Parts | Days | Selection | Search | Updates | Downloads | Help

AKBP: Arbeitskreis Beschleunigerphysik

AKBP 8: Advanced IT Tools

AKBP 8.4: Talk

Wednesday, March 22, 2023, 16:30–16:45, CHE/0183

Sensitivity Analysis and Online Surrogate Construction at the S-DALINAC Using Polynomial Chaos and Neural Networks — •Dominic Schneider, Michaela Arnold, Jonny Birkhan, Ruben Grewe, Norbert Pietralla, and Felix Schliessmann — Institut für Kernphysik, Technische Universität Darmstadt, Darmstadt, Germany

Particle accelerators are complex systems that coincide with their ideal design within the tolerances of its large number of technical components. Quantitative understanding of the beam dynamics and the analysis of their sensitivity to various components are challenging tasks. Machine learning methods provide the potential for the optimized operation of particle accelerators. In this contribution, the application of so-called surrogate models to the electron accelerator S-DALINAC will be discussed. This machine learning technique gives access to predict future behavior and an extensive set of characteristics that can be extracted by analyzing the trained model. The talk will include the presentation of a series of measurements performed in the injector section of the S-DALINAC to study the behavior of beam-influencing elements. Surrogate models, constructed and based on the acquired data, are being evaluated to reveal the behavior of these elements. Based on the information obtained, optimizations of the alignment of magnets as well as the beam dynamics simulations at the S-DALINAC will be discussed.

Supported by the State of Hesse and the Research Cluster ELEMENTS (Project-ID 500/10.006).