Abstract:Artificial intelligence (AI) and machine learning (ML) have the potential to significantly improve particle accelerator operations, with applications in diagnostics, control, and modelling. Experimentally testing AI/ML methods before deployment to user facilities remains a challenge. The capacity to swiftly generalise and adapt these algorithms to different operational configurations inside or between facilities remains a difficulty, requiring a combination of model-independent adaptive feedback and classic machine learning technologies. These techniques can also be used to detect, classify, and avoid operational abnormalities that can result in accelerator damage or excessive beam loss during atypical operations. Broadening AI/ML approaches for early identification of a wide variety of accelerator component or subsystem problems is an opportunity. The optimization of a large number of connected accelerators is required in modern accelerator architecture.
|
DOI:
10.17148/IJARCCE.2022.114153
