Paper | Title | Page |
---|---|---|
WEPNEC07 |
Comparison of Two Pareto Optimization Tools Using OPAL and ASTRA for a Dedicated BERLinPro Injector Optimization. | |
|
||
BERLinPro is an Energy Recovery Linac (ERL) Project, currently being set up at the HZB. The accelerator consists of an injector part (SRF photo injector and booster section) and the ERL part (LINAC and recirculator section). Until realization of the final ERL setup further beam applications are under development using only the injector part of BERLinPro, like ultrafast electron diffraction (UED) or tomography experiments. For those cases a dedicated beam optic for the complete injector is required that differ from the standard ERL optics. Especially for UED experiments, an extreme short electron bunch and a tiny transverse emittance are needed. For the optimization of the injector two multi-parameter Pareto optimization tools were used. On the one hand, ASTRA with an external MATLAB optimizer, on the other hand, OPAL** with its new internal optimization tool. In this paper we will present both generic methods and compare their results. | ||
Export • | reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml) | |
WECOYBS04 | Commissioning of theBERLinPro Diagnostics Line using Machine Learning Techniques | 123 |
|
||
Funding: Work supported by German Bundesministerium für Bildung und Forschung, Land Berlin and grants of Helmholtz Association BERLinPro is an Energy Recovery Linac (ERL) project currently being set up at HZB, Berlin. Commissioning is planned for early 2020. HZB triggered and supported the development of release 2.0 of the particle tracking code OPAL, that is now also applicable to ERLs. OPAL is set up as an open source, highly parallel tracking code for large accelerator systems and many particles. Thus, it is idially suited to serve attempts of applying machine learning approaches to beam dynamics, as demonstrated in [1]. OPAL is used to calculate hundreds of randomized machines close to the commissioning optics of BERLinPro. This data base will be used to train a neural network, to establish a surrogate model of BERLinPro, much faster than any physical model including particle tracking. First steps, like the setup of the sampler and a sensitivity analysis of the resulting data are presented. The ultimate goal of this work is to use machine learning techniques during the commissioning of BERLinPro. Future steps are outlined. [1] A. Edelen, A. Adelmann, N. Neveu, Y. Huber, M. Frey, ’Machine Learning to enable orders of magnitude speedup in multi-objective optimization of particle accelerator systems’ |
||
DOI • | reference for this paper ※ https://doi.org/10.18429/JACoW-ERL2019-WECOYBS04 | |
About • | paper received ※ 30 October 2019 paper accepted ※ 07 November 2019 issue date ※ 24 June 2020 | |
Export • | reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml) | |