Authorship analysis relies on identifying linguistic and stylistic characteristics that can distinguish an author's writing. However, these characteristics can vary with factors such as topic, domain, text length, and the amount of available authorship data, making reliable authorship analysis challenging. This thesis investigates authorship analysis, focusing on the tasks of authorship attribution (AA) and authorship verification (AV), as well as textual relationships relevant to literary influence. It examines textual features and representations and different approaches based on machine learning (ML), deep learning (DL), semantic similarity, and large language models (LLMs).

The thesis first examines the current landscape of AA and AV through a systematic literature review of machine learning (ML), deep learning (DL), and LLM-based approaches. The review analyzes methods, textual features and representations, datasets, and languages, and identifies current challenges, research gaps, and directions for further investigation.

The thesis then investigates open-source LLMs and style embeddings for zero-shot AA and examines how different author representation strategies affect their performance. The experiments use different strategies for providing author-related representations to examine the impact of additional authorship-related information on attribution performance. The results highlight the challenging nature of the task and show that the investigated LLMs struggle to reliably distinguish authors.

Finally, the thesis investigates textual analysis for identifying relationships and similarities between texts in the study of potential literary influence. Using Herman Melville as a case study, BERTScore is applied to measure semantic similarity between selected passages from his writings and texts documented as part of his reading. The analysis identifies similarities previously recognized by literary experts and highlights additional passages for further qualitative examination.

Ensuring adequate durability of mass concrete requires controlling its early-age thermal cracking risk. For this purpose, proper curing techniques must be rationally selected based on concrete’s temperature, stress and strength developments. Hence, understanding the evolution of young concrete strength, heat of hydration and basic creep, is mandatory. Moreover, the decarbonization of the construction industry requires increasing the knowledge on hardening Low Carbon Concrete (LCC). In Sweden, interest exists on testing the early-age properties of modern LCC based on Volcanic Pozzolan (VP), Limestone Filler (LF) or Ground Granulated Blast Furnace Slag (GGBFS). Additionally, the existing material models must be evaluated and modified if required. 

In response, this study tested and modelled the strength, heat, temperature and basic creep of hardening LCC containing GGBFS, VP and LF. In total ten concrete mixes were studied, including reference Portland Cement Concrete mixes for comparison. All mixes compressive and tensile strength developments were tested at various ages and bath curing temperatures. The heat of hydration was tested using semi-adiabatic and isothermal calorimetries. In addition, the compressive basic creep at different loading ages was also tested. Novel formulations for the maturity, strength, heat and basic compliance, were proposed as improvements to those currently used in Sweden. Then, the potential of the studied LCC for the winter cast of a ground slab was numerically evaluated, considering insulating formwork and heating cables as control measures. For this purpose, simulations of the temperature, maturity and strength fields were combined to a novel surface cracking risk index proposal. 

The results show that the GGBFS mitigates the strength reduction due to thermal-damage and promotes a better strength gain under cold curing. The VP was found increase the heat generation, which can be beneficial for winter cast under controlled conditions. Regarding the basic creep, the GGBFS increased the compliance at early loading ages but reduced its value at later loading ages. Differently, the VP increased the instantaneous compliance but reduced the creep rate. On the modelling, the tensile strength model proposed here was found more accurate than the Swedish and fib Model Code proposals. Also, the heat model proposed here remarkably performed in predicting the heat flow kinetics of concretes with extended dormant periods. Moreover, for the maturity, the novel aging activation energy formulation proposed here, was found beneficial to transform the heat measured in the semi-adiabatic test to isothermal conditions. In addition, the basic compliance model proposed here, based on the solidification theory, was found to provide the best fitting for the LCC studied here. Finally, the winter cast simulations showed the potential of the heating cables in control the early risk of surface cracking on VP based concrete. 

Froth flotation is a processing technique at the heart of mineral processing, with widespread use and a long history. However, achieving reliable advanced control and optimisation of the process has remained a challenge for the process control community. This can be explained by the multivariable and nonlinear nature of the process. Combined with the difficulty of measuring metallurgical performance in an industrial flotation setting, this makes the problem particularly challenging. 

This licentiate thesis investigates data-driven approaches to improve the control and optimisation of froth flotation. The work considers both model-free optimisation methods and Gaussian process models to further enhance the ability to operate the process optimally. 

First, a geometrically constrained extremum-seeking control method is studied to optimise the process while maintaining stable operation. The results support the potential of extremum-seeking control as a control methodology for driving the process to an optimum while accounting for interactions with the regulatory level of froth flotation control.

Second, a Gaussian process regression model of froth flotation is investigated as a probabilistic method for modelling the dynamics of an industrial flotation bank. The proposed model was evaluated against a previously derived parametric model, showing improvements in predictive capability, thus demonstrating the method's potential for froth flotation control. 

Third, a Gaussian process state space model was developed to further explore methods for training such models in the presence of noisy datasets. Illustrating moving horizon methods as a way to learn Gaussian process models in a state space format. Thus, further enabling the methods to be used in noisy industrial data settings such as Froth flotation processes.

Together, these studies indicate that data-driven methods can help bridge the gap between process complexity and instrumentation availability. Further work will investigate how these models can be combined to tackle industrial challenges in froth flotation and beyond.

In the context of autonomous systems, public safety missions involve deploying robotic agents to assist in tasks such as search and rescue, surveillance, and situational assessment in environments that pose risks to human responders. These missions pose fundamental challenges for autonomous robotic systems operating in both single-agent and multi-agent settings, as they require reliable and robust operation in unstructured and dynamic environments under partial observability and high uncertainty, while satisfying strict task requirements and safety constraints. In multi-agent deployments, these challenges are further complicated by the need for coordination among agents under limited or unreliable communication. 

This thesis devises reinforcement learning-based autonomy for robotic systems across increasing levels of complexity, ranging from safety-aware single-agent operation to decentralized multi-agent coordination under communication and dynamic constraints. First, the thesis addresses single-agent exploration in unknown, cluttered environments. The proposed approaches integrate reinforcement learning–based behavioral policies for waypoint selection, leveraging graph-based representations to enable structured spatial reasoning over candidate waypoints and frontiers via attention-based graph neural networks. Safety is explicitly enforced via hierarchical decision-making, where a learned exploration policy is coupled with a safety filter that prevents the execution of movements leading to collision with non-traversable regions. The thesis then extends to decentralized multi-agent exploration under limited connectivity. Exploration is formulated as a decentralized role-assignment problem in which structurally homogeneous agents learn independent policies that adaptively switch between exploration and communication with nearby agents, balancing local map expansion with information sharing through proximity-based networks. However, real-world missions often involve heterogeneity among agents, where teams are composed of platforms with differing dynamics, sensing modalities, endurance, and operational constraints, and where effective coordination requires leveraging these asymmetries. In this direction, the thesis initiates a study on cooperative target acquisition involving structurally heterogeneous agents characterized by differing system dynamics. A decentralized reinforcement learning framework is introduced in which agents learn individual policies supported by graph-based communication and trajectory-aware safety filters. In conclusion, this thesis demonstrates that safe decentralized multi-agent reinforcement learning enables autonomous agents to effectively coordinate exploration, communication, and task execution under constraints such as partial observability, limited connectivity, and dynamic environments. 

A free-burning thermal plasma jet is a poorly characterised medium. It is far hotter than a flame, chemically active, strongly non-uniform, and unsteady on timescales set by the motion of the arc and by the turbulence of the jet, and almost none of its interior is accessible to a physical probe. Basic questions therefore remain open: what the thermochemical state of the jet actually is, how the gas temperature and composition are distributed and how they respond to the way the torch is operated, how strongly the jet fluctuates and how that unsteadiness is coupled to the electrical behaviour of the discharge, and which measurements can reach these quantities at all in an environment that is at once intensely luminous, spectrally congested and steeply non-uniform. These questions are of interest in their own right, and they are also what currently limits the design and scale-up of plasma burners intended to replace fossil-fuel burners in high-temperature process industries, in processes such as iron-ore pelletising, cement and lime production and steel reheating, where the associated nitrogen-oxide formation is a further open problem.

This licentiate thesis addresses these questions by developing optical diagnostics for thermal plasma jets and for the high-temperature gases they produce, and applying them to a non-transferred plasma torch. Mid-infrared laser absorption spectroscopy was first established for a target species, nitric oxide, in plasma-heated and combusting gases, where the medium is hot and rich in water vapour but comparatively well behaved; the same approach was then carried into the plasma jet itself, where carbon monoxide serves both as the dominant carbon-bearing species of a carbon-dioxide-fed discharge and as an indicator of the dissociation, and hence the thermal, state of the gas, and where the measured absorption feature, formed by two transitions whose lower states differ in energy by 3065 cm⁻¹, yields the gas temperature simultaneously with the concentration through the temperature dependence of the two line strengths. The same line pair was recorded in premixed flat flames of known and computable state, which fixes the accuracy attainable when the medium does not vary along the beam. In parallel, spontaneous Raman scattering was used to obtain the gas temperature at a point inside the luminous core of the jet, which is spatially resolved and therefore complementary to the line-of-sight character of absorption. Because absorption necessarily reports a quantity averaged along the beam, while the jet possesses steep gradients in temperature and composition, a central concern throughout is what such an averaged value represents and how it can be related to the local thermodynamic properties of the plasma. The high repetition rate of the laser scan, together with synchronous recording of the discharge current and voltage, additionally samples the fluctuation of the jet rather than averaging over it, which is the precondition for relating its thermochemical unsteadiness to the behaviour of the arc.

Taken together, the appended papers establish a common measurement chain that extends from plasma-heated gases at pilot scale to the plasma jet at torch scale, and they deliver absolute, calibration-free temperatures and species concentrations of a kind that has so far been reported only to a very limited extent for this class of jet. They also identify what such measurements can and cannot resolve, in particular the interpretation of path-averaged quantities in a strongly non-uniform medium and the accuracy limits imposed by high-temperature spectroscopic parameters. The resulting methods and data provide the experimental basis against which magnetohydrodynamic and chemical-kinetic models of the torch can be validated, and thereby an initial step toward the design of large-scale, low-emission plasma burners for energy-intensive industry.

Thermal exposure from hydrogen flames is determined by the spectral properties of the combustion products, the optical thickness of the flame, and the attenuation of the emitted radiation during propagation through the surrounding atmosphere. This licentiate thesis investigates the effect of flame thickness on emitted spectral radiance, the fractional radiative contribution of species present in hydrogen flames, and the propagation of the emitted radiation through humid air, using line-by-line spectral calculations based on the HITRAN/HITEMP spectroscopic databases and Voigt line profiles.

The effect of flame thickness on the emitted spectral radiance is analysed by evaluating the spectral optical thickness, , and determining the fractions of the spectrally integrated radiance associated with optically thin, intermediate, and optically thick regimes. For flame thicknesses 𝐿 ≤ 0.1 m, more than 90% of the emitted radiation satisfies the optically thin criterion ≤ 0.1. As flame thickness increases, most of the  radiation that no longer satisfies the optically thin criterion enters the intermediate regime, whereas only a very small fraction reaches the optically thick limit.

The fractional radiative contributions of the species present in hydrogen flames are determined. The emitted radiation is dominated by water vapour, which contributes more than 96.6% of the total, followed by CO₂ at approximately 3.0% and OH at approximately 0.4%. Each of the remaining species contributes less than 0.01%.

The propagation of the emitted radiation through humid air is analysed. Since H₂O is the dominant radiating species in the flame and an absorbing species in humid air, the attenuation is determined by the spectral mismatch between the hydrogen flame emission spectrum and the absorption spectrum of ambient water vapour. Depending on relative humidity (RH), humid air absorbs approximately 10–20% of the emitted radiation within the first metre and approximately 20–40% over a 10 m propagation path.

The freezing of water droplets on cold surfaces is a fundamental process in many natural phenomena and engineering applications. The freezing process involves coupled heat transfer, phase change, moving interfaces, and internal liquid motion. Experimental studies have reported changes in the direction of internal circulation during freezing, but the physical mechanisms responsible for this behavior are not yet completely understood. This thesis investigates the thermal behavior and internal flow of sessile water droplets by combining experiments with numerical modeling.

The experimental investigation uses coupled monitoring and infrared cameras to com- pare two freezing regimes. In the supercooled regime, the droplet cools below the equilibrium freezing temperature before nucleation and subsequently undergoes recalescence. In the immediate-onset regime, freezing begins shortly after the droplet contacts a cold substrate. Observations show differences in droplet transparency and surface temperature evolution between the two regimes.

For numerical part, two OpenFOAM solvers are developed to investigate selected mechanisms separately. The first model combines the enthalpy-porosity method with a nonlinear temperature-dependent relation for water density. The nonlinear density formulation captures the experimentally reported reversal of internal circulation. The reversal occurs when the maximum liquid temperature approaches the density-maximum temperature of water. Quantitative differences from the experiments indicate that natural convection alone cannot fully explain the measured internal flow.

The second model combines VOF-based free-surface tracking with phase change and a modified mass-conservation formulation. The density difference between water and ice is introduced as a volumetric source during solidification, allowing the model to represent expansion, displacement of the remaining liquid, and deformation of the liquid-gas interface. Verification cases demonstrate that the implemented source term and phase-change treatment reproduce the expected interface motion and volume change. Simulations of sessile droplets show that expansion generates an additional internal velocity field directed away from the advancing solidification front.

The results demonstrate that internal motion during droplet freezing can arise fromseveral mechanisms with different physical origins. Natural convection determines circulation through temperature-dependent buoyancy, while volume expansion producesflow directly through phase change and liquid displacement. Treating these mechanismsseparately clarifies their individual contributions and provides a foundation for futuredevelopment of a unified model.

The increasing demand for safe, efficient, and sustainable energy-storage technologies has urged the need for advanced electrolytes for next-generation batteries. Although fluorinated salts, solvents, and ionic liquids (ILs) are widely employed as electrolyte components because of their favorable electrochemical properties, several concerns associated with their flammability, lower chemical and electrochemical stabilities, hazardous decomposition products, environmental persistence, and end-of-life management have motivated the development of non-flammable and fluorine-free alternatives. This thesis is focused on the structural design, synthesis, and electrochemical characterization of new fluorine-free ILs and electrolytes, with a particular emphasis on understanding how molecular structure and ionic interactions influence their physicochemical, transport, thermal, and electrochemical properties.

The first part of the thesis investigates a series of fluorine-free ILs based on methane sulfonate and ethane sulfonate anions combined with pyrrolidinium, piperidinium, imidazolinium, morpholinium, and ether-functionalized pyrrolidinium cations. The ILs and their electrolytes exhibited thermal decomposition temperatures of 300–360 °C and ambient-temperature ionic conductivities ranging from 0.04 to 8.5 mS cm⁻¹. Among these systems, pyrrolidinium-based ILs exhibited favorable ionic conductivities, ion diffusivities, and electrochemical stabilities. The second part of the thesis addresses ion association through an anion-receptor strategy. A fluorine-free neutral bis-thiourea anion receptor, α,α′-bis(N′-phenylthioureido)-m-xylene (mXyl-PhTU), was found to selectively interact with anions through non-covalent interactions and thereby facilitate lithium salt dissociation and thereby elevating the Li⁺ mobility. The third part of the thesis introduces a series of new fluorine-free ILs with anions derived from caffeine; the theobromine and theophylline anions coupled with phosphonium and N-heterocyclic cations. Overall, this thesis highlights the importance of new molecular structures and the role of ion association for optimizing electrolyte behavior and provides a basis for the rational development of fluorine-free ILs and electrolytes for next-generation batteries.

Ore pass degradation is a critical challenge in underground mining, affecting operational reliability, safety, and production efficiency. While existing ore pass longevity models emphasize geological, geometrical, and operational factors, the contribution of groundwater and groundwater control to ore pass performance has not been extensively investigated. Cementitious grouting is widely used to reduce groundwater inflow in underground excavations; however, its influence on the mechanical behaviour of rock fractures and the resulting implications for ore pass longevity are not yet fully understood. This thesis investigates the effect of cementitious grouting on the shear strength of fractured rock and examines how this may contribute to improved ore pass performance. 

A laboratory testing programme was conducted to evaluate the influence of fracture roughness, normal stress, grouting coverage, and grout curing time on fracture shear strength behaviour. Artificial fractures with controlled roughness were subjected to direct shear testing under different confinement levels and grouting configurations. The results demonstrate that the mechanical effect of grouting is highly conditional. Under favourable conditions, grouting increased fracture shear strength by providing additional cohesion and load transfer across fracture surfaces. In other cases, particularly when grout maturity or fracture conditions were unfavourable, reductions in shear strength were observed. The interaction between normal stress, fracture roughness, grout distribution, and curing time was found to govern the overall mechanical response. 

The findings indicate that grouting influences ore pass longevity through both hydraulic and mechanical mechanisms. Beyond reducing groundwater inflow, cementitious grout may strengthen fractures and improve rock mass stability when properly designed and applied. The findings indicate that grouting should be evaluated not only as a groundwater control measure but also for its effect on fracture strength, which can vary depending on the fracture characteristics and stress conditions. These results contribute to a better understanding of groundwater related degradation processes and provide new insight into the potential role of cementitious grouting as a strategy for improving ore pass longevity in underground mines.

Large-scale underground hydrogen storage is expected to play an important role in balancing variable renewable energy supply and future hydrogen demand. In regions dominated by hard rock, lined rock caverns (LRCs) provide a potential storage solution. Repurposing existing underground oil storage caverns could further reduce the need for new excavation; however, these caverns were not originally designed for high-pressure gas storage and commonly have shallow cover and elongated geometries that differ substantially from conventional LRCs. This thesis investigates the feasibility and key engineering considerations of converting such oil storage caverns into LRCs for underground hydrogen storage.

The research is developed through three interconnected numerical studies. The first provides a geological foundation by investigating how in-situ stress and rock mass discontinuities influence anisotropic cavern deformation during excavation. The results demonstrate that joint orientation and stiffness can substantially modify both the magnitude and spatial distribution of cavern deformation, highlighting the importance of geological structures in assessing existing caverns. The second study serves as a pilot geometric-design investigation, examining the effects of cavern-bottom geometry and sliding layer stiffness on steel liner stress and interface displacement. Smoother bottom transitions generally reduce local liner demand, particularly when interface restraint is high, although reductions in stress and interface displacement do not necessarily coincide. These two studies provide geological and geometric understanding for the central investigation of oil-cavern repurposing.

Building on these preliminary studies, the main investigation evaluates two concepts for converting shallow, elongated oil storage caverns into hydrogen-storage LRCs: a Multi-cavern scheme, in which the existing excavation is divided into multiple lined sub-caverns, and a Single-cavern scheme, in which one continuous lined volume occupies the existing cavern. A modified uplift failure model is used to obtain a preliminary maximum operating gas pressure estimate, after which numerical analyses evaluate liner stress, yielding of the concrete and surrounding rock, storage capacity, material demand, and sensitivity to key geological and design parameters. Short-term cyclic thermo-mechanical analyses are additionally performed to examine the implications of hydrogen injection and withdrawal.

The results reveal a clear trade-off between structural performance and storage/material efficiency. At the benchmark pressure of 3.3 MPa, the Multi-cavern scheme reduces the maximum steel liner tensile stress from 32.4 MPa to 25.4 MPa and has a total plastic-zone volume approximately 18% of that of the Single-cavern scheme. In contrast, the Single-cavern scheme provides approximately twice the effective gas-storage volume and a substantially lower indicative material cost per unit mass of stored hydrogen. Rock mass quality strongly influences structural response, demonstrating the importance of geological conditions in retrofit assessment. The cyclic analyses further show that static loading at the maximum operating gas pressure does not conservatively bound all response quantities: compared with the cyclic cases examined, it produces higher steel liner stress and concrete yielding but lower rock mass and total yielding.

Overall, the thesis demonstrates that repurposing existing oil storage caverns using LRC technology is a technically promising option, but that retrofit feasibility cannot be assessed from storage volume or pressure capacity alone. Geological conditions, LRC geometry, global uplift resistance, liner response, material demand, and operational loading need to be considered together. The work establishes a preliminary assessment framework and identifies the principal issues requiring further development toward project-specific design. In particular, the continuing PhD research will extend the geological investigation to construction and pressurized operational stages and develop more detailed performance-based and coupled analyses for shallow LRC hydrogen storage.