Machine learning applications for beam diagnostics and automated tuning at CLEAR

Gilardi, Antonio; Bergamaschi, Michele; Burger, Stephane; Corsini, Roberto; Farabolini, Wilfrid; Franek, Ondřej; Gamba, Davide; Leygonie, Johan Pierre; Malyzhenkov, Alexander; Moore, Ciara Marie; Roncarolo, Federico; Simonsson, Sara Jenny; Tangari, Giacomo; Trad, George; Qiyuan, Xu; Wang, Sabrina

doi:10.18429/JACoW-LINAC2026-MOPO010

Journals of Accelerator Conferences Website (JACoW)

JACoW is a publisher in Geneva, Switzerland that publishes the proceedings of accelerator conferences held around the world by an international collaboration of editors.

URL	https://doi.org/10.18429/JACoW-LINAC2026-MOPO010
Title	Machine learning applications for beam diagnostics and automated tuning at CLEAR
Authors	A. Gilardi, M. Bergamaschi, S. Burger, R. Corsini, W. Farabolini, D. Gamba, J. Leygonie, F. Roncarolo, G. Trad European Organization for Nuclear Research O. Franek Czech Technical University in Prague, European Organization for Nuclear Research A. Malyzhenkov European Organization for Nuclear Research, Swiss International Institute C. Moore University of Exeter, European Organization for Nuclear Research S. Simonsson Lund University, European Organization for Nuclear Research G. Tangari Sapienza University of Rome, European Organization for Nuclear Research X. Qiyuan University of Liverpool, European Organization for Nuclear Research S. Wang European Organization for Nuclear Research, University of Oxford
Abstract	Machine learning (ML) is increasingly recognized as a fundamental tool in modern accelerator physics, offering new capabilities to complement and extend traditional analytical approaches. A structured programme of ML studies has been initiated at CLEAR (CERN Linear Electron Accelerator for Research), covering three principal directions: beam diagnostics, charge forecasting, and automated tuning via reinforcement learning. Deep learning models were explored to reconstruct transverse beam profiles from indirect measurements, offering a pathway toward radiation-resistant diagnostics. Classical and deep learning approaches were benchmarked for charge prediction, demonstrating the potential for accurate real-time beam characterization. A reinforcement learning framework was interfaced with the CERN control infrastructure, enabling proof-of-principle autonomous beam steering through iterative magnet corrections. These results demonstrate the feasibility of integrating ML into routine CLEAR operations and position the facility as an attractive testbed for the development and validation of machine learning techniques for particle accelerators.
Paper	download: MOPO010.pdf
Cite	BibTeX, LaTeX, Text/Word, RIS, EndNote
Conference	33rd International Linear Accelerator Conference (LINAC'26)
Series
Location	Daejeon, Korea
Date	16-21 Aug 2026
Publisher	JACoW Publishing, Geneva, Switzerland
Editorial Board	Dong-Eon Kim (Pohang Accelerator Laboratory), Jangwon Kwon (Institute for Basic Science), Yujung Ahn (Institute for Basic Science), Jae-yu Lee (Pohang Accelerator Laboratory), Lin Bian (Institute of High Energy Physics, Chinese Academy of Sciences), Thakonwat Chanwattana (Synchrotron Light Research Institute), Yung-Sen Cheng (National Synchrotron Radiation Research Center), Manuel J. Gutiérrez (IFMIF-DONES España), Maurizio Montis (Istituto Nazionale di Fisica Nucleare), Volker RW Schaa (GSI Helmholtzzentrum für Schwerionenforschung GmbH), Akito Uchiyama (RIKEN Nishina Center)
Online ISBN	978-3-95450-275-2
Online ISSN	2226-0366
Received	14 August 2026
Revised	14 August 2026
Accepted	16 August 2026
Issued	23 October 2026
DOI	10.18429/JACoW-LINAC2026-MOPO010
Pages	49-52
Copyright	Published by JACoW Publishing under the terms of the Creative Commons Attribution 4.0 license. Any further distribution of this work must maintain attribution to the author(s), the published article's title, publisher, and DOI.