Given stricter regulations and the transfer of sensitive data to the cloud, there is a clear need to further strengthen cloud security. The latest advances fall under the term confidential computing, which complements existing methods of protecting data during storage and transfer with memory encryption and remote attestation. The introduction of these countermeasures significantly raises the security bar for both remote attackers who operate malicious virtual machines and exploit vulnerabilities in cloud infrastructure, and malicious actors with physical access. AMD SEV-SNP and Intel TDX are the latest developments implementing confidential computing for server-grade processors. For wider adoption of this technology, effective management of confidential virtual machines, i.e., virtual machines utilizing the protection provided by confidential computing chips, is essential. To facilitate the lifecycle management of confidential virtual machines, the Secure VM Service Module (SVSM) has been introduced as a common layer that can be used across different vendors.

This thesis investigates live migration of confidential virtual machines running under AMD SEV-SNP using the SVSM module. First, the current state of the art is investigated. Since there is no solution for migrating confidential machines with the SVSM module, a migration design is developed and proof of concept is provided for the most time-consuming part of the migration process, random access memory (RAM) migration. The proposed solution is analyzed and the steps needed to increase its scope and functionality are outlined. A new methodology for evaluating incomplete migration is developed and used to assess the upper limit of the overhead that AMD SEV-SNP confidential machines would represent for the live migration process. Our single-threaded proof-of-concept resulted in a tenfold slowdown in memory page transfers.

Spatial Active noise control (ANC) aims to manipulate sound fields within target regions through active control methods, with applications including noise cancellation and high-fidelity audio reproduction. The placement of secondary sources and sensors has a significant impact on system performance; however, traditional approaches optimise these placements independently, resulting in suboptimal results. This thesis addresses the joint source and sensor placement optimisation problem in spatial ANC systems. The challenge lies in formulating an effective cost function that simultaneously optimises secondary source placement for sound field synthesis and sensor placement for accurate spatial interpolation, while maintaining reasonable computational complexity. A novel joint cost function formulation was developed that explicitly incorporates interpolation error alongside synthesis error by comparing interpolated field estimates against the true desired field across the entire control zone. This approach addresses fundamental limitations in existing methods, including clustering behaviour and convergence issues, by properly accounting for estimation uncertainty inherent in limited sensor measurements. A comprehensive simulation environment was implemented to evaluate four placement algorithms: Random (baseline), Regular (geometric), Greedy (optimisationbased), and Matching Pursuit (signal processing-based). The algorithms were systematically compared using normalised mean square error and computational efficiency metrics across multiple resolution configurations. Results demonstrate that optimisation-based approaches provide meaningful performance improvements over simple placement strategies, with the Greedy algorithm achieving the best acoustic performance. However, significant computational trade-offs were revealed, with optimisation methods requiring substantially longer execution times. The novel cost function formulation successfully resolved convergence issues and demonstrated superior robustness across varying acoustic conditions.

This thesis explores the potential of gamification to enhance the onboarding experience in a factory simulation tool designed for the corrugated fiber industry. The tool, developed for Van Den Bos Robotics, allows users to model transport systems and simulate production flows. While rich in functionality, the tool’s complexity may pose a challenge for new users during onboarding. To address this issue, a gamified onboarding system was developed, incorporating game design elements such as missions, achievements, levels, and in-app rewards to guide users through key features of the tool. Most gamification work focuses on employee performance and motivation; very little work is done on learning complex simulation software. This is a crucial gap provides a strong justification for the present study. By examining gamification in the context of digital tool onboarding, this thesis contributes new insights into improving learnability and user experience in industrial simulation environments. To evaluate the impact of this approach, a between-subjects experiment was conducted with 16–20 participants of diverse academic backgrounds. One group used a traditional onboarding method (verbal intro and manual), while the other group experienced the gamified onboarding. Task performance, usability (SUS), engagement, and cognitive load (NASA-TLX) were measured and compared.

Lightweight composite structures have become essential in modern aerospace engineering, where increasing demands for fuel efficiency, reduced emissions, and improved operational reliability place new requirements on both materials and manufacturing. As composite components grow more advanced, featuring co-cured components, complex geometries, and thinner design margins, the need for improved insight into their internal behaviour becomes critical. Existing sensing technologies struggle to provide local, in-situ information from the composite’s interior during manufacturing or throughout its service life, without compromising structural integrity. This creates a gap between the capability of current sensing approaches and the monitoring demands required by the complexity of next-generation composites.

This thesis addresses this gap by investigating the feasibility of embedding nanomaterial-based sensing structures, primarily vertically aligned carbon nanotube (VACNT) forests, into fibre-reinforced polymer composites. The overarching aim is to explore how such sensors can be integrated with minimal structural intrusion, from where their sensing behaviour originates, and how they can provide reliable, multifunctional monitoring both during manufacturing and in the cured state. The work spans the development of embedding and contacting strategies, bottom-up characterisation to investigate sensing mechanisms, and the exploration of both direct current (DC) and alternating current (AC) measurement approaches. Collectively, the research seeks to expand the understanding of how nanomaterial sensors interact with composite materials and how they can support the design of future multifunctional aerospace structures.

The findings demonstrate that VACNT forests can be embedded into composite laminates without compromising the composite’s mechanical structure, while providing robust and reproducible sensing capabilities. A bottom-up analysis helps determine that the embedded VACNT forests’ thermoresistive behaviour is governed by fluctuation-assisted tunnelling, and their linear piezoresistive response originates in the intrinsic piezoresistivity of individual CNTs. The VACNT forests enable local in-situ cure monitoring of prepreg laminate, detecting key process transitions. Strategies for sensing in conductive carbon fibre environments are established, as well as comparisons with alternative nanomaterial-based sensors such as graphene coatings. Finally, by transitioning from DC resistance to AC impedance measurements, the work shows that embedded CNT structures can detect high transverse pressures and exhibit frequency-dependent sensing sensitivity.

Together, these results establish VACNT forests as a promising, multifunctional, and structurally compatible sensing concept for advanced composite structures, offering new pathways for embedded process monitoring, structural health monitoring, and the development of next-generation multifunctional aerospace components.

As algorithmic systems increasingly shape human experiences, ensuring fairness and diversity has become a central challenge. This thesis studies fairness and diversity through the lens of algorithm design and optimization theory, providing formal frameworks and efficient algorithms across three domains: ranking-based recommendation, streaming recommendation, and graph analysis.

The first part of the thesis investigates diversity maximization in recommender systems with stochastic user engagement. We first study how to rank items in recommendation systems, where users engage with content sequentially and probabilistically. We introduce two novel diversity measures, sequential sum diversity and sequential coverage diversity, which account for uncertainty in user engagement. Our goal is to find a ranking of items that maximizes these sequential diversity measures. We show that sequential coverage diversity is ordered submodular, enabling a greedy 1/2-approximation. For sequential sum diversity, we provide polynomial-time constant-factor approximation algorithms. Separately, we study a streaming setting where items arrive continuously and users may visit the system multiple times at arbitrary moments. For this setting, we aim to design a streaming algorithm that maximizes a stochastic coverage diversity measure. We show that a classic greedy algorithm achieves a tight 1/2-competitive ratio but requires memory linear in the stream length. With sublinear memory and an upper bound T' on the number of user visits T, we propose STORM, which achieves a 1/4(T'-T+1)-competitive ratio. We further propose STORM++, improving the competitive ratio to 1/8delta, where the integer parameter delta controls the tradeoff between solution quality and computational cost.

The second part of the thesis studies diversity as a constraint in densest subgraph discovery and addresses the problem of finding dense communities in networks with heterogeneous relationship types. We model relationship types as edge colors and formulate the At Least h Colored Edges Densest Subgraph problem (ALHCEDGESDSP), which seeks subgraphs that are both dense and contain at least h_i edges of each color i. We prove that even the simplest variant of this problem is NP-hard and develop constant-factor approximation algorithms. Our key technical contribution links the edge-constrained and node-constrained versions of the densest subgraph problem. We first show that algorithms for the At Least k Nodes Densest Subgraph problem (DalkS) can approximate the At Least h Edges Densest Subgraph problem (ATLEASTHEDGESDSP), and then extend the algorithm for DalkS to handle colored edge constraints for solving ALHCEDGESDSP.

The third part of the thesis studies graph interventions for fairness in networks. We examine two fairness measures, PageRank fairness and hitting-time fairness, developing methods to balance influence and improve accessibility across groups. For each demographic group, the sum of PageRank scores within it quantifies the influence of that group. PageRank fairness measures how far the current group-wise influence deviates from a given target. We formulate the PageRank fairness problem as an optimization problem that adjusts edge weights such that the resulting graph achieves a group-wise influence distribution as close to the target as possible. The optimization problem involves a nonconvex objective over a convex feasible set under practical constraints, such as not adding new edges and limiting the magnitude of weight changes. We solve this PageRank fairness maximization problem using efficient projected gradient descent, proving convergence to a stationary point. For hitting-time fairness in bipartite graphs, we formulate two problems, minimizing the average (BMAH) and the maximum hitting time (BMMH) from one group to another via strategic edge additions. We provide a (2+epsilon)-approximation for BMAH by combining fast random walk simulation with greedy supermodular minimization. For the more challenging BMMH problem, we develop two approaches: the first leverages its connection to the BMAH problem, and the second employs a method based on the asymmetric k-center problem. Both approaches yield provable approximation guarantees for BMMH.

The algorithms and analysis techniques presented in this thesis contribute to the growing body of work on fairness and diversity in algorithmic systems. By formalizing new problem variants that capture realistic constraints in interactive and networked settings, and by providing approximation algorithms with provable guarantees, this work expands the toolkit available for addressing fairness and diversity challenges in computational systems.

Waste management is particularly challenging for Small Island Developing States (SIDS) due to their high per-capita infrastructure costs, remoteness, narrow resource bases and high dependence on tourism. The lack of integrated planning frameworks considering these SIDS-characteristics has stalled progress on sustainable waste management. To address this challenge, this paper proposes an integrated methodology for long-term waste management planning to deliver on the United Nations’ Sustainable Development Goals (SDGs) in SIDS. This explicitly combines multi-level participatory SDG visioning and back-casting with waste infrastructure modelling. This methodological development is piloted using a national-scale demonstration on Curacao. Three island-specific waste management portfolios (Inaction, Circular Economy, Technology-led), developed through stakeholder back-casting, are modelled for SDG delivery using a national accounting model under different socio-economic futures. The results highlight the importance of waste prevention and material re-use strategies within islands that engage local populations. Evidence-based identification and evaluation of waste management strategies, grounded in participatory processes, can itself contribute to SDG delivery.

We explore a novel problem in streaming submodular maximization, inspired by the dynamics of news-recommendation platforms. We consider a setting where users can visit a news website at any time, and upon each visit, the website must display up to k news items. User interactions are inherently stochastic: each news item presented to the user is consumed with a certain acceptance probability by the user, and each news item covers certain topics. Our goal is to design a streaming algorithm that maximizes the expected total topic coverage.

To address this problem, we establish a connection to submodular maximization subject to a matroid constraint. We show that we can effectively adapt previous methods to address our problem when the number of user visits is known in advance or linear-size memory in the stream length is available. However, in more realistic scenarios where only an upper bound on the visits and sublinear memory is available, the algorithms fail to guarantee any bounded performance. To overcome these limitations, we introduce a new online streaming algorithm that achieves a competitive ratio of 1/(8δ), where δ controls the approximation quality. Moreover, it requires only a single pass over the stream, and uses memory independent of the stream length. Empirically, our algorithms consistently outperform the baselines.

In this paper, two strategies to isolate resistive vertically aligned carbon nanotube (VACNT) forestsfrom the conductive carbon fibre environment are presented, enabling embedded sensing with resistivecarbon nanotube sensors in carbon fibre laminates. VACNT forests are used due to their already proventemperature and strain sensing capabilities and ease of placement in prepregs, enabling localisedsensing. To achieve this, a non-permeable separator and a permeable separator are used and compared.Performing cure-monitoring on the VACNT forests during the sample manufacture, it can be concludedthat short-circuits of the resistive sensor are avoided. After manufacture, the temperature and strainsensing capabilities of the VACNT forests when using the two isolation strategies are evaluated. Fromthese measurements, differences in temperature sensing range and sensitivity to strain are observed.

Link-analysis algorithms, such as PageRank, are instrumental in understanding the structural dynamics of networks by evaluating the importance of individual vertices based on their connectivity. Recently, with the rising importance of responsible AI, the question of fairness in link-analysis algorithms has gained traction.

In this paper, we present a new approach for incorporating group fairness into the PageRank algorithm by reweighting the transition probabilities in the underlying transition matrix. We formulate the problem of achieving fair PageRank by seeking to minimize the fairness loss, which is the difference between the original group-wise PageRank distribution and a target PageRank distribution. We further define a group-adapted fairness notion, which accounts for group homophily by considering random walks with group-biased restart for each group. Since the fairness loss is non-convex, we propose an efficient projected gradient-descent method for computing locally-optimal edge weights. Unlike earlier approaches, we do not recommend adding new edges to the network, nor do we adjust the restart vector. Instead, we keep the topology of the underlying network unchanged and only modify the relative importance of existing edges. We empirically compare our approach with state-of-the-art baselines and demonstrate the efficacy of our method, where very small changes in the transition matrix lead to significant improvement in the fairness of the PageRank algorithm.

Diversification is a useful tool for exploring large collections of information items. It has been used to reduce redundancy and cover multiple perspectives in information-search settings. Diversification finds applications in many different domains, including presenting search results of information-retrieval systems and selecting suggestions for recommender systems.

Interestingly, existing measures of diversity are defined over sets of items, rather than evaluating sequences of items. This design choice comes in contrast with commonly-used relevance measures, which are distinctly defined over sequences of items, taking into account the ranking of items. The importance of employing sequential measures is that information items are almost always presented in a sequential manner, and during their information-exploration activity users tend to prioritize items with higher ranking.

In this paper, we study the problem of maximizing sequential diversity. This is a new measure of diversity, which accounts for the ranking of the items, and incorporates item relevance and user behavior. The overarching framework can be instantiated with different diversity measures, and here we consider the measures of sum diversity and coverage diversity. The problem was recently proposed by Coppolillo et al. [11], where they introduce empirical methods that work well in practice. Our paper is a theoretical treatment of the problem: we establish the problem hardness and present algorithms with constant approximation guarantees for both diversity measures we consider. Experimentally, we demonstrate that our methods are competitive against strong baselines.

In this paper, a graphene-coating on glass fiber weave has been evaluated as a multifunctional sensingmaterial based on previous work performed by the authors on embedded vertically aligned carbonnanotube forests. First, the graphene-coating has been assessed as a resistive cure-monitoring sensorwhen embedded in thermosetting glass fiber/epoxy laminate to monitor its production. The graphene-coating showed similar results to the resistive cure monitoring of carbon nanotubes. However, thegraphene-coating diverges in its resistive signature upon reaching cure temperature, showing acontinuous resistance decrease in this phase, suggesting a sensitivity to cure-induced shrinkage of theepoxy, not seen in the carbon nanotube sensor. Later, in the cured state, the embedded graphene-coatingfunctions as an excellent temperature sensor, possessing a negative thermoresistive effect. However, asa strain sensor, the graphene-coating does not perform as well as the embedded carbon nanotube sensor,possessing an initial drift in resistance upon its first load cycle and additional drift during constant strainconditions, and when unloaded.

Diffusion models have shown great potential in generating trajectory plans for agents in environments with unknown dynamics. However, such models provide no safety guarantees. In this work, we focus on risk-aware planning with respect to safety constraints and introduce a probabilistically risk-aware variant of Diffuser (PRA-Diffuser). The diffusion model initially learns a distribution over trajectories that may or may not be unsafe. We then fine-tune this model to reduce the probability of sampling such unsafe trajectories. We analyze the proposed solution and introduce a provable lower bound on risk of safety violation leveraging concentration inequalities for conditional Value-at-Risk. Our approach can be applied to models that have been pre-trained, potentially from datasets containing unsafe trajectories. Our empirical results demonstrate that our approach significantly reduces unsafe trajectories generated by the diffusion model across multiple environments.

We study multi-agent reinforcement learning (MARL) under temporally extended Signal Temporal Logic(STL) objectives, which require reasoning over both long-horizon dynamics and inter-agent relations. Wepropose TD-MAT, a transformer-based architecture with multivariate positional encodings, causal temporalmasking, and a decomposed reward based on arithmetic–geometric mean robustness with variance regularization. Experiments on coordination tasks ranging from unstructured multi-objective problems to strict temporalsequencing show that TD-MAT learns effective long-term behaviors and generalizes to heterogeneous agentsettings. Ablation studies highlight the necessity of temporal masking, positional encodings, and reward decomposition, while comparisons to MAPPO, RMAPPO, and MAT reveal that transformers provide the greatestbenefit on unstructured, long-horizon tasks.

Global floor area is increasing every year which is subsequently leading to an increase in electricity and heating demand in buildings. Residential buildings have huge potential for energy savings and there is an immediate need to decarbonize them by the end of 2050. Machine learning is finding application across all fields and will thus have an important role to play in the building sector also. One of the important challenges that building owners need to tackle is occupancy detection in residential apartments which can help save considerable amounts of energy and costs. However, occupancy is highly variable, and it is difficult to quantify and predict occupancy because of the random and individualistic nature of humans. In addition, scalable approaches for occupancy detection should prioritize data from common and cost-effective sensors like temperature sensors. In contrast to existing literature which has stated that occupancy detection based on the data from a single environmental sensor is not appropriate for obtaining good results, this paper aims to detect occupancy in a real residential building using only indoor temperature as the feature to train the model. Different machine learning models and techniques are studied and tested to understand how the accuracy of occupancy detection can be increased. With the right techniques, it has been possible to obtain promising results in the form of an accuracy of 95% using machine learning models and only indoor temperature to train it.

construction sector, deeply entrenched in a linear model of resource extraction, consumption, and disposal, significantly contributes to environmental degradation. In response, the circular economy paradigm has emerged as a key alternative. Given their high resource intensity, large-scale urban development projects call for a reconfiguration of their material flows. These interventions offer a strategic opportunity to develop closed loops at a fine spatial scale. This study explores the development of circular economy strategies for construction and demolition materials in French urban renewal projects led by the National Agency for Urban Renewal (ANRU). A qualitative research design is adopted, combining a literature review, document analysis, ten semi-structured interviews, a site visit and participant observation. Following a contextual analysis of the ANRU Framework, two case studies of NPNRU projects in the Est Ensemble area are examined. The literature review engages with conceptual approaches to the circular economy in the built environment and identifies urban metabolism as a key theoretical concept at the macro scale. The subsequent analysis is informed by this framework, complemented by the Multi-Level Perspective (MLP). The empirical results first identify opportunities associated with intense material flows and conditions conducive to innovation, as well as barriers related to the spatial, temporal and material characteristics of this specific context. The case studies further reveal key implementation mechanisms, particularly financial instruments, physical platforms, and monitoring tools. Finally, the findings point to conditions for scaling up circular practices, including institutional changes, enhanced synergies and a broader paradigm shift in urban production. In brief, this thesis contributes to both the literature on circular economy in the built environment and to research on ANRU projects. The integration of circular economy theory with empirical evidence highlights the potential of these urban renewal programmes in the transition to a circular regime, while also acknowledging the diverse associated limitations.

The study of pattern avoidance stems from a question in computer science: which sequences of distinct numbers can be ordered by a single pass through a stack? Knuth (1968) found that these sequences are characterized by having no subsequence a, b, c, such that c < a < b. Such a subsequence has the same relative order as the permutation 231, so we say that our original sequence avoids 231.

In more general terms, pattern avoidance is a natural way to restrict the structure of permutations by forbidding subsequences with a certain relative order. This has become a popular topic in enumerative combinatorics, and it has connections to various other fields.

Determining the number of 1324-avoiding permutations of length n is the most important open problem in pattern avoidance. This thesis is comprised of two papers contributing to the inversion monotonicity conjecture by Claesson, Jelínek and Steingrímsson (2012), according to which av_n^k(1324), the number of 1324-avoiders of length n with a fixed number k of inversions, is weakly increasing in n. If the conjecture is true, it improves our understanding of the asymptotic behavior of the number of 1324-avoiders.

In Paper A, we provide an explicit formula for av_n^k(1324) for all n ≥ (k + 7)/2. The proof relies on a novel structural characterization of 1324-avoiders with few inversions. As a byproduct, we show that the inversion monotonicity conjecture holds when n ≥ (k + 7)/2.

In Paper B, we study the inversion monotonicity of classes of permutations avoiding multiple patterns. We show the sets {1324, 231} and {1324, 2314, 3214, 4213} are inversion monotone via explicit injections, and introduce a general procedure for constructing large inversion-monotone sets. We also analyze the limiting structure of large permutations with a fixed number of inversions avoiding 1324 and another pattern of length four, and prove several half-monotonicity results similar to Paper A.

Understanding the degradation mechanisms in lithium-ion batteries is essential for the development of next generation batteries. Herein, a post-mortem protocol described in “Quantifying lithium lost to plating and formation of the solid-electrolyte interface in graphite and commercial battery components” by Fang et.al is used to track the redistribution of metallic lithium in two types of carbonaceous anode materials: graphite, and hard carbon. In this report, two different commercially available graphite anodes were used, one power optimized while the other energy optimized. The hard carbon anode was prepared in house.

Metallic lithium in plated samples displayed an immediate redistribution to ionic compounds at low chemical shifts on a complex manner. Graphite displayed a behaviour that could be modelled by a double-exponential function whilst hard carbon displayed a single-exponential function decaying to a baseline. There was no observed difference between the power optimized and energy optimized graphite. “Dead” lithium in delithiated samples showed a period of stability followed by a fast redistribution into ionic compounds. No attempt was made to identify these ionic compounds due to their large NMR linewidths. 

The plating kinetics of hard carbon differed significantly from graphite, requiring a much greater overcharge to reach the deposition underpotential which was the same for both anode materials.

Ultra-endurance cycling exposes athletes to prolonged static postures, leading to a high prevalence of overuse injuries, particularly in the lower back and knees. While aerobars are widely used to reduce aerodynamic drag, their internal musculoskeletal loads remain poorly understood. This study aims to quantify the biomechanical impact of aerobar use compared to standard road positions and evaluate the specific influence of stack and inclination settings on lumbar and lower limb joint reaction forces. Ten experienced cyclists performed trials on an instrumented ergometer at 70% of their Functional Threshold Power (FTP) across six conditions: hoods, drops, and four aerobar configurations varying in stack (high/low) and inclination (0°/15°). A custom automated musculoskeletal modeling workflow using OpenSim was developed to compute compressive and shear forces for lumbar and lower limb joints. Results revealed that the flat aerobar position (0°) significantly reduced L4-L5 compressive forces (17.5 ± 10.9 N/kg) compared to the standard hoods position (20.9 ± 4.8 N/kg), likely due to effective skeletal support. However, tilting the aerobars to 15° significantly increased spinal loading. A mechanical trade-off was identified: aerobar use shifted the load distally, more than doubling hip compression forces (~46 N/kg) and significantly increasing knee forces compared to road positions. These findings suggest that while flat aerobars reduce spinal joint loading, they impose severe stress on the lower limbs, highlighting the need for injury-specific bike fitting strategies in ultra-endurance.

Photocuring is central to many biomedical polymer applications because light-activated photoactivators (PAs) generate radicals that rapidly convert acrylate functions into a crosslinked network, enabling fast, localised curing. For clinical devices the PA must therefore combine efficient, rapid crosslinking under the clinical light source with an acceptable safety profile.

Following a regulatory change, three candidate photoactivators (A, B and C) were evaluated as possible replacements for the current PA in a medical polymer formulation. Absorption at 405 nm (the emission of the clinical light source) was confirmed for all candidates, and formulations were prepared by mixing the base polymer with PA solutions in ethanol at the highest feasible loading. Target concentrations of 2500 ppm and 5000 ppm were tested. Crosslinking kinetics and efficiency were characterized using real-time FTIR, ATR-FTIR and photo-DSC; efficient crosslinking was defined as ≥90% acrylate conversion within 15 s at the Lowest Acceptable Polymerization Irradiance (70 mW・cm⁻2).

Here we show that Candidate A, despite slower polymerization kinetics, offers the best balance of performance and safety: minimal concentrations to meet the 90%/15 s criterion were ~2500 ppm for the current PA and ≥2500 ppm for Candidates B and C, whereas Candidate A required ≥5000 ppm. All formulations gave similar depth of cure and were stable to ambient light for 7h except Candidate C, which hardened more rapidly. Regulatory database review indicated Candidate A had the lowest toxicity, Candidate C was more cytotoxic and classified as skin sensitizer 1A, and less data were available for Candidate B.

These findings revise the expectation that all candidate PAs would perform like the incumbent PA at the same dose: structural and photophysical differences alter required operating concentrations and residual risk. Practically, the study identifies Candidate A as the preferred replacement and also shows that reducing the current PA below labelling thresholds is infeasible (900 ppm failed to achieve effective crosslinking even at higher irradiance and longer cure). Ongoing work evaluates POL004 formulations with Candidate A after supercritical CO₂ purification to confirm crosslinking efficiency and overall performance in an industrial, clinically relevant context.

Autonomous systems have become integral to industry and society, spanning applications from robotic platforms to autonomous vehicles and beyond. Most contemporary engineering systems exhibit complex nonlinear dynamics and are often subject to time-varying uncertainties, making high-fidelity modeling difficult. At the same time, these systems are increasingly required to execute complex tasks that extend beyond classical objectives such as set-point tracking and stabilization. In many real-world scenarios, they must satisfy spatiotemporal specifications, requirements that depend jointly on space and time. Such specifications can be formulated as time-varying constraints in control systems, and enforcing them is crucial for enhanced performance, guaranteed safety, and reliable, timely task execution. 

Funnel-based control methods provide closed-form feedback laws that enforce certain classes of time-varying constraints for uncertain nonlinear systems. The main focus of this thesis is to develop new robust closed-form control schemes, rooted in the core ideas of funnel-based control, to address broader classes of time-varying constraints that cannot be treated directly by conventional funnel-based methods. In addition, the thesis investigates distributed coordination of multi-agent systems under spatial constraints, as well as the application of funnel-based methods to multi-agent formation control under transient performance requirements.

The first part of the thesis is devoted to extending funnel-based control methods, particularly prescribed performance control (PPC), to address time-varying hard (safety) and soft (performance) funnel-type specifications. We then revisit the standard PPC design to highlight its limitations and to motivate the need for a new control framework. Building on the PPC design philosophy, we propose a novel robust closed-form control scheme that enforces generic time-varying set invariance for high-relative-degree, multi-input multi-output uncertain nonlinear systems, thereby accommodating classes of time-varying constraints beyond those handled by standard PPC. Finally, we extend the proposed design to treat potentially conflicting generalized time-varying hard and soft specifications, further broadening the applicability of the method.

In the second part of the thesis, we shift the focus to multi-agent coordination problems. First, we present a novel coordinate-free formation control scheme for directed leader–follower multi-agent systems that achieves almost-global convergence to a desired shape. Fully decentralized robust controllers are synthesized by leveraging the PPC framework to impose prescribed transient and steady-state performance on the agents’ formation errors, while ensuring robustness to system uncertainties. A key ingredient of the approach is the use of bipolar coordinates to obtain orthogonal (decoupled) formation-error coordinates for each follower. This not only promotes almost-global convergence to the desired shape but also enables a systematic and effective application of PPC. Finally, we introduce a distributed, task-based implicit formation determination and control problem in which each agent is subject to spatial constraints with respect to other agents and the environment. We reformulate the problem as a distributed optimization scheme and, based on this formulation, develop a control protocol for kinematic agents.

PurposeThe study investigates how openBIM workflows can support standardised digital validation processes, aiding the transition towards circular construction. Specifically, it examines the use of the Information Delivery Specification (IDS) standard to validate Industry Foundation Classes (IFC) models for the reuse of precast hollow core slabs, in accordance with the Norwegian standard NS 3682:2022 Hollow Core Slabs for Reuse.Design/methodology/approachThis study proposes an Automated Checking Compliance (ACC) method to verify the compliance of IFC models with reuse-driven information contents for precast hollow core slabs. To achieve this, the IDS standard was selected to develop and test an openBIM validation workflow. The methodology includes three main steps: (1) identifying the minimum set of information requirements derived primarily from NS 3682:2022; (2) implementing these as IDS specifications linked to IFC entities, and (3) applying the workflow to a case study of a precast hollow core slab modelled in Autodesk Revit and exported to IFC4x3. The validation is performed using the open-source Bonsai add-on for Blender, while a complementary buildingSMART Data Dictionary (bsDD) is developed to ensure semantic consistency and standardised property definitions.FindingsThe results confirm that IDS enables effective automation of compliance checking for data presence and structure within IFC models. The IDS-based workflow reliably identifies missing information, highlighting the specific objects that fail the validation. By aligning rule-based validation with recognised standards, the proposed approach supports quality assurance for the digital representation of reusable hollow core slabs. Furthermore, the study establishes a standardised, machine-readable database of reuse requirements. The approach can be adapted and applied to other building components, promoting interoperability and reliability in digital marketplaces for reclaimed materials.Research limitations/implicationsWhile the proposed ACC workflow guarantees consistency and completeness of digital information, it does not assess the correctness or validity of underlying physical test results and of the initial data entry, such as mechanical properties or service life parameters. The applicability of the approach also depends on the digital maturity of stakeholders and the completeness of IFC models. Broader applicability will require further harmonisation of reuse-related standards and increased awareness of information requirements.Practical implicationsThe study illustrates how openBIM standards can be used within reuse-driven design and procurement processes through automated data validation. The combined use of IDS and bsDD enhances regulatory compliance, data transparency, and reliability of digital inventories of reclaimed components, lowering barriers to reuse, especially for small and medium enterprises.Originality/valueThis paper is among the first to apply the IDS framework to the context of building component reuse, translating reuse-oriented standards into machine-readable validation rules. It extends the use of openBIM standards from model verification to circular construction practices, supporting both digitalisation and sustainability efforts in the built environment. Specifically, the study contributes to improved interoperability and trust in digital marketplaces for reclaimed construction products.

This thesis investigates the thermal-hydraulic phenomena during a LBLOCA ina Nordic BWR with a focus on the temperature and pressure response in thewetwell gas space. The phenomena in the pressure suppression system weresimulated using GOTHIC thermal-hydraulic code. Reliable response of a pressuresuppression system is crucial for containing main steam line break accidents inBWRs to avoid leakage of radioactive materials to the environment. Therefore,it is a thoroughly researched subject but a gap seems to exist in evaluating theaccident with a maximum possible break area for the steam line. The objectiveof this thesis is to model the accident with maximum break size and evaluate theresults and suitability of chosen methodology.Two models with different approaches regarding steam venting were developedto compare which one leads to a better alignment with expected results oftemperature and pressure. Sensitivity studies were performed on the mainsimplifications of the model to study the impact on the response.Comparison with other studies shows that a model with lumped blowdownpipes yields more conservative results than the model with one large subdividedblowdown pipe in GOTHIC. However, the model with lumped pipes lacks incapability of monitoring processes in the pipe and modeling the heat gradientaccurately throughout the length of the pipe. Sensitivity studies emphasized theimportance of wetwell gas space grid resolution, number of modeled blowdownpipes and incorporating the heat radiation from blowdown pipe to the wetwellgas space into the model. The pressure safety criteria for the containment was notexceeded in any of the simulations in this study.Additional studies could be done on the condensation phenomena in the wetwellpool with total number of pipes and horizontal asymmetry of temperaturedistribution in the wetwell containment.

Generative video editing has enabled several intuitive editing operations for short video clips that would previously have been difficult to achieve, especially for non-expert editors. Existing methods focus on prescribing an object's 3D or 2D motion trajectory in a video, or on altering the appearance of an object or a scene, while preserving both the video's plausibility and identity. Yet a method to move an object's 3D motion trajectory in a video, i.e. moving an object while preserving its relative 3D motion, is currently still missing. The main challenge lies in obtaining paired video data for this scenario. Previous methods typically rely on clever data generation approaches to construct plausible paired data from unpaired videos, but this approach fails if one of the videos in a pair cannot easily be constructed from the other. Instead, we introduce TrajectoryAtlas, a new data generation pipeline for large-scale synthetic paired video data and a video generator TrajectoryMover fine-tuned with this data. We show that this successfully enables generative movement of object trajectories.

One of the most important problems in society today is the abundance of plastics and microplastics causing harm in nature. Because of this, the current need for a switch to wood fibre-based solutions to replace plastics is urgent, and one promising emerging process is the dry-forming of cellulosic fibres. Dry-forming is advantageous since it can produce biobased, recyclable products in complex shapes with mechanical properties in par with plastics. Since dry-forming is an emerging field, additional research is required to better understand the process. In this work the effect of different conditions for forming and the effect of hydrophobisation using Alkyl Ketene Dimer (AKD) has been investigated. It was found that an increase in tensile properties could be observed when increasing the pressing temperature from 100 to 140 °C with a difference in the type of tensile failure between these temperatures. Below 100 °C the main mechanism is fibre pull-out, and above 140 °C the failure is more chaotic indicating stronger interfibre bonding. The results from analysis of AKD-treatment were unexpected with small differences in tensile properties, where previous experiences show decreased mechanical properties when material is treated with AKD. The reason for this is not yet understood, but an interesting area for further research.

The transition toward sustainable energy systems requires the development of environmentally benign and resource-efficient materials, including membranes used in electrochemical and humidification processes. This thesis focuses on the sustainability assessment and optimization of Cellfion’s pilot-scale production of PFAS-free((per- and polyfluoroalkyl substances-free)), cellulose-based humidifier membranes, positioning the work at the intersection of sustainable energy engineering, materials innovation, and circular economy principles.

The study begins by mapping the pilot plant’s resource consumption through a Material and Energy Flow Analysis (MEFA), quantifying inputs such as water, energy, and chemical solvents. This baseline assessment identifies critical environmental and operational hotspots, with drying emerging as the largest energy consumer, washing stages driving significant water demand, and solvent losses contributing to hazardous waste generation. A Techno-Economic Assessment (TEA) complements this analysis by linking these environmental hotspots to cost structures, revealing that while fixed costs dominate at pilot scale, solvent losses and waste disposal remain non-trivial variable costs.

To address these specific challenges (excess energy use in drying, high water consumption, and solvent wastage), a scenario-based approach is employed, developing three optimization pathways: Energy Optimization through waste-heat recovery in drying, Water Optimization via process-water recycling, and Material Reduction through improved solvent management. These scenarios are informed by analogous studies in sustainable materials manufacturing and modeled using MEFA and TEA frameworks to assess their environmental and economic impacts.

The results show that targeted circular economy interventions can achieve notable sustainability gains without increasing production costs. Specifically, the Water Optimization scenario reduced freshwater use by over 35% and lowered material intensity by 34%, the Energy Optimization scenario decreased energy intensity by approximately 7%, and the Material Reduction scenario cut hazardous solvent waste by around 30%. All scenarios maintained unit production costs within ±1% of the baseline, confirming cost neutrality. Trade-off analysis reveals minimal cross-impacts, with the most notable being a slight energy increase in the water-recycling case.

In conclusion, the integrated assessment demonstrates that PFAS-free membrane production can be made significantly more resource-efficient while retaining economic viability. The findings provide actionable guidance for Cellfion’s scale-up strategy, offering a replicable methodological framework for early-stage sustainable process design in the energy materials sector.

With the growing renewable energy power production, Battery Energy Storage Systems (BESS) have appeared on the French electricity grid. To monitor those batteries and ensure its safety Battery Management Systems have been developed as an interface between the battery cells and the Energy Management System (EMS) that dictates the energy flows between the BESS and the grid. Developing a robust Battery management System (BMS) is at the core of economical optimization since controlling the state of health and state of charge while meeting energy requirements is the main interest. This thesis focuses on the development of an academic BMS tool based on a 2-RC Equivalent Circuit Model (ECM) coupled with an aging and a thermal model. The ECM has been built on to replicate the hysteresis behavior of the Lithium iron Phosphate (LFP) chemistry by representing surface particles with a probabilistic approach. A focus on reversible and irreversible heat generation has been done with a feedback effect on the electrical circuit following an Arrhenius law. The aging model is physically-driven but with the lack of experimental data, the parameters could only have been mathematically determined. With a decent accuracy at the single cell level, it has been scaled up to the battery level with multiple cells in series and strings. The effects of cell variability in terms of capacity and resistances has been tackled. Computation time and Error compared with commercial data will be studied in a sensitivity analysis that focus on the number of particle of the hysteresis algorithm, as well as the time step of simulation. The safety measures implemented in the BMS have been tested in different configurations to observe the different mechanisms at stake. A simulation of a world-leader’s battery cell has been done to determine the capacity fade over 20 years of daily cycling. The accuracy of the prediction was of 0.138% RMSE. The tool simulates and monitors Voltage, State Of Health (SOH), State Of Charge (SOC) and temperature of a battery pack as would a digital twin in a commercial BMS.

Graph Neural Networks (GNNs) have emerged as the standard paradigm for machine learning on graph-structured data, demonstrating remarkable success in diverse applications such as molecular design, anomaly detection within networks, and recommendation systems. However, despite their effectiveness in learning meaningful representations for nodes and graphs, GNNs remain vulnerable to adversarial attacks. These attacks, which are small strategically crafted perturbations to the input graph, can result in unreliable predictions. This specific vulnerability raises serious concerns regarding the deployment of GNNs in safety-critical domains like finance and healthcare, where ensuring robustness is crucial. Consequently, understanding and enhancing the adversarial robustness of GNNs has become a critical research focus, involving both the design of potent attack strategies and the development of resilient defense mechanisms.

Many existing defense methods rely on pre-processing techniques or modifications to the message-passing framework to mitigate attacks, often by discarding or re-weighting parts of the input graph. Although these defenses have shown great results, they are frequently based on heuristic reasoning and lack strong theoretical guarantees. Specifically, given the input graphs' rich topological aspect, a deeper understanding of their vulnerabilities and internal behaviors is essential, especially regarding how an attack can propagate through the network. Moreover, current defense methodologies are typically evaluated only against the state-of-the-art attacks available at the evaluation time; in the absence of theoretical guarantees, these defenses remain susceptible to more advanced or previously unseen attack strategies. This gap underscores the need for mechanisms that not only exhibit robust empirical performance but also provide certifiable robustness for long-term effectiveness. Furthermore, most current approaches entail high computational overhead, limiting their practical feasibility in real-world applications.

In this thesis, we address key challenges in GNN adversarial robustness, focusing on the aforementioned drawbacks. First, we introduce defense mechanisms that are both empirically effective and grounded in solid theoretical analysis, thereby offering provable robustness against evolving attacks. Second, we investigate how to reconcile strong defense performance with computational efficiency, which is an essential requirement in multiple domains such as applications in the mobile and online platforms. Achieving this balance is critical for broadening the deployment of robust GNNs in practical settings. Finally, we explore often overlooked factors related to the training dynamics, such as weight initialization and the number of training epochs, that can substantially influence a model’s underlying robustness, illustrating how effective parameter selection can bolster resilience with very limited costs.

The contributions of this thesis are organized around four core pillars. In the first, we propose an adaptation of Graph Convolutional Networks (GCNs) using orthogonal weight matrices, showing both theoretically and empirically that this design can significantly enhance model robustness. In the second contribution, we present a simple yet powerful technique for injecting noise into hidden representations during training, which substantially improves robustness with minimal additional computational cost, consequently offering a more lightweight alternative to many existing, high-complexity defense methods. The third work examines the neglected interplay between training dynamics (e.g., number of epochs, initialization strategies) and model vulnerability, demonstrating how careful tuning of these parameters can enhance a model's underlying robustness. Finally, we propose a novel adversarial attack approach that generates adversarial graphs from scratch via a learnable generator, rather than merely perturbing existing graphs, thereby introducing new perspectives on attack methodologies.

Through these contributions, the current thesis aims to provide theoretical insights and tools that could help advance the current understanding of adversarial attacks in the context of GNNs. These contributions and insights can advance the development of robust GNNs, paving the way for safer and more reliable graph-based machine learning systems.

Due to the increasing urbanisation of the African continent, the urban population of the continent is expected to double by 2050, putting a strain on road networks. It is therefore essential for transport planners to develop their public transport offer while remaining affordable for travellers, in order to ease general traffic congestion, which already causes significant financial losses nowadays. The master's thesis below aims to provide an overview of various aspects of transport in different African cities: the municipalities selected for benchmarking are Nairobi, Abidjan and Kigali. After describing and mapping the network of formal operators in each municipality, the thesis describes and maps their informal networks. Two networks are then evaluated in terms of fare and geographical equity, using national surveys to establish the average expenditure of residents according to their area of residence and income class. It is showed the weight of transport costs in monthly expenditure increases with poverty and distance from the city centre. In a second step, a multimodal model is used to assess the impact of implementing fare integration on the formal and informal transport networks on another city of the continent. The results indicate that the reduction has a negligible impact on non-public transport users and is not sufficient to create a modal shift towards public transport. In the case of public transport users, the implementation of the measure may even prove negative if the integration of informal lines into the formal network is accompanied by an alignment of their fares with those of the formal network. This observation is potentially caused by a network design that favours direct connections, or by the input parameters of the model itself penalising connections.

As in many other countries, Swedish defense during the Cold War was primarily organized around the perceived threat of war, including the potential use of nuclear weapons. This article shifts attention from such scenarios to the ways in which civil contingencies and natural disasters were conceptualized as knowledge objects within the emerging field of defense and disaster medicine. Their incorporation into Sweden’s preparedness agenda signaled a broader and more multifaceted understanding of protection and security within the scientific advisory system of the Swedish defense. By centering medical knowledge production on these hazards, the article offers new insights into the role of medical expertise in Swedish preparedness, while simultaneously shifting focus away from a war-centered narrative of Cold War defense investments. The empirical exploration spans the period from the late 1950s, when a comprehensive governmental inquiry into Swedish defense medicine led to the establishment of the Delegation for Applied Medical Defense Research [Försvarsmedicinska forskningsdelegationen] and subsequently the Organizing Committee for Disaster Medicine [Katastrofmedicinska organisationskommittén] (Kamedo) in the first half of the 1960s. The study concludes in the mid-1970s, when the Delegation was incorporated into the Swedish Defense Research Establishment [Försvarets forskningsanstalt] (FOA).

This thesis project investigated whether single-port robotic surgery (SP) provides meaningful ergonomic advantages over traditional multi-port laparoscopy (LAP) for main and assistant surgeons. Work-related musculoskeletal disorders (WRMSDs) are highly prevalent among minimally invasive surgeons, yet objective, role-specific biomechanical comparisons between these two modalities and roles remain scarce. The aims of the study were to quantify and compare head, trunk, and upper-arm postural demands in main and assistant surgeons during LAP  and SP robotic procedures, and to evaluate whether their exposure exceeded evidence-based action levels for neck and upper-extremity MSD prevention. 

An observational design was used at Karolinska University Hospital Solna, where thirteen experienced surgeons contributed to, in total 43 observed procedures (laparoscopic: 11 mains, 10 assistants; single-port robotic: 12 mains, 10 assistants). Wearable inertial measurement units continuously recorded head and trunk sagittal inclination (positive forward), and bilateral upper-arm elevation from incision to skin closure. For each role and modality, 10th, 50th, and 90th percentile angles were calculated and compared using t-tests (LAP vs SP within role). The postural results were interpreted in relation to established ergonomic thresholds for increased MSD risk, through one-sample one-tailed tests against recommended action levels (α=0.05). 

Single-port robotic surgery was associated with partial, role-specific ergonomic advantages: for main surgeons, singleport robotic surgery showed a trend toward less extreme backward head sagittal inclination compared to laparoscopic surgery: the 10th percentile head sagittal inclination was −11.9° in SP versus −18.9° in LAP (p = 0.32), closer to the −10° action level. The median head angles remained near neutral (LAP −2.0°, SP −0.6°) and 90th percentile forward head angles stayed well below the 50° action level (LAP 15.5°, SP 10.5°). Trunk sagittal inclination was modest for all roles (10th90th percentiles: −9.0° to 11.6°), although the 10th-percentile trunk angles were negative in all groups (SP main −1.8°, LAP main −5.3°, SP assistant −9.0°, LAP assistant −7.2°), hence below the 0° action level. In contrast, SP main surgeons showed significantly higher left-arm elevation than LAP main surgeons at the 10th (15.5° vs 11.9°, p=0.02) and 50th percentiles (26.6° vs 19.3°, p<0.001), while the 90th percentile was not significantly different (p=0.28). For the right arm, the medians were near the 30° action level in both modalities (LAP 26.0°, SP 32.0°), with 90th percentile values below 60° (LAP 42.3°, SP 47.4°). Assistants exhibited no significant LAP–SP differences in arm elevation, yet both groups were more backward than the −10° head sagittal inclination limit at the 10th percentile (LAP −16.0°, SP −22.0°) and worked with median arm angles clustered around 30° (left: 26.7°–28.7°; right: 23.6°–28.7°). Collectively, these statistics indicate that WRMSD risk in these procedures is dominated by neck and shoulder loading, particularly sustained backward head inclination and midrange arm elevation rather than lumbar forward inclination, and that SP technology only partially mitigates this burden for main surgeons while leaving assistant exposures largely unchanged. 

The findings demonstrate that single-port robotic technology does not provide a comprehensive ergonomic solution but redistributes biomechanical load within the surgical team. While console operation moderates extreme neck loading for main surgeons, shoulder elevation remains high, and assistant surgeons continue to experience pronounced backward head inclination and near-threshold arm elevations regardless of modality. Benchmarking against action levels confirms that many postural exposures fall in zones associated with elevated MSD risk, consistent with the high WRMSD prevalence reported in minimally invasive surgeons. Consequently, the thesis concludes that technological innovation must be complemented by multi-level ergonomic strategies including equipment and room redesign, optimized monitor and console positioning, task reallocation, role rotation, and ergonomics training aligned with Swedish work-environment regulations and the hierarchy of controls to sustainably reduce musculoskeletal risk in laparoscopic and single-port robotic surgery. 

The densification of wireless networks toward fifth- and sixth-generation standards has intensified the demand for reliable high-throughput connectivity in indoor deployment scenarios (IDS), such as aircraft cabins, metro wagons, and stadiums. Although millimeter-wave (mmWave) communication offers the spectral resources needed to meet this demand, its sensitivity to propagation loss and blockages severely limits its performance, particularly in IDS. Metasurfaces have emerged as a promising means of extending mmWave coverage through manipulating the propagation environment. Advanced investigations have been conducted on metasurface-featured system performance enhancement. However, the operating cost, which is a practical and critical concern of metasurface deployment, has received insufficient attention in the literature. Deploying a reconfigurable metasurface in practice requires cabling, power supply, and control infrastructure, costs that represent a real barrier to scalable deployment, particularly in indoor environments like IDS, where infrastructure installation is physically limited or tightly regulated.

This thesis investigates the design of sustainable metasurface-assisted indoor wireless communication systems, placing operating cost alongside performance as a primary design criterion. The work examines different types of metasurfaces that differ in the metasurface gain they provide and the operating cost they incur. By identifying and verifying an optimal design choice among these alternatives, this thesis advances a sustainable metasurface-assisted system that addresses the performance-cost dilemma inherent to IDS deployments.

The first contribution studies the trade-off between operating cost and performance enhancement by optimizing a mixed static metasurface (SMS) and reconfigurable intelligent surface (RIS) deployment in an mmWave IDS. Using a fractional programming penalty-based successive convex approximation (FPPSCA)-based iterative algorithm, the results reveal a diminishing-returns relationship. While replacing two SMSs with RISs already yields a 13 Mbps gain, increasing the RIS count beyond 16 out of 22 surfaces produces less than 1 Mbps of additional gain, confirming that full reconfigurability is unnecessary and motivating a more cost-effective middle-ground solution. The second contribution proposes and evaluates a self-sustainable RIS (ssRIS)-assisted mmWave system for IDS, where ssRIS achieves self-sustainability through power harvesting via a codebook-based element splitting scheme, eliminating the need for cabling and external power. A two-stage iterative algorithm jointly optimizes phase shifts, user equipment (UE)-to-ssRIS associations, and time allocation. The results show that ssRIS outperforms SMS by up to 19.8 Mbps in compact environments, confirming a favorable position within the gain-cost trade-off, with coverage advantages diminishing as deployment distances grow. The third contribution conducts a feasibility study of ssRIS across diverse scenarios, analyzing how element count scales with transmit power, data rate demands, and outage constraints under element splitting (ES) and time switching (TS) schemes. TS benefits from stronger channel hardening under moderate conditions, but scales exponentially with harvesting difficulty, whereas ES scales only linearly, offering greater robustness in challenging environments. Together, these findings provide actionable guidance for practical ssRIS deployment.

In the upcoming era of communication systems, there is an anticipated shift towards using lower-grade hardware components to optimize size, cost, and power consumption. This shift is particularly beneficial for multiple-input multiple-output (MIMO) systems and Internet of Things devices, which require numerous components and extended battery lives. However, using lower-grade components introduces impairments, including various non-linear and time-varying distortions affecting communication signals. Traditionally, these impairments have been treated as additional noise due to the lack of a rigorous theory. This thesis explores a new perspective on how the structure of impairments can be exploited to optimize communication performance. To address these challenges, this thesis presents impairments-aware beamforming in various scenarios. 

Initially, we investigate the systems with limited fronthaul capacity. We propose an optimized linear precoding for advanced antenna systems (AAS) operating at a 5G base station (BS) within the constraints of a limited fronthaul capacity, modeled by a quantizer. The proposed novel precoding minimizes the mean-squared error (MSE) at the receiver side using a sphere decoding (SD) approach. 

After analyzing MSE minimization, a new linear precoding design is proposed to maximize the sum rate of the same system in the second part of this thesis. The latter problem is solved by a novel iterative algorithm inspired by the classical weighted minimum mean square error (WMMSE) approach. Additionally, a quantization-aware low-complexity algorithm expectation propagation (EP) is presented for large massive MIMO setups, which is more practical for nowadays systems. Besides, the heuristic quantization-aware precoding method with lower computational complexity is presented, showing that it outperforms the quantization-unaware baseline. This baseline is an optimized infinite-resolution precoding, which is then quantized. This study reveals that it is possible to double the sum rate at high SNR by selecting weights and precoding matrices that are quantization-aware. 

Next, we adopt a splitting precoding architecture tailored to fronthaul-constrained systems for practical deployments. In modern systems, the AAS can perform part of the beamforming locally, for example, through beam-space selection. The remaining lower-dimensional interference-cancelation precoder can then be transmitted over the limited-capacity fronthaul link. Compared to the previous fully centralized setup under the same fronthaul constraint, this approach enables higher quantization resolution for the precoder coefficients. Moreover, since both the uplink pilot signals used for channel estimation and the downlink precoding matrix must be transmitted over the limited-capacity fronthaul link, we design a joint uplink–downlink bit allocation scheme to determine the optimal distribution of fronthaul resources between the two directions.

In the final part of this thesis, we focus on the signaling problem in mobile millimeter-wave (mmWave) communication. The challenge of mmWave systems is the rapid fading variations and extensive pilot signaling. We explore the frequency of updating the combining matrix in a wideband mmWave point-to-point MIMO under user equipment (UE) mobility. The concept of beam coherence time is introduced to quantify the frequency at which the UE must update its downlink receive combining matrix. The study demonstrates that the beam coherence time can be even hundreds of times larger than the channel coherence time of small-scale fading. Simulations validate that the proposed lower bound on this defined concept guarantees no more than 50 \% loss of received signal gain (SG). Based on these results, beam-coherence-aware two-stage digital combining is proposed for the mmWave single-user point-to-point MIMO and multi-user MIMO systems. We also propose time-domain channel estimation.

Global efforts to achieve the Sustainable Development Goals and operate within Planetary Boundaries require comprehensive strategies that account for non-territorial impacts, particularly negative spillover effects driven by consumption in wealthy nations. This highlights a critical disconnect in climate mitigation in high-consumption economies like Sweden mainly focuses on territorial emissions, creating a significant oversight in addressing the true global environmental impact driven by urban consumption patterns.This thesis evaluates Stockholm, Gothenburg, and Malmö urban strategies and identifies different policy instruments including policy instruments from the Nordic Cooperation report’s longlist to pinpoint local policy needs. The analysis revealed a significant implementation gap: municipal strategies remain largely based on territorial accounting, resulting in insufficient integration of policies that effectively address consumption-related emissions and international spillover effects. This thesis concludes with a set of recommendations for local policies aimed at minimizing the impact of Swedish cities on the global climate and promoting positive global effects.

Measurement-induced entanglement transitions arise in quantum circuits of unitary gates and measurements, referred to as hybrid circuits. On average, unitary dynamics generate entanglement, and measurements suppress entanglement. In one-dimensional hybrid circuits, this competition leads to a transition between an area-law phase when the probability of measurements is higher, and a volume-law entangled phase when the probability is lower. In this thesis, the phase transition is studied using two different approaches. First, state vector simulations are performed for hybrid circuits composed of Haar-random unitary gates and random Clifford gates. This allows for direct access to the state vectors and entanglement entropy, but is only feasible for smaller size systems (up to 12 qubits). Second, a traffic flow model is implemented as an approximation to stabiliser dynamics of Clifford circuits, which is studied for larger systems (up to 2048 qubits). Scaling analyses are carried out for both approaches to investigate the dependence of the steady-state entropy on system size and measurement probability. Data collapses of simulation data for different circuit sizes are obtained using both a logarithmic form S−αlnN and by also including an additional correction S−αlnN+βN^(−ω) where ω = 1, leading to a more accurate collapse for smaller sizes. Both the state vector model and traffic flow model lead to a phase transition in entanglement entropy, but with different critical parameters. This indicates that the traffic flow model and the quantum circuit models do not belong to the same universality class. The results clarify the efficiency but also limitations of classical models for describing measurement-induced entanglement transitions.

Chemoselective hydrogenation of α, β-unsaturated aldehydes (UAL) is an important reaction in the chemical and petrochemical industry producing high valuable intermediate products. Over the years, extensive studies has been made on chemoselective heterogeneous catalysts for this type of process to obtain the desired product. Butyraldehyde, is a saturated aldehyde (SAL), which is traditionally produced through fossil-based hydroformylation of propylene, using homogeneous catalysts. These types of processes require complex separation processes and catalyst recycling, which is energy demanding and costly. As an alternative, liquid-phase hydrogenation of crotonaldehyde, an UAL, over heterogeneous catalyst offers a more sustainable process, especially when crotonaldehyde is bio-derived.

In this study, liquid-hydrogenation of crotonaldehyde to butyraldehyde was investigated in a trickle-bed reactor using a commercial Pd/Al2O3 catalyst with a Pd loading of 0.5 wt%. The effects on reaction temperature (30-60◦C) and H2 pressure (2.7-5.25 barg) on conversion and product selectivity were examined. Catalyst characterization was performed with temperature programmed reduction and BET method.

Complete conversions of crotonaldehyde was achieved for all conducted experiments, indicating high catalytic activity. The highest mean selectivity towards butyraldehyde, 62.92%, was at 45◦C and H2 pressure of 3.5 barg. Over-hydrogenation to butanol was unavoidable and showed low effect towards the process parameters, suggesting an excess of H2 for the system.

The conventional approach to machine learning, that is, collecting data, training models, and performing inference in a single location, faces fundamental limitations, including scalability and privacy, that restrict its applicability. To address these challenges, recent research has explored collaborative learning approaches, including federated learning and decentralized learning, where individual agents perform training and inference locally, with limited collaboration. Most collaborative learning research focuses on Euclidean data with regular, grid-like structure (e.g., images, text). However, these approaches fail to capture the relational patterns in many real-world applications, best represented by graphs. Learning on graphs relies on message-passing mechanisms to propagate information between connected nodes, making it conceptually well-suited for collaborative environments where agents must exchange information. Yet, the opportunities and challenges of learning on graph-structured data in collaborative settings remain largely underexplored. This survey provides a comprehensive investigation of collaborative learning from Euclidean to graph-structured data, aiming to consolidate this emerging field. We begin by reviewing its foundational principles for Euclidean data, organizing them along three core dimensions: learning effectiveness, efficiency, and privacy preservation. We then extend the discussion to graph-structured data, introducing a taxonomy of graph distribution scenarios, characterizing associated statistical heterogeneities, and developing standardized problem formulations and algorithmic frameworks. Finally, we systematically identify open challenges and promising research directions. By bridging established techniques for Euclidean data with emerging methods for graph learning, our survey provides researchers and practitioners with a well-structured foundation of collaborative learning, supporting further development across a wide range of scientific and industrial fields.

The development of total freight transport performance is closely linked to the development of industry and has increased more or less continuously. During the post-war period until 1985, both rail, truck and shipping increased rapidly. Since 1985, almost all of the increase in freight transport work in Sweden has taken place by truck. This means that rail and maritime transport have reduced their share, contrary to the objectives of the EU and national transport policies.

The cost of transport (price) and quality are decisive for the choice of means of transport. Better environment is also a request but must not cost much. This is evident from all the research both in terms of models for the choice of means of transport and in studies of how customers choose the means of transport. Competition is fierce and customers change carriers if they can save money, which is natural.

Switching from rail to truck is often easier than the other way around. There are roads everywhere but no tracks. If rail transport to a place is shut down, the industrial tracks will also be closed down sooner or later, and then it will not be easy to start again if the situation changes. Admittedly, there are combined transports by truck and rail, but the transshipment costs money and takes time.

Until 2020, transport costs developed relatively similarly by rail and truck. Between 2021 and 2023, the cost of trucks increased through higher diesel prices, but from 2023 to 2025, the cost decreased through a lower diesel price, while the cost of freight trains increased through higher track fees. The cost difference between rail and truck has increased by 20% in favor of trucks in just a few years. This is due in roughly equal parts to increased track fees and reduced diesel prices.

Delays in train traffic have increased over the years, although there are variations. A method has been developed to calculate the socio-economic additional costs for delays. In 2023 and 2024, the additional costs were approximately SEK 9 billion/year as there was low punctuality for both passenger and freight traffic. The largest share of costs is made up of passenger and customer costs, which account for 65% of the total costs.

For several reasons, greater redundancy is needed in the railway network so that trains can run a different route than the one planned, if necessary:

·       To make room for track work

·       In the event of disruptions and traffic interruptions

·       In case of lack of capacity so that another route can be chosen

·       To create alternative transport routes in crisis and war

One problem is that it is not always easy to divert trains at a junction In many places where the tracks intersect, it is only possible to drive in one direction. Then you need triangle tracks that are shortcuts between different tracks.

One example is the ongoing overhead contact line replacement between Gothenburg and Alingsås. When the line is closed, trains between Gothenburg and Stockholm will have to be diverted via Borås-Herrljunga or Vänersborg-Herrljunga. The distance is longer and it takes 40-90 minutes longer to drive and turn around in Herrljunga. A significant part of this time could have been saved if triangular tracks had been built in Herrljunga and it could also be used in case of disturbances.

For the needs of the Armed Forces, redundancy in the railway network in the form of alternative routes and triangular tracks is of crucial importance. There is also a need for opportunities to distinguish between military and civilian transport, backup plans for manual traffic control and a railway corps with trained personnel for crisis and war.

To analyze the opportunities for the business community to use rail transport, we have conducted a number of in-depth interviews with about 25 transport customers, operators and transport companies. The interviews consistently show that the respondents feel that the cost picture for railways has changed negatively over the past ten years. Together with our analyses, the results indicate that the following measures are required to better exploit the potential of freight rail:

The cost of running freight trains must be reduced by 20%Interviews with customers and transport companies show that costs need to be reduced by 20% if rail is to increase its market share

2.      The maintenance backlog must be worked off and quality increased

Delays for passenger and freight trains cost society SEK 9 billion per year – invest SEK 9 billion a year in maintenance and develop new methods and the maintenance backlog can be worked off in 10 years

3.      Build redundancy in the rail network through triangular tracks and alternate routes 

Facilitate the diversion of trains through better connections between lines and alternative routes – this is needed both to deal with disruptions, for track works and for the needs of the military

4.      Introduce fair track charges for freight traffic

Freight traffic cannot pay for poorer train paths and lower priority – track charges must take into account the competitiveness of industry and the fact that Sweden has long distances compared to other countries 

5.      Set a goal for increased freight traffic by rail by 2030 where everyone contributes

·       Government: Set a target for 2030, eliminate the maintenance backlog over 10 years and decide on competitive track charges for railways and industry

·       The Swedish Transport Administration: Develop railway expertise and plan and prioritise for increased freight traffic by 2030

·       Transport companies: Collaborate in a disturbed situation, develop new transport systems and market rail transport

·       Transport customers: Collaborate in industry-wide transport systems and develop knowledge of rail logistics

Trucks have accounted for most of the increase in the freight transport market since 1985. This means that rail and shipping have reduced their share, which is contrary to the objectives that exist in the EU and also nationally in many countries. However, developments are different in different countries.

Switzerland and Austria have the most positive development for freight rail with a high and stable market share. This is due both to the fact that they have invested in the railways and to have a conscious transport policy with instruments for trucks. Switzerland is the only country that has implemented the EU's transport policy even though it is not part of the EU. Austria and Switzerland also have the highest market share for passenger transport, which shows that there need be no contradiction between passenger and freight transport. Sweden and Germany are also quite well placed with a relatively high market share.

The liberalisation of the railways is an important part of the EU's railway policy. The aim of liberalisation is not only to create new railway undertakings, but also, above all, to put pressure on the old railway undertakings to become more efficient and customer-oriented. This will make the railway more attractive and more people will choose to travel or transport by train. Rail is the most energy-efficient and environmentally friendly means of transport, so if rail had a larger share of the market, energy consumption and emissions would also decrease.

There are still major obstacles to freight traffic by rail. Deregulation has meant that new authorities have been created so that a neutral party can handle what the old national railways did themselves before. However, this has meant extensive bureaucracy that entails both increased costs and difficulties in establishing new traffic. New vehicles can take several years to get approved at high costs. In the end, it affects rail customers. The difference in bureaucracy between driving freight trains and trucks is sometimes almost insurmountable today.

One way to increase the capacity for freight traffic on the existing network is to run faster freight trains. If freight trains could run at 140 km/h instead of 100 km/h as today, the number of freight trains on the main lines during the day could be doubled. For this, new wagons are required while existing locomotives can be used.

The other measure that both increases capacity and reduces costs is longer trains. In Sweden, we have shorter freight trains than in Germany and Denmark, 630 m instead of 750 m, while we have longer trucks. To run longer trains, longer meeting and passing tracks are sometimes required, but it should be possible to run longer trains at night on double tracks when passenger trains do not need to overtake the freight trains. Between Denmark and Germany, even 835 m long trains are running, which would be efficient to also run from Hallsberg in connection with the opening of the fixed link across the Fehmarn Belt.

The industry has pointed out that there is a need for better redundancy in the railway network through more triangular tracks and better lines for diversion. It is also something that is needed for our new defense and that can be used in the event of major maintenance operations and traffic interruptions.

What is the potential of the railways in the future market? The railway is a high-capacity and efficient means of transport that also requires little energy and has minimal emissions. The electric railway has been around for 100 years and is constantly evolving. The rolling resistance with steel wheels against steel rails is extremely low and it is possible to drive trains fast in a safe way. We do not need high-speed trains for freight, but freight trains of up to 160 km/h already exist today and, if developed, could speed up transport and expand the industrial market in the future.

Electric trucks will gradually replace fossil-fuelled vehicles, but energy consumption will always be higher. Rubber wheels against asphalt have about 15 times higher rolling resistance than railway steel wheels against steel rails. Road traffic also causes a lot of particulate emissions, which are a health problem. And heavy trucks wear down the roads and contribute to the road network also having a maintenance debt.

What measures are needed for the freight railway to have a greater role in the future? We are proposing here in the first place measures that do not cost a lot of money. We know that new railways need to be built to increase capacity, and we know that we need increased maintenance to achieve better quality in rail traffic. These problems are well known and discussed a lot in the industry and politics, but we do not believe that we can come up with any new solutions to this. In this analysis, we therefore concentrate on other measures that may not have received as much attention before. Behind many of these proposals are also views that we have received from interviews with 25 representatives of the industry.

1.       Create a cohesive force for rail freight traffic in Sweden

2.       Deregulate liberalisation

3.       Fully implement deregulation in all countries

4.       Create more capacity by better operation planning

5.       Develop new maintenance methods and organize maintenance better

6.       Creation of a common information system and database for rail freight transport

7.       Facilitating intermodal transport through incentives

8.       Develop new technologies for more efficient transport

9.       Let Traffic Management take into account the customers' needs

10.   Forming a Railway Administration

Our interviews show strong criticism of how the Swedish Transport Administration works both from transport companies and from customers. One could summarize it as the Swedish Transport Administration not being customer-oriented enough. The Swedish Transport Administration is there for the customers and the Swedish Transport Administration's customer is the train companies and the end customer is the industry. The expertise needed for the railways would be better utilised if the railway-specific functions were to be located in a separate authority. Then a sector responsibility for railways can also be created. 

Despite rapid advances in whole-genome sequencing (WGS), translating genomic findings into individualized insights remains challenging. We present GenRiskPro, a clinical decision-support and research platform, which automates WGS variant calling, annotation, prioritization, and reporting to deliver actionable findings and facilitate precision wellness. (To test the GenRiskPro platform, log on to https://www.phenomeportal.org/dashboard using the following credentials: Username: user@test.com; Password: test.) GenRiskPro integrates rare and common variant prioritization in a unified pipeline and in-house database, enabling both rare and complex disease and trait association analyses. Variant reporting is supported via LongevityCloud, which features a web portal for clinicians to review, adjust, and authorize the return of results in tabular and PDF formats, alongside a mobile app with artificial intelligence (AI) integration for sequenced individuals. Case studies using Turkish (TR, n = 275) and Swedish (SW, n = 101) WGS data assessed platform performance and variant prioritization: (a) predefined gene panels yielded a 1.82% positive rate for actionable findings per American College of Medical Genetics and Genomics (ACMG) secondary findings guidelines; (b) phenotype-driven support diagnosed cases including muscular dystrophy and microcephaly; (c) cohort-level ClinVar reassessment identified potentially misclassified pathogenic variants; (d) rare variant burden analysis revealed enrichment in ABCA4 for TR and SMPD1 in SW; and (e) population analysis highlighted carrier differences in trait-associated SNPs (rs12913832 and rs4988235) and PGx variants (CYP2B64 and CYP2B66). GenRiskPro unifies databases, literature, web development, and AI for rapid, user-friendly genomic analysis and reporting, which fosters collaboration among hospitals, researchers, clinicians, and patients.

Despite rapid advances in whole-genome sequencing (WGS), translating genomic findings into individualized insights remains challenging. We present GenRiskPro, a clinical decision-support and research platform, which automates WGS variant calling, annotation, prioritization, and reporting to deliver actionable findings and facilitate precision wellness. (To test the GenRiskPro platform, log on to https://www.phenomeportal.org/dashboard using the following credentials: Username: user@test.com; Password: test.) GenRiskPro integrates rare and common variant prioritization in a unified pipeline and in-house database, enabling both rare and complex disease and trait association analyses. Variant reporting is supported via LongevityCloud, which features a web portal for clinicians to review, adjust, and authorize the return of results in tabular and PDF formats, alongside a mobile app with artificial intelligence (AI) integration for sequenced individuals. Case studies using Turkish (TR, n = 275) and Swedish (SW, n = 101) WGS data assessed platform performance and variant prioritization: (a) predefined gene panels yielded a 1.82% positive rate for actionable findings per American College of Medical Genetics and Genomics (ACMG) secondary findings guidelines; (b) phenotype-driven support diagnosed cases including muscular dystrophy and microcephaly; (c) cohort-level ClinVar reassessment identified potentially misclassified pathogenic variants; (d) rare variant burden analysis revealed enrichment in ABCA4 for TR and SMPD1 in SW; and (e) population analysis highlighted carrier differences in trait-associated SNPs (rs12913832 and rs4988235) and PGx variants (CYP2B64 and CYP2B66). GenRiskPro unifies databases, literature, web development, and AI for rapid, user-friendly genomic analysis and reporting, which fosters collaboration among hospitals, researchers, clinicians, and patients.

Starlink provides satellite internet connectivity to customers worldwide using Low Earth Orbit (LEO) satellites connecting to ground stations and user equipment. Precipitation,hourly variability, and the use of different transport protocols,all have impact on throughput. The study was conducted in Stockholm, Sweden, at a latitude of 59.3 degrees north, which is well north of the main coverage area of Starlink. Higher latitudes are covered by fewer satellites compared to Central Europe and the main regions of the United States. The study consists of throughput measurements with the network performance measurement tool iPerf3 using two different transport protocols:Transmission Control Protocol (TCP) and User Datagram Protocol(UDP). Precipitation (rainfall) measurements were conducted simultaneously. The results show a notable decrease in the throughput when moderate rainfall (about 1 mm per hour) is present, about 16 percent for UDP and 28 percent for TCP. The data also show that the throughput varies during different hours of the day, with around 21 percent for UDP and 32 percent for TCP. The highest throughput is received at night and early mornings for both transport protocols. The throughput achieved through the Starlink network with the TCP protocol fluctuates more than on 4G mobile networks. In conclusion, our study provides further knowledge about the effects of precipitation and hourly variability with TCP and UDP on Starlink’s performance, specifically when operated at latitudes outside of Starlink’s main coverage area.

Modeling the pressure in the containment building of a Pressurized Water Reactor following a Loss Of Coolant Accident is a critical topic, especially in the medium and long term, where experimental data is scarce. This document aims at comparing results obtained with 0D and 3D GOTHIC calculations, to ensure 0D predictions used in safety analyses remain conservative. Because manually generating 3D models in GOTHIC is highly time-consuming and error-prone, the development of an automated tool to generate 3D models – the VOXELS method – had begun before this present work. Throughout this work, methodological enhancements were integrated into this tool to enable the automated generation of reliable 3D simulations. A method allowing to lump thermal conductors together and to span them locally was then chained to the VOXELS approach. The comparisons finally performed between results obtained with 0D and 3D calculations show that the long-term pressure tends to be underestimated by 0D analyses. This mainly results from an excessively high prediction of heat absorption of thermal conductors located in the lower floors in the 0D analyses. Hence, it is proposed to remove thermal conductors located below a certain height of the containment building. This corrects the underestimation of the pressure and ensures that 0D analyses remain conservative.

The rapid growth of mobile data traffic demands increasingly energy-efficient error correction solutions. Traditional decoding based on Belief Propagation (BP) often relies on the approximate Min-Sum (MS) algorithm for efficiency reasons, leading to performance degradation. While Machine Learning (ML) approaches like Neural Normalized Offset Min-Sum (NNOMS) mitigate this, they also introduce complexity inherent to artificial neural networks. To address this trade-off, this thesis designs and evaluates a Spiking Normalized Offset Min-Sum (SNOMS), a Spiking Neural Network-based decoder compatible with Digital Neuromorphic Hardware (DNH) such as Intel’s Loihi 2. SNOMS translates the parameter-learned decoding of NNOMS into the event-driven domain. Simulation results show SNOMS achieving performance competitive with its NNOMS baseline, incurring a maximum penalty of only 0.05 dB on short Low-Density Parity-Check codes when measuring Block Error Rate. Importantly, the analytical energy consumption model found that read and write operations to memory are the dominant energy constraint, accounting for over 90% of total consumption. With simulated deployment on DNH, the SNOMS architecture demonstrates superior energy efficiency over its nonspiking counterpart deployed on a traditional hardware architecture. These results points us toward both more efficient trainable decoding algorithms and specialized, energy-efficient neuromorphic decoding hardware, building a foundation for future sustainable wireless communication.

Green hydrogen is gaining attention as a clean energy carrier for the decarbonization of transport and industry. Yet, its widespread adoption remains limited by high production costs, primarily driven by electricity prices. This thesis presents an integrated approach that combines AI and mathematical optimization to improve the performance and cost-efficiency of a Hydrogen Refueling station (HRS) supplied by a renewable-based atmospheric alkaline electrolyser. The project is positioned within Sweden’s ongoing efforts to expand hydrogen use across both industry and transport. It considers hydrogen as a pathway to support decarbonization of energy-intensive sectors while enabling the transition toward cleaner mobility. In this context, the study addresses how hydrogen production, storage, and refueling infrastructure can be planned and operated to align with these industrial goals of Ovako with a broder context to contribute to Hydrogen economy of Sweden. The system is built around two intelligent forecasting models, a Bi-LSTM network that predicts regional electricity prices, and a Random Forest-based Multi-Output Regressor that estimates the CP or operational cost estimated at that hour. These forecasts are generated using features such as electricity prices from different zones (SE1–SE4), physical power flows, fuel prices (LPG), market data for FCR-D, and time-based indicators. The forecasted data is then used in an mathematical optimization model developed in Pyomo to plan the hourly operation of the electrolyzer. The model determines when the electrolyzer should run, how much hydrogen should be directed to steel making process or the refueling station, and with how much of the capacity it can participate in balancing services like FCR-D. The goal is to minimize the total operational cost while ensuring hydrogen availability and flexibility. The model is tested under different real-world scenarios, and key results such as LCOH, hydrogen storage behavior, and market revenues are analyzed. Sensitivity studies are also carried out to understand how electricity prices, fuel costs, and demand changes affect the system’s performance. This thesis demonstrates how combining AI-based forecasting with optimization can lead to smarter, more economical hydrogen systems. The approach offers a practical roadmap for designing intelligent hydrogen infrastructure in future energy markets.

The ubiquity of image files and their use across digital platforms have made them an attractive medium for malicious intent. Image files are often considered harmless, but can be used to hide malicious data with different steganography techniques. Previous research has been conducted on PDFbased malware, but the research on image-based threats remains comparatively unexplored. The thesis aims to evaluate which methods can be used to insert malicious data into common image file formats, create malicious image files based on the identified methods, and finally validate the effectiveness of state-of-the-art antivirus software in detecting said malicious files. The chosen steganographic methods were the least-significant bit method (LSB), end-of-file appending (EOF) and METAdata injection. The methods were then used to insert real malware in each of the four chosen file formats: GIF, PNG, JPEG and WebP. The malicious image files were scanned with state-of-the-art antivirus software, including the widely used Windows Defender. Additionally, the files were also analysed by steganalysis tools. The antivirus softwares proved inconsistent results, depending on file format and steganographic method, but nevertheless performed underwhelmingly. The steganalysis tools performed better in detecting the use of steganographic methods, but did not evaluate whether the image file was malicious or not. Neither of the two detected WebP at all, suggesting the lack of support for newer file formats in both antivirus software and steganalysis tools. This thesis can be used as the foundation for future work aimed at other file types. Furthermore, as steganalysis tools outperformed any antivirus software in detecting the use of steganographic methods, it could be of interest to investigate how steganalysis can be incorporated in malware detection pipelines.

The growing adoption of Cooperative Adaptive Cruise Control (CACC) in vehicle platooning raises new cybersecurity challenges, particularly from insider threats. While traditional cryptographic methods defend against external attacks, they fail to detect malicious behavior from compromised yet authenticated vehicles. Such insider attacks, especially false data injection, can destabilize platoons and pose serious safety risks. This thesis proposes a real-time Misbehavior Detection System (MDS) that identifies threats based on behavioral patterns rather than message authenticity. We present a Transformer-based MDS tailored for CACC platoons, using a sliding window strategy to capture temporal dynamics with low inference latency. The model is evaluated on simulated datasets covering diverse platoon configurations, attack types, and positions. We also examine the impact on the predictions based on the step size and compare the global and individual models to assess the generalization capabilities of our framework. Our approach achieves strong performance, with F1-scores up to 0.95 for a one-second prediction horizon. Individual models often outperform the global model, especially for vehicles with distinct behavior patterns. These results demonstrate the potential of attention-based architectures for robust, real-time detection in Intelligent Transportation Systems (ITS). This thesis highlights Transformer models’ suitability for vehicular security applications and opens avenues for future research, including lowlatency training, real-world validation, and fine-grained attack analysis. Enabling timely and accurate detection of insider misbehavior contributes to a safe and resilient operation of autonomous vehicle platoons.

Advances in 5G/6G hardware have pushed reverberation chambers to the forefront of antenna and RF‐module testing, yet early‐time reflections from metal walls and stirrers lead 𝑆-parameters to be overestimated and to distort derived figures such as radiation efficiency. Full-wave solvers capture this behaviour but become impractically slow for electrically large cavities. To bridge the gap, this thesis develops a high-frequency ray-tracing framework that reproduces the deterministic pre-mixing response of a pseudo-twodimensional reverberation chamber in minutes. Rays are launched from the measured far-field pattern of an 𝐻-plane horn, propagated via specular reflections, and weighted by a ray-tube power-conservation law that delivers both amplitude and phase at a reception arc. Two extraction paths are then implemented: (i) a pure time-domain route that forms an impulse response and converts it to 𝑆₂₁ and 𝑆₁₁, and (ii) a reaction-theorem route that evaluates the mutual impedance 𝑍₂₁ from the captured rays and maps it, together with CST-imported self-impedances, to the full scattering matrix. Comparison with CST over 26–32 GHz shows the ray-tracing model can accurately predict peak responses of 𝑆₂₁ and 𝑆₁₁ while reducing run-time by three orders of magnitude. Because the antenna pattern enters only through sampled farfield data, the method adapts readily to other feed geometries and offers a fast, versatile tool for early-time correction and rapid 𝑆-parameter evaluation in practical reverberation chamber measurements.

With current advancements in robot technology, it is essential that improvements to mechanical parts do not fall behind. A drawback of present-day robot designs concerns motors. Robots with numerous degrees of freedom end up with many motors since each integrated conventional cylindrical motor is only able to move in one degree of freedom. Multiple motors increase weight, amplify complexity and decrease efficiency. To mitigate this, spherical motors have been studied for quite some time. A spherical motor with three degrees of freedom could potentially replace three conventional cylindrical motors in a robot joint. Yet, they are still rare in the industry due to design difficulties and relatively low torque densities. Therefore, in this master’s thesis study conducted at KTH Royal Institute of Technology, in collaboration with ABB, a spherical motor for robot applications has been investigated. More precisely, this study presents the design and simulation of a spherical induction motor capable of three degrees of freedom. The methods of choice consisted of a literature review, modelling and simulations. In detail, a parameter estimation (analytical model) of the spherical induction motor was performed in MATLAB, based on a conventional design process of rotating electrical machines. Furthermore, a computer-aided design (CAD) model of the spherical induction motor was constructed in SOLIDWORKS. Finally, output torque results of the rotation around the z-axis (yaw motion) and y-axis (pitch motion) were obtained through finite element analysis (FEA) simulation in Ansys Maxwell. Specifically, plots of torque as a function of slip frequency were produced, and the result of the rotation around the z-axis shows a torque of about 4% of the torque used in the analytical model, and a preliminary result of the rotation around the y-axis also indicates a relatively low torque. However, magnetic flux density plots demonstrate that the spherical induction motor can, to a great degree, be further optimised due to the steel being far from saturation. Lastly, since the rotation around the x-axis (roll motion) is generated similarly to the pitch motion, the conclusion of a developed spherical induction motor capable of three degrees of freedom can be inferred.

Autonomous driving perception is commonly developed and evaluated under a set of enabling assumptions: that multi-sensor evidence is physically consistent at the frame level, that geometry is sufficiently dense to support reliable inference about other traffic participants and the surrounding environment, and that learning can rely on either abundant human labels or self-supervised objectives derived from the sensor stream. This thesis examines what remains feasible when these assumptions no longer hold, and develops methods and design principles for perception under asynchronous sensing, long-range sparsity, and weak or unreliable supervision.

We first study physical inconsistency in multi-sensor data. We show that rolling and asynchronous acquisition, motion during aggregation, and annotation practices that implicitly assume temporal coherence can render the perception problem ill-posed before any representation choice is made. We therefore treat data preparation, motion compensation, and annotation consistency as integral parts of the perception pipeline, since errors at this stage can propagate directly into annotation, training, and evaluation.

We then examine representation under long-range sparsity. We show that long-range performance is limited not only by model capacity, but by the representations used to encode and expose ambiguous evidence. In particular, object-centric outputs and dense internal representations can force premature commitment when available evidence collapses at distance. To study this, we present results on long-range 3D object detection and sparse long-range scene flow, showing both the limits of object-centric perception under weak observability and the value of motion-centric estimation as range increases.

Finally, we study learning signals when labels and geometry-derived self-supervision become unreliable. We show that motion supervision can be recovered by importing physically grounded constraints from complementary modalities, using radar Doppler to guide LiDAR scene flow learning. We further show that scalable semantic supervision can be obtained from foundation-model priors through curriculum-based synthetic-to-real adaptation, which anchors language-aligned representations to real LiDAR characteristics.