Cellulose nanocrystals (CNCs) have emerged as highly effective solid stabilizers for Pickering emulsions and latexes, offering a sustainable alternative to conventional surfactants. Their amphiphilic nature allows partial wettability at the oil-water interface, creating robust colloidal stabilization in fully waterborne systems. Because CNCs originate from renewable lignocellulosic biomass and can be surface-modified through abundant hydroxyl groups, they provide a versatile platform for designing green, bio-based polymer composites with tailored interfacial and mechanical properties.

This thesis develops and investigates CNC-stabilized waterborne non-isocyanate polyurethane (WNIPU) and polyacrylate latexes with the goal of achieving high colloidal stability and excellent material performance. The work focuses on optimizing latex synthesis, engineering CNC surface functionalities, elucidating CNCs-polymer matrix interactions, and assessing the structural and mechanical properties of the resulting latexes and composites.

In the first part, pristine CNCs were utilized to prepare WNIPUs through CNC-stabilized suspension polymerization, eliminating hazardous isocyanates, volatile organic compounds (VOCs), and external surfactants. In Chapter 2, CNC-mediated Pickering stabilization of WNIPU latexes was successfully achieved for the first time, with systematic optimization of CNCs concentration, monomer composition and reaction parameters. The obtained WNIPUs also exhibited florescence, indicating potential for sensing or anti-counterfeiting applications. Chapter 3 demonstrated the dual role of CNCs as both stabilizer and reinforcing agent for the WNIPU latex. The incorporation of 17 wt% CNCs increased probe tack adhesion strength by 680% and lap-shear strength by 340%, enabled by their uniform dispersion within the WNIPU matrix.

The second part of the thesis focused on functionalized CNCs as advanced stabilizers for waterborne latexes. In Chapter 4, CNCs were covalently modified with octylamine (oCNCs) to increase surface hydrophobicity. The oCNCs-stabilized WNIPU showed improved stabilization efficiency and smaller sizes in monomer emulsion droplets. The resulting WNIPU/oCNC composites showed up to 34% and 57% increases in probe tack adhesion strength and work of adhesion, respectively, compared to pristine CNC-stabilized WINPU. Chapter 5 introduced a non-covalent surface modification using charge coupling method, which rendered CNCs sufficiently hydrophobic to stabilize polyacrylate latexes with solid contents up to 20 wt%. The resulting polyacrylate/CNC composites demonstrated coating performance comparable to surfactant-stabilized references.

Overall, this thesis establishes CNC-stabilized WNIPU latexes as a new class of Pickering polymer systems and extends the concept to polyacrylate matrices. It elucidates how CNC surface chemistry governs dispersion stability, interfacial stabilization, and composite performance. These findings provide a basis for designing scalable, surfactant-free, and sustainable waterborne polymer systems, advancing the use of functionalized nanocellulose in environmentally friendly coatings and adhesives.

Construction of pedestrian timber bridges is an important step towards creating a more sustainable future. However, bridges with resonance frequencies close to the pedestrian pacing frequencies can be susceptible to uncomfortable vibrations. Previous research work has shown that numerical models often require specific adjustments to agree with experimental results. The purpose of the present research project has therefore been to perform experimental and numerical dynamic analyses of five pedestrian timber bridges at different construction stages to increase the general knowledge regarding the implementation of more accurate finite element models.

The results showed that calibration of the longitudinal stiffness at the supports were required for the girder and truss bridges in Papers I-II. The connection stiffnesses between the timber members were required to calibrate the truss and arch bridges in Papers III-IV. The stiffness of the pile foundations was implemented in Paper III to calibrate the first bending mode. A simplified model of the pile foundations by modelling the soil with springs provided adequate results compared to a detailed model with solid elements. The partial composite action of the mechanically connected arch segments and vertical web members was quantified from laboratory experiments and was subsequently implemented in the numerical model of the finished arch bridge in Paper IV, which reduced the stiffness of the first lateral and torsional modes. The reduced axial stiffness of the stays due to their deformed catenary shape was implemented to fine-tune the first bending mode for the cable-stayed bridge in Paper V. The asphalt could generally be modelled as a mass at warm temperatures, but consideration of the asphalt stiffness was required at cold temperatures. Certain structural aspects such as the asphalt continuity at bridge ends and continuity between individual cross-laminated timber elements were also introduced. Railings with in-plane stiffness affected the mass and stiffness of the bridges equally much. The damping ratios typically increased with an asphalt layer on the bridge, especially for modes of vibration with large deformation of the asphalt. These damping ratios were in many cases considerably higher than the values from technical guidelines.

Several model uncertainties were identified and discussed such as the variability in material properties and stiffness definitions as well as climate variations between the construction stages. However, the aforementioned main factors that affected the dynamic properties of each bridge were established. The main conclusion is that most bridges required detailed consideration of certain structural aspects to achieve calibrated results. 

Membrane proteins constitute 27% of the human proteome and play important roles in cellular processes. Dysfunction of membrane proteins causes and contributes to diverse diseases. The study of the structure and function of membrane proteins will help us to a deeply understand them. Cryo-electron microscopy (cryoEM) single-particle analysis (SPA) can provide us not only with high-resolution structures of membrane proteins but also give us the possibility to study protein’s dynamics and heterogeneous complexes they form. In this thesis, we applied SPA together with electrophysiology, surface plasmon resonance (SPR) and Microscale thermophoresis (MST) to investigate the structural and functional properties of several membrane proteins and protein complexes. 

In Paper I, we studied Pannexin 1 (PANX1), an ATP release channel involved in neurological disorders and inflammation. Electrophysiology and mass spectrometry identified Y308 as a key phosphorylation site. CryoEM structures revealed that phosphorylated PANX1 and the PANX1 R75A mutant possess conformational flexibility in the transmembrane domain. Our data show that PANX1 transits between a narrow and a wide state. This transition causes the N-terminal region to be ordered (wide) or disordered (narrow). The phosphomimetic Y308E mutant only shows a wide conformation with an ordered N-terminus. Electrophysiology shows this mutation converts PANX1 into a constitutively open channel. This suggests Y308 is a phosphorylation site that locks PANX1 into the wide conformation, enabling ATP release. 

Paper II presents a modified workflow for identifying and determining the structures of membrane proteins in a heterogeneous sample; without prior knowledge of the protein sequence or a model. We used a conventional suing SPA workflow to obtain cryoEM maps. These maps were modelled unsupervised with ModelAngelo and a HMMER search was done to obtain sequence information. Our data allowed us to generate three different maps from a heterogenous sample. We were able to identify each protein only based on the maps. The results were supported by mass spectrometry. The three E. coli membrane proteins were cytochrome bo3 oxidase (2.72 Å), AcrB (3.27 Å), and a previously uncharacterized E. coli ArnC protein (2.72 Å). Another important finding of this study is that MSP2N2 nanodiscs adapt their assembly around different proteins, implying that scaffold architecture depends on protein interactions. 

In Paper III, we examined SMCT1 interactions with PDZK1. SMCT1 is an important membrane protein that transports monocarboxylate substrates like butyrate, lactate or niacin. It was thought that PDZK1 is an important regulator that influences the transport rate of SMCT1. We tested the affinity for SMCT1 with various assays. Our pull-down assay, SPR, and MST assays revealed only weak binding, suggesting the interaction is likely physiologically irrelevant except under conditions of local PDZK1 enrichment. 

Overall, this work demonstrates the power of SPA for resolving dynamic membrane proteins, providing structural and functional insights, and establishing workflows for studying unknown or heterogeneous protein complexes. 

In this thesis, we develop and extend information-theoretic methods for analyzing reinforcement learning, focussing particularly on bandit problems. Our aim is to understand how learning algorithms manage the fundamental trade-off between exploration and exploitation, and how information-theoretic quantities such as entropy and mutual information can be used to characterize and bound their performance. Building on the framework of Russo and Van Roy, which relates Bayesian regret of an algorithm to its learning efficiency, measured by the information ratio, we derive new results that broaden the scope of this approach to more complex and structured learning settings.

The first part of the thesis revisits the analysis of linear bandits, a model in which rewards depend linearly on the selected action via an unknown parameter. While existing information-theoretic analyses provide general regret bounds, they all assume finite action sets. We close this gap by introducing refined analyses and show that near-optimal regret bounds can be obtained even for infinite or continuous action spaces. 

The second part of the thesis extends the information analysis framework to more complex bandit models. In the contextual bandit setting, we build on the lifted information ratio and extend its application to more general reward classes. Additionally, we provide alternative proofs based on classical information-theoretic tools. This new derivation reveals simplifies the analysis and enables sharper guarantees for structured problems. We then study the Thompson Sampling algorithm for logistic bandits, a widely used model for binary rewards. We derive the first regret bounds that avoid exponential dependence on the logistic slope parameter and are independent of the number of actions, resolving a long standing open question in the field. 

The third part of the thesis takes a step beyond bandit models and investigates how information-theoretic principles can be extended to more general reinforcement learning problems. We study the theoretical performance limits and learning guarantees of model-based reinforcement learning, establishing a framework for analyzing its Bayesian regret. In parallel, we derive PAC-Bayesian bounds for offline decision-making, including improved guarantees for offline bandits that are parameter-free and achieve near-optimal rates.

Together, the contributions of this thesis provide a coherent extension of information-theoretic analysis to a wide range of learning settings; from linear to nonlinear, from discrete to continuous, and from bandits to reinforcement learning. 

How early should assembly considerations shape modular product architecture?

This thesis addresses the question by reframing Design for Assembly (DfA) from a downstream optimisation task into a design logic embedded at the start of modular product development. In many industrial settings, modularisation is shaped by customer values and market variety, while production and assembly remain implicit or postponed. When these factors are left until later, the chance to create automation-ready, lifecycle-resilient architectures and to enable reconfigurable manufacturing systems (RMS) creation is often lost.

The research develops an expanded Modular Function Deployment (MFD) method that integrates DfA into the earliest stages of decision-making. MFD is well established as a way to structure architectures around customer needs, but it has usually treated assembly as a concern for later. This has led to DfA being applied reactively, once architectural choices are already locked, which limits its ability to influence strategic or system-level design.

To change this, the thesis introduces a new dual framing: module-level DfA (mDfA) for guiding the early selection and grouping of technical solutions, and architecture-level DfA (aDfA) for evaluating spatial layout, sequencing logic, and assembly complexity once a candidate architecture exists.

This framing is made practical through a set of lightweight, prescriptive tools designed to fit within the standard MFD process.

1. DfA-based internal evaluation criteria for concept selection.
2. Assembly-oriented module drivers within the Module Indication Matrix.
3. A coded interface taxonomy to structure and retain assembly knowledge.
4. The Assembly Directions and Connections Draft (ADCD) for improving the planning of spatial logic and insertion directions.
5. The Module Set Assembly Strategy Matrix (MSASM) for the evaluation of module-set complexity and automation potential.

These supports allow teams to analyse assembly implications before geometries are fixed, making it easier to align modularity with production realities.

The research follows a design research methodology (DRM), combining literature synthesis, industrial case studies, expert workshops, and applications in graduate-level engineering education. The tools were tested in both greenfield and brownfield contexts, in sectors ranging from professional equipment to consumer products. Results show that they help bring assembly consequences into view earlier, improve interface considerations, and strengthen cross-functional alignment.

The contribution is twofold. Theoretically, it introduces the dual framing of module-level and architecture-level DfA, extending assembly reasoning from part-level simplification to architectural planning. Practically, it delivers a workflow that supports production-aware modularisation without requiring high digital maturity or large resource investments. By enabling adaptable, automation-ready architectures that align with lifecycle goals, the work contributes to long term manufacturing resilience and as a consequence connects to United Nations’ Sustainable Development Goals (SDGs) 7, 9, and 12.

Graph representation learning has become a powerful paradigm for modeling structured data, enabling machine learning systems to reason over relationships, spatial dependencies, and topological patterns. However, its potential in wireless networks remains underexplored, particularly in learning native representations of complex and dynamic wireless environments. This thesis addresses the challenge of applying graph representation learning (such as graph neural networks and transformer architectures) to wireless systems, where topology, domain heuristics, and physical constraints critically impact optimization performance and generalization.

The core problem investigated is how to construct and exploit graph representations that faithfully encode the native structure of wireless networks to enable scalable, topology-aware optimization. This includes coverage relations, interference patterns, and environment-specific propagation effects. Existing solutions in wireless machine learning often overlook these structural priors, resulting in brittle models that generalize poorly across different network deployments.

This thesis introduces a graph-centric methodology to bridge this gap. By representing wireless elements—such as base stations, links, and coverage zones as nodes and their interactions as graph edges, we develop learning architectures that integrate attention mechanisms, domain-aware features, and physics-inspired constraints. Four studies demonstrate the approach across key tasks: routing latency prediction, antenna tilt optimization, real-time radio coverage estimation, and neural ray tracing for link-level modeling.

Our results suggest that these graph-based models significantly outperform traditional baselines, achieving near-simulator accuracy with improved generalization across unseen topologies and user scenarios. They also uncover a correspondence between engineering practices and graph spectral properties, offering a new lens for understanding network design. The proposed methods reduce supervision needs and support scalable deployment across variable network configurations.

Overall, this thesis establishes graph representation learning as a foundational tool for wireless intelligence, enabling structure-informed, optimization-driven modeling across diverse network conditions. These advances pave the way towards future wireless foundation models capable of supporting a wide range of optimization, sensing, and decision-making tasks with minimal retraining.

Environmental data serves as one of the principal mechanisms for making sense of and organizing sustainable futures in the face of rampant climate change. In the context of everyday life, such data involves notions such as carbon footprints and energy usage metrics. These metrics operate across various scales - from the macro level of United Nations and the Sustainable Development Goals to the micro level of eco-feedback systems for individuals. The underlying assumption is that presenting more data will prompt action. However, this approach has proven inadequate for achieving the scale of change required to address climate crisis and overconsumption. The limitation stems partly from how we relate to and make meaning with environmental data. The presumed objective nature of data invites particular relations to, and sensemaking processes with the data that often fail to connect abstract metrics with lived experience.

Sustainable Human-Computer Interaction (SHCI), a subfield of HCI, initially approached everyday sustainability through technological solutions and persuasive technologies following these assumptions. Scholars have increasingly criticized these limited framings - recognizing that environmental data can create objective but meaningless representations of environmental phenomena and advocate for more relational approaches that embrace individual subjectivity and agency. Given data’s pervasiveness in sustainability discourse, this thesis argues for re-examining our predominantly cerebral relations with abstract environmental metrics. Instead, an alternative possible path could be fostering new relational, embodied connections and sense-making processes that build shared understandings about everyday consumption and encourage participation in change-making.

Building on the four central pillars of environmental data, everyday life, design, and sense-making, I conceptualise Data Anchoring - a design concept for reimagining and recontextualising environmental data in the everyday. Through design exemplars, I articulate how Data Anchoring operates through specific mechanisms - embodiment, social, affect, quotidian, and frame-based strategies that enable new forms of knowing and relating to data. I position its contribution within the broader landscape of SHCI, as a means to transform abstract environmental metrics into meaningful, experiential encounters.

This thesis investigates wave propagation in waveguides with different cross-sectional geometries, filled with graded metamaterial structures. The growing interest in graded metamaterials in electromagnetic applications and models is motivated by their realism, mathematical simplicity, and versatility, compared to the conventional materials that are in use today. The majority of the research community resorts to the use of numerical implementations and solvers for obtaining a solution for the field distribution and propagation characteristics. However, these methods do not provide explicit physical insight into the connection between the steepness of the graded-index profiles and their respective field solutions. Thus, the motivation for the research in this thesis is to focus on analytically solving the wave equation for several graded-index profiles to gain physical insight into the field solutions and what phenomena may be predicted from these results.

The research in this thesis focuses on the graded impedance-matched RHM-LHM profile, which is a composite material involving an impedance-matched transition from a right-handed material to a left-handed material. The profile possesses a variable transition steepness of the relative material parameter values and dispersive characteristics. These variable properties act as degrees of freedom, thereby introducing a high degree of modeling flexibility and allowing for the study of wave propagation under more general conditions. Both non-periodic and periodic impedance-matched RHM-LHM transition profiles have been theoretically studied here using analytical functions. It is of interest to study these RHM-LHM composite materials, as the interaction between a regular material and a metamaterial may lead to newly discovered phenomena, novel analytical expressions of electromagnetic fields, and provide a deeper understanding of the underlying principles at the interface between such media.

The solution method is based on describing these graded metamaterial structures by their relative permittivity and permeability functions, where either one or both of these properties have a graded transition profile. The wave equation for each metamaterial structure is derived using Maxwell's equations and solved using the boundary conditions imposed by the waveguide geometry. The field distribution and propagation characteristics for a given electromagnetic mode are analytically expressed and visualized using the numerical software tool MATLAB. Furthermore, numerical results are obtained using the numerical software tool COMSOL Multiphysics, which are then compared with the analytical results to validate the analytical expressions derived from the wave equation.

Powder metallurgy is used in the manufacturing of cutting tool inserts to achieve the desired material properties. Tungsten carbide, WC, mixed with a metallic binder, such as Cobalt, Co, is a common cemented carbide used for cutting tool inserts. In large scale production, the powder is pressed and sintered. During the manufacturing process, the volume of the cutting tool significantly decreases. During the sintering process, very little can be done to change the final shape, and the shrinkage during sintering must be accounted for during pressing. This thesis will investigate the mechanical modeling of shrinkage that occurs in the sintering process.

The constitutive model at issue can capture the isothermal stage in the sintering process and can be adjusted to a specific powder. To increase its usability, a sensitivity analysis was performed in Paper A to determine if all parameters must be determined when using a new powder mixture. It was found that some of the parameters were more sensitive than others when optimizing parameters using experimental results. The less sensitive parameters could be constant, reducing the number of necessary experiments to determine all adjustable parameters. 

The different stages of sintering were also investigated in Paper A, where the model had difficulty describing the shrinkage both in the initial stage and in the liquid phase, where the Cobalt starts to melt. The latter was expected, since the constitutive model was explicitly developed for the solid stage of sintering.

To evaluate the model, all the experiments were performed on a specific WC-Co powder blend. For adherence to match the industrial process for cutting tools made from this powder, the debinding phase was included early in the sintering cycle. This was not done in the previous development of the sintering model. The influence of the debinding process was experimentally investigated in Paper B, where it was shown that the shrinkage, as well as the final microstructure, is influenced by the exclusion of the debinding stage. 

In Paper C, multiple dilatometer and sintering furnace experiments were performed to further develop the model. Complements to the initial stage and liquid phase were introduced, and the effect of changing the initial density, post-compaction, was investigated. Different sintering cycles were used to ensure the robustness of the constitutive parameters of the model. The deviatoric influence was tuned with bending experiments. 

To verify the model, Paper D compares simulations to experiments. This was done using two different press dies, where the same amount of powder was pressed to different heights. The powder blanks were sintered to different maximum temperatures and measured. The compaction and sintering process was simulated and compared to the experimental values, showing that the sintering model captured the process well. A sintering furnace was used, where one of the sintering cycles was representative of the industrial production of cutting tools.  

The global environmental crises continue to worsen, approaching irreversible thresholds. While much sustainability research focuses on policy, technological solutions and behaviour change interventions, the role of professionals—the people actually doing the work—remains largely unexplored as potential actors in sustainability transitions. This thesis addresses this gap by asking: how can Human-Computer Interaction (HCI) be practiced otherwise at the end of the world as we know it, to enable liveable futures?

Rather than asking what HCI as a field should do, I centre on HCI professionals themselves—researchers, practitioners, designers, educators, students—and our potential for agency in responding to what I term the socio-environmental predicament. This predicament encompasses interconnected environmental and social crises that cannot be "solved" but require thoughtful, situated responses. I have explored this question through first-person research grounded in my own experiences as a PhD student practicing HCI otherwise.

Through six papers, this thesis makes three contributions. First, I demonstrate the value of centring HCI professionals rather than HCI as a field, shifting from abstract calls for change to concrete possibilities for individual agency and responsibility. Second, I provide two practical resources: a framework for surfacing assumptions about sustainability in our work, and an application of the Two Loops model that identifies multiple sites of agency within both dominant and emerging systems. Third, I show what action-oriented second-order transition research can look like in HCI, demonstrating how such research can support transformations towards liveable futures.

My exploration reveals that meaningful responses include: 1) learning from communities that are already living with the knowledge and/or experience of collapse; 2) using and designing speculative methods to support desirable futuring; 3) engaging in action-oriented community-led work; and 4) paying attention to, promoting and hosting emotions and care in your interactions with others, in the face of uncertainty, complexity and loss. Ultimately, this thesis argues that the transition ahead requires us as HCI professionals to engage in critical reflexivity about our assumptions, values, and practices, and to find new ways of using our skills and positions in service of life and the living.

Microelectromechanical systems (MEMS) offer powerful solutions for drug delivery where biological barriers limit the potential of advanced therapeutics. This thesis demonstrates how unorthodox applications of microfabrication techniques can create novel platforms to overcome drug delivery challenges, enhancing the delivery of potent and fragile biologics.

The first part of this work focuses on implantable systems. An ultrasonically actuated micro-implant is presented, which exploits mechanical resonance not for sensing but for the selective, on demand destruction of reservoir membranes. This enables remotely triggered drug release without onboard power or electronics. Building on this, a miniaturized ultrasonic energy harvester is developed, integrating a high-performance, bulk piezoelectric material (PZT-5H) via a novel low-temperature bonding process, creating a robust power source for future active implants.

The second part explores two-photon polymerization (2PP) to fabricate complex 3D microstructures for non-invasive delivery. First, rolling ultra-miniaturized microneedle spheres (RUMS) are introduced. Unlike traditional flat microneedle patches, these 3D particles are suspended in topical formulations to gently and repeatedly disrupt the skin’s stratum corneum, enabling the effective transdermal delivery of biologics. Second, a micro-swirl nozzle, a design typically found in internal combustion or agricultural applications, has been developed to aerosolize fragile biologics. This geometry generates a fine mist suitable for deep lung deliverythrough a low-shear mechanism, preserving the integrity of sensitive payloads like lipid nanoparticle (LNP)-encapsulated mRNA.

Collectively, this work showcases a versatile approach to biomedical engineering, where the precise control of micro-scale geometry and physics is leveraged to solve persistent challenges in therapeutic delivery.

Safe disposal of high-level radioactive waste is crucial for protecting human health and the environment over geological timescales. A detailed understanding of the chemical processes that govern fuel corrosion is therefore essential for reliable safety assessments. A scenario of particular concern involves groundwater intrusion following the failure of the engineered barriers designed to contain the spent fuel. In this case, groundwater in contact with the fuel becomes exposed to ionizing radiation, producing radicals and molecules that can shift the conditions from reducing to oxidizing. Among the radiolytic oxidants, hydrogen peroxide (H₂O₂) is expected to have the greatest impact on the rate of fuel oxidation. The presence of carbonate in groundwater enhances the dissolution of oxidized UO₂—the main constituent of spent nuclear fuel—through the formation of soluble uranyl–carbonate complexes. This thesis examines several aspects of the oxidative dissolution of UO₂. One aspect is the effect of uranyl ion (UO₂²⁺) accumulation in the near-surface solution on the overall oxidation rate. It was found that increasing concentrations of UO₂²⁺ suppress H₂O₂-induced oxidation by reducing the concentration of free, reactive H₂O₂ through the formation of inert uranyl–peroxo–carbonate complexes. UO₂²⁺ was also found to affect the stability of H₂O₂ in irradiated solutions. In the presence of O₂, UO₂²⁺ can suppress the concentration of H₂O₂ through selective reduction of uranyl–peroxo–carbonate complexes by the superoxide radical. Under anoxic (N₂) atmosphere, UO₂²⁺ scavenges reducing radicals (e^-_aq and H^●),which increases the H₂O₂ concentration of the γ-irradiated solution. The kinetics of carbonate-facilitated dissolution were also re-evaluated, as previous models were based on systems where oxidation and dissolution occurred simultaneously. In the absence of oxidants, dissolution itself was found to be a multistep process with an apparent activation energy of (34.8 ± 3.2) kJ mol^-1. The reaction order with respect to bicarbonate varied between 0.5 and 1.5 within the temperature range (283–333) K and bicarbonate concentrations of (1.3–15) mM. The dependence on the oxidation state of uranium oxide showed three distinct stages: (1) a rapid initial release of a small fraction, (2) a slower, nearly constant releases rate (independent of the remaining oxidized fraction), and (3) a gradual rate decrease as the oxidized product approached depletion. Finally, comparative studies of Pd-doped and undoped UO₂ thin films in the presence of H₂ confirmed that any uncatalyzed H₂ effect is several orders of magnitude less efficient at inhibiting oxidative dissolution. Palladium was found to reduce the oxide only partially, to a U(V) intermediate, identified as UO₂OH, based on a calculated U:O atom ratio of 1:3 and an average uranium oxidation state of +5. 

This thesis dives into the world of autonomous underwater robotics, specifically oriented at solving the relative navigation problem that arises during an underwater multi-agent proximity operation (prox-op). The present document starts with an introduction to the background theoretical and practical concepts required for the reader to follow the contributions outlined within. We define the concept of proximity operations in the underwater domain and highlight a factor-graph-based robotic state estimation framework used to intuitively model arbitrary prox-ops as Simultaneous Trajectory Estimation and Relative Navigation (STERN) problems. We continue by outlining the attached scientific contributions which carefully address the different elements of the general factor graph representation in a procedural manner: we start by isolating the two navigation-dependent phases of the prox-op and solve them independently; subsequently, we study the full scenario from end to end. The document is redacted such that it provides the story in hindsight surrounding the scientific contributions that are part of this compilation thesis.

 Stainless steels are employed in a wide range of applications, spanning everyday household items, such as kitchen appliances, to advanced aerospace technologies, including space rockets. This range of applications is connected to the intimate link between the structure and properties of a material, influenced by the processing conditions. Duplex stainless steels, which consist of a mixture of the body centred cubic ferrite phase and the face centred cubic austenite phase, are used in highly corrosive environments under high stress. These alloys are prone to changes in the microstructure during service due to phase separation of the ferrite at elevated temperatures which limits their service life. Strategies to mitigate these microstructural changes caused by decomposition and delay embrittlement are of great interest for both economic and sustainability reasons. In this thesis, the effect of intermediate heat treatments on phase separation kinetics has been explored using in situ small-angle neutron scattering (SANS) to follow the evolution of the nanostructure. SANS is an established method for studying phase separation, and is well-suited for characterization of phase separation in stainless steels. During phase separation, a characteristic correlation peak develops in the SANS signal. The peak position and intensity are related to the wavelength and amplitude of decomposition in the sample, and the peak profile is often extracted with the help of models or peak fitting. To capture the earliest stages of the decomposition, the methodology was refined to account for the change of the structure factor induced by the changes in morphology during the ageing treatment. Using this new approach, the change in phase transformation kinetics due to the applied intermediate heat treatments was quantified, showing up to 60 % decrease in the SANS correlation peak amplitude, directly related to the amplitude of decomposition in the alloy. During the intermediate heat treatments, sigma phase precipitation was observed during some of the conditions. Sigma phase is a brittle intermetallic phase that needs to be avoided during processing, in order to not compromise the mechanical properties of the alloy. To investigate the kinetics of sigma phase formation in order to avoid it during the proposed heat treatments, it was studied in a model alloy as well as in a commercial alloy using in situ neutron diffraction. The study highlighted the complexity of predictive modelling and the limitations of the precipitation theory approach, in the same time showing good agreement with equilibrium calculations regarding formation of sigma phase but not other intermetallics. To gain further understanding of the mechanism of the delay of embrittlement caused by the intermediate heat treatments, pair distribution function (PDF) analysis of neutron total scattering data on model binary alloys was utilised. Due to the bulk characteristics of the material, texture effects were found during the analysis. The effect of texture on PDF analysis was investigated, showing a large influence over the results that could be extracted from a large-box model, limiting the conclusions that could be drawn from the local structure investigation.

Conventional pyrometallurgical techniques for metal production rely heavily on carbon-based reduction and high-temperature operations such as melting, leading to significant CO₂ emissions and high energy consumption. Additionally, industrial by-products containing valuable metals remain underutilized for metallurgical applications, highlighting the need for more sustainable and efficient processes. To address these challenges, this thesis investigated a solid-state hydrogen reduction approach to produce metal alloy powders directly from mixed metal oxide mixtures. Copper–nickel and iron–nickel alloys were successfully synthesized using this method by reducing synthetic Cu₂O-NiO and Fe₂O₃-NiO powder mixtures, respectively, at 700 °C for 45 minutes in a horizontal tube furnace using hydrogen. Off-gas analysis results confirmed a complete reduction by monitoring the water vapor content. X-ray diffractometry confirmed the absence of residual oxides and the presence of solid phases consistent with the respective binary phase diagrams. Scanning electron microscopy combined with energy-dispersive X-ray spectroscopy revealed microscopic compositional fluctuations that remained within the variability typically observed in conventionally cast alloys. These findings demonstrate the potential of hydrogen-based processing as a low-carbon alternative for alloy production.

To improve chemical homogeneity at the microscopic scale and enable direct use of the alloy powder, iron–nickel powders obtained through hydrogen reduction of 50 wt% Fe₂O₃-NiO mixtures at 700 °C for 45 minutes were subjected to homogenization at 1100 °C under an argon atmosphere. Diffusion models constructed using the diffusion module in Thermo-Calc guided the selection of homogenization durations of 5, 10, and 15 hours. Experimental homogenization trials revealed an increasing chemical homogeneity over time. Furthermore, the 15-hour sample demonstrated a chemically uniform microstructure along with extensive neck growth and minimal porosity. X-ray diffractometry confirmed the absence of oxides after homogenization. In addition, energy-dispersive X-ray spectroscopy revealed impurities located along particle boundaries, identified as iron- and nickel-free oxide phases originating from the Fe₂O₃ feedstock. These inclusions did not interfere with the reduction or homogenization processes.

The reduction process was further scaled up and extended to multicomponent metal oxide systems to evaluate its robustness and applicability to industrial by-products. A 250 g batch of 50 wt% Fe₂O₃-NiO powder was reduced in a horizontal Fe-Cr-Al tube furnace by heating non-isothermally from room temperature to 700 °C over 30 minutes, followed by an isothermal reduction at 700 °C for 3 hours. The resulting alloy exhibited stable body-centred cubic and face-centred cubic phases, and melt extraction analysis confirmed only trace oxygen levels. X-ray diffraction peaks consistent with a nickel–hydrogen phase were observed in the product from one upscaling trial but disappeared after 30 days, indicating that the phase is thermodynamically unstable at room temperature. As a preliminary step towards more complex chemistries, reduction trials were conducted using 0.1 g of a synthetic oxide mixture composed of equal mass proportions of Fe₂O₃, NiO, Cu₂O, WO₃, CoO, and MoO₃. The reduction trials conducted at 700 °C for 45 minutes demonstrated that all constituent oxides in the synthetic mixture could be completely reduced under the established process conditions.

In small- and large-scale by-product reduction trials, involving 0.1 g and 250 g samples of roasted spent catalyst, selective reduction of WO₃ and NiO was achieved within an unreduced Al₂O₃-SiO₂ matrix by applying the same thermal profile used in the successful upscaling trial with the 50 wt% Fe₂O₃-NiO synthetic mixture. X-ray diffractometry confirmed complete reduction of WO₃ and NiO, and scanning electron microscopy showed tungsten- and nickel-rich sites embedded in the Al₂O₃-SiO₂ matrix. Mass spectrometry of the off-gas in all trials detected only water vapor, with no hazardous emissions, confirming the environmental safety of the process. These findings demonstrate that the approach not only offers a scalable and environmentally friendly route for alloy production from primary ores but also provides an effective means to recover metals from by-products, supporting a more sustainable and circular materials economy.

Dendrimers have emerged as versatile platforms in nanomedicine due to their well-defined architecture, monodispersity, and multivalent nature. However, conventional designs confine functional groups to the periphery, limiting their chemical and therapeutic diversity. This thesis presents heterofunctional polyester dendrimers (HFDs) based on a novel bromide-functional AB₂C monomer as a multifunctional platform, combining molecular precision with clinical potential. The synthetic route follows divergent anhydride-mediated esterification, yielding monodisperse dendrimers (polydispersity ≤1.03) with internal bromide or azide groups and external hydroxyl functionalities. The biodegradable polyester scaffold undergoes controlled degradation over 30–96 h, balancing stability with biocompatibility. Orthogonal post-functionalization (azide-alkyne click chemistry and thiol-bromo coupling) enables selective incorporation of diverse payloads. Ammonium groups or apolar drugs like diclofenac can be embedded internally, while cationic groups or PEG chains are introduced onto the surface via esterification.

Systematic biological evaluation across antibacterial, gene delivery, and anticancer applications reveals a unifying principle: the therapeutic efficacy is governed by spatial distribution of functionalities rather than dendrimer generation. In antibacterial applications, dual-charge dendrimers display potent activity (MIC: 10–21 μM) against Gram-positive and Gram-negative pathogens with enhanced E. coli sensitivity, maintaining >85% mammalian cell viability (3–100-fold improvement over homofunctional counterparts). For gene therapy, the same cationic architecture achieves 93% RNase protection and 62% gene silencing in glioblastoma cells with minimal cytotoxicity. In anticancer applications, covalent diclofenac conjugation with PEGylation induces selective ROS-mediated cytotoxicity at 16–32-fold lower concentrations than free diclofenac, preserving >95% normal cell viability.

This work demonstrates that therapeutic performance can be decoupled from generation number through rational architectural design. The HFD platform defines transferable design principles for next-generation nanomedicine, uniting polymer versatility with molecular precision.

The explosive growth of distributed data generation — spanning data centers, sensor networks, massive IoT, and edge learning places an unsustainable burden on modern infrastructure, where the energy and latency costs of moving raw data often outstrip those of processing it. While analog over-the-air computation (OAC) promises a solution by exploiting the natural superposition of wireless waveforms to aggregate data in-channel, it remains fragile against noise and fundamentally incompatible with the ubiquitous digital hardware that powers all modern communication systems.

This thesis introduces a digital-native framework that unifies communication and computation at the physical layer. Rather than treating channel interference as an obstacle, we engineer the geometry of digital constellations so that the superposition of signals directly yields the desired function value. This paradigm shift transforms the communication link from a passive data pipe into an active computational engine, applicable to any multiple-access channel—whether wired or wireless—without requiring the decoding of individual inputs.

We generalize this framework along three axes to ensure scalability and reliability across diverse network environments. First, we develop noise-aware constellation designs that optimize inter-symbol geometry for non-Gaussian and heavy-tailed interference, ensuring robustness beyond standard Euclidean metrics. Second, we introduce a sampling-based reduction strategy that leverages the symmetry of aggregation functions to cut design complexity by orders of magnitude, enabling deployment in massive-scale networks. Third, we extend the framework to vector-valued computation, utilizing spatial degrees of freedom to perform complex, multi-dimensional aggregations in a single transmission shot without relying on perfect channel state information.

Finally, to bridge the gap to immediate deployment, we present a closed-form algebraic coding scheme for exact summation. The proposed solution integrates seamlessly with standard quadrature amplitude modulation, eliminating the need for complex optimization and offering a plug-and-play solution for digital aggregation. We validate these contributions through the lens of Federated Edge Learning, demonstrating that computation-by-communication is not only feasible using standard digital protocols but significantly outperforms traditional orthogonal transmission. Collectively, these works prove that computation-by-communication is not only feasible on digitally modulated signals but superior to analog alternatives, paving the way for the next generation of compute-aware networks, enabling energy efficient, scalable, and robust intelligence across any digital infrastructure.

Global navigation satellite system (GNSS) and other assisted positioning infrastructures provide ubiquitous, precise locations for cyber-physical system (CPS), from autonomous vehicles to location-based service (LBS) applications on mobile phones in daily lives. Combining multiple satellite constellations, network infrastructures, and onboard sensors typically makes the position solutions more accurate and robust than any single source alone. 

However, civilian GNSS signals, Wi-Fi beacons, and cellular pilot signals lack cryptographic protection and are therefore vulnerable to signal spoofing attacks. Even if they can be upgraded to support authentication, meaconing or wormhole attacks can relay and falsify the wireless signals and then manipulate the localisation. More seriously, an attacker can selectively jam the wireless signals from specific infrastructures to force CPS to downgrade to less secure signals, which are later spoofed; coordinated adversaries can also target multiple infrastructures simultaneously to manipulate the positioning result. 

This thesis is in the broad area of data trustworthiness for CPS, focusing on the security and resilience of localisation. Emphasis is given on securing the localisation based on GNSS, as they are relevant to a multiplicity of modern systems (e.g., connected vehicles, smartphones, and other Internet-of-Things (IoT) platforms). Significant efforts are dedicated to detecting attacks on position and providing secure and reliable location information, even in the presence of adversaries and benign faults (e.g., challenging propagation environments). Where perfect recovery is unlikely, the proposed methods aim for a best-effort position estimation by opportunistically fusing the remaining available benign signals. 

These efforts are concerned with designing, analysing, implementing, and evaluating diverse protocols that address GNSS-specific attacks, other positioning signal attacks, and simultaneous GNSS with other signal attacks. The approaches are theoretically rigorous, are evaluated through detailed simulations, real-world experiments, and system implementation, proposing concrete defense mechanisms.

The practical deployment of terahertz systems for future applications requires a comprehensive toolkit of high performance, compact passive components that overcome several limitations innate to terahertz waves. A primary challenge is minimizing attenuation in signal routing, a problem especially critical when terahertz power is scarce. At the same time, effective signal management and device characterization require components that can efficiently absorb power to minimize reflections and crosstalk. This thesis addresses both of these needs, presenting novel low-loss all-silicon devices for terahertz applications and introducing advanced loss engineering techniques to create highly effective integrated and free space absorbers.

The first part of this thesis introduces high performance, all-silicon passive devices fabricated using silicon micromachining techniques. A perforation-free, mechanically robust planar parabolic reflector antenna is presented utilizing slab optics and total internal reflection to achieve a flat gain response over a broad bandwidth of 220-330 GHz. Furthermore, a very low crosstalk waveguide crossing is demonstrated by applying transformation optics to a Maxwell fisheye lens. This approach resolves the fundamental mode mismatch problem inherent to conventional lens designs, enabling dense and complex terahertz circuit integration.

The second part focuses on lossy structures for both dielectric and metallic waveguides. For open dielectric waveguides, ultrathin single-walled carbon nanotube films are deposited to create compact, broadband and reflectionless terminations without altering the geometry of the guide, drastically attenuating the propagating waves over short distances by evanescent field interaction. For enclosed hollow metallic waveguides, integrated absorbers are developed using highly porous nanomaterials, including randomly oriented and aligned graphene-coated nanofibers, as well as carbon nanotube aerogels. Characterized using a novel multi-band measurement methodology, these materials demonstrate broadband stealth and shielding performance across a wide frequency range (67-500 GHz). The investigation is also extended to free-space applications, demonstrating a hierarchically porous carbon-silica composite as a low reflectivity absorber in 140-220 GHz.

Collectively, this thesis expands the component toolkit for terahertz integrated systems, providing practical and high performance solutions for waveguiding, radiation and termination that are crucial for the next generation of high-frequency applications.

The ongoing proliferation of intelligent systems, driven by artificial intelligence (AI) and 6G, is leading to a surge in closed-loop inference tasks performed on distributed compute nodes.These systems operate under strict latency and energy constraints, extending the challenge beyond achieving high accuracy to enabling timely and energy-efficient inference.This thesis examines how distributed inference can be optimised through two key decisions: when to sample the environment and when to offload computation to a more accurate remote model.These decisions are guided by the semantics of the underlying environment and its associated costs.The semantics are kept abstract, and pre-trained inference models are employed, ensuring a platform-independent formulation adaptable to the rapid evolution of distributed intelligence and wireless technologies.

Regarding sampling, we studied the trade-off between sampling cost and detection delay in event-detection systems without sufficient local inference capabilities. The problem was posed as an optimisation over sampling instants under a stochastic event sequence and analysed at different levels of modelling complexity, ranging from periodic to aperiodic sampling. Closed-form, algorithmic, and approximate solutions were developed, with some results of independent mathematical interest.Simulations in realistic settings showed marked gains in efficiency over systems that neglect event semantics. In particular, aperiodic sampling achieved a stable improvement of ~10% over optimised periodic policies across parameter variations.

Regarding offloading, we introduced a novel Hierarchical Inference (HI) framework, which makes sequential offload decisions between a low-latency, energy-efficient local model and a high-accuracy remote model using locally available confidence measures. We proposed HI algorithms based on thresholds and ambiguity regions learned online by suitably extending the Prediction with Expert Advice (PEA) approaches to continuous expert spaces and partial feedback. HI algorithms minimise the expected cost across inference rounds, combining offloading and misclassification costs, and are shown to achieve a uniformly sublinear regret of O(T^2/3).The proposed algorithms are agnostic to model architecture and communication systems, do not alter model training, and support model updates during operation. Benchmarks on standard classification tasks using the softmax output as a confidence measure showed that HI adaptively distributes inference based on offloading costs, achieving results close to the offline optimum. HI is shown to add resilience to distribution changes and model mismatches, especially when asymmetric misclassification costs are present.

In summary, this thesis presents efficient approaches for sampling and offloading of inference tasks, where various performance metrics are combined into a single cost structure. The work extends beyond conventional inference problems to areas with similar trade-offs, advancing toward efficient distributed intelligence that infers at the right time and in the right place. Future work includes conceptual extensions like joint sampling-offloading design, and integration with collaborative model-training architectures.

In nuclear reactor safety analysis, we label an accident as severe once a partial core meltdown and material relocation begin. Researchers use simulation tools such as ANSYS and MELCOR to study these events safely, producing vast and complex datasets. In this work, we applied machine learning explainability and interpretability to extract insights from severe accident simulations for the Nordic boiling water reactor (BWR) through five iterative studies. First, we examined the explainability of the decision tree classification algorithm to distinguish between accident types using time-wise pressure vessel external temperature. Second, we generalised the model to create a more statistically robust and generic framework, introducing the open-source Decision Tree Insights Analytics (DTIA) framework (https://github.com/KHossny/DTIA), which combines explainability, interpretability, and statistical robustness. Third, we applied DTIA to high-dimensional MELCOR COR package data for a station blackout combined with a loss-of-coolant accident (SBO + LOCA) in a Nordic BWR, revealing new findings. Fourth, we used DTIA to compare structural variables of the reactor pressure vessel lower head under SBO and SBO + LOCA conditions. Finally, we coupled DTIA with K-Means clustering to address its need for labelled data, uncovering previously overlooked events such as canister melting. We concluded that the patterns identified by machine learning in mapping inputs to outputs can uncover insights that were previously overlooked, particularly in high-dimensional and complex datasets.

This thesis consists of three papers, an individual summary of each paper, and an introduction. The papers are all related to existence, uniqueness, or regularity theory of local and nonlocal partial differential equations (PDEs).

In Paper A, we establish uniqueness for viscosity solutions of the inhomogeneous nonlocal infinity Laplace equation Lu = f, where the right hand side f is a bounded, continuous, and nonpositive function. Uniqueness is proven through a comparison principle.

In Paper B, we use Perron's method to construct viscosity solutions to the equation ∂u/∂t = L u in Ω, and u = g in the complement.

In Paper C we study regularity of a minimizer of the expression J(u) := ∫ F(∇u) dx, where F(x) is a strongly convex function whose second derivatives might jump at |x| = 1. The specific form of F gives rise to a free boundary Γ, and the resulting Euler-Lagrange equation varies over Γ. In this paper we only consider two-phase flat points. We show that under some regularity and non-degeneracy assumptions the asymptotic expansion of a minimizer u can be written as u(x) = a + ν · x + p(x) + q(x), where a ∈ R, ν ∈  R^n. The function p is a broken polynomial that is defined as a C^1 function consisting of one polynomial in the upper half space and another polynomial in the lower half space, and the function q is a rest term. We derive the PDEs that are satisfied by p and q, respectively, and show many regularity properties for the terms in the expansion. This paper is intended to be the first part of a project that aims at establishing regularity of the free boundary Γ.

Membrane proteins are proteins that are embedded in the lipid bilayers of

organisms. Roughly a fourth of all human proteins are estimated to be

membrane proteins and about 60 % of human-approved medications target

membrane proteins. The correct function of membrane proteins is essential to

all organisms.

This thesis is made up of two parts. First, the biochemistry and function of

dolichol phosphate mannose synthases (DPMS) are investigated. These

enzymes are responsible for the transfer of mannose from a nucleotide sugar

donor to the acceptor lipid dolichol phosphate, forming dolichol phosphate

mannose (Dol-P-Man). In eukaryotes and archaea, Dol-P-Man is the key

mannose donor for mannosylation reactions inside the endoplasmic reticulum

(ER) lumen or on the extracellular leaflet of cell membrane, respectively. As

the synthesis of Dol-P-Man is known to take place on the cytoplasmic side of

the ER membrane in eukaryotes or the cell membrane in archaea, the question

remains how Dol-P-Man is transported onto the other side of the membrane

to serve as a mannose donor. This thesis presents a hypothesis in which the

DPMS itself is responsible for the flipping of its own product. The hypothesis

is supported by crystallographic data that shows Dol-P-Man bound to a DPMS

in a “flipped” orientation that could enable the transport to the other side of

the membrane. This thesis also covers the recombinant expression,

purification, and in vitro characterization of DPMS from the zebra fish Danio

rerio. This DPMS is similar to the human enzyme and can therefore yield

mechanistic details behind DPMS-related diseases.

The second part covers the development of scattering-type scanning near-field

optical microscopy (s-SNOM) to study proteins in solution. The method is

capable of collecting images and infrared spectra from samples at nanometerscale

lateral resolution. The method is not readily applicable for the study of

objects in solution, but this limitation can be circumvented by the use of a

liquid cell. The liquid cell is first used to probe the stretching vibrations of

water in nanoscale and the method is then further developed and is applied to

collect images and spectra from purple membranes, a model membrane

comprising tightly packed bacteriorhodopsin molecules and associated lipids.