% IMPORTANT: The following is UTF-8 encoded.  This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.

@INPROCEEDINGS{Tuennermann:599237,
      author       = {Tuennermann, Henrik and Pressacco, Federico and Tavakol,
                      Hamed and Hartl, Ingmar},
      title        = {{M}achine {L}earning {E}nhanced {B}eam {P}ointing
                      {S}tabilization for the {P}ump-{P}robe {L}aser at the
                      {FLASH} {FEL} {F}acility},
      reportid     = {PUBDB-2023-07244},
      year         = {2023},
      abstract     = {Pump-probe lasers for FELs must provide stable pulse
                      energy, timing, and beam position. Here, we show active
                      stabilization of beam pointing fluctuations using a
                      combination of classic control, artificial intelligence, and
                      machine learning techniques. As our laser system operates in
                      10 Hz burst mode, fast feedback is not possible. Therefore,
                      we have to utilize the available information as efficiently
                      as possible. Beam pointing fluctuations of laser beams can
                      be described by 4 parameters – as the actuators (motorized
                      mirrors) are not orthogonal we need a model to calculate the
                      required actuator movements. As effects such as motor
                      acceleration are not easy to capture in a physical model, we
                      use an automated data-driven approach. The measurement of
                      the beam position is noisy, so we use a Kalman-Filter, which
                      also integrates our feedback actions to smooth the output.
                      Finally, we use an integrating controller to control the
                      beam. The final transport of the beam to the pump-probe
                      experiment introduces additional drifts, but during user
                      operation, the beam position at the interaction point cannot
                      be measured. We, therefore, measure correlated properties
                      such as temperature, humidity, and air pressure and trained
                      a machine learning model to predict its location.
                      Integrating this model in a feed-forward loop could improve
                      the RMS error of the beam position by $63\%$ in the x-axis
                      and $8\%$ in the y-axis.},
      month         = {Jan},
      date          = {2023-01-28},
      organization  = {Photonics West / LASE 2023, San
                       Francisco (USA), 28 Jan 2023 - 2 Feb
                       2023},
      cin          = {FS-LA},
      cid          = {I:(DE-H253)FS-LA-20130416},
      pnm          = {6G2 - FLASH (DESY) (POF4-6G2) / InternLabs-0011 - HIR3X -
                      Helmholtz International Laboratory on Reliability,
                      Repetition, Results at the most advanced X-ray Sources
                      $(2020_InternLabs-0011)$},
      pid          = {G:(DE-HGF)POF4-6G2 / $G:(DE-HGF)2020_InternLabs-0011$},
      experiment   = {EXP:(DE-H253)F-PG1-20150101 /
                      EXP:(DE-H253)FLASH2020p-20221201},
      typ          = {PUB:(DE-HGF)6},
      url          = {https://bib-pubdb1.desy.de/record/599237},
}