IPAC2019 - List of Authors (Bai, M.)

Paper	Title	Page
MOPGW026	Transfer Line Optics Design Using Machine Learning Techniques	139
	D.M. Vilsmeier IAP, Frankfurt am Main, Germany M. Bai, M. Sapinski GSI, Darmstadt, Germany
	Optimization of transfer line optics is essential for delivering high quality beams to the experimental areas. This type of optimization is usually done by hand and relies on the experience of operators. The nature of this task is repetitive though highly complex. Besides optimizing the beam quality at the experiments this task is often accompanied by secondary objectives or requirements such as keeping the beam losses below an acceptable threshold. In the past years Deep Learning algorithms have experienced a rapid development and gave rise to various advanced software implementations which allow for straightforward usage of corresponding techniques, such as automatic differentiation and gradient backpropagation. We investigate the applicability and performance of these techniques in the field of transfer line optics optimization, specifically for the HADES beamline at GSI, in form of gradient-based differentiable simulators. We test our setup on results obtained from MADX simulations and compare our findings to different gradient-free optimization methods. Successfully employing such methods relieves operators from the tedious optimization tasks.
DOI •	reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2019-MOPGW026
About •	paper received ※ 15 May 2019 paper accepted ※ 20 May 2019 issue date ※ 21 June 2019
Export •	reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)

Paper

Title

Page

Transfer Line Optics Design Using Machine Learning Techniques

139

D.M. Vilsmeier
IAP, Frankfurt am Main, Germany
M. Bai, M. Sapinski
GSI, Darmstadt, Germany

Optimization of transfer line optics is essential for delivering high quality beams to the experimental areas. This type of optimization is usually done by hand and relies on the experience of operators. The nature of this task is repetitive though highly complex. Besides optimizing the beam quality at the experiments this task is often accompanied by secondary objectives or requirements such as keeping the beam losses below an acceptable threshold. In the past years Deep Learning algorithms have experienced a rapid development and gave rise to various advanced software implementations which allow for straightforward usage of corresponding techniques, such as automatic differentiation and gradient backpropagation. We investigate the applicability and performance of these techniques in the field of transfer line optics optimization, specifically for the HADES beamline at GSI, in form of gradient-based differentiable simulators. We test our setup on results obtained from MADX simulations and compare our findings to different gradient-free optimization methods. Successfully employing such methods relieves operators from the tedious optimization tasks.

DOI •

reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2019-MOPGW026

About •

paper received ※ 15 May 2019 paper accepted ※ 20 May 2019 issue date ※ 21 June 2019

Export •

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