Search

One of the big challenges in developing fusion energy is controlling the plasma. Disruptions of the burning plasma can halt the fusion reaction and damage the walls of the fusion device. Scientists therefore are keen to learn how to predict plasma disruptions.

Researchers at the Princeton Plasma Physics Laboratory (PPPL) employ artificial intelligence to improve predictive capability. They developed predictive software, the Fusion Recurrent Neural Network (FRNN) code, which is a form of what is called "deep learning"—a powerful version of modern machine-learning software. (Research team: William Tang, to the left; Eliot Feibush; and Alexey Svyatkovskiy, seated)

Drawing from data of the JET facility in the UK, the team has demonstrated the tool's ability to predict disruptions more accurately than previous methods. Turning their attention now to ITER, the team aims to improve the correct prediction of disruptions and reduce the number of false alarms.

The deep learning software is also a challenge for computing hardware. Several tests with the FRNN on modern powerful supercomputing devices, so-called GPU clusters, in the United States, Europe and Asia showed promising results.

Read the full article on the PPPL website.

At the joint meeting of the 26th International Toki Conference (ITC) and the 11th Asia Plasma and Fusion Association (APFA) conference—held from 5 to 8 December in Toki, Japan—ITER Science & Operations Department head David Campbell gave a plenary talk on ITER progress.

It was the occasion to recognize the many contributions of Dr Campbell to fusion research over the course of his career. The chair of the international program committee, Yutaka Kamada, and the chair of the international advisory committee, Yasuhiko Takeiri, both presented tributes. Dr Campbell retired in December from the ITER Organization.

The theme this year of the ITC meeting was "A new era in plasma and fusion research." The launch of deuterium experiments in Japan's Large Helical Device, and the first experimental results, were highlighted at the meeting, which also brought together academics and researchers from Japan, China, Korea and India in the context of the biennial APFA meeting. Three hundred participants in all took part in the four-day event.