EGErwan Grelier

Erwan Grelier

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Erwan Grelier has a total of 142 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Cross‐device and cross‐modal arc detection: transfer learning between EAST visible and WEST infrared diagnostics

    by Zhongfang Guan, Erwan Grelier, Bin Zhang, Sebastien Vives, Baoguo Wang, Victor Moncada, Léo Dubus, Jinping Qian, Kedong Li, Jian Liu, et al.

    This work presents an engineering-oriented data processing and model training pipeline for Experimental Advanced Superconducting Tokamak (EAST) visible and (W) Environment in Steady-state Tokamak (WEST) infrared data, enabling a systematic study of cross-device and cross-modal electric arc detection. A lightweight convolutional neural network is adopted as a cross-device arc detection model, and joint training and transfer learning experiments are conducted on EAST visible (VIS) and WEST infrared (IR) data. Single‐modality baselines are first trained on EAST and WEST, then evaluated zero‐shot on the opposite device, followed by a comparison of frozen‐backbone and fine‐tuned transfer for VIS → IR and IR → VIS. The results show that models trained from scratch on each diagnostic channel already achieve high precision and recall for arc detection, whereas direct zero-shot cross-modal transfer drives the arc recall on the opposite domain almost to zero. In contrast, starting from the single-modality baselines and modestly fine tuning higher convolutional layers and the classification head on the target domain can significantly restore-cross device, cross-modal detection performance, with positive class precision and recall approaching or even surpassing the single-domain baselines. Class activation map analysis further indicates that, under fine-tuned and joint training configurations, the models consistently focus on the antenna, the arc, and its surrounding scattering region, supporting physically meaningful explanations and suggesting that cross-device, cross-modal training can capture arc related structures that generalize across devices while providing a lightweight, scalable route to reusing imaging-based diagnostic knowledge with reduced annotation and retuning effort. Given that machine protection is safety-critical and that imaging diagnostics are never identical across devices, the transferability demonstrated in this work is directly relevant to equipping next-generation tokamaks such as ITER and BEST with a reliable arc-detection capability before a large native training set becomes available. On a newly commissioned device, arc events are rare, sporadic and safety-relevant, so deferring detection until a sufficiently large native training corpus has accumulated would leave the machine unprotected during the early operational phase, when limited operating experience makes the associated risks most difficult to characterize; cross-machine and cross-modal transfer instead enables a new device to inherit a usable detection capability from the outset. Accordingly, the present work is intended not merely as a benchmark of cross-domain accuracy, but as evidence that scalable, transferable arc monitoring can constitute a machine-protection layer of future fusion reactors.

  • Open Access

    WEST advanced wall protection achievements toward long pulse operation

    by R. Mitteau, M.H. Aumeunier, L. Dubus, J. Gerardin, V. Gorse, E. Grelier, V. Moncada, S. Vives, X. Litaudon, M. Jakubowski, et al.

    Long pulse operation in magnetic fusion devices requires well controlled plasma power exhaust to the divertor & wall, and avoidance of wall hot spots that could evolve in wall damage. At WEST, 10 major plasma facing components are monitored using 10 series of temperature/power indicators, based on multiple diagnostic systems, among which the infrared viewing system is especially relevant. These indicators span from the most basic ones (temperatures, power and energy from deterministic models) to advanced processes using artificial intelligence acquired through machine learning. Some advanced processes do operate in real time, and feedback on power actuators through the plasma control system, providing active control toward remaining within the safe operational domain. Other advanced processes intervene as forensic tools post discharge to identify possible dangerous situations regarding the power loading to the wall, so that the discharge run plan is adjusted to avoid running into aggravating wall events. No critical wall power event happened during the campaigns C9 to C11 (2024–2025), totalling about 13 h of plasma, that would have affected the campaign plan. While it cannot be demonstrated that the active & intelligent wall protection enabled the new plasma duration record of 1337 s, the wall protection system as a whole plausibly contributed to obtaining these record durations by preventing wall hot spots to become critical during the campaigns.


  • Learning with tree-based tensor formats : Application to uncertainty quantification in vibroacoustics

    by Erwan Grelier

    Apprentissage statistique avec des formats de tenseurs basés sur des arbres : Application à la quantification d’incertitudes en vibroacoustique: De nombreux problèmes nécessitent l’évaluation de modèles paramétrés complexes pour de nombreuses valeurs des paramètres, en particulier pour la quantification d’incertitudes. Quand le modèle est coûteux à évaluer, il est souvent approximé par un autre modèle, moins coûteux à évaluer. L’objectif de cette thèse est de développer des méthodes d’apprentissage statistique utilisant des classes de fonctions au format de tenseurs basés sur des arbres pour l’approximation de fonctions en haute dimension, pour l’apprentissage supervisé et non supervisé. Ces classes de fonctions, qui sont structurés par rangs et paramétrées par un réseau de tenseurs de faible ordre à structure d’arbre, peuvent être interprétées comme des réseaux de neurones profonds avec une architecture et des fonctions d’activation particulières. L’approximation est obtenue par minimisation du risque empirique sur l’ensemble des fonctions au format de tenseurs basés sur des arbres. Pour l’approximation de fonctions en haute dimension, ou quand peu d’information sur la fonction est disponible, la classe de fonctions doit être soigneusement choisie. Nous proposons des algorithmes d’apprentissage stables qui adaptent l’arbre et les rangs et sélectionnent le modèle en s’appuyant sur des estimateurs de validation croisée. De plus, certaines fonctions peuvent n’exhiber une structure de faible rang qu’après un changement de variables adapté. Dans de tels cas, nous proposons des algorithmes d’apprentissage adaptatifs avec des classes de fonctions combinant formats de tenseurs basés sur des arbres et changements de variables. Les algorithmes proposés sont appliqués à la quantification d’incertitudes en vibroacoustique. Cette thèse est incluse dans le Joint Laboratory of Marine Technology entre Naval Group, Centrale Nantes et l’Université de Nantes, et dans le projet Eval-PI.


  • Deep Learning for Intelligent Monitoring of the WEST Tokamak First Wall Using Infrared Imaging: An Overview

    by Erwan Grelier, Valentin Gorse, Raphael Mitteau, Victor Moncada, Leo Dubus, Sebastien Vives, Julien Marot

    First wall protection is a critical challenge in fusion machine operation, requiring fast and precise diagnostics. Infrared imaging in long-pulse devices, such as the WEST tokamak, generates vast amounts of thermal video data, making both real-time and postpulse analysis demanding for human operators. Traditional real-time protection has relied on thresholding and conventional image processing, which, while robust, require predefined regions of interest (ROIs) and lack adaptability to new hot spots or camera displacements. Deep learning methods have therefore emerged as promising alternatives, offering greater flexibility and improved detection capabilities. This work presents the comprehensive framework developed over several years for monitoring the WEST’s first wall using infrared imaging. The tools fall into two main categories: active real-time systems integrated with plasma control and decision-support tools for postpulse analysis. In real time, the developed methods include a CPU-optimized model that detects electric arcs on the grill of the heating antennae, which has been successfully deployed with the plasma control system (PCS) to regulate the injected power during the C10 and C11 experimental campaigns (October 2024–April 2025). This model achieved 98% accuracy on C10 data, with only a single false positive occurring after a disruption. Another key component is the GPU-based thermal event detection model, which detects, tracks, and characterizes thermal events in real time, overcoming the limitations of fixed ROIs and enabling the tracking of emerging hot spots. Postpulse decision-support tools are equally important. One example is a detection tool designed to identify and track moving tungsten particles in infrared data, supporting studies of correlations between these particles and plasma disturbances. Additionally, a multimodal large language model (LLM) has been developed to automate pulse summary analysis. This LLM has been actively used by infrared experts between pulses during campaigns. Expert evaluations indicate that over 62% of the model’s outputs were satisfactory. Overall, integrating deep learning into the infrared diagnostic workflow significantly enhances first wall monitoring and protection for fusion devices.


  • Open Access

    Protection of the plasma facing components in the WEST tokamak, progress and development in view of ITER

    by M. Houry, M-H. Aumeunier, Y. Corre, X. Courtois, R. Mitteau, Th. Loarer, L. Dubus, E. Gauthier, J. Gerardin, V. Gorse, et al.

    The protection of the inner walls of magnetic confinement fusion research reactors is a crucial issue, particularly in this research context where plasma scenarios are explored to reach high power performance, thus leading to high temperature of the plasma facing components (PFCs), possibly close to their technological limitation. The aim is to protect the PFCs from damage during experimental campaigns, whilst enabling the expansion of the operational domain toward long duration and high power performances. With nearly 35 years of operation of Tore Supra and now WEST, CEA’s magnetic fusion research institute, the IRFM, has deployed a system combining thermal instrumentation, modeling of the heat transfer and photonic emission, signal processing and understanding of the physics of plasma-wall interaction to provide an optimized and controlled protection of the PFCs in metallic environment (with tungsten, bore, copper and stainless steel materials). In this context, the WEST Tokamak is a relevant Fusion facility capable of combining steady-state (Vloop ∼ 0) 1000 s long pulse operation with up to 6 MW m−2 on its divertor together with an advanced first wall protection system that could be deployed on ITER and future Fusion machines. The paper describes the wall protection system installed on WEST, highlighting its particular features and recent results.


  • Open Access

    EUROfusion contributions to ITER nuclear operation

    by X. Litaudon, U. Fantz, R. Villari, V. Toigo, M.-H. Aumeunier, J.-L. Autran, P. Batistoni, E. Belonohy, S. Bradnam, M. Cecchetto, et al.

    ITER is of key importance in the European fusion roadmap as it aims to prove the scientific and technological feasibility of fusion as a future energy source. The EUROfusion consortium of labs within Europe is contributing to the preparation of ITER scientific exploitation and operation and aspires to exploit ITER outcomes in view of DEMO. The paper provides an overview of the major progress obtained recently, carried out in the frame of the new (initiated in 2021) EUROfusion work-package called ‘Preparation of ITER Operation’ (PrIO). The overview paper is directly supported by the eleven EUROfusion PrIO contributions given at the 29th Fusion Energy Conference (16–21 October 2023) London, UK [www.iaea.org/events/fec2023]. The paper covers the following topics: (i) development and validation of tools in support to ITER operation (plasma breakdown/burn-through with evolving plasma volume, new infra-red synthetic diagnostic for off-line analysis and wall monitoring using Artificial Intelligence techniques, synthetic diagnostics development, development and exploitation of multi-machine databases); (ii) R&D for the radio-frequency ITER neutral beam sources leading to long duration of negative deuterium/hydrogen ions current extraction at ELISE and participation in the neutral beam test facility with progress on the ITER source SPIDER, and, the commissioning of the 1 MV high voltage accelerator (MITICA) with lessons learned for ITER; (iii) validation of neutronic tools for ITER nuclear operation following the second JET deuterium–tritium experimental campaigns carried out in 2021 and in 2023 (neutron streaming and shutdown dose rate calculation, water activation and activated corrosion products with advanced fluid dynamic simulation; irradiation of several materials under 14.1 MeV neutron flux etc).


  • Detection of Thermal Events by Semi-Supervised Learning for Tokamak First Wall Safety

    by Christian Staron, Hervé Le Borgne, Raphaël Mitteau, Erwan Grelier, Nicolas Allezard

    This paper explores a semi-supervised object detection approach to detect thermal events on the internal wall of tokamaks. A huge amount of data is produced during an experimental campaign by the infrared (IR) viewing systems used to monitor the inner thermal shields during machine operation. The amount of data to be processed and analyzed is such that protecting the first wall is an overwhelming job. Automatizing this job with artificial intelligence (AI) is an attractive solution, but AI requires large labelled datasets that are not readily available for tokamak walls. Semi-supervised learning (SSL) is a possible solution to being able to train deep learning models with a small amount of labelled data and a large amount of unlabelled data. SSL is explored as a possible tool to rapidly adapt a model trained on an experimental campaign A of tokamak WEST to a new experimental campaign B by using labelled data from campaign A , a little labelled data from campaign B and a lot of unlabelled data from campaign B . Model performance is evaluated on two labelled datasets and two methods including semi-supervised learning. Semi-supervised learning increased the mAP metric by over six percentage points on the first smaller-scale database and over four percentage points on the second larger-scale dataset depending on the method employed.