Heriot-Watt University
ROS: The Research Output Service. Heriot-Watt University EdinburghNot a member yet
4689 research outputs found
Sort by
Fast hyperparameter optimisation of graph neural network for molecular property prediction
Full text linkIn the evolving domain of graph neural networks, there is a growing effort focused on
predicting molecular properties. However, a noticeable gap persists, as much of the
research overlooks the comprehensive exploration of hyperparameters—a crucial aspect
for achieving positive outcomes in graph neural network applications. This underscores
the vital role of hyperparameter optimisation, despite the challenge posed by resource-intensive procedures. To address this gap and overcome the challenge, our study achieves
significant advancements. Firstly, we summarise graph neural networks for molecular
property prediction into a structured framework, systematically identifying key hyperparameters for optimisation. Secondly, we introduce an innovative hierarchical evaluation
strategy embedded in a genetic algorithm named HESGA, expediting optimisation by
early elimination of unpromising solutions. This approach demonstrates improved efficiency and cost-effectiveness compared to traditional Bayesian optimisation. Thirdly, we
propose the implementation of a binary tree to model the hyperparameter space, further
enhancing HESGA’s effectiveness. Lastly, guided by empirical insights, we present a hybrid evaluation strategy that surpasses advanced optimisation methods, demonstrating
reduced computational costs and accelerated optimisation. Overall, our research not
only addresses the challenge of elevated computational expenses in hyperparameter optimisation but also enhances graph neural network performance, effectively bridging the
research gap in hyperparameter optimisation for graph neural networks in the context
of predicting molecular properties
Optical metasurfaces for generating and manipulating unusual optical vortex beams
Full text linkOptical vortices (OVs) carrying orbital angular momentum (OAM) have attracted considerable
interest in the field of optics and photonics owing to their peculiar optical features and extra
degree of freedom for carrying information. Although there have been significant efforts to
realize OVs using conventional optics, it is limited by large volume, high cost, and lack of design
flexibility. In recent years, optical metasurfaces have attracted tremendous interest due to
their unprecedented capability in the manipulation of the amplitude, phase, polarization, and
frequency of light at a subwavelength scale. Optical metasurfaces have revolutionized design
concepts in photonics, providing a new platform to develop ultrathin optical devices for the
realization of OVs at subwavelength resolution.
Inspired by plant grafting, grafted vortex beams can be formed through grafting two or more
helical phase profiles of optical vortex beams. Recently, grafted perfect vortex beams (GPVBs)
have attracted much attention due to their unique optical properties and potential
applications. This thesis proposes and experimentally demonstrates a compact metasurface
approach for generating and manipulating GPVBs in multiple channels which eliminates the
need for such a complex optical setup. For the first time, a single metasurface is utilized to
realize various superpositions of GPVBs with different combinations of topological charges in
various channels, leading to asymmetric singularity distributions. The positions of singularities
in the superimposed beam can be further modulated by introducing an initial phase difference
in the metasurface design.
Next, the concept of GPVBs is extended to grafted perfect vector vortex beams (GPVVBs), a
new type of perfect vector vortex beams (PVVBs) with inhomogeneous polarization and spiral
phase profiles. This approach overcomes the limitation of the number of topological charges
(TCs) in the involved vortex beams. Hybrid GPVVBs are generated through the superposition
of new hybrid GPVBs with a novel multifunctional metasurface. The generated hybrid GPVVBs
possess spatially variant rates of polarization change in 2D space due to the involvement of
more TCs. Remarkably, each hybrid GPVVB features multiple different GPVVBs in the same
beam, adding more design flexibility. Furthermore, these beams are dynamically tuned with
a rotating half waveplate, making the metasurface function as a dynamic optical device.
Various generations and superpositions of Ince-Gaussian beams (IGBs) are also realized.
Multiple phase and polarization singularities are observed in the superimposed IGBs through
the superposition of IGBs with even and odd modes, a task that is extremely challenging to
achieve with conventional OVs and GVBs. The polarization profile of resultant vector beams
can be further modulated by controlling the polarization direction of the incident light.
The compactness of the developed metasurface devices and the unique properties of the
generated beams have the potential to impact many practical applications such as particle
manipulation, OAM spectrum manipulation, singular optics, quantum science and optical
communications, and optical encryption.James-Watt Scholarshi
Pressure profile prediction in building drainage system using artificial neural network (ANN) modelling
Full text linkThe mechanisms of fluid flow phenomena found in Building Drainage Systems
(BDS) in the analysis of transient flow of air and water even though are grounded,
needs attention with respect to its modelling, as it is quite relevant with respect to
safe removal of the waste from the building and to avoid the trap seal depletion.
This research seeks to predict the classical pressure profile for a range of BDS
configurations using a number of available parameters such as pipe diameter, water
discharge height, water discharge flow rate and overall building height using
Artificial neural network (ANN) algorithms, optimised for BDS systems for the first
time. Experimental data from peer reviewed literature and data from a unique 32-
storey building drainage test rig have been used as pressure profile data (Target data)
for an Artificial Neural Network (ANN) model. Discharge flow rate and height are
considered to be the two independent input parameters, and the pressure along the
vertical stack is considered to be the output parameter. In this work, both the Feed
Forward Back Propagation (FFBP) ANN model and the Radial Basis Function
(RBF) ANN model have been used to train, test, and validate the respective
algorithms. Subsequently, a FF-PSO algorithm has been employed to reduce the
intrinsic error in the Feed-Forward model by refining the weights and biases. The
work has confirmed the applicability of all the tested models for steady two-phase
fluid flow phenomena in BDS in different configurations. Further, the FFBP ANN
model has been employed to establish a relationship among pressure profiles for the
same discharge coming from different floors. The prediction of pressure profile of
a BDS has also been modelled using the pressure profile of other BDS. In order to
capture the change in characteristics of the data, a hybrid model with segregated
data for dry-stack and wet-stack zone has also been employed. It is surmised that
this model could be trained with a database of real-world system data in the future
Examining ERP enhancement management in Israeli SMEs
Full text linkExisting literature shows that there are gaps between the standard functionality provided
by Enterprise Resource Planning (ERP) systems and the specific functionality required
by implementing companies. These gaps, known as misfits, are closed by means of
enhancements – changes enabled in the system beyond the initial implementation
configuration. As business requirements are constantly evolving, the longer a company is
in the post-implementation phase of the ERP life cycle, the greater is the need for further
enhancements that were not envisaged during the initial implementation. These
enhancements can vary in complexity from innovative uses of existing functionality
through to the addition of complete, bespoke, modules. Each enhancement should
produce a benefit that leads to the achievement of strategic or tactical objectives.
This research presents the case for defining the discipline of ERP Enhancement
Management by investigating how it is implemented within one company. This research
is carried out by means of the Action Research approach that permits the collaboration
between the researcher and those being researched.
From the analysis of three case studies is developed a checklist consisting of 17 steps that
need be taken for the successful management of the development and deployment of
enhancements. The model suggests three Critical Success Factors: top management
support, active user participation, along with training and documentation. The model is
then validated by the successful management of a fourth case study.
This research is applicable to manufacturing, retail and service companies
Why institutional investor expectations on the speed of the energy transition matter : a complex systems perspective on sustainable investment behaviour
Full text linkThe sustainable finance gap continues to widen despite the urgent need for a sustainable
energy transition. Solutions to closing this gap tend to overlook the complexities
associated with the energy transition and the role of investor expectations on the speed of
the energy transition in investment behaviour. To address these issues, this thesis firstly
aims to develop a framework linking investor expectations on the speed of the energy
transition and investment behaviour, from a complex systems perspective in a UK
context. This can be used for understanding key sustainable finance issues and creating
appropriate solutions and policies. This thesis also aims to develop a method for testing
the effects of potential policies on expectations and behaviour in a complex environment,
under various scenarios. To fulfil these aims, a unique combination of methods is used
in the context of sustainable finance involving causal mapping and gamification.
The results show that investor expectations can significantly impact investment behaviour
in some cases, particularly when energy transition expectations turn negative. This occurs
when expectations work simultaneously along multiple pathways and/or feedback loops
are initiated, which can trigger non-linear effects in the system. The results also show
that significant change in investor expectations and investment behaviour can be induced
under scenarios connected to integrating sustainability into valuation, a reversal of
support for the energy transition, and through particular scenario combinations. These
findings imply that investor expectations on the speed of the energy transition need to be
managed carefully when implementing policies and regulations, as there can be adverse
effects to sustainable finance flows if expectations turn negative. It also highlights
potential scenarios and policies that can trigger changes in sustainable investment
behaviour
Vibroacoustic sensing of the mechanical properties of skin
No full textRestricted access theses until 30/04/2027
High-velocity compressible gas flow modelling in porous media
Full text linkCompressible gas flow through porous media is a complex phenomenon with
significant implications in various energy-related processes. This thesis addresses the
need for accurate numerical simulations to capture the complexities associated with
compressible flows, particularly focusing on the behaviour of choked flow associated
with high-velocity compressible gas flows. Choked flow occurs when gas velocity
approaches or exceeds the velocity of sound during which the mass flow rate reaches a
limiting value with no further increase with an additional increase in pressure drop across
the flow conduit. The research aims to deepen the understanding of gas transport in porous
media, which is important for various energy production and storage scenarios such as
carbon capture utilisation and storage (CCUS), cyclic hydrogen injection and production,
natural gas production, coalbed methane (CBM) resources, and compressed air energy
storage (CAES).
The research begins with an overview of the importance of gas flow in porous
media and the challenges it presents due to gas compressibility and pressure-dependent
thermophysical properties. The study investigates the dynamics of choked flow and
shockwave phenomena in porous media at both microscopic and macroscopic scales.
Microscopic behaviours pertain to fluid flow at the pore scale, whereas macroscopic
behaviours refer to the average flow characteristics observed over larger scales. The
research delves into the interplay between driving mechanisms, pore-throat shape,
boundary conditions, and pressure gradients, aiming to develop accurate numerical
simulators and predictive equations.
The thesis presents novel methodologies for quantifying and incorporating the
effects of choked flow and shockwaves in numerical simulations and experimental
analyses. It introduces modified quantification methods for pore-scale choked flow and
proposes core-scale analytical solutions for transient highly compressible gas flow
through porous media. In the proposed revised pore scale quantification, the gas
compressibility was determined by considering the minimum pressure value within the
capillary, leading to a redefined parameter referred to as "modified alfa." Subsequently,
the modified alfa parameter was utilised to evaluate the permeability. The core scale
analytical solutions incorporated quadratic pressure gradient and variable compressibility
terms to enhance applicability across various high-pressure gas flow scenarios. The
analytical solutions are validated through laboratory experiments, ensuring the consistency and accuracy of the findings. A comparison with previous simplified models
revealed an approximate 45% error in very high-pressure scenarios (80-10 bar).
This research further validates the analytical results by conducting improved
transient core scale experiments that eliminate slip flow and net stress effects. Additional
steady-state permeability measurements were performed to confirm the consistency of
the transient pressure pulse decay (PPD) measurements.
The research further explores the potential application of the findings here for
Pore Network Modelling (PNM) to incorporate the complexities of choked flow. A
dimensionless analysis is conducted to determine the dominant parameters affecting the
choked flow conditions. Accordingly, numerous 2D and 3D simulations are performed to
present separate formulations for the calculation of critical pressure ratio for choked flow,
and the corresponding limiting mass flow rate. The integrity of these formulations is
verified by comparing them against both the data used in their development and the data
not used. The application of these formulations is demonstrated using a simple 3D PNM.
The results show a significant reduction in numerical errors compared to traditional PNM
approaches.
In summary, this thesis provides valuable insights into the behaviour of
compressible gas flow through porous media and offers avenues for further research to
improve our understanding and prediction capabilities in this important field
Implications of reactive transport on the hydraulic and frictional properties of granites in geothermal systems
Full text linkOne of the biggest challenges for humanity in the 21st century is climate change for which
one of the driving factors are high carbon dioxide emissions from the energy sector. For
this reason, a lot of effort and research aims to accelerate the energy transition towards
renewable energy sources. In addition to wind or solar energy, harnessing the Earth’s
geothermal heat can help diversify national green energy portfolios, particularly as
geothermal energy projects can deliver a continuous supply of heat and/or power. Faults
in high-heat-producing granites are targeted in Cornwall as reservoirs for the production
of heat and electricity. These faults have been subject to natural geochemical alteration.
In addition, geothermal operations will disturb the geochemical equilibrium between
fluids and the host rock following fluid production and reinjection, resulting in additional
alterations.
To investigate the effects of these geological and engineered alterations, we collected
granite samples from a mine in the Carnmenellis granite in central Cornwall, a granite
that is targeted in a major deep geothermal project. The granite samples were collected
from different fault zones in the mine and underwent different degrees of argillic
alteration. From three consolidated rocks we prepared intact and fractured plugs that we
characterised for porosity, tensile strength, and permeability. Surprisingly, we found that
increasing degrees of alterations increase matrix and decrease fracture permeability. This
potentially results in a shift of preferential flow pathways from fracture to matrix
controlled. Furthermore, the altered sample showed higher porosity, which indicates a
higher storage potential for fluids than in the unaltered granite.
To study the effect of alterations on fault stability, we prepared rock powders from five
rock samples – four from one fault and one from another parallel fault for comparison.
We conducted direct shear experiments on the powder samples to measure friction
coefficient and Rate-and-State Friction parameters. We found that with a higher degree
of argillic alteration frictional strength decreases and frictional stability increases.
Consequently, alterations increase the likelihood of aseismic slip and potentially destress
geothermal systems. The fault core sample diverged from this trend due to an additional
alteration mechanism, which showed fault systems to be complex.
For geothermal systems in faulted granites, these results imply that it can be beneficial to
explore for zones of argillic alteration for two major reasons:
1. Low to intermediate degrees of argillic alteration increase porosity and matrix
permeability in the granite and therefore increase reservoir storage potential. If
alteration zones of two parallel faults intersect it can further increase reservoir
connectivity.
2. At higher degrees of alteration, the additional porosity might collapse, but the
change in frictional properties reduces risk of induced seismicity. Such altered
zones in proximity to the actual reservoir fault can help destress the system
through promoting aseismic slip.
Lastly, a new experimental apparatus is presented that allows more in-depth investigation
of thermo-hydro-mechanical-chemical (THMC) couplings in subsurface media, by
discussing its potential and limitations. We used various case studies conducted on granite
and sandstone to present its measurement capabilities. We further conducted initial
leaching experiments on Carnmenellis granite using deionised water over long periods of
time. Fracture permeability recovered, while the fluid compositions showed evidence for
biotite and plagioclase dissolution. This work showed the successful measurement of a
slow chemical effect on granite permeability in the laboratory while hopefully inspiring
future experiments in the new setup that aim to understand complex interactions in
geothermal systems.Engineering and Physical Sciences Research Council (EPSRC) funding
Company investment announcements and the sustainability agenda
Full text linkThis thesis empirically investigates the impacts of environmental (E), social (S), and
governance (G) credentials, managerial tone, and information complexity on the stock market’s
valuation of corporate investment decisions, with and without a sustainability agenda, in the
UK from 2013 to 2021. The motivation behind this study is derived from the exposure of listed
companies to investor scrutiny of sustainability strategies and disclosures to identify whether
or not markets encourage the sustainability agenda. Using company investment announcements
as the basis of investigations, this study examines the stock market reaction to a set of
sustainable and non-sustainable investments. An event study is conducted to evaluate the
abnormal returns to the investment announcements. The study addresses how a set of firm-specific ESG credentials affect the stock market’s valuation of investment decisions.
Additionally, this study examines the effect of the tone conveyed by managers and the
information complexity of investment announcements on the stock market’s valuation of
investment decisions. Textual analysis using the Loughran and McDonald dictionary is
employed in examining managerial tone and information complexity. The empirical tests of
hypotheses are conducted using pooled ordinary least squares (OLS) regressions.
The findings reported in this thesis indicate that the stock market positively values sustainable
investment, although slightly lower than their non-sustainable counterparts. The discount on
market valuation is pronounced for environmental and social credentials and weak ESG
engagements increase the market’s valuation of corporate investment decisions. Of the two key
environmental measures (resource use efficiency and emissions scores) examined, weaker
emission credentials increase the stock market valuation of corporate investment decisions with
a sustainability agenda, whereas weaker emission credentials reduce the stock market valuation
of corporate investment decisions without a sustainability agenda. Furthermore, the effect of
governance credentials, particularly governance scores (G), on the stock market’s valuation of
investment decisions is impacted by profitability, leverage, growth opportunities, and
sustainability-based compensation. Finally, the tone employed in investment announcements
by managers contains incremental information that the stock market responds to, particularly
in investments with a sustainability agenda. However, high information complexity reduces the
market’s valuation of investment decisions, but slightly less so for sustainable investments. The
evidence documented in this thesis has implications for allocative efficiency and suggests that
new sustainable investments should be encouraged for pecuniary reasons
Machine learning solutions for intelligent reflecting surface-assisted communications
Full text linkIntelligent reflecting surface (IRS) has been recognised as a promising technology for future wireless systems. Consequently, several conventional optimisation
schemes have been proposed for solving IRS-assisted communication problems.
However, these conventional optimisation methods come with high computational
complexity, especially for the multiple-input-multiple-output (MIMO) communication systems. Fortunately, it has been shown in the literature that machine
learning (ML)-based schemes have been successful in tackling diverse problems in
the wireless communications domain with reduced complexity. Motivated by this,
in this Thesis, ML-based solutions are proposed for IRS-assisted MIMO communication systems aiming to maximise the spectral efficiency and/or secrecy rate
of the proposed systems. First, a deep learning (DL) based model is proposed to
jointly optimise the hybrid beamformers and the IRS reflection matrix for an IRS-aided MIMO communication system wherein a two-stage neural network is trained
sequentially to estimate the IRS matrix (from a formulated effective channel), as
well as the hybrid beamformers while aiming to maximise the spectral efficiency of
the system. In the second and final contributions, a deep reinforcement learning
(DRL) algorithm is proposed to optimise the phase shift of the reflecting elements
of an IRS for IRS-assisted MIMO communication systems aiming to maximise the
system’s spectral efficiency and secrecy rate respectively. Specifically, the deep
deterministic policy gradient (DDPG) framework is proposed to tackle the formulated non-convex problems following its success in handling high-dimensional
continuous action spaces and tackling non-convex optimisation problems. Numerical simulations are provided to validate the competitive performance of the
proposed ML-based solutions
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!