MOP2WA —  WG-A   (05-Mar-18   16:00—18:00)
Chair: L. Giannessi, Elettra-Sincrotrone Trieste S.C.p.A., Basovizza, Italy
Paper Title Page
MOP2WA01
Report on the First ICFA Mini-Workshop on Machine Learning for Particle Accelerators  
 
  • D.F. Ratner, X. Huang, C.E. Mayes, T.O. Raubenheimer
    SLAC, Menlo Park, California, USA
 
  Machine learning techniques are playing a growing role in operation of particle accelerators around the world. To facilitate collaborations and discussions at this early stage, we have organized the first ICFA mini-workshop on machine learning applications at SLAC National Accelerator Laboratory: https://conf.slac.stanford.edu/icfa-ml-2018/. Here we report on the outcome of the meeting.  
slides icon Slides MOP2WA01 [4.130 MB]  
Export • reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)  
 
MOP2WA02
Automated Optimization of Machine Parameters at the European XFEL  
 
  • S.I. Tomin
    XFEL. EU, Hamburg, Germany
  • L. Froehlichpresenter, M. Scholz
    DESY, Hamburg, Germany
 
  In today's single-pass free-electron lasers, a lot of time and effort is spent on the manual optimization of machine parameters with the goal of improving the photon beam. Identifying and automating these tuning procedures holds the promise of faster set-up times and higher machine efficiency. This talk gives an overview of various optimization routines used for this purpose at the European XFEL. Our first attempts at exploiting machine learning algorithms are also included.  
slides icon Slides MOP2WA02 [6.927 MB]  
Export • reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)  
 
MOP2WA03
The Feasibility of Neuron Network-based Beam-based Alignment  
 
  • L. Zeng
    SINAP, Shanghai, People's Republic of China
 
  Artificial neuron networks which inspired by biological neural networks have been widely used in various domains, including computer vision, machine translation, pattern/speech recognition, medical diagnosis and so on, due to its overwhelming superiorities. But it's not until recently that intelligent algorithms have been introduced in light source field. M.P. Ehrlichman, Yi Jiao, Juhao Wu and A. Sanchez-Gonzalez did some work in this respect and got commendable results. Considering Shanghai X-ray Free-Electron Laser (SXFEL) conditions, we are urgent to improve the FEL performance, and fundamental technique turns out to be beam-based alignment. But it's difficult to implement this means in SXFEL due to the low electron beam energy resulting in uncontrollable orbit disturbance. Thus, a new method which is suitable for SXFEL is an eager desire. Here, we discuss the feasibility of neuron network-based beam-based alignment, and try to take it into reality in SXFEL. In fact, Hornik have proved, as early as 1989, that a single hidden layer feedforward networks can approximate any measurable function arbitrarily well, which provides the theoretical evidence to our suggestion.  
slides icon Slides MOP2WA03 [5.439 MB]  
Export • reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)  
 
MOP2WA04
FEL Optimization Through BBA with Undulator Spectrum Analysis and Undulator Optics Matching  
 
  • H.-S. Kang, H. Yangpresenter
    PAL, Pohang, Kyungbuk, Republic of Korea
 
  The use of electron-beam-based alignment incorporating undulator radiation spectrum analysis has allowed reliable operation of PAL-XFEL with unprecedented stability in terms of orbit, energy, and timing. The hard XFEL operation procedure established using e-BBA and incorporating undulator radiation spectrum analysis proved to be highly reliable and robust, and essential for the variable-gap undulators. The undulator optics matching using the wire-scanner based emittance measurement is extensively used to maximize the FEL intensity even at higher photon energy up to 14.5 keV.  
slides icon Slides MOP2WA04 [17.465 MB]  
Export • reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)