Publications | Joris Paret

“Design and implementation of machine learning-based anomaly detection in the ITER Tokamak Systems Monitor” - Frontiers in Physics

Mon, 01 Jun 2026 00:00:00 +0000

Authors

Joris Paret
Brian Sammuli
Victor Costa Perez
Nathaniel Saura
Laura Hernandez Cubo
Daniel Iglesias

Abstract

The Tokamak Systems Monitor (TSM) is a software suite under development at ITER that provides operators with an integrated view of the tokamak’s engineering health based on operational instrumentation. A key functionality of TSM is anomaly detection, aimed at identifying unexpected behaviors across a wide range of systems. A dedicated anomaly-detection module integrates multiple machine-learning-based algorithms, ranging from intershot classification of complete pulses to online detection of localized events. The article illustrates the framework with a gyrotron-pulse classifier and a time-localized approach for monitoring magnet power supplies. Automated warnings generated by the module support operators in evaluating anomalies and improving ITER reliability.

Article availability

This open-access paper is published in Frontiers in Physics.

“Machine Learning-Based Anomaly Detection for ITER's Tokamak Systems Monitor: A Gyrotron Case Study” - IEEE Transactions on Plasma Science

Fri, 08 May 2026 00:00:00 +0000

Authors

Joris Paret
Daniel Iglesias
Daniel Sabio Ruiz
Dilin Meloni
Andrea Antonione
Ruggero Bertazzoni
Giuseppe Carannante
Martino Ferrari
Maria Ortiz de Zuniga
Mario Cavinato
Francisco Sanchez Arcos
Alfredo Portone

Abstract

The Tokamak Systems Monitor (TSM) software will provide ITER operators with timely assessments of machine health, component lifetime, and early warnings of potential faults. Among its key functions, the TSM will employ data-driven anomaly-detection methods to identify unexpected or abnormal behavior across a broad range of systems and diagnostics. This work presents an initial proof of concept for anomaly detection applied to gyrotron pulses, leveraging data from the European gyrotron prototype. The method combines dimensionality reduction and clustering techniques to identify deviations from expected operational patterns. While this first demonstration focuses on intershot anomaly detection for gyrotrons, the methodology is designed to be adaptable to other systems and signals within the TSM.

Article availability

This open-access paper is published in IEEE Transactions on Plasma Science.

“Preliminary machine learning-based calibration strategy for the ITER Tokamak Systems Monitor” - Fusion Engineering and Design

Tue, 21 Oct 2025 00:00:00 +0000

Authors

Joris Paret
Daniel Iglesias
Hyogeon Bak
Matthew Clough
George Vayakis
Michael Walsh

Abstract

A machine learning-based calibration strategy for the ITER Tokamak Systems Monitor (TSM) is presented, focusing on the optimization of dynamic finite-element method (FEM) models essential for monitoring and assessing the structural health of the tokamak. Traditional finite-element model updating techniques are unsuitable for the ITER environment due to constraints such as limited sensor placement and the inability to perform controlled excitations. To overcome these challenges, an automated calibration procedure is proposed, leveraging Bayesian optimization with Gaussian Process regression. The parameter space is efficiently explored by sequentially sampling model configurations, minimizing the discrepancy between simulated and reference modal characteristics. This preliminary work establishes a foundation for calibrating other numerical models within the TSM.

Article availability

This paper is published in Fusion Engineering and Design.

“Dimensionality reduction of local structure in glassy binary mixtures” - The Journal of Chemical Physics

Mon, 24 Oct 2022 00:00:00 +0000

Authors

Daniele Coslovich
Robert L. Jack
Joris Paret

Abstract

We consider unsupervised learning methods for characterizing the disordered microscopic structure of supercooled liquids and glasses. Specifically, we perform dimensionality reduction of smooth structural descriptors that describe radial and bond-orientational correlations and assess the ability of the method to grasp the essential structural features of glassy binary mixtures. In several cases, a few collective variables account for the bulk of the structural fluctuations within the first coordination shell and also display a clear connection with the fluctuations of particle mobility. Fine-grained descriptors that characterize the radial dependence of bond-orientational order better capture the structural fluctuations relevant for particle mobility but are also more difficult to parameterize and to interpret. We also find that principal component analysis of bond-orientational order parameters provides identical results to neural network autoencoders while having the advantage of being easily interpretable. Overall, our results indicate that glassy binary mixtures have a broad spectrum of structural features. In the temperature range we investigate, some mixtures display well-defined locally favored structures, which are reflected in bimodal distributions of the structural variables identified by dimensionality reduction.

Article and data availability

This paper is published The Journal of Chemical Physics and is also available on arXiv. The computation of descriptors and most of the dimensionality reduction analysis were performed using the partycls package. Data analysis has been carried out using a reproducible workflow, deposited in the Zenodo public repository.

“partycls: A Python package for structural clustering” - The Journal of Open Source Software

Mon, 08 Nov 2021 00:00:00 +0000

Presentation

partycls is a Python framework for cluster analysis of systems of interacting particles. By grouping particles that share similar structural or dynamical properties, partycls enables rapid and unsupervised exploration of the system’s relevant features. It provides descriptors suitable for applications in condensed matter physics and integrates the necessary tools of unsupervised learning, such as dimensionality reduction, into a streamlined workflow. Through a simple and expressive interface, partycls allows one to open a trajectory file, perform a clustering based on the selected structural descriptor, and analyze and save the results with only a few lines of code.

A short paper presenting the code, written in collaboration with Daniele Coslovich, was published in The Journal of Open Source Software.

Short example

As a simple example, we consider the detection of the grain boundaries in a polycrystal formed by differently oriented FCC crystallites. This is easily achieved even with a simple radial descriptor, since the average radial distribution of particles at the boundaries is different than the one of the crystal in the bulk. The following short piece of code opens the input trajectory stored in the file grains.xyz, computes the local radial distribution functions of the particles, applies a standard Z-Score normalization on the data, and finally performs a clustering using the Gaussian mixture model (GMM) with $K = 2$ clusters (default):

from partycls import Workflow

wf = Workflow('grains.xyz',
 descriptor='gr',
 scaling='zscore',
 clustering='gmm')
wf.run()

Each of these steps is easily tunable, so as to change the workflow with little effort. The labels are available as a simple attribute of the Workflow instance. Optionally, a set of output files can be produced for further analysis, including a trajectory file with the cluster labels. Quick visualization of the clusters, as in the following figure, is possible within partycls through optional visualization backends.

(a) A polycrystalline material with differently oriented FCC crystallites.
(b) Using the individual radial distributions of the particles as structural descriptor, the algorithm identifies the crystalline domains (blue, $k = 0$) and the grain boundaries (red, $k = 1$).
(c) The radial distribution functions restricted to these two clusters display a marked difference, with higher peaks for the crystals. The 3D visualization was performed with OVITO.

Publications | Joris Paret

“Design and implementation of machine learning-based anomaly detection in the ITER Tokamak Systems Monitor” - Frontiers in Physics

Authors

Abstract

Article availability

“Machine Learning-Based Anomaly Detection for ITER's Tokamak Systems Monitor: A Gyrotron Case Study” - IEEE Transactions on Plasma Science

Authors

Abstract

Article availability

“Preliminary machine learning-based calibration strategy for the ITER Tokamak Systems Monitor” - Fusion Engineering and Design

Authors

Abstract

Article availability

“Dimensionality reduction of local structure in glassy binary mixtures” - The Journal of Chemical Physics

Authors

Abstract

Article and data availability

“partycls: A Python package for structural clustering” - The Journal of Open Source Software

Presentation

Related publication

Short example