Heriot-Watt University
ROS: The Research Output Service. Heriot-Watt University EdinburghNot a member yet
4689 research outputs found
Sort by
From pastime to paycheque : a longitudinal study of hobby-based entrepreneurship and the dynamics of motivation
Full text linkThis thesis set out to investigate why individuals turn a hobby into a business and whether their
narratives of hobby-based entrepreneurship (HBE) remain consistent over time. Little previous
research has explored this phenomenon, despite the relatively common monetisation of
hobbies, and none has done so longitudinally; moreover, no research has explored this
phenomenon in the context of COVID-19. The purpose of this thesis was threefold: 1) to gain
insight into a sample of American HBE founders’ motivations for starting a business during
COVID-19; 2) to explore any changes in their stated motivations over time; and 3) to
investigate whether existing theories of motivation may apply in the study context.
The thesis was underpinned by a social constructionist philosophy and employed a qualitative
research approach. Data were collected from 10 American HBE founders in 2020 and 2022
using unstructured interviews; resulting data were subject to thematic analysis.
The main finding of this study is that sample HBE founders’ stated motivations change with
time and experience, shifting from a pragmatic to an affective focus. The research offers three
major contributions. First, sample HBE founders have mixed motivations for engaging in
HBE. Second, entrepreneurial motivations are not fixed, but rather fluid depending on
individual context and experience. Third, Herzberg’s hygiene theory is useful as a conceptual
lens for entrepreneurial motivation in this context. In addition, an adaptation to this model is
proposed, using Deci and Ryan’s self-determination theory as a supplement to help
understand the admixture of pragmatic and affective motivations in HBE
Behavioural disequilibrium
Full text linkThis thesis examines the role of stock misvaluation in explaining the dynamics of
corporate financing strategies, investment decisions, earnings management practices,
and earnings performance of firms. It aims to contribute to the literature on the effect
of stock price misvaluation on corporate financial decisions and outcomes. The first
study examines how equity misvaluation affects firm equity issuance, debt issuance
and the proportion of earnings retained. Further, it examines the financing channels
through which firms invest in capital expenditures and research and development in
response to equity misvaluation. The second study investigates the role of equity misvaluation and investor psychological reference points embodied in earnings targets in
driving earnings management behaviour. The third study explores the effect of misvaluation on future firm profitability measured by growth in earnings.
The results indicate firms react to misvaluation of their shares primarily through
the issuance of equity. Issued equity is used to finance increases in capital expenditures as well as research and development investments. Further, the findings suggest
the element of stock misvaluation which is idiosyncratic to the firm rather than shared
amongst industry peers is associated only with research and development investment.
The second study presents evidence that misvaluation-induced earnings management
is positively associated with the probability that the firm reports earnings above the
profitability and analyst forecast targets. This suggests that managers cater to investor psychological reference points through earnings manipulation. Robustness tests
also show that a positive relationship between misvaluation and earnings management
holds for firms reporting earnings above and below their targets, implying capital market motivations are not the only incentive driving misvalued firms to manage earnings,
as assumed in prior literature. In the third study, findings indicate misvaluation is positively associated with future growth in earnings per share. Furthermore, the results
evidence that misvaluation is more strongly related to future earnings changes in the
long-run, rather than the short run. These results support the findings that misvaluation has a positive effect on long-term investments such as research and development.
Further, the findings indicate that the relationship between misvaluation and earnings
growth varies with level of growth opportunities. Firms with higher growth opportunities exhibit a stronger link with misvaluation than do firms with lower growth
opportunities. These findings align with the theoretical expectation that growth firms
are more sensitive to equity valuations in their corporate financing and investment decisions.Edinburgh Business School scholarship and fundin
Empirical Bayesian image restoration by Langevin sampling with total variation and denoising diffusion priors
Full text linkBayesian statistics is a cornerstone of imaging sciences, underpinning a variety of methodologies and approaches for tackling image inverse problems. These problems are highly
challenging due to their ill-posed nature and the significant impact that the used models can have on the estimated results. To effectively address these challenges, this thesis
adopts a Bayesian approach tailored for dealing with ill-posed problems and partially unknown models. This approach allows for the incorporation of incomplete prior knowledge
and uncertainty about model parameters directly into the Bayesian model, enhancing the
robustness and accuracy of the solutions.
In this thesis, we make three contributions to image restoration within the Bayesian
framework. Our first approach introduces a stochastic optimisation methodology for
performing empirical Bayesian inference in semi-blind image deconvolution problems.
Our approach automatically calibrates the parameters of the blur model using maximum marginal likelihood estimation, followed by non-blind image deconvolution through
maximum-a-posteriori estimation conditional on the estimated model parameters. Additionally, it calibrates the measurement noise level and any regularisation parameters.
Given the computational intractability of the marginal likelihood of these parameters,
our method employs a stochastic approximation proximal gradient optimisation scheme,
iteratively solving the required integrals using a Moreau-Yosida regularised unadjusted
Langevin Markov chain Monte Carlo algorithm. This strategy can be efficiently applied
to any log-concave model, utilising the same gradient and proximal operators required for
computing the maximum-a-posteriori solution through convex optimisation. We provide
convergence guarantees under realistic and easily verifiable conditions and validate our
approach with a series of deconvolution experiments and comparisons with state-of-the-art methods.
Our second contribution seeks to address image restoration tasks by combining a
pre-trained foundational score-based diffusion model as prior, with a likelihood function
specified during test time. While existing methods like denoising diffusion probabilistic models (DDPM) and denoising diffusion implicit models (DDIM) produce realistic
images, they suffer from likelihood intractability issues. Alternatively, Langevin diffusion processes avoid these issues but often result in inferior quality and longer computing
times. Inspired by recent works at the interface of these two approaches, the proposed
method embeds a DDPM denoiser within an empirical Bayesian Langevin algorithm,
which jointly calibrates key model hyperparameters while estimating the minimum mean
squared error solution, given by the model’s posterior mean. Extensive experiments on
image deblurring, super-resolution, and inpainting tasks demonstrate that our approach
outperforms current state-of-the-art strategies in both mean square error image estimation
accuracy and computational efficiency.
Lastly, we introduce a novel Monte Carlo method to assess the accuracy of existing Bayesian imaging methods for uncertainty quantification tasks. More precisely, this
Monte Carlo method aims to investigate whether the probabilities delivered by existing
Bayesian imaging methods are meaningful under the replication of an experiment, or if
they only serve as subjective measures of belief. Leveraging this method, we conducted
an extensive experiment requiring 1,000 GPU-hours to probe the accuracy of five canonical Bayesian imaging methods that are representative of some of the main Bayesian
imaging strategies from the past decades. We find that, in a few cases, the probabilities
reported by modern Bayesian imaging techniques are in broad agreement with long-term
averages as observed over a large number of replication of an experiment. However, existing Bayesian imaging methods are generally not able to deliver reliable uncertainty
quantification results.
In summary, this thesis makes three significant contributions to the Bayesian imaging
literature and opens new avenues for future research in the field. We have tested and
validated the effectiveness and efficiency of the proposed methods across various inverse
problems, including image deblurring, inpainting and super-resolution. Furthermore, we
compared the proposed methods with cutting-edge methods in the literature.Supported by the UKRI EPSRC projects BOLT and BLOOM (EP/T007346/1,
EP/V006134/1
Laboratory surface astrophysics of cyclohexane and decalin
Full text linkThis thesis investigates the thin film growth, desorption kinetics, and interaction
behaviour of simple molecules on interstellar dust grain analogues and water ice using
surface science techniques. An amorphous silica (aSiO2) substrate was employed to
mimic bare interstellar dust grains, while a thin film of amorphous solid water (ASW)
represented icy grain surfaces. The study focused on the interactions of these surfaces
with thin adsorbed films of cyclohexane (C6H12), cis-decalin (C12H10), and trans-decalin.
Temperature programmed desorption (TPD) and reflection-absorption infrared (RAIR)
spectroscopy were utilized to analyse the desorption kinetics and surface interactions of
pure solid C6H12, cis-C12H10, and trans-C12H10 adsorbed on aSiO2. Additionally, RAIR
spectroscopy was employed to study the surface interactions of C6H12 on ASW. Ab initio
quantum chemical calculations supported the interpretation of the data, highlighting key
interactions between the species for each system and comparing them with systems
incorporating dehydrogenated aromatic species, benzene (C6H6), and naphthalene
(C10H8).
The desorption of C6H12 from the aSiO2 surface exhibited zero-order kinetics due to rapid
two-dimensional gas-solid equilibrium, consistent with de-wetting and the formation of
three-dimensional islands. Low exposure deposition of C6H12 at 26 K indicated
interaction with the aSiO2 surface via blue shifting of the C-H stretching frequency.
Higher dosage spectra revealed peak broadening at low temperatures and evidence of
peak splitting nature after annealing, suggesting a low-temperature phase change around
45 K.
RAIR spectra of C6H12 on ASW showed red-shifted C-H stretching modes, with the
asymmetric mode being more influenced than the symmetric. Ab initio investigations
revealed significant interactions between equatorial H of C6H12 and H2O, with binding
energies estimated at -5.93 kJ mol-1
for the most strongly bound complex and -8.93 kJ
mol-1
for interactions with a H2O cluster. Contrasting behaviours were observed between cis and trans isomers of C10H18 on the
aSiO2 surface. Trans-C10H18 formed a distinctive monolayer prior to multilayer growth,
while cis-C10H18 did not wet the substrate. The desorption kinetics of the trans-C10H18
monolayer followed a first-order process with an Edes of 67.8 kJ mol-1
, while multilayer
desorption was zero-order for both isomers, with Edes of 62.6 kJ mol-1
Using scenario planning to build strategic options to boost the performance of League of Ireland football
Full text linkThis thesis considers how a strategy can be developed by applying scenario planning
methodologies. The aim is to use a qualitative approach to expand on existing theory
relating to scenario planning and to consider how it feeds into the strategy process. This
research was conducted in relation to League of Ireland football, which has historically
underperformed.
Existing academic research on scenario planning, futures studies and strategy was
reviewed to gain an understanding of existing scenario planning methodologies that are
widely applied in practice, leading to a conceptual framework that develops the Strategy
Identification Framework (SIF).
The SIF was applied to LOI football in a three step process. First, a range of secondary
data was analysed to identify topics that were previously highlighted as areas to improve.
Second, these topics were used as a basis for semi-structured interviews with stakeholders
within the industry, or tangentially linked. Using thematic analysis, a number of
improvement drivers were identified and used to create a matrix of potential scenarios.
These scenarios were reduced in number to three based on a review for possibility and
plausibility, and then developed in detail. Finally, the three scenarios were presented to
an expert panel and a future strategy for the LOI was identified. This practical application
led to a revised SIF that can be used in other domains.
The key contribution of the research is the development and application of the SIF, which
reflected the learnings gained from the practical application to the LOI. This study has
closed the gap between theory and practice in scenario planning, making it more
accessible for organisations to use to create more robust strategies for the future, whilst
engaging all relevant stakeholders at various stages in the process.
The application of the SIF to the LOI highlighted the issues that remain within football
in Ireland. The recommendation of Stadia 2031 using the financial recommendations
contained in the present study, highlighted how key stakeholders are reticent to take
ownership and believe that the gap should be closed by the Irish government. This is not
a likely outcome. Unless significant changes are made within the structure of the game in
Ireland, this study is consigned to being another lost opportunity
Exploring the adoption of a risk management system for fiscal space expansion in Botswana
Full text linkThe purpose of this study is to examine how civil servants in Botswana's healthcare sector
perceive the adoption of risk management systems. Risk management systems (RMS)
adoption factors are examined through technology adoption theories. Reviewing
documents and conducting interviews reveals barriers such as little understanding,
resistance to change, resource limitations, and system integration challenges. RMS
adoption remains positive despite these obstacles, emphasizing the importance of
financial resources, strong organizational support, active user involvement, and alignment
with existing processes. Investing in comprehensive training programs to build
confidence and competence within users is key to overcoming these barriers. Implications
suggest significant policy changes to enhance RMS adoption in Botswana's healthcare
sector. Implementing these findings could enhance accountability, transparency,
efficiency, and safety. Applying the technology adoption model provides an
understanding of RMS adoption factors in Botswana. Practically, it facilitates the
implementation of RMS through actionable insights. Future research should examine the
long-term impact of RMS adoption, compare it to other regions or sectors, and explore
advanced technologies. Such research could improve the effectiveness of RMS in
healthcare settings by providing deeper insights into best practices. Using this study,
informed decisions can be made, and strategic planning can be implemented, enhancing
healthcare quality
Multi-frequency bandwidth Empirical Market Factors in regularised covariance regression
No full textWe survey and test non-constructive basis decomposition algorithms capable of analysing
the structures present in financial security times series data. We name these implicit financial structures Empirical Market Factors (EMFs) as a homage to Huang et al. (1998)
and Empirical Mode Decomposition (EMD) upon which part of this work is based. The
EMF covariates are isolated via implicit factor extraction (IFE) which is a decomposition algorithm or feature engineering technique. `Implicit' is used to differentiate these
covariates from explicit (easily observable or contructable) covariates such as the return
of a market portfolio, ratios of market capitalisations, and book-to-market ratios such
as in Fama and French (1993). The forthcoming investment period's covariance structure is forecast using these estimated EMFs in a regularised covariance regression (RCR)
framework from Hoff and Niu (2012) to which we made very modest extensions.
We present a real-world case study in which we test our method in forecasting the covariance of the potential investments before we weight the portfolio accordingly. The
strategies assessed are also restricted to Long/Short Equity (LSE) and Risk Premia Parity (RPP) weighting strategies in which there are cumulative weight shorting restrictions
(speci cally the 130/30 strategy) as opposed to restrictions on the individual weights -
this mimics real-world shorting limitations. All these techniques and technologies (IFE,
RCR, RPP, and LSE) are combined to construct risk-conscious leveraged RPP portfolios
using EMFs in a lagged RCR framework
Sounding out blue carbon : a review of integrated data collection using acoustic techniques to support cost-effective quantification, and subsequent accreditation, of blue carbon
Full text linkBlue carbon was first introduced as a term to highlight the importance of marine and intertidal habitats
for climate change mitigation. Adopted into policy mechanisms, the need for integration of ecologic,
economic, and societal values in eco-social-economics became evident as a key requirement of blue
carbon research. This thesis explores, details, and critically evaluates existing approaches to blue
carbon research, data gaps, and quantification techniques. Subsequently, blue carbon research in this
thesis is explicitly framed with the relevance of habitat stock and sequestration quantification to future
research needs, carbon accreditation processes, and policy mechanisms.
This thesis comprises seven chapters. The first chapter, Introduction, presents the importance of blue
carbon to climate change mitigation, key components of marine carbon cycles, current blue carbon
research gaps, and research goals of the thesis. The second chapter, redefining blue carbon with
adaptive valuation for global policy, further delves into the policy context, eco-social-economics, and
carbon accreditation criteria by critically evaluating blue carbon definitions. In this chapter a
redefinition of blue carbon is proposed with additional recommendations to support the uptake of blue
carbon research in policy frameworks.
There are three central data chapters, focusing on seagrass, maerl, and horse mussel habitats, that assess
the suitability of Sub-Bottom Profiling (SBP) data for quantification of blue carbon sediment thickness.
These chapters provide novel data on seagrass, maerl, and horse mussel bed sediment thickness and
use core data to ground-truth SBP data and support image analysis and sediment identification. Drop-Down Video (DDV) is also used to align measures of biodiversity, as habitat % coverage, to sediment
thickness. Structural Equation Modelling (SEM) pathway analysis is then used to determine the key
factors, and differences between abiotic and biotic variables, on seagrass, maerl, and horse mussel bed
sediment thickness.
These central data chapters, and the data presented in them, are then used to support novel predictive
models in Chapter 6, Adaptive Stacked Species Distribution Models (AS-SDMs), which update areal
extents, carbon stocks, and sequestration potential of seagrass, maerl, and horse mussel habitats in
Orkney. Key areas for conservation are identified and valuated to support engagement with policy
makers and community stakeholders. The key outputs and findings of the thesis are summarised in
Chapter 7, Conclusions
The impact of Forward Recovery Voltage on power MOSFET reliability, and power MOSFET Remaining Useful Life prediction using Machine Learning
No full textIncreasing energy demands, the integration of renewable energy systems, and distributed generation have resulted in a high level of participation from academia and
industry in the pursuit of new and/or improved power efficiency and reliability solutions. Power converters have an extremely important part to play in the functions
of power conversion and control in the modern age, often known as industry 4.0. As
a result of a growing demand for electrical power, there is also an increase in the
requirements imposed on the reliability of power converters.
The research in this thesis concentrates on the experimental investigation of the
Three-Level Hybrid Active Neutral Point Clamped (3L-Hybrid ANPC) converter
topology. Combining 2-Silicon Carbide (SiC) Metal Oxide Semiconductor Field
Effect Transistor (MOSFET)s and 4-Silicon (Si) Insulated Gate Bipolar Transistor (IGBT)s in (3L-Hybrid ANPC) topology provides comparable performance, decreased volume, and cost advantages in comparison to its equivalent Si ANPC structure. Recent research have demonstrated, however, that Forward Recovery Voltage
(FRV) occurring in Si IGBT and its anti-parallel diodes may be a possible limiting
factor for these advantages. This thesis investigates the impact of FRV in (3L-Hybrid ANPC) topology, and demonstrates its impact on SiC MOSFET reliability.
The analysis reveals that the forward recovery effect in IGBT and its anti-parallel
diode have two major shortcomings. First, the FRV energy loss is substantial and
decreases the total system efficiency. When switched at high switching rates, SiC
MOSFETs undergo voltage stress, which lowers their long-term reliability in this
topology.
In this thesis, further, a hybrid power MOSFET Remaining Useful Life (RUL) estimation method based on Machine Learning (ML) classification and regression algorithms is investigated. It is vital to understand wear-out degradation and parameter
changes in power MOSFETs in order to implement condition monitoring systems
that are capable of identifying failure precursors before occurring and enhancing
system reliability. This thesis experimentally demonstrated an ML-based RUL prediction model using on-state resistance between Drain-Source terminals of a power
MOSFET (R(DS(on)) as a precursor. To obtain the necessary data for the validation of the proposed technique, an Accelerated Life Test (ALT) setup was constructed.
After collecting pertinent data, the suggested RUL estimation technique using classification and regression algorithms was implemented on both Si and SiC MOSFET
devices. It has been shown that this hybrid technique delivers promising accuracy
on both technologies. Furthermore, this hybrid system can be used not only on
power MOSFETs but also on any other electronic systems that have exponential
degradation behaviour.
Overall, the original research objectives of this thesis are two pieces. This thesis
suggests an application-level loss mechanism referred to as FRV occurring in a Three-Level Hybrid ANPC (3L-Hybrid ANPC) converter topology and a Data-Driven (DD)
approach for predicting the RUL of a power MOSFET using ML
The development of photoactive microparticles in continuous flow for the controlled release of agrochemicals
Full text linkThe following research aims to address one of the challenges posed by industrial
agriculture. The inefficient use of agrochemicals (fertilisers, pesticides, herbicides and
fungicides) often results in a loss of >99% of applied agrochemical to the environment.
These losses are a result of pathways such as soil-leeching and photolysis. A well-established method for preventing the unnecessary loss of applied agrochemicals is found
in ‘controlled release’, which has been most prevalent in medical research, but has also
seen increasing applications in agriculture. Controlled release refers to a method by which
a synthetic barrier to diffusion slowly releases stored agrochemicals over an extended
period of time. Commonly, this synthetic barrier to diffusion takes the form of nano/micro
particles, in which the agrochemical is encapsulated. The materials from which these
particles can be manufactured are often of biological origin, such as lignin or cellulose,
but increasingly the vast toolkit of synthetic ‘biodegradable’ polymers are also being put
to use.
As an alternate facet of controlled release, this thesis explores the concept of ‘stimulated
release’, referring to the release of an encapsulated agrochemical in the presence of a
specific external stimulus. Typical stimuli can be biotic or abiotic in nature such as
enzymatic activity or changes in temperature or pH respectively. A promising but
underutilised stimulus is the abundant and renewable source of energy provided by
sunlight. Through an in-depth literature review on the topic of photo-stimulated release
of molecules from polymers, a dearth in research focused on wavelengths of light above
the ultraviolet (UV) region and into the benign and deeply penetrating wavelengths of
the visible region of the electromagnetic spectrum was identified. The rational for this
lack of research being that chemical transformations such as bond breaking, or structural
isometric changes inherently require more energy than what is provided by photons in the
visible region.
As a means to address this energetic discrepancy, an indirect method of photo-stimulation
was identified and developed. Use of organic photocatalysts based on the
benzothiadiazole (BTZ) core moiety were found to be an effective means for transferring
the dependence on photon wavelength away from photo-active functional groups
typically used for photo-stimulated release. In this way, the wavelength of absorption required for a chemical transformation was entirely dependent on the properties of the
photocatalyst. It was shown that the absorption wavelength of the BTZ photocatalyst
could be altered from 380 nm to 450 nm by simple structural and electrical alterations
made during the photocatalyst’s synthesis. The greatest effect was found to be increasing
the extent of conjugation in the photocatalyst and the introduction of electron donating
groups. The photocatalysts were also assessed for their efficiency in the photosensitised
production of reactive oxygen species (ROS), which could then go on to selectively
oxidise functionality within a polymeric particle and thus result in its degradation (i.e. the
mechanism for stimulated release).
A suitable photocatalyst was then developed for anchoring to a polymeric chain. To that
end, a monomeric analogue of ɛ-caprolactone was synthesised for covalent bonding to
the photocatalyst and subsequent co-polymerisation with unmodified ɛ-caprolactone.
Anchoring the photocatalyst to a polymer chain was seen to increase the photosensitised
production of ROS, as quantified by the exclusive conversion of α-terpinene to
ascaridole. The conversion increased from 23% conversion (free BTZ) to 72% (anchored
BTZ) over 120 minutes.
A method for the production of polymeric particles up to 500 µm in diameter was also
established using microfluidic devices. The method of preparation was highly robust to
changes in flow rate and feed composition, producing particles of acceptable dispersity
in size and shape.
Finally, to compliment the synthesised photocatalysts, a new monomer, (DMODT) was
developed to provide a sacrificial linker in a polymer chain which would break in the
presence of ROS. This linker was shown to undergo selective photosensitised oxidation,
effectively breaking the polymeric backbone and hence providing a means to degrade a
polymeric particle. This process was quantified using a novel means of continuous atline
analysis of polymer molar mass in flow using gel permeation chromatography. The
accelerated degradation of copolymers of ɛ-caprolactone was thus demonstrated over a
24-hour period.
This thesis has provided the groundwork for a fairly broad area of study and has shown
that there is great potential for selectively oxidisable functionality of biopolymers by
means of photocatalysis. The materials and compounds described herein have contributed to the growing research surrounding sustainable intensification and have provided a
means to shift photo-responsive materials away from a dependence on UV light
Access Repository Dashboard
Do you manage ROS: The Research Output Service. Heriot-Watt University Edinburgh ? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!