Speaker
Description
The application of Artificial Intelligence (AI) and Machine Learning (ML) to particle accelerator systems has emerged as an effective strategy for managing complex operations and enhancing performance. At INFN-Legnaro National Laboratories (INFN-LNL), both offline and online AI/ML-driven approaches have been developed to improve beam dynamics, reduce setup times, and increase overall accelerator efficiency.
Offline efforts focus on surrogate modeling of complex facilities such as ANTHEM BNCT, as well as on virtual diagnostics implemented using supervised neural operators. By combining these tools with AI/ML optimization algorithms, new design and commissioning strategies are being explored to further enhance beam quality and operational performance.
In parallel, online real-time optimization strategies based on Bayesian Optimization (BO) has delivered promising results. Notably, at the PIAVE-ALPI superconducting accelerator, the application of BO improved beam transmission up to 85%, a significant increase compared to the typical operational average of 35%. These advances demonstrate the growing impact and future potential of AI/ML technologies in accelerator science and operations.