A research team from the United States is using artificial intelligence to better understand the turbulent behavior of plasma inside nuclear fusion devices like ITER. The MIT team believes that unlocking the secrets of plasma behavior is key to advancing fusion energy, and they are applying AI-enhanced simulations to keep fusion research on track while discovering more efficient ways to generate energy.
Nathan Howard, a principal research scientist at MIT’s Plasma Science and Fusion Center, emphasizes the significance of fusion’s scientific potential and its promise as a clean energy source. As a member of the Magnetic Fusion Experiments Integrated Modeling group, Howard collaborates with group leader Pablo Rodriguez-Fernandez to forecast how different fusion technologies or configurations might perform before they are tested in real-world conditions.
The team’s research, recently published in a scientific paper, highlights a breakthrough in fusion efficiency. By analyzing different operational setups, they discovered that ITER could potentially produce the same energy output while requiring less energy input. This finding could lead to more efficient fusion devices in the future.
The team’s AI-driven approach relies on CGYRO, a computer code developed in collaboration with General Atomics. CGYRO applies a complex plasma physics model to predefined fusion operating conditions, generating detailed simulations of plasma behavior within a reactor. These simulations, though highly accurate, are time-intensive. To streamline the process, MIT researchers developed the PORTALS framework, which uses machine learning to build “surrogate” models. These surrogates quickly mimic the results of CGYRO’s complex simulations, significantly speeding up the predictive process.
Rodriguez-Fernandez explains that PORTALS refines its accuracy by comparing the surrogate model’s predictions with CGYRO’s full simulations. If discrepancies arise, the model undergoes further training to enhance precision. Once fully trained, the surrogate model can predict plasma behavior under different conditions with minimal reliance on additional high-fidelity simulations.
The AI-enhanced simulations provided valuable insights into ITER’s expected performance. Using 14 iterations of CGYRO, Howard confirmed that ITER’s current baseline configuration could achieve a power output ten times greater than the energy input into the plasma. The surrogate-enhanced modeling also revealed that reducing power input had minimal impact on plasma core temperature, suggesting that ITER could operate more efficiently than previously thought.
Howard describes the team’s modeling as the highest-fidelity plasma simulation currently possible, with their findings likely representing the most advanced published research in this area.
