New AI hits 94% accuracy in predicting nuclear fusion plasma failures

Source: interestingengineering
Author: @IntEngineering
Published: 8/21/2025
To read the full content, please visit the original article.
Read original articleResearchers at the Hefei Institutes of Physical Science, Chinese Academy of Sciences, led by Professor Sun Youwen, have developed two advanced AI systems aimed at enhancing the safety and efficiency of nuclear fusion experiments. The first AI tool uses interpretable decision tree models to predict plasma disruptions—sudden events that can damage fusion reactors—achieving a 94% accuracy rate and providing warnings approximately 137 milliseconds before disruptions occur. This early detection focuses on identifying ‘locked modes,’ a common plasma instability, and offers transparent insights into the physical signals behind disruptions. The second AI system employs a multi-task learning model to monitor plasma states, accurately classifying different operating modes and detecting edge-localized modes (ELMs) with a 96.7% success rate, thereby supporting smoother and safer reactor operations.
These AI innovations address critical challenges in nuclear fusion, a promising clean energy source that could provide nearly limitless power without carbon emissions or long-lived radioactive waste. As global energy demand rises and the environmental impact of fossil
Tags
energynuclear-fusionAI-in-energyplasma-monitoringfusion-reactor-safetyclean-energyfusion-energy-technology