Single Nanoparticle Catalysis- Implementing Machine Learning Algorithms for the Identification of Kinetic Phase Transitions on the Surface of Individual Catalyst Nanoparticles during Surface Chemical Reactions through Nanoplasmonic Spectroscopy
Typ
Examensarbete för masterexamen
Program
Complex adaptive systems (MPCAS), MSc
Publicerad
2020
Författare
Klein Moberg, Henrik
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
The importance of catalysis within industrial and scientific endeavours can hardly be
overstated, playing a crucial role in most processes that contain chemical reactions
of any kind. This ubiquitous nature of catalysts makes them very relevant for
study, yet there is a clear divide between scientific understanding and practical
purpose of catalysts - too little is known of the nature of catalysts at the nanoscale
within atmospheric conditions. Research within single nanoparticle catalysis,
a hitherto elusive field of chemical physics, aims to bridge this gap by analyzing
the surface of individual nanoparticles, during reaction conditions that are relevant
for practical applications, using a relatively cheap spectroscopy setup. However,
the resulting data is so complex and multivariate that standard analysis methods
are incapable of elucidating the relevant physical correlations between nanoparticle
surface kinetics and spectroscopic results. To remedy this, this thesis’ focus is the
implementation of a deep densely connected neural network and a long short-term
memory recurrent neural network to identify a specific surface kinetic behaviour,
known as the kinetic phase transition, within each individual nanoparticle, given
only the results of spectroscopy. We find that both networks are reliable, and with
a few reservations and admonissions, give generalizable results that can be applied
across samples of nanoparticles.
Beskrivning
Ämne/nyckelord
nanoparticle , catalysis , nanoplasmonic spectroscopy , surface catalysis , heterogeneous catalysis , machine learning , recurrent network , deep learning