Heriot-Watt University
ROS: The Research Output Service. Heriot-Watt University EdinburghNot a member yet
4689 research outputs found
Sort by
Ecosystem restoration and habitat management : blue carbon and bivalve shellfish
Get PDFHistorically, considerations of the carbon budget of bivalve shellfish have
disproportionately focused on the cycling of carbon in shell alone, overlooking the
potential role of bivalve shellfish habitats in the stabilisation of sediment and therefore
carbon. With the aim of providing essential evidence to inform evaluation of
management strategies and the business case for restoration of flat oyster habitats, the
purpose of this study was to examine the bivalve shellfish carbon budget at both the scale
of the individual, and at the ecosystem level. At the individual scale, components of the
major pathways of carbon deposition and release were identified and, where possible,
quantified. Through feeding alone, European flat oyster (Ostrea edulis) were shown to
significantly enhance bentho-pelagic coupling and carbon transport. In consideration of
the bivalve carbon budget, the inclusion of the deposition of sedimentary carbon, as well
as carbon stored in shell, balanced with the release of carbon through respiration and
calcification suggested the recovery of oyster beds is likely to facilitate the accretion of
substantial carbon stocks, though these habitats are unlikely to be significant carbon sinks
in the context of global climate change mitigation. At the ecosystem level, organic carbon
content of on-bed sediments of blue mussel (Mytilus edulis) beds was nearly twice that
of off-bed sediments. The evidence presented within this thesis demonstrates the
importance of a holistic approach to understanding carbon cycling in marine ecosystems
Developing an implementation framework for Lean Six Sigma in high-value and low-volume industries
Get PDFHigh-value low-volume (HVLV) industries hold a strong relevance in Germany, with a
focus on complex engineering and large projects conducted at low frequencies.
Methodologies for continuous improvement (CI), such as Lean Six Sigma (LSS), have
been used in mass production industries, such as automotive, to promote operational
excellence and are now relevant in HVLV industries to survive amid growing
international competition. The implementation of LSS in the HVLV industries has not
yet been studied to much extent. Therefore the purpose of the present study is to develop
an implementation framework for Lean Six Sigma in the wind power industry in
Germany.
To develop the conceptual framework, a systematic literature review of critical success
factors (CSFs) and critical failure factors (CFFs) was conducted. The review identified
similarities between the CSFs and CFFs which often reflected opposite conditions of the
same variable. The present study connects the success and failure factors as critical
influencing factors (CIFs), which include reasons for both failure and success. An
analysis of five relevant implementation frameworks of LSS shows that none of the
frameworks fully includes all CIFs and do not provide a complete answer concerning LSS
implementation in terms of what should be done to secure success and avoid failure in
the process.
The chosen research paradigm was critical realism and the research method was action
case in combination with action learning. The outcome of the study was an
implementation framework for LSS, the 3D framework house, as the essential result of
the research, which was validated by LSS experts.
The present study contributes to scholarship with the 3D framework house and a cycle
approach in 18 steps, as well as with the detailed description of the newly defined CIFs
with the focus dimensions and with the HVLV-specific focus dimensions.
It contributes to practice with the improved situation in the research organisation and with
the clear guideline for practitioners, giving answers to the questions “what needs to be
done” to improve the situation of the CIFs of LSS, “how this can be implemented” and
“who is responsible” in the HVLV industry context
High-dimensional Bayesian methods for interpretable nowcasting and risk estimation
Get PDFThis thesis presents new models for nowcasting and macro risk estimation using frontier Bayesian methods that enable incorporating Big Data into policy relevant prediction problems. We propose variable selection algorithms motivated from Bayesian
decision theory to make model outcomes interpretable to the policy maker.
In chapter 2, we propose a Bayesian Structural Time Series (BSTS) model for nowcasting GDP growth. This model jointly estimates latent time trends to capture
slow moving changes in economic conditions along-side a high dimensional mixed
frequency component that is extracted from higher frequency (monthly) cyclical information. We extend on previous implementations of the BSTS with priors and
variable selection methods which facilitate selection over latent time trends as well
as mixed-frequency information that remain tractable to the policy maker. Empirically, we provide a novel nowcast application where we use a large dimensional
set of Internet search terms to gain advance information about supply and demand
sentiment for the US economy before more commonly considered macro information
are available to the nowcaster. We find that our proposed BSTS model offers large
improvements over competing models and that Internet search terms matter for
nowcasts before hard information about the macro economy have been published.
A simulation exercise confirms the good performance of the proposed model.
Chapter 3 presents the T-SV-t-BMIDAS (Bayesian Mixed Data Sampling) model
for nowcasting quarterly GDP growth. The model incorporates a long-run time-varying trend (T) and t-distributed stochastic volatility accounting for outliers (SV-t) into a Bayesian multivariate MIDAS. To address the high-dimensionality of the
model, to account for group-correlation in mixed frequency data, and to make the
model interpretable to the policy maker, we propose a new combination of group-shrinkage prior with sparsification algorithm for variable selection. The prior flexibly
accommodates between-group sparsity and within-group correlation and allows to
communicate the joint importance of predictors over the data release cycle. We
evaluate the model for UK GDP growth nowcasts covering also the time-span of
the Covid-19 recession. The model is competitive prior to the pandemic relative to
various benchmark models, while yielding substantial nowcast improvements during
the pandemic. Contrary to many previous nowcasting approaches, the model reads
in sparse group signals from the data. Simulations show competitive performance
of the variable selection methodology, with particularly good performance to be
expected for highly correlated data as well as dense data-generating-processes.
Chapter 4 presents a new Bayesian Quantile Regression (BQR) model for high dimensional risk estimation. It extends the horseshoe prior to the BQR framework
and provides a fast sampling algorithm for computation that makes it efficient for
high-dimensional problems. A large scale simulation exercise reveals that compared
to alternative shrinkage priors, the proposed methods yield better performance in
coefficient bias and forecast error, especially in sparse data-generating processes
and in estimating extreme quantiles. In a high dimensional Growth-at-Risk forecasting application, we forecast tail risks as well as complete forecast densities using
a database covering over 200 variables related to the U.S. economy. Quantile specific and density calibration score functions show that the horseshoe prior provides
the best performance compared to competing Bayesian quantile regression priors,
especially at short and medium run horizons.
Bayesian quantile regression models with continuous shrinkage priors are known to
predict well but are hard to interpret due to lack of exact posterior sparsity. Chapter 5 bridges this gap by extending the idea of decoupling shrinkage and sparsity.
The proposed procedure follows two steps: First, the quantile regression posterior is
shrunk via state of the art continuous shrinkage priors; then, the posterior is sparsified by taking the Bayes optimal solution to maximising a policy maker’s utility
function with joint preference for predictive accuracy as well as sparsity. For the
sparsification component, we propose a new variant of the signal adaptive variable
selection algorithm that automates the choice of penalization in the integrated tility
through a quantile specific loss-function that works well in high dimensions. Large
scale simulations show that, compared to the un-sparsified regression posterior, the
selection procedure decreases coefficient bias irrespective of the true underlying degree of sparsity in the data, and goodness of variable selection is competitive with
traditional variable selection priors. A high dimensional Growth-at-Risk forecasting
application to the US shows that the method detects varying degrees of sparsity
across the conditional GDP distribution and that the sources to downside risk vary
substantially over time.
Inspired by the work of Giannone et al. (2021) on the “illusion of sparsity” from
sparse modelling techniques, this chapter (6) investigates whether the recently popularised global-local priors, firstly, are implicitly informative about sparsity and,
secondly, whether they are able to communicate the true degree of sparsity from
the data. We consider two methods of analysis: implicit model size distributions
and sparsification techniques which are tested on a host of economic data sets and
simulations. The findings motivate a new horseshoe type model to which we add a
prior that makes it a-priori agnostic about the degree of sparsity and is shown to be
competitive to the spike-and-slab of Giannone et al. (2021) for forecasting as well
as sparsity detection.
Chapter 7 concludes with summaries, limitations of the thesis, as well as directions
for future research
Robust optimal control using computationally efficient deep reinforcement learning techniques
Get PDFWe investigate current challenges in the application of reinforcement learning (RL) to solve
subsurface flow control problem which is the subject of intensive research in the field of reservoir
management. In typical subsurface flow control problems, the system is partially observed
because the data is often only available at well locations. Furthermore, the model parameters
are highly uncertain as a result of the sparsity of available field data. As a result, we begin
by presenting an RL framework to solve the stochastic optimal control for predefined model
uncertainty and partially observable system. The numerical results are presented using two
state-of-the-art model-free RL algorithms, proximal policy optimization (PPO) and advantage
actor-critic (A2C), on two single-phase subsurface flow test cases representing two distinct flow
scenarios. We identify that computational intractability is one of the major limitations for
the proposed RL framework. This is because the model-free RL algorithms are by definition
sample inefficient and require thousands if not millions of samples to learn optimal control
policies. For subsurface control problems, this corresponds to performing a large number of
simulations, which is computationally quite expensive. Our aim is to build a more generalized
framework that can help alleviate this problem of computational complexity for the proposed RL
framework. This is achieved by employing multiple levels of models. Here, the level refers to the
accuracy or fidelity of the discretization of the domain grid of the underlying partial differential
equations. We propose two distinct approaches that can be used in the most generalized manner.
The first approach involves a more explicit modification of the proposed RL framework. In this
approach, a multigrid framework is proposed that essentially takes advantage of the principles of
sequential transfer learning. The second approach implicitly modifies the classical reinforcement
learning framework itself to take advantage of information from lower-level models. This is
achieved by modifying the classical framework of RL algorithms so that they use an approximate
multilevel Monte Carlo estimates as opposed to Monte Carlo estimates of policy and/or value
network objective functions.Engineering and Physical Sciences Research Council (EPSRC) Funding
Mechanistic studies of CO2 reactive transport in deep saline aquifers
Get PDFThe increasing in concentration of greenhouse gases (GHG) has been considered as the
main cause of global warming and carbon capture, usage and sequestration (CCUS) is an
option for addressing this challenge. Deep saline aquifers are known to have the highest
capacity for storing captured CO2. The CO2 injection in geological formations disturbs
resident phases equilibrium and therefore, increasing uncertainties about CO2 fate and
leakage risks arising from interactions with CO2 dissolved brine and reservoir rock.
However, there is not sufficient amount of information available in the literature for
investigating the effect of aqueous phase chemistry e.g., salinity and ion type/valency of
formation water on the extend of geochemical rock-fluid interactions. Additionally, the
intricate nature of the rock phase, characterised by its mineralogical composition and
porosity, coupled with the variability in the aqueous phase, specifically in terms of the
type and concentration of dissolved ions, have led to varying behaviours and, in some
cases, contrasting outcomes obtained concerning triggering or inhibiting geochemical
rock-fluid interactions. Hence, further exploration of these intricate environments and the
acquisition of comprehensive data on geochemical rock-fluid interactions within CO2
storage conditions would be a significant advantage. In this regard, this PhD study aims
to evaluate the effect of aqueous phase chemistry in terms of ionic strength (ranging from
0 – 0.65 M) and role of the most common types of dissolved cations (e.g., Na
+
, K+
and
Mg2+) on geochemical rock-fluid interactions. The outputs of this study can open a door
to a better understanding of the key drivers on CO2-saturated brine induced geochemical
reactions.
To fill this gap in the literature, the following investigations were carried out in this PhD
study:
1. Examining the impact of rock type on geochemical rock-fluid interactions.
2. Assessing the reactivity of different minerals with brines of varying ionic strength
under conditions resembling geological reservoirs.
3. Investigating the role of dissolved ion type/valency in the aqueous phase on the
geochemical reactions between formation brine and reservoir rocks.
4. Integrating geochemical modelling and laboratory batch experiment data to
further investigate the water chemistry alteration over time particularly when
sampling of test fluid is impractical.
In this regard, experiments were designed to be conducted using hydrothermal batch
vessels for determining mechanisms that induce rock dissolution and precipitation at
elevated pressure and temperature similar to subsurface reservoirs. This study is
conducted by utilising a non-destructive method, which is micro–Computed Tomography
(micro-CT) imaging while providing the opportunity to study the alteration of pore
structure and porosity in even submicron scale. Furthermore, Powder X-Ray Diffraction
(PXRD) has been conducted to analyse the type of minerals dissolved/precipitated and
Environmental Scanning Electron Microscopy (ESEM) in conjugate with Energy
Dispersive Spectroscopy (EDS) has been applied to obtain information on morphology
(e.g., physical structure, form, and appearance) and constituent elemental distribution on
rock surface, respectively. Two different kinds of rock samples which are representative
of subsurface formation were selected and prepared for experiments in order to evaluate
rock type-induced geochemical reactions. Initial analysing methods were conducted to
characterise the rocks’ properties (e.g., mineralogy, porosity and micro-scale pore
structure) using various software e.g., Qualx for analysing raw data of PXRD and Avizo
and Fiji for processing images acquired by micro-CT scanner and ESEM-EDS.
This research work illustrates the effect of chemistry of brine on geochemical rock-fluid
interactions under realistic conditions and respectively the following conclusions were
drawn:
a. In contact with different chemistries of brines, Berea sandstone preserved its
original structure, while Indiana limestone experienced dissolution of its outer
layer upon contacting CO2 dissolved brine due to higher reactivity of calcite.
b. Investigations showed that the degree of ionic strength of aqueous phase is
effective on the rate of rock dissolution and precipitation through increasing the
ion activity of H+
ion.
c. The enhanced dissolution of rocks was observed in the presence of ions with
higher valency or similar valency but smaller ion radii, related to the increased
electrical force.
d. The modelling results aligned well with experiments, indicating its potential in
forecasting CO2-induced rock-fluid interactions.
The present study focused on micro-scale batch tests to investigate geochemical rock-fluid interactions. For future research, dynamic in-situ micro-CT studies are suggested to
explore these interactions at elevated pressure and temperature conditions. Additionally, the validation of proposed mechanisms and theories can be achieved through molecular
level research using density functional theory and molecular dynamic simulations.James Watt Scholarshi
Production of whisky using specialty malts
Get PDFMost Scotch malt whiskies are produced using only pale pot still malt as a cereal
ingredient. Recently there has been interest for use of high-colour malts in whisky
production for the potential to develop novel spirit aroma properties, however, knowledge
around the challenges associated with this remains limited. The presented work
investigated the consequences for using high-colour malts across the whisky production
process. Malt roasted at laboratory-scale was shown to impact development of spirit
aroma volatiles when incorporated into a distiller’s grist at 30 % inclusion (w/w). Spirits
derived from roasted malt were elevated in heat-derived aroma volatile compounds such
as pyrazines, furans and phenolic compounds. Raw material selection proved important
with grain nitrogen correlating positively to pyrazine content in spirit and negatively to
furans. Use of roasted malts was found to impact development of fermentation congeners
by yeast and high-colour wash had increased levels of higher alcohols and fatty acid ethyl
esters as compared to pale malt wash. Industry-scale trials confirmed that roasted malt
volatiles could be recovered into spirit and retained during maturation (monitored for one
year). Spirits were sensorially distinct, with Brown and Chocolate-type malts producing
spirits diverging most clearly from a pot still malt reference. The present work contributes
substantially to the knowledge of applying roasted malts to the whisky production process
and highlights a potential role for these materials as a tool for expanding the range of
aroma properties achievable in whisky
Digital Mindset in the accounting of German medium-sized group companies and the implications for software implementations
Get PDFDigitisation is one of the biggest challenges for today’s economy and enterprises, resulting in the adaption of business models and change activities. However, numerous Digitisation projects fail. The main identified underlying reasons for these project failures are
essentially the rejection of change projects by employees, missing application of Critical
Success Factors (CSFs), and misleading change management.
This research builds on the Digital Mindset concept of Solberg, Traavik and Wong (2020)
and connects it with CSFs for software implementation projects. Thereby, it aims to assess the perception of finance and accounting professionals in Germany whether Digitisation is an opportunity or threat, and how individuals cope with it. Thus, in a deductive
approach, a quantitative online survey resulting in 394 data sets of finance and accounting
professionals, working for organizations across Germany, was conducted to assess the
Digital Mindset 2.0 distribution of professionals working in the field of finance and accounting in Germany. The data was analysed taking a critical realist stance.
The key findings of this particular research reveal the following. First, the distribution
shows that Digitisation is perceived as principally beneficial and that this population feels
capable of developing the necessary abilities. Second, Top Management Support and
Commitment, Project Resources and Objectives, and Employee Training are perceived as
the most important CSFs for software implementation projects. Third, the perceived importance differs across the different Digital Mindsets 2.0, age groups, and hierarchical
levels.
This research contributes to practise by the development of a standardised tool to assess
the Digital Mindset 2.0 of involved project team members to increase the success probability of the software implementation project. Furthermore, this research contributes to
theory by extending the initial Digital Mindset model into further sub-sets and extending
the database for the Digital Mindset concept
Exploring levers for agility and their inter-relations in the German energy industry via neo-configurational theory
Get PDFOrganisational agility describes firms’ ability to proactively and reactively handle external changes like the COVID and Ukraine crises. This thesis researches how levers
like culture (in this thesis = mindset) or strategy impact agility. Existing research shows
agility’s outcome but neglects its origin and its levers’ interactions. Since mindsets guide
employees and leaders, research was requested for how organisational culture influences
other levers’ effects. Therefore, this thesis developed a literature-based framework of
levers, tailored it to the studied context, proposing that strategy, technology, linkages, and
structures, filtered through employees’ and leaders’ mindsets, interact to lead to agility.
Neo-configurational theory (NCT) provided the theoretical underpinning for lever inter-relations, basing this research in wider organisational theory. As critical realist work, the
thesis recognised agility’s context-specificity and examined the recently turbulent German
energy industry as exemplary context. 36 semi-structured interviews in 15 purposefully
sampled companies were analysed in three steps: All data were thematically analysed.
Fuzzy-values were derived using the Generic Membership Evaluation Template (GMET).
The concluding fuzzy set Qualitative Comparative Analysis (fsQCA) determined pathways to agility and non-agility, levers’ interdependencies, and mindset’s role.
The results show that agility presupposes an implemented agile strategy (i.e. strategy
filtering agility) but not necessarily a very agile culture, while non-agility comes with
a very non-agile employee mindset (i.e. culture filtering non-agility). Three strategy-dependent paths to agility exist for energy companies: one builds on internal and external linkages, one on lacking technological capabilities with improvement spirit, and one
couples agile employee mindsets with decentralised structures. Three employee mindset-dependent paths describe non-agility: one builds on lacking linkages and supportive leadership, one on lacking technological capabilities, supportive leadership and strategy, and
one on lacking technology capabilities reflecting in inadequate structures. This thesis’
major methodological contributions are refining the GMET as new tool to transform qualitative data into fuzzy-values and further establishing fsQCA in management research.
Academics gain a sound theoretical basis for agility in form of NCT and practitioners
and academics a view on agility levers’ role, especially on culture and strategy. Utilities’
managers can use this to prioritise levers facing sudden changes
Toward deep monocular view generation and omnidirectional depth estimation
Get PDFThis thesis proposes new strategies for obtaining environmental depth representations from monocular perspective and omnidirectional vision. This research is
inspired by the necessity for mobile autonomous systems to be able to sense their
surroundings, which is frequently abundant in vital data necessary for planning,
decision-making and action.
The methodologies presented here are primarily data-driven and based on machine learning, specifically deep learning.
Our first contribution is the generation of top-down, “bird’s eye view” representations of detected vehicles in a scene. This was achieved using only monocular,
perspective view images. The novelty here was via an adversarial training scheme,
which our experiments showed resulted in more robust models versus a strictly supervised baseline.
Our second contribution is a novel method for adapting view synthesis-based
depth estimation models to omnidirectional imagery. Our proposal comprise three
important facets. Firstly, a "virtual" spherical camera model is integrated into the
training pipeline to facilitate model training. Secondly, we explicitly encode information of the spherical nature of the image format by adopting spherical convolutional layers to perform convolution operations, consequently compensating for the
significant distortion. Thirdly, we propose an optical flow-based masking strategy
to reduce the impact of undesired pixels during training, such as those originating
from large, challenging visual areas of the image such as the sky. Our qualitative
and quantitative findings indicate that these additions result in improved depth
estimations versus earlier methods.
Our final contribution, broadly, is a method for incorporating LiDAR information into the training pipeline of an omnidirectional depth estimation model. We
introduce a Bayesian optimisation-based extrinsic calibration method to match LiDAR returns with equirectangular images. Primarily, we weight the incorporation
of this data via a frequency-based scheme dependent on the number of detected
LiDAR projections. The results from this show that there is a tangible quantitative
benefit in doing the aforementioned
CyPhER : a digital thread framework towards human-systems symbiosis
Get PDFCyber-physical twinning is an important area of study across multiple diverse fields. Creating
more symbiotic human-machine partnerships facilitates extended reality. This thesis presents
a flexible digital thread framework, CyPhER (Cyber Physical Extended Reality), as a platform
and application agnostic solution for human-systems symbiosis. This framework includes
software, techniques, and a reference architecture to allow for implementation in any field
where cyber-physical twinning is possible.
This thesis contains case studies carried out with industry partners in the domains of vocational
education and robotics. These case studies demonstrate extended reality enabling human-systems symbiosis within their fields. When moving between these fields, CyPhER itself
evolved, improving in terms of performance and capability. These applications required
CyPhER to be deployed on a range of platforms spanning operating systems and form factors,
which influenced its performance across these devices. Having flexibility in this approach
allows CyPhER to address barriers in terms of computing apparatus in each field, such as edge
devices.
A cyber-physical extended reality is beneficial as a teaching aid, supporting a symbiotic
process where both students and tutors can benefit from a teaching environment which utilises
both the real and virtual worlds. It also benefits the field of automation, allowing for a symbiotic
partnership between the human operator and systems. This is achieved through bidirectional
interactions between robots and humans to enable enhanced operational decision support.
Approaching these applications with a cyber-physical solution has enabled gains in usability,
flexibility, and scalability in each field, abstracting complex systems with extended reality
features to enable symbiosis between systems and the humans that control them. This is
demonstrated in the consideration of control display gains in human-system interaction, which
addresses the interaction barrier between the human and the system.Funded by Heriot-Watt Universit
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!