To ensure the operational continuity of its experimental power source, Japan’s Nippon Telegraph and Telephone Corporation (NTT) has adapted an AI tool originally designed for monitoring company networks to forecast irregularities in a nuclear fusion furnace.
Collaborating with the International Thermonuclear Experimental Reactor (ITER) research project since May 2020, the two entities have been striving to develop alternative energy sources.
NTT states that the Deep Anomaly Surveillance (DeAnoS) AI could enhance the “seamless operation of experiments” by averting equipment failures when integrated with ITER’s fusion facility.
DeAnoS scans ICT systems for excessive data using autoencoders and then employs limited marketing to pinpoint the root cause of anomalies.
Given the extremely high operating temperatures of fusion reactors—reaching millions of degrees Celsius—ensuring smooth operations is paramount. Any errors could lead to equipment damage or other issues resulting in prolonged downtime.
In the event of equipment malfunctions, especially in environments with high-intensity neutron or gamma ray radiation, the repair process is time-consuming and can significantly disrupt integration experiments, as highlighted by NTT.
After identifying furnace deficiencies and anomaly rates, DeANoS utilizes long-term data to analyze their impact on procedures.
Var will oversee the provision of administrative data and the testing environment, while NTT will concentrate on utilizing and enhancing the accuracy of DeANoS.
NTT envisions expanding the program to encompass large-scale systems like plants, provided that the execution proceeds as planned.
Nuclear fusion, unlike its more volatile counterpart nuclear fission, involves heating plasma to extreme temperatures within a powerful magnetic field to emulate the energy production processes of the Sun. This method holds the promise of generating substantial amounts of clean energy.
While ITER aims to achieve a key reactor and first plasma by 2025, fusion power plants are still in the experimental phase. Despite originating in the 1950s, critics argue that significant advancements are still necessary, with some suggesting that large-scale energy production from nuclear fusion may only be viable by 2050.
Noteworthy entities like Google-backed DeepMind have speculated that AI could accelerate this progress. In collaboration with EPFL’s Swiss Plasma Center (SPC) in 2022, DeepMind ventured into managing superheated plasma within a fusion reactor.
