Choose timezone
Your profile timezone:
Powered by Indico
v3.3.5
Maximilian Schütte: First Steps towards data-based IPC for the EuXFEL optical synchronization System
We describe the effort at DESY MSK/MCS to enable extensive data collection for the EuXFEL optical synchronization system and first use of the data for system understanding and fault analysis. We also give an outlook on what we think can be achieved with the data in combination with IPC in the long run and discuss foreseen challenges.
Jan Timm: Trip Event Logger - Fault Analysis and HPC
The Trip Event Logger is a fault diagnosis tool to detect errors, inform the operators and trigger automatic supervisory actions. Further goals are to provide information for a fault tree and event tree analysis as well as a database of labeled faulty data sets for offline analysis. The tool is based on the C++ framework ChimeraTK Application Core. With this close interconnection to the control system it is possible not only to monitor but also to intervene as it is of great importance for supervisory tasks. The core of the tool are the fault analysis modules ranging from simple ones (e.g., limit checking) to advanced ones (model-based, machine learning, etc.).
Sören Jalas: Bayesian optimization of laser-plasma accelerators
The high-dimensional and non-linearly coupled parameter spaces that define laser-plasma accelerators calls for advanced optimization strategies to reach their full potential. Here we report on the latest result of applying Bayesian optimization to the LUX accelerator.
Manuel Kirchen: Prediction of the electron beam quality in a laser-plasma accelerator
The quality and stability of laser-plasma accelerated electron beams is highly sensitive to fluctuations of the laser pulse that drives the plasma wave. Here, we demonstrate the use of machine learning to predict the shot-to-shot electron beam quality of the LUX accelerator as a function of key drive laser properties.
Jun Zhu: Application of multi-task deep learning at the injector of European XFEL
Deep neural networks can be trained based on simulated or experimental data to rapidly predict a variety of beam properties at different locations of a beamline, using machine control parameters as input. Preliminary studies at the injector of European XFEL showed very promising result in predicting beam parameters such as emittance and twiss parameters at the main beamline as well as images for longitudinal phase-space measurement at the diagnostic beamline. The approach of multi-task learning aims at producing a more generalized model by sharing representations between these related tasks, which could also facilitate learning new tasks in a data-efficient way.
Gianluca Martino: Self-healing of the LLRF system at firmware level
Self-healing is increasingly becoming a promising approach to designing reliable cyber-physical systems, and it refers to the ability of a system to detect faults or failures and fix them through healing or repairing. Its application in highly distributed and complex systems poses new and interesting challenges.