Contribution to the 31st IEEE Symposium on Fusion Engineering

Thu, 03 Jul 2025 00:00:00 +0000

SOFE 2025 – Machine learning-based anomaly detection for ITER’s Tokamak Systems Monitor: a gyrotron case study

From June 23 to 26, 2025, I attended the 31st IEEE Symposium on Fusion Engineering (SOFE 2025), hosted by the Plasma Science and Fusion Center (PSFC) at MIT in Boston, USA. This major event gathered 700 participants and featured a wide array of talks and posters covering all aspects of fusion engineering. There was also a strong presence of private fusion companies, including Commonwealth Fusion Systems, Proxima Fusion, and many more, reflecting the growing industrial momentum around fusion energy.

I presented a poster titled “Machine learning-based anomaly detection for ITER’s Tokamak Systems Monitor: a gyrotron case study”. The work focuses on applying unsupervised learning techniques to detect anomalies in gyrotron pulses, using data from the European industrial prototype for ITER. The proposed pipeline extracts statistical features from each pulse, reduces dimensionality via PCA, and applies DBSCAN clustering to identify outliers in an entirely data-driven way. It enables the detection of subtle irregularities without requiring labeled examples or predefined failure modes.

This method is designed to integrate with ITER’s Tokamak Systems Monitor (TSM) software, which will support operators by providing early warnings and health assessments across multiple subsystems. While demonstrated on intershot gyrotron data, the approach is generalizable and can be adapted to other systems and subsystems covered by the TSM.

A conference proceedings paper associated with this contribution will be published in IEEE Transactions on Plasma Science.

See the poster below or download it as a PDF here.

Contribution to the 33rd Symposium on Fusion Technology

Sun, 22 Sep 2024 00:00:00 +0000

SOFT 2024 - Preliminary machine learning-based calibration strategy for the ITER Tokamak Systems Monitor

From September 22 to 27, 2024, I had the opportunity to attend the 33rd Symposium on Fusion Technology (SOFT 2024) at Dublin City University, Ireland. This premier event brought together researchers, engineers, and industry experts from around the world to discuss the latest advancements in fusion energy. The conference covered a wide range of topics, from reactor technology and materials science to plasma diagnostics and control systems, fostering valuable exchanges within the international fusion community.

As part of my research at ITER, I presented a scientific poster on a machine learning-based calibration strategy for the ITER Tokamak Systems Monitor (TSM), a software framework that aggregates data from multiple sensors installed across ITER to assess the machine’s health. This work focuses on improving the accuracy and reliability of numerical models that monitor the ITER tokamak’s structural dynamics. This work explores the use of sequential model-based optimization (SMBO) for the calibration of finite-element models. The goal is to refine structural models of the tokamak using a limited dataset, addressing the challenge of calibrating complex engineering systems under restrictive conditions—a common limitation in fusion experiments due to the harsh operational environment and limited sensor accessibility.

See the poster below or download it as a PDF here.

Conferences | Joris Paret

Contribution to the 31st IEEE Symposium on Fusion Engineering

SOFE 2025 – Machine learning-based anomaly detection for ITER’s Tokamak Systems Monitor: a gyrotron case study

Contribution to the 33rd Symposium on Fusion Technology

SOFT 2024 - Preliminary machine learning-based calibration strategy for the ITER Tokamak Systems Monitor