Explainable observers and interpretable AI for superconducting accelerators and radioactive isotope identification
GANIL’s SPIRAL1 and SPIRAL2 facilities produce complex data that remain hard to interpret. SPIRAL2 faces instabilities in its superconducting cavities, while SPIRAL1 requires reliable isotope identification under noisy conditions.
This PhD will develop observer-based interpretable AI, combining physics models and machine learning to detect, explain, and predict anomalies. By embedding causal reasoning and explainability tools such as SHAP and LIME, it aims to improve the reliability and transparency of accelerator operations.
Beam dynamics for a multi-stage laser-plasma accelerator
Laser–plasma wakefield accelerators (LWFAs) can provide accelerating gradients exceeding 100 GV/m, providing a pathway to reduce the size and cost of future high-energy accelerators for applications in synchrotron radiation, free-electron lasers, and emerging medical and industrial uses.
Scaling this technology to higher beam energies and charges requires both technological maturity and innovative acceleration schemes. Multi-stage configurations — connecting several plasma acceleration stages — offer key advantages: increasing beam energy beyond single-cell limits and enhancing total charge and/or repetition rate. These systems aim to overcome single-stage limitations while maintaining or improving beam quality at higher energies.
Designing an accelerator delivering stable, reproducible, high-quality beams requires comprehensive understanding of plasma acceleration physics and beam transport between successive stages.
Building on expertise at CEA Paris–Saclay's DACM, this PhD will focus on physical and numerical studies to propose a fully integrated multi-stage LWFA design, with particular attention to optimizing all components — plasma accelerating section and transport lines — to preserve beam quality in terms of transverse size, divergence, emittance, and energy spread.