Author: 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)