By launching 15 new research projects, EUROfusion is engaging data science experts across Europe to apply Artificial Intelligence and Machine Learning techniques to fusion energy. These projects will leverage the world's largest and most diverse dataset of fusion experiments to identify optimal methods for understanding and controlling the fusion process, in order to develop optimal methods for understanding and controlling the hydrogen isotope fusion process, thus accelerating the path to its practical application in the energy industry.

New technologies in Polish research on nuclear Fusion

One of the projects was awarded to a research team from the Institute of Plasma Physics and Laser Microfusion (IPPLM), which is involved in the development of LIBS (Laser Induced Breakdown Spectroscopy) diagnostics in the context of nuclear fusion research. Researchers, who previously participated in key LIBS experiments as an application of a remotely operated diagnostic head on an FTU tokamak in Italy, will now focus on implementing machine learning methods for mass processing of spectroscopic data. Their goal is to use models based on convolutional networks, which are great for image analysis, for new challenges in spectral diagnostics. Although spectroscopic data require specific pre-processing methods and network architecture, the main author of the project, Dr. Paweł Gąsior from the IPPLM, emphasizes that the potential of artificial intelligence can bring similar breakthrough results to those observed in other fields, such as image recognition. In addition to the IPPLM team, researchers from CU (Slovakia) and FZJ (Germany) are also involved in the research. The first test of the novel approach will be conducted using data from the ongoing LIBS for JET experiment on the JET tokamak in Culham, UK, opening new perspectives for the application of LIBS technology in nuclear fusion research.

Fusion energy promises to deliver safe, sustainable, and low-carbon baseload power, complementing other clean energy sources like solar and wind. To achieve this, we need to address complex physics and engineering challenges, including understanding the collective movements of charged particles in magnetic fields, mitigating disruption events, analysing material erosion effects, and processing data rapidly enough for use in control loops. Artificial Intelligence and Machine Learning offer tools which allow for a substantial progress in all these seemingly diverse research areas.

Dr. Paweł Gąsior emphasizes that the use of artificial intelligence can bring breakthrough results: "Machine learning, particularly convolutional neural networks, has demonstrated remarkable proficiency in recognizing patterns within large datasets. Consequently, they can be significantly beneficial when handling spectral data, which, despite being sensitive to experimental conditions, still retains information too deeply embedded for traditional data processing methods."

Artificial intelligence will accelerate progress in research

"With new research projects on Artificial Intelligence and Machine Learning, EUROfusion aims to accelerate progress towards fusion energy and support the ongoing efforts in its work packages", explains Sara Moradi of the EUROfusion Programme Management Unit. "Machine learning and Artificial Intelligence are powerful tools for extracting insight from data, uncovering patterns and suggest control schemes that are too computationally intensive to identify with traditional computer models."

EUROfusion’s extensive dataset of fusion experiments spans decades of research, from the earliest fusion machines to the most advanced systems currently in operation. This unparalleled resource positions EUROfusion uniquely to drive forward Artificial Intelligence applications in fusion research.

Fusion is a great sandbox for Artificial Intelligence and Machine Learning, agrees José Vicente (University of Lisbon), the principal investigator of one of the fifteen projects. "As a very complex system, it has many open questions. We can already address those with today's large amounts of experimental data and realistic numerical simulations of the key physics, but not all of them — that is the gap that Artificial Intelligence may help close."

The 15 projects will receive a total amount of €2.659 million, of which half is provided by collaborative co-funding from the researchers' home institutes and half from EUROfusion. The research projects will run for a period of two years.

Projects supported by EUROfusion, including the one implemented by the team from the IPPLM, underscore the potential of Artificial Intelligence and Machine Learning to address key challenges in fusion research, paving the way for more efficient and effective control strategies as we move closer to realizing fusion energy.

Artist’s impression of Artificial Intelligence research for fusion. Credit: Pexels / GoogleDeepMind

Supported projects:

David Zarzoso (CEA / CNRS, France)
Artificial Intelligence augmented Scrape Off Layer modelling for capturing impact of filaments on transport and PWI in mean field codes simulations.

Feda Almuhisen (CEA / Aix-Marseille Université, France)
Towards Tokamak operations Conversational Artificial Intelligence Interface Using Multimodal Large Language Models

Augusto Pereira (CIEMAT, Spain)
Testing cutting-edge Artificial Intelligence research to increase pattern recognition and image classification in nuclear fusion databases

Sven Wiesen (DIFFER, the Netherlands)
Machine learning accelerated Scrape Off Layer L simulations: SOLPS-NN

Gergő Pokol (EK-CER, Hungary)
Fast inference methods of advanced diagnostics for real-time control

Riccardo Rossi (ENEA / Università di Roma Tor Vergata, Italy)
Artificial Intelligence-assisted Causality Detection and Modelling of Plasma Instabilities for Tokamak Disruption Prediction and Control

Michela Gelfusa (ENEA / Università di Roma Tor Vergata, Italy)
Development of Physics Informed Neural Networks (PINNs) for Modelling and Prediction of Data in the Form of Time Series

Alessandro Pau (EPFL, Switzerland)
Artificial Intelligence-assisted Plasma State Monitoring for Control and Disruption-free Operations in Tokamaks

Pawel Gasior (IPPLM, Poland)
Laser Induced Breakdown Spectrocopy data-processing with Deep Neural Networks and Convolutional Neural Networks for chemical composition quantification in the wall of the next step-fusion reactors

Jose Vicente (IST, Portugal)
Deep Learning for Spectrogram Analysis of Reflectometry Data

Geert Verdoolaege (LPP-ERM-KMS / Ghent University, Belgium)
Identification and confinement scaling of hybrid scenarios across multiple devices

Marcin Jakubowski (IPP, Germany)
Leveraging Generative Artificial Intelligence Models for Thermal Load Control in High-Performance Steady-State Operation of Fusion Devices

Daniel Böckenhoff (IPP, Germany)
Surrogate modelling of ray-tracing and radiation transport code for faster real-time plasma profile inference in a magnetic confinement device

Antti Snicker (VTT, Finland)
Applying Artificial Intelligence/Machine Learning for Neutral Beam Injection ionization and slowing-down simulations using ASCOT/BBNBI

Aaro Järvinen (VTT, Finland)
Machine learning accelerated pedestal Magneto Hydro Dynamics stability simulations

Source: EUROfusion