China's nuclear fusion startup applies AI to spot plasma anomalies

China’s nuclear fusion startup Startorus Fusion, collaborating with Tsinghua University, has developed SUNIST-2, a compact spherical tokamak designed to advance fusion research by leveraging artificial intelligence (AI). The company aims to improve fusion reactor operation, plasma control, and diagnostic reconstruction through intelligent, real-time sensing and anomaly detection. Unlike traditional nuclear fission, fusion fuses light elements like hydrogen to release energy without producing radioactive waste, but sustaining stable fusion reactions remains challenging. Startorus Fusion’s approach includes allowing plasma to dissipate heat before reigniting, inspired by automotive engine principles, to overcome these challenges and move closer to commercialization. Startorus Fusion employs AI-driven anomaly detection systems using supervised and unsupervised learning to monitor critical systems such as power supplies, which require stable high currents from multiple devices prone to imbalance and failure. This AI monitoring enables early warnings and automatic alarms, improving safety and maintenance. Additionally, the startup uses data-driven and physics-based models combined with reinforcement learning to optimize plasma control,