Page tree

Present: Lars Froehlich Olaf Hensler Raimund Kammering @Torsten Limberg Sascha Meykopff Matthias Scholz @Sergey Tomin Josef Wilgen

Discussion of the task list

We adjust priorities and discuss new projects (see global task list and action items below).

Machine learning workshop (PSI)

Sergey gives another brief report on the machine learning workshop:

Talk can be found here https://indico.psi.ch/event/6698/timetable/#20190227

I selected a few but the others are also very interesting: 
1. Safe optimization with GP by Johannes (PSI)  presentation.pdf 
2. Joe (SLAC) had talk about fast optimization (about 4 times faster then simplex) with Bayesian optimization with GP if you pre-calculate from physical model correlation between devices e.g. quad scan versus average beam size in undulator Duris_ICFA-PSI-ML_BayesOpt_talk.pptx 
3. Xiaobiao reported that they analysed an injection history and found significant correlation between ground temperature and ideal orbit for injection  injection_data_modeling_SPEAR3_xhuang.pdf 

A side effect of a tour to the SwissFEL control room: SwissFEL Operation Messages seems interesting for communication with instruments:

5-way meeting (SLAC)

Sergey highlights a talk from PAL that is quite interesting from an operations point of view:
https://conf.slac.stanford.edu/hxrcollab-2019/sites/hxrcollab-2019.conf.slac.stanford.edu/files/5-Way-2019_VGU_Wavelength_Tuning_of_PAL-XFEL.pdf

Magnet ML server

The XFEL TLD kickers are now integrated into the magnet ML server. There was a meeting on the future of magnet controls at FLASH between Piotr Bartkiewicz, Lars Froehlich, Christian Gruen, @Matthias Vogt, and Marcus Walla in which necessary features were discussed. Major points: The magnet ML server will need support for cycling groups, and the power supply server will need a way to protect interlocked magnets.

Undulator phase shifters

The wavelength control server now correctly calculates the phase shift between cells for all three undulator beamlines.

BAMs

We need some server to calculate arrival time differences between two BAMs. The current state of the BAM servers is unclear.

Action items

  • Bunch pattern server: Extend beam limit functionality to allow discrimination of SA1/SA3 bunches Lars Froehlich  
  • Magnet ML server: Implement cycling groups Lars Froehlich  
  • Orbit feedbacks: Make usable with alternating bunch patterns Raimund Kammering  
  • Gain curve measurement: Develop tool @Lyuba  
  • IntelliPhase: Implement parasitic scan of L3 stations vs. energy manager  
  • Investigate state of BAM servers and push implementation of arrival time difference