IPAC2021 - List of Authors (Frey, M.)

Paper	Title	Page
WEPAB203	RFQ Beam Dynamics Optimization Using Machine Learning	3100
	D. Koser, J.M. Conrad, L.H. Waites, D. Winklehner MIT, Cambridge, Massachusetts, USA A. Adelmann, M. Frey, S. Mayani PSI, Villigen PSI, Switzerland
	To efficiently inject a high-current H²⁺ beam into the 60 MeV driver cyclotron for the proposed IsoDAR project in neutrino physics, a novel direct-injection scheme is planned to be implemented using a compact radio-frequency quadrupole (RFQ) as a pre-buncher, being partially inserted into the cyclotron yoke. To optimize the RFQ beam dynamics design, machine learning approaches were investigated for creating a surrogate model of the RFQ. The required sample datasets are generated by standard beam dynamics simulation tools like PARMTEQM and RFQGen or more sophisticated PIC simulations. By reducing the computational complexity of multi-objective optimization problems, surrogate models allow to perform sensitivity studies and an optimization of the crucial RFQ beam output parameters like transmission and emittances. The time to solution might be reduced by up to several orders of magnitude. Here we discuss different methods of surrogate model creation (polynomial chaos expansion and neural networks) and identify present limitations of surrogate model accuracy.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2021-WEPAB203
About •	paper received ※ 20 May 2021 paper accepted ※ 01 July 2021 issue date ※ 30 August 2021
Export •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

Paper

Title

Page

RFQ Beam Dynamics Optimization Using Machine Learning

3100

D. Koser, J.M. Conrad, L.H. Waites, D. Winklehner
MIT, Cambridge, Massachusetts, USA
A. Adelmann, M. Frey, S. Mayani
PSI, Villigen PSI, Switzerland

To efficiently inject a high-current H²⁺ beam into the 60 MeV driver cyclotron for the proposed IsoDAR project in neutrino physics, a novel direct-injection scheme is planned to be implemented using a compact radio-frequency quadrupole (RFQ) as a pre-buncher, being partially inserted into the cyclotron yoke. To optimize the RFQ beam dynamics design, machine learning approaches were investigated for creating a surrogate model of the RFQ. The required sample datasets are generated by standard beam dynamics simulation tools like PARMTEQM and RFQGen or more sophisticated PIC simulations. By reducing the computational complexity of multi-objective optimization problems, surrogate models allow to perform sensitivity studies and an optimization of the crucial RFQ beam output parameters like transmission and emittances. The time to solution might be reduced by up to several orders of magnitude. Here we discuss different methods of surrogate model creation (polynomial chaos expansion and neural networks) and identify present limitations of surrogate model accuracy.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2021-WEPAB203

About •

paper received ※ 20 May 2021 paper accepted ※ 01 July 2021 issue date ※ 30 August 2021

Export •

reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)