% 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{Bedei:1018815,
author = {Bedei, Julian},
othercontributors = {Andert, Jakob Lukas and Trimpe, Johann Sebastian},
title = {{S}ichere {S}ystemidentifikation und {R}egelung der
ottomotorischen {S}elbstzündung durch maschinelles
{L}ernen},
school = {Rheinisch-Westfälische Technische Hochschule Aachen},
type = {Dissertation},
address = {Aachen},
publisher = {RWTH Aachen University},
reportid = {RWTH-2025-07998},
pages = {1 Online-Ressource : Illustrationen},
year = {2025},
note = {Veröffentlicht auf dem Publikationsserver der RWTH Aachen
University; Dissertation, Rheinisch-Westfälische Technische
Hochschule Aachen, 2025},
abstract = {Climate change is one of the greatest global challenges.
Despite the increasing electrification of the transportation
sector, combustion engines remain significant worldwide.
Enhancing their efficiency is essential for reducing
greenhouse gas emissions. Homogeneous charge compression
ignition (HCCI) offers the potential to improve efficiency
while simultaneously reducing nitrogen oxide emissions (NOx)
through low combustion temperatures. However, controlling
HCCI is challenging due to non-linearities, stochastic,
autoregressive characteristics and multiple input multiple
output (MIMO) behavior. Since the combustion depends on the
thermodynamic, chemical state of the mixture, minimizing
state fluctuations through closed-loop control is necessary
for process stabilization. In this work, to account for
innercyclic fluctuations of the mixture state, in addition
to a cyclic control scale, the potential of an additional
innercyclic control loop is demonstrated. To achieve this,
machine learning algorithms are employed, requiring large
data sets generated through interaction with the real
process to capture all cross-couplings between control and
state variables. For safe exploration of the experimental
space, a measurement algorithm is implemented to interact
with the process on both time scales, enabling the
identification of stochastic process limitations. The
generated data is used to train artificial neural networks
(KNN), which are integrated into the multiscale control
through a real-time capable implementation. Thus, the
benefits of the multiscale approach are demonstrated
experimentally for the first time. Compared to a purely
cyclic control, the standard deviation of the combustion
phasing, which is a measure of combustion stability, is
reduced by $19,7\%.$ Furthermore, the process limitations
identified during measurement allow for the application of
Reinforcement Learning (RL) to interact with the
safety-critical system of an HCCI test bench for the first
time. The potential of RL is demonstrated through
load-transient process control while adhering to
safety-relevant boundary conditions. Additionally, the
adaptability of RL is leveraged to reduce emissions by
partially substituting gasoline with ethanol. This work
highlights the potential of machine learning approaches for
HCCI control to achieve both improved stability and reduced
emissions. Moreover, given its adaptability, the developed
RL toolchain provides a framework for systematic follow-up
studies.},
cin = {412330 / 422320},
ddc = {620},
cid = {$I:(DE-82)412330_20140620$ / $I:(DE-82)422320_20210420$},
pnm = {DFG project G:(GEPRIS)277012063 - FOR 2401:
Optimierungsbasierte Multiskalenregelung motorischer
Niedertemperatur-Brennverfahren (277012063)},
pid = {G:(GEPRIS)277012063},
typ = {PUB:(DE-HGF)11},
doi = {10.18154/RWTH-2025-07998},
url = {https://publications.rwth-aachen.de/record/1018815},
}
guest ::