Meeting Organization:

Ideas
WhoApplicationMethod




Cavity Piezo tuningBayesian Optimization
Drift Compensation with Beam-Based Feedbacks

problem: distributed feedback loops in a coupled system. current implementation is too simple, static and prone to errors because it does not account for dynamic coupling between loops.

1st approach:
model based learning algorithm

link in confluence:
https://confluence.desy.de/x/Eca7EQ
Electron Bunch Trajectory Optmization & Feedback

problem: current implementation is not suited for a dynamic system and uses a too simple error estimation. 

ideas:
RL algorithm (beam optics model could be included)
or
safe bayesian optimization

link in confluence:
https://confluence.desy.de/x/Dca7EQ
Simulation and control of charged particle beam dynamics using Polynomial Neural Networks (PNN)

idea: Implementation of the target NN on FPGA

Main quiestion: How to convert PNN to CNN?


Fast linear particle simulation for machine learning applicationsCheetah Python packagehttps://github.com/desy-ml/cheetah
Reinforcement learning for particle accelerator optimisation (transverse beam parameter tuning at ARES)Reinforcement learning, domain randomisation, differential actions, logarithmic rewards, TD3, PPOhttps://proceedings.mlr.press/v162/kaiser22a.html
Virtual LPS diagnostics at EuXFEL with adaptive resolution Supervised learning, neural networks
Tuning of the LbSync System and Cavity Piezo TuningMulti-Information Source Bayesian Optimization
Data collection in DOOCSDxMAF python toolhttps://gitlab.desy.de/msk-ipc/dxmaf
Arne Gruenhagen Laser Oscillator Fault detection and DiagnosisFeature Extraction and Fault analysis
Quench detection at the EuXFELAnomaly detection, Residuals, Clustering








Schedule

WhenWhoTopicInformationSlides

 

Simulation and control of charged particleReferences

 




 

Tuning of the LbSync System

 

Bianca Veglia ID Gap Compensation and Extremum Seeking at PETRA III
















  • No labels