that can terminate the plasma discharge and potentially damage the reactor's internal components. The ability to suppress these modes before they fully develop is crucial for maintaining the stable, high-pressure conditions required for net energy gain in devices like ITER and future commercial reactors. Source: Interesting Engineering
Neoclassical tearing modes are a persistent challenge in high-performance tokamak operation.
The PACMAN system operates by continuously monitoring plasma parameters and feeding the data into a machine learning model trained to recognize the precursors of tearing modes. This predictive capability allows the system to act proactively rather than reactively. Upon predicting an imminent instability, PACMAN triggers a localized injection of electron cyclotron waves into the plasma. This targeted heating alters the plasma current profile in the specific region where the magnetic island would form, thereby removing the conditions necessary for the instability's growth. The entire predict-and-actuate cycle's speed is a key feature, enabling intervention well within the instability's growth timescale. Source: Interesting Engineering
This achievement by General Atomics, which operates DIII-D for the U.S. Department of Energy, provides a proof-of-concept for AI-driven control schemes that will be essential for steady-state tokamak operation. While demonstrated at DIII-D, the underlying principles and control architecture are designed to be adaptable to other magnetic confinement devices. Future work will likely focus on expanding the system's capabilities to predict and control a wider range of plasma instabilities simultaneously. Integrating such intelligent, autonomous systems is a critical path item for optimizing the performance and ensuring the operational reliability of commercial-scale fusion energy. Source: Interesting Engineering