% 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”.
@PHDTHESIS{Jalas:605571,
author = {Jalas, Soeren},
othercontributors = {Maier, Andreas and Leemans, Wim},
title = {{M}achine {L}earning {B}ased {O}ptimization of
{L}aser-{P}lasma {A}ccelerators},
school = {Universität Hamburg},
type = {Dissertation},
address = {Hamburg},
reportid = {PUBDB-2024-01513},
pages = {156},
year = {2023},
note = {Publish under creative commons Attribution-NonCommercial
4.0 license
(https://creativecommons.org/licenses/by-nc/4.0/);
Dissertation, Universität Hamburg, 2024},
abstract = {Laser-plasma accelerators (LPA) utilize intense laser
pulses to drive plasma waves, generating strong electric
fields capable of accelerating electrons to relativistic
energies. This technology promises accelerator facilities
that are orders of magnitude smaller than their conventional
counterparts. However, the compactness of LPA systems and
the non-linearity of the laser-plasma interaction create a
highly coupled environment, where various input and output
parameters intricately interconnect. Consequently, manually
identifying suitable working points providing the desired
beam quality and stability as well as precise tuning of beam
parameters to match the demands of specific applications
becomes a tedious task, often leading to sub-optimal
performance.The aim of this thesis is to explore the
application of machine learning-based methods, particularly
Bayesian optimization, within the realm of laser-plasma
accelerators. The study involves the implementation of
Bayesian optimization to fine-tune the parameters of the Lux
accelerator, encompassing simulations and real-time
experimentation.The proposed approach is initially examined
through particle-in-cell simulations of the accelerator.
After optimizing the system for beam quality, the predictive
model built by the optimizer is employed to identify sources
of instability and explore potential avenues for
improvement. Subsequently, Bayesian optimization is
implemented for online control of the experiment. Various
methods to address experimental noise and parameter
fluctuations are investigated, leading to the identification
of techniques emphasizing ultimate beam quality or
stability. This way the machine can autonomously tune itself
to generate electron beams with sub-percent energy spread
and significant charge. Furthermore, a robust operational
regime with less than five percent energy spread for 90 $\%$
of all shots is identified and studied.Finally, Bayesian
optimization targeting multiple beam parameters enables the
identification of tuning curves allowing precise adjustments
to specific beam properties while maintaining optimal beam
quality. These complex tuning curves describe the intricate
balancing of multiple laser and plasma parameters to achieve
the most favorable tuning of beam parameters. This approach
facilitates charge tuning over nearly 100 pC for various
beam energies, while keeping energy spreads below 5 $\%.In$
combination, the methods presented in this thesis provide
valuable tools for effectively managing the inherent
complexity of LPAs, spanning from the design phase in
simulations to real-time operation, potentially paving the
way for LPAs to cater to a wide array of applications with
diverse demands.},
cin = {MLS},
cid = {I:(DE-H253)MLS-20210107},
pnm = {621 - Accelerator Research and Development (POF4-621) /
PHGS, VH-GS-500 - PIER Helmholtz Graduate School
$(2015_IFV-VH-GS-500)$},
pid = {G:(DE-HGF)POF4-621 / $G:(DE-HGF)2015_IFV-VH-GS-500$},
experiment = {EXP:(DE-H253)LUX-Beamline-20221201},
typ = {PUB:(DE-HGF)11},
urn = {urn:nbn:de:gbv:18-ediss-117247},
doi = {10.3204/PUBDB-2024-01513},
url = {https://bib-pubdb1.desy.de/record/605571},
}