Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-Driven Optimisation of Superconducting Magnets at CEA Paris-Saclay

Damien F. G. Minenna, Guillaume Dilasser, Robin Penavaire, Valerio Calvelli, Thibault de Chabannes, Thibault Lecrevisse, Thomas Achard, Jason Le Coz, Christophe Berriaud, Benoît Bolzon, Antomne Caunes, Phillipe Fazilleau, Hélène Felice, Clément Genot, Antoine Guinet, Nikola Jerance, François-Paul Juster, Thibaut Lemercier, Gilles Lenoir, Clément Lorin, Yann Perron, Camille Pucheu-Plante, Étienne Rochepault, Damien Simon, Francesco Stacchi, Michel Segreti, Vincent Trauchessec, Olivier Tuske, Hajar Zgour

Abstract

Superconducting magnets for particle accelerators are particularly challenging to design because they involve a large number of coupled physical phenomena and the management of complex datasets. Artificial Intelligence (AI), including machine learning and advanced optimisation techniques, offers promising approaches to address these challenges and accelerate the design process. This paper presents a new AI-based optimisation and data management platform, and highlights several ongoing applications of AI methods carried out at CEA Paris-Saclay, including multiphysics optimisation using active learning, topology optimisation, holistic modelling of an Electron Cyclotron Resonance (ERC) ion source, and anomaly detection in quench events.

Data-Driven Optimisation of Superconducting Magnets at CEA Paris-Saclay

Abstract

Superconducting magnets for particle accelerators are particularly challenging to design because they involve a large number of coupled physical phenomena and the management of complex datasets. Artificial Intelligence (AI), including machine learning and advanced optimisation techniques, offers promising approaches to address these challenges and accelerate the design process. This paper presents a new AI-based optimisation and data management platform, and highlights several ongoing applications of AI methods carried out at CEA Paris-Saclay, including multiphysics optimisation using active learning, topology optimisation, holistic modelling of an Electron Cyclotron Resonance (ERC) ion source, and anomaly detection in quench events.

Paper Structure

This paper contains 22 sections, 9 figures.

Table of Contents

  1. Introduction
  2. Data Analysis And Management
  3. Project Manager
  4. Databases
  5. Interface With Physics Software
  6. Multiparametric Optimisers
  7. Supervised Learning Techniques
  8. Data Analysis Tools
  9. In-House Modelling
  10. Magnet Optimisation Design
  11. Multiphysics Optimisation
  12. Shim Coils
  13. Topology Optimisation
  14. HTS optimisation
  15. Beyond particle accelerator magnets
  16. ...and 7 more sections

Figures (9)

  • Figure 1: Main modules of the alesia platform.
  • Figure 2: Cross section of the 6.2 m-long SPIN ROTATORS magnet. The colormap represents the magnetic field distribution within the coil. The design was obtained using the alesia platform.
  • Figure 3: Automated workflow of the design of an Nb$_3$Sn solenoid magnet with the new platform alesia.
  • Figure 4: Parity plots with 2131 test points (not used during model training), evaluating a surrogate model predictions. Upper left: integrated magnetic field along the beam propagation axis (reflects beam propagation needs). Upper right: peak magnetic field in the coil (up to 18 T with an Nb$_3$Sn conductor). Lower left: Maximum von Mises stress in the coil windings (relevant to mechanical integrity under Lorentz forces and cooling down). Lower right: Maximum hotspot temperature during a quench event.
  • Figure 5: Example of shim coils (racetrack and solenoid geometries). For clarity, the coil positions are shifted along the longitudinal axis. The detector magnet is not represented. The system is composed, from left to right, by a dipole (with 8 loops), a quadrupole (with 6 loops), an hexadecapole (with 3 loops) and 6 solenoids.
  • ...and 4 more figures