NTT adapts AI network analysis tool to detect faults in nuclear fusion reactors

Here comes the Sun

Japan’s Nippon Telegraph and Telephone Corporation (NTT) has taken an AI tool created to monitor telco networks and adapted it to predict anomalies in a nuclear fusion reactor to help keep its experimental power source running.

The work is currently being done in partnership with the International Thermonuclear Experimental Reactor (ITER) research project. The duo have collaborated since May 2020 to develop alternative means of energy.

NTT hopes its Deep Anomaly Surveillance (DeAnoS) AI, applied to ITER's fusion facility, could prevent equipment failure and contribute to "smooth experimental operation."

DeANoS uses autoencoders to find abnormal data in ICT systems. Sparse optimization can then be applied to estimate which input dimension caused the anomaly, thereby identifying the root cause.

Smooth operations are vital for fusion reactors because the theoretical machines operate at enormous temperatures – millions of degrees Celsius – and errors could damage equipment or cause other problems that lead to days or weeks of downtime.

"In the event of equipment failure, especially in a high-intensity, high-volume neutron or gamma-ray radiation environment, it will take time to repair the equipment, which will greatly affect the operation of fusion experiments," NTT explained [PDF].

DeANoS first detects reactor faults and rates anomalies, then analyzes the impact on operations using long-term data.

While NTT focuses on the use and validity of DeANoS, ITER will be in charge of providing operational data as well as the testing environment.

Assuming everything goes to plan, NTT expects the duo will "expand the system to large-scale systems such as plants."

Nuclear fusion plants mimic the processes that power the Sun – cooking plasma at extremely high temperatures while containing it in a powerful magnetic field. The technique has potential to produce vast amounts of clean energy, without many of the risks associated with its temperamental sibling, nuclear fission.

Fusion power plants are still experimental, with ITER's main reactor and first plasma planned for 2025. But the technology has been in development since the 1950s, and many skeptics think there's still a long way to go. Indeed some have predicted 2050 as the earliest year by which large-scale energy from nuclear fusion may be possible.

Other interested parties, like Google-backed DeepMind, have bet on AI accelerating the process. In 2022, it teamed with EPFL's Swiss Plasma Center (SPC) to control the superheated plasma inside a fusion reactor. ®

More about


Send us news

Other stories you might like