With the increasing use of railway transportation, efficient techniques for solving related personnel scheduling problems are becoming increasingly important. Traditionally, such problems have been solved by decomposing them into two smaller problems: the Crew Scheduling Problem (CSP) and the Crew Rostering Problem (CRP). This thesis investigates the possibility of achieving better overall solutions by combining the CSP and CRP into an Integrated Scheduling Rostering Problem (INT) for realistic railway instances based on data from the Swedish railway operator SJ.

Column Generation models for the CSP, CRP, and INT are proposed. The resulting subproblems take the form of resource-constrained shortest path problems, for which tailored labeling algorithms are developed. Several acceleration strategies are introduced to improve computational performance, and different initialization schemes for the INT are explored.

The results demonstrate improved computational performance of the column generation approach compared with the search-based method, as well as significant speedups from the tailored labeling algorithms relative to commercial solvers. The INT is also shown to achieve better solutions than the sequential approach. The main advantage of the INT is its ability to find solutions requiring fewer rosters. For the largest INT instances, long computation times remains a challenge. The results, however, indicate that early termination of the column generation procedure, together with extrapolation from shorter planning horizons, can provide high-quality solutions while significantly improving computation times.

In recent years, artificial intelligence (AI) has emerged as a prominent topic of interest, attracting growing attention across both industry and academia. In the rise of rapid advancements in digitalization, data availability, and computing power, organizations are increasingly exploring the potential for integrating AI in organizational decision-making processes. This opens up for a new landscape of decision-making, when humans and AI continuously interact to make decisions towards reaching organizational goals. However, there is still limited understanding of how such human-AI decision-making unfolds in practice, particularly in complex organizational contexts such as personalized medicine, where currently the integration of AI into decision-making remains at an early stage.

This thesis explores human-AI decision-making as an organizational process occurring in complex contexts. To do so, it employs a phenomenon-driven strategy, considering human-AI decision-making in organizations as a continuously emerging phenomenon in its center. The empirical evidence is gathered in the field of personalized medicine and data-driven healthcare context in Sweden, including cases of breast cancer diagnostics and personalized medicine development.

Based on the embedded case study performed, this thesis contributes to a multi-level understanding of human-AI decision making, moving beyond perspectives that view AI mainly as a decision support tool. At the organizational level, AI influences attention structures, search processes, decision premises, routines, and governance arrangements. As a result, decision-making is shown to be distributed between human and AI in ways that are neither fixed at design nor evenly spread, but allocated according to specific contextual conditions.

At the interactional level, AI becomes meaningful through ongoing sensemaking processes in which actors interpret what the technology does, what its outputs mean, and how those outputs should be incorporated into organizational action. Sensemaking for AI anticipation is done by relating AI’s output to something already familiar and already legitimate like an established professional norms, and transparent decision environment. Additionally, situated sensemaking is grounded on temporality, which means that for each decision in a point in time humans and AI can take different roles depending on the context and situatedness.

At the individual level, AI alters the conditions under which professional judgment is formed, exercised, challenged, and legitimized. At the same time, AI outputs are themselves interpreted through existing heuristics, professional frames, and organizational expectations.

Finally, practical implications focus on AI integration that is grounded on distributed effort and fluid roles within humans and AI configurations, as well as organizations periodically reflecting on which aspects of reality AI solutions are making more visible and which may be receiving less attention as a result in decision-making processes. In this way, organizations can develop forms of human–AI collaboration that support informed decisions in complex environments.

Auditory distance perception relies on multiple acoustic cues, with sound level serving as the dominant cue. While near-field distance perception under air conduction (AC) stimulation has been extensively studied, comparatively little knowledge is known about distance perception under bilateral bone conduction (BC) stimulation. The present study investigated near-field auditory distance perception under AC and BC stimulation using virtual sound sources presented at distances between 0.2 and 1.0 m and at azimuths of 45°, 90°, and 135°. Two experiments were conducted, one with natural distance-dependent level cues preserved and one with level cues removed via normalization. Distance estimates were quantified using power-function exponents as a measure of perceptual distance compression. When level cues were preserved, distance perception did not differ significantly between AC and BC stimulation. Under level-normalized conditions, perceived distance ranges were significantly reduced for BC compared with AC at lateral azimuths. These results are consistent with the interpretation that sound level is the primary cue for auditory distance perception under both stimulation modalities, and that binaural cue degradation under BC stimulation limits distance accuracy when level cues are unavailable. The findings have implications for the design and signal processing of BC auditory devices intended to support spatial hearing. 

The aim of this study was to investigate how voice familiarity influences neural processing

during language comprehension using EEG and time-frequency analysis. The study focused

on neural oscillations within the theta and alpha frequency bands, as previous research has

linked these frequencies to language comprehension, attention, and predictive processing.

The study was based on EEG data from a previous experiment in which participants listened

to sentences spoken by both familiar and unfamiliar voices. EEG data were analysed in

MATLAB and FieldTrip using wavelet-based time-frequency analysis within the 2–20 Hz

frequency range. The analysis mainly focused on activity at electrode Cz. Data from ten

participants were included in the analysis.

The results showed visual differences in oscillatory activity between familiar and unfamiliar

voices, particularly within the 6–10 Hz frequency range and during the 0.5–1.5 second

interval after speech onset. An exploratory paired t-test analysis was conducted, but no

statistically significant difference was observed between conditions, t (9) = 0.97, p ≈.36.

Visual differences could be observed in the time-frequency plots, particularly within the 6–

10 Hz frequency range. However, because the results were not statistically significant, it is

not possible to determine whether these differences reflect a genuine effect of voice

familiarity or random variation within the sample.

The findings are interpreted as preliminary indications that voice familiarity may influence

neural processing during language comprehension. The study supports previous theories

describing language comprehension as a predictive process influenced by prior experience

and social information. However, larger participant groups and more advanced statistical

analyses are needed to draw firmer conclusions.

Denna studie syftade till att undersöka hur ett digitalt smittspårningsverktyg för hepatit B och C kan utformas för att stödja vårdpersonalens kognitiva processer. För att undersöka detta har  denna kandidatuppsats utforskat frågeställningen: Hur kan ett digitalt smittspårningsverktyg utformas för hepatit B och C för att stödja vårdpersonals kognitiva processer vid smittspårning? Studien använde forskning genom design som metodansats tillsammans med en designprocess beskriven av Arvola (2020). Designprocessen innefattade tre steg: konceptfas, bearbetningsfas och en detaljeringsfas, dessa anpassades utifrån studiens begränsningar och syfte. Under designprocessens gång genomfördes en semistrukturerad intervju med en före detta smittspårare för att identifiera vilka behov som fanns för ett digitalt verktyg. Resultatet låg till grund för de designkoncept som identifierades, vilket inkluderade enhetlig design över diagnoser, diagnosanpassad informationsvisning, stöd för överlämning mellan användare, avgränsad och relevant funktionalitet, struktur som stödjer smittspårningsflödet samt en kombination av fritext och fördefinierade alternativ. Designkoncepten användes vid utvecklingen av prototyper i studien. Först skapades en Lo-Fi-prototyp, denna användes som grund för vidare utveckling av en Hi-Fi-prototyp. Den framtagna Hi-Fi-prototypen utvärderades genom användartester, vilket resulterade i ytterligare insikter kring vad som fungerade väl, vilka delar som saknades samt vilka förbättringar som önskades för framtida utveckling. Resultaten från användartesterna och de semistrukturerade intervjuerna som genomfördes i samband med testerna tydde på att ett smittspårningsverktyg med den föreslagna designen kunde stödja vårdpersonalens kognitiva processer genom att stödja arbetsflödet, minska informationsöverflöd och anpassa information utifrån val av diagnos och relevanta alternativ. Detta besvarar därmed forskningsfrågan genom att indikera att ett strukturerat och diagnosanpassat smittspårningsverktyg kan stödja vårdpersonalens kognitiva processer. Sammantaget bidrog studien med insikter som kan vara värdefulla vid utveckling av digitala smittspårningsverktyg för hepatit B och C.

Quantum key distribution (QKD) is sometimes marketed as secure against computational attacks, as its security is based on the laws of quantum mechanics for detecting eavesdroppers, rather than the difficulty of mathematical problems. However, this security guarantee applies only to the quantum channel itself, and the physical hardware implementing these systems remain vulnerable. The electrical components controlling these systems generate electromagnetic emissions when active, and if those emissions carry information such as the polarization states being encoded, an attacker could recover key-relevant information without ever interacting with the quantum channel, bypassing the security guarantees of the protocol entirely. This work investigates whether such a vulnerability exists in a commercial QKD transmitter unit by applying various unsupervised and supervised machine-learning methods to near-field electromagnetic measurements, with the aim of determining whether device-level parameters and the prepared polarization states can be extracted from the observed emissions alone. The developed unsupervised machine-learning methods were unable to recover the distribution of polarization states from the electromagnetic measurements, partitioning a single dominant cluster into arbitrary groupings, rather than finding separate clusters. When trained on files with various distributions of key and check bases, the Ridge regression model exhibited strong predictive capability, accurately following the proportions of D-states during leave-one-out cross-validation, with a mean absolute error (MAE) of 4.9% and an R² score of 0.946. During testing on new, unseen files, the MAE increased to 11.3% and the R² score dropped to -0.12, largely attributed to a single file, with the remaining test files producing results more consistent with the cross-validation performance. Similarly, a convolutional neural network trained on fixed five-symbol sequences achieved an accuracy of 41.13% and a F1-score of 36.46%, which for the given symbol space is significantly above random guessing of 33.33%.

The transition towards a circular economy requires improved recycling of plastic components in heavy trucks to achieve circular flows of plastic. Large material variety and complex product designs present significant challenges. This master’s thesis investigates the factors which affect the recyclability of plastic components in heavy trucks, explores how circular product design can improve the recyclability of them, and evaluates how workers’ mental workload during sorting for recycling can be reduced.

A qualitative research approach has been conducted following a double diamond process, including literature review, interviews and study visits with actors across the value chain, observations, and material process flows mapping. Furthermore, design methods such as brainwriting, braindrawing, a focus group, and a usability test were conducted to develop and evaluate solutions.

The results show that the recyclability of plastic components in heavy trucks is affected by a wide range of factors. In total, 57 factors were identified, including 10 enablers, 23 overarching barriers and 24 design barriers. A key finding is that current design practices significantly hinder the recyclability of plastic components in heavy trucks.

To address these issues, circular product design guidelines for increased recyclability was developed, synthesizing recommendations from design for recycling, design for disassembly, and other sustainable recommendations. The guidelines focus on material selection, structure, fastening methods, disassembly, minimizing contamination, and improving material identification. Validation through a focus group indicated that the guidelines are relevant, practical, and valuable for integration into the product development process.

In addition, a new marking system for sorting plastic components was proposed and evaluated through a usability test. Compared to the current system, the new marking concept significantly improved sorting accuracy and reduced workers’ perceived mental workload. The results highlight that simplifying information and reducing decision-making complexity are key to improving manual sorting.

The thesis concludes that improving recyclability of plastic components used in heavy trucks requires both the implementation of circular product design and systemic changes among the actors within the heavy truck value chain. Implementing circular product design guidelines for increased recyclability and user-centered marking systems can substantially enhance recycling efficiency while reducing cognitive strain for workers. However, successful implementation also requires collaboration across the value chain and alignment with technical and economic conditions.

Different state management approaches and standby configurations can affect how traffic is handled during failovers in a Radio Access Network (RAN). This thesis investigates state externalization in a RAN-inspired environment and how different state management approaches affect throughput, packet loss, and recovery time during normal operation and during failover. To evaluate this, a prototype was implemented using evgen to simulate a RAN-like traffic handling function. The implementation compares a local state approach, where state is maintained within the function instance, with an externalized state approach using a shared state store. Different failover scenarios were also evaluated by comparing cold standby and hot standby takeovers.

The results showed that the local state had slightly higher throughput during normal operation, while the externalized state showed higher packet loss. During failover, the externalized state showed slightly lower throughput degradation along with decreased packet loss relative to its baseline compared to the local state. When comparing cold standby and hot standby takeovers, hot standby showed a shorter recovery time. Since the evaluation was performed in a controlled prototype environment, the results should be interpreted within the limitations of the implementation. Ideas for future research are also presented.

The design of road infrastructure plays a decisive role in traffic safety and is strengthened through elements such as refuge islands, speed humps and roundabout elements. Two approaches are currently applied, prefabricated and cast-in-place solutions, but a clear basis for selecting between methods is lacking. The study, conducted in collaboration with Trafikverket, aims to compare the methods in terms of construction time, cost and work environment. The study was conducted through semi-structured interviews with industry representatives including clients, contractors, facility managers and manufacturers, complemented by a literature review. The results show that prefabricated elements are faster to install, as processes such as curing and stone setting are already completed in the factory. Although prefabricated elements initially entail higher material costs compared to cast-in-place alternatives, this perception shifts when the full picture is considered. When factors like construction time, machinery costs and long-term maintenance are taken into account, prefabricated elements are regarded as the more cost-effective solution over time, particularly under high traffic loads. From a work environment perspective, prefabricated elements are more advantageous because shorter exposure time in traffic reduces the risk of accidents and the most physically demanding tasks are carried out under ergonomically better conditions in the factory.

This master’s thesis aims to investigate why IT investments are made in Swedish municipalities, how decisions regarding IT investments are made in Swedish municipalities, and what challenges Swedish municipalities face regarding IT investments. The study is therefore guided by three research questions: “Why are IT investments made in Swedish municipalities?”, “How are IT investment decisions made by decision-makers in Swedish municipalities?”, and “What challenges do Swedish municipalities face regarding IT investments?”. The theoretical framework is based on a combination of institutional theory, public value theory, and bounded rationality theory to analyze IT investments as something complexed shaped by external pressures, internal value aspirations, and cognitive limitations. The study employs a qualitative, interpretivist approach and was conducted through semi-structured interviews with seven key individuals involved in IT investments from five different Swedish municipalities. The data was analyzed using thematic analysis. 

The results show that IT is not viewed as a goal in itself, but rather as a mean and a support function for enabling the municipality’s core mission and delivering public value. The primary objectives of IT investments are to achieve more efficient operations, better service, and to contribute to equitable welfare for residents. IT investments in this study are defined in line with previous research. Decisions regarding IT investments are made in environments characterized by bounded rationality, where decision-makers seek sufficiently satisfactory solutions rather than optimal ones. This is exacerbated by the difficulty of quantifying the actual benefits and impact of IT investments, where formal benefit-realization calculations are used in an attempt to manage uncertainty. The main challenges identified were financial constraints, decision-making processes, vendor lock-in and dependency, as well as expertise and size.

Planning of future capabilities in complex societal-bearing organisations depends on the ability to understand large amounts of complex data. This thesis project includes the development and evaluation of a web-based visualization tool where the underlying sociotechnical data is formatted as a node-link diagram. This thesis aims to implement a visualization tool to help users with different experiences to understand aspects of complex sociotechnical data and relations within the data for planning and decision-making in a military organisation. The implementation of the visualization tool was guided by a literature review as well as the user requirements and desires which were identified at Swedish Defence Research Agency (Swe. Totalförsvarets Forskningsinstitut (FOI)) through the use of user stories. Furthermore, the implementation was conducted through two iterations of prioritising milestones created from the user stories, sketching potential mock-ups and developing the necessary tools. The final evaluation was done using a form, a test and a pre task interview before each user was tasked to complete a set of ten tasks using the visualization tool. After completing the ten tasks another interview was conducted before each user were asked to complete the System Usability Scale (SUS) and ICE-T questionnaires. The results indicate that the prototype developed has an overall usability value for the intended end-users and domain application through the visual representations, design choices and interaction techniques implemented. Naturally, there are factors and aspects affecting the degree of this usability value and these are addressed and discussed as well. An outlook of future development to extend the prototype to align even better with the aim and the identified user tasks and user requirements is also conducted.

Den snabba framväxten och ökade användningen av generativ artificiell intelligens (GAI) inom högre utbildning har skapat ett pedagogiskt dilemma mellan effektiv kodgenerering och djupinlärning. Denna studie undersöker hur universitetsstudenter integrerar och tillämpar GAI i sina självstudier inom programmering. Syftet är att förstå hur promptingstrategier och iterativa interaktionsmönster påverkar det självstyrda lärandet. Detta är en kvalitativ studie som har tillämpat en deduktiv ansats. Empiriskt material har samlats in genom semistrukturerade intervjuer med sju studenter från system- och kognitionsvetenskapliga programmen vid Linköpings Universitet. Empirin analyserades tematiskt utifrån etablerade promptingstrategier och QED-ramverket (Questioning, Exploring, Discovering). Resultaten visade att studenterna är aktiva i sin GAI-användning. De tillämpar medvetna strategier, såsom att begränsa systemets output för att få konceptuella förklaringar snarare än färdig kod, samt att bryta ner problem i sekvenser för att undvika kognitiv överbelastning. Även om snabb kodgenerering lockar, uttrycker studenterna en stark ambivalens inför risken att förlora kontrollen över sin egen problemlösning. Särskilt framträdande är användningen av reflekterande och utforskande promptingstrategier, där studenter ger systemet konstruktiv feedback eller byter roll för att låta AI:n agera förhörsledare. Studiens huvudsakliga kunskapsbidrag är insikten att GAI i sig självt varken garanterar eller hindrar lärande. Utfallet är helt beroende av studenternas förhållningssätt och val av inmatningsstrategi. Genom att synliggöra hur studenter använder GAI som en interaktiv och pedagogisk partner bidrar studien till en ökad förståelse för hur framtidens utbildningar bör formas. Praktiskt bidrar studien med värdefulla insikter för universitetslärare och utbildningsansvariga vid utformningen av pedagogiska uppgifter, samtidigt som den stöttar studenter i att utveckla mer reflekterande och självständiga studiestrategier.

In hydrogeological assessments for construction projects, the consulting firm WSP monitors groundwa- ter levels using project-specific observation wells. Because the resulting measurement series are often irregular, they are typically supplemented with data from reference wells in the permanent monitoring network operated by the Geological Survey of Sweden (SGU). In WSP’s current workflow, a hydro- geologist selects a suitable reference well based on expert judgement and visually aligns the two time series a process that is subjective.

This study examines whether quantitative methods can match or outperform the visual approach. Three alternatives are compared against it: a simple linear regression using a single reference well, a multiple regression combining several reference wells, and a state space model that relies solely on the observation well’s own time series. The methods are applied to three observation wells whose measurement series differ in density and continuity: one is relatively dense and continuous, another is infrequent, and the third is highly irregular with a gap exceeding three years. All methods are evaluated using MAE, RMSE, and the Durbin-Watson statistic, complemented by residual diagnostics.

The quantitative methods consistently have lower MAE and RMSE than the visual method across all three wells. For the two wells with denser measurement series the state space model performs best overall, producing the lowest RMSE and residuals with little remaining autocorrelation. For the well with the most irregular series the multiple regression model shows better results.

The study shows that quantitative methods can complement WSP’s visual method and contribute to a more systematic and reproducible workflow. Which method is most suitable depends on the data characteristics of the measurement series, where the state space model benefits from dense series and the regression models have an advantage in the presence of longer time gaps.

The company designs and assembles large-scale isostatic press systems, across both stan-

dardized and highly customized configurations, managing an assortment of over 50 000

articles that range from standard fasteners to multi-ton steel forgings. Because production

is organized around project deliveries, demand is triggered by project schedules and sales

forecasts rather than continuous output.

Many articles, particularly those tied to engineered-to-order projects, are consumed only

a handful of times over several years, leaving insufficient historical data for traditional

statistical inventory models. The same warehouse simultaneously supplies aftermarket

spare part operations, where demand is largely unpredictable, compounding the challenge

of setting efficient safety stock levels across a shared assortment.

This study develops a methodology for determining safety stock levels to balance mate-

rial availability, lead time performance, and inventory cost in such a context. A single-case

study was conducted at an industrial company operating within project-based manufactur-

ing, where a central warehouse supports both production and aftermarket operations. The

research combined a current state analysis, stakeholder interviews, and a semi-systematic

literature review.

A multi-criteria article classification methodology was developed, integrating ABC, XYZ,

and VED analyses to capture consumption value, demand variability, and criticality for

production and aftermarket separately. Safety stock levels were determined using differ-

entiated approaches based on demand frequency and criticality at the article level, and a

set of key performance indicators (KPIs) was proposed to enable follow-up and evaluation.

The results indicate that distinguishing between production and aftermarket criticality and

demand, enables more context-adapted inventory control. The proposed methodology im-

proves prioritization of critical items and supports a standardized way of setting safety

stock levels based on their criticality for articles with low and intermittent demand.

The study contributes with a practically applicable framework for determining article crit-

icality and safety stock setting in environments characterized by high product variety and

low demand per article, where standard inventory models are insufficient.

Material handling vehicles such as forklifts continuously generate operational data on their Controller Area Network (CAN) bus. Engineers analyse these recorded CAN logs to diagnose faults and reconstruct sequences of events, but conventional tools present this data as time-series graphs that lack any spatial or contextual representation of the vehicle's behaviour. This thesis investigates whether a game-engine-based 3D visualization, reconstructed from CAN log data, can support troubleshooting in an industrial context. Following design science research methodology, a proof-of-concept artifact was developed in two iterations using the Godot game engine and a Python-based decoder server to visualize a Toyota Material Handling BT Reflex Reach Truck. Requirements were derived from an initial set of functional goals and extended through a questionnaire (n=25) sent to the R\&D department at Toyota Material Handling. The questionnaire revealed three main use-cases: troubleshooting, virtual verification and safety analysis. It also revealed five visualization needs. The completed artifact reconstructs forklift motion, fork height and reach. It collects error codes and detects an extendable set of safety-critical events. It provides synchronized graphs and integrates with the existing Main Information Display simulator. The finished artifact was evaluated conceptually and evaluated empirically through semi-structured interviews with three expert participants. Results indicate that the artifact provides a faster and more accessible overview of forklift behaviour than current graph-based tools. Key limitations are an unquantified open-loop positional drift and a small evaluation sample. The thesis contributes a working proof-of-concept, insight into users' perceived use-cases and the requirements needed to support them, as well as an evaluation of the artifact and an analysis of limitations of CAN-based 3D visualization.

Risk theory studies stochastic surplus processes and is often used to analyse ruin probabilities, that is, the probability that a surplus process reaches a critical lower level. This thesis studies the classical continuous-time dual risk model, in which a company incurs deterministic running costs and receives random gains at random times. The aim is to investigate how Bayesian inference can be used to estimate the ultimate ruin probability when the model parameters are unknown.

The study is based on simulated data and should therefore be interpreted as a methodological investigation rather than an empirical company analysis. The data are generated from a compound Poisson dual risk model with exponentially distributed gain sizes. The arrival intensity and the parameters of the gain-size distribution are treated as unknown and estimated using Bayesian posterior distributions. For each posterior draw, the ultimate ruin probability is calculated, yielding a posterior distribution for the risk of ruin rather than a single point estimate.

The results show that parameter uncertainty can lead to substantial uncertainty in the estimated ruin probability, particularly when the observed dataset is limited. Prior assumptions have an especially strong influence when the data contain little information, partly through their effect on the posterior probability that the net profit condition is not satisfied. The results also show that misspecification of the gain-size distribution can affect the estimated ruin probability, but the effect does not always operate in the same direction. In this study, the misspecified lognormal model produced a lower estimated ruin probability in some cases than the correctly specified exponential model.

Overall, the thesis shows that Bayesian inference is a useful framework for quantifying uncertainty when estimating ruin probabilities within the dual risk model. At the same time, the results should be interpreted with caution when the data are sparse, the model assumptions are strong, or the gain-size distribution is misspecified.

At high levels of automation, autonomous vehicles remove an important source of passengersupport. At SAE level 5,the absence of a driver eliminates a form of adaptive accessibilitythat traditionally assists passengers in navigating challenges during travel.Thedriver constantly perform situational judgment by observing the passengers, adapting vehiclebehavior, providing reassurance, and responding to unexpected situations. Many ofthese actions are informal, context dependent and difficult to fully translate into automatedsystems.This creates new challenges related to passenger support and interaction, particularly forelderly users who often experience age related physical, sensory, cognitive, behavioral,and psychological limitations that influence their perceived control, perceived safety, andtrust.This thesis addresses this gap by investigating how autonomous vehicle interiors can supportelderly passengers during critical events and by developing a framework for translatingdriver support functions into autonomous vehicle support systems. Using a Researchthrough design approach grounded in inclusive design principles, the study combinedliterature reviews, field observations, task analysis, user journey mapping, empathy simulations,scenario exploration, concept development, and prototype evaluation to identifyelderly user needs and driver support functions in public transportation.The findings revealed that elderly passengers rely not only on physical accessibility but alsoon clear orientation cues, predictable interactions, understandable communication, andhuman reassurance during unexpected situations. Based on these insights, a multimodalinterior support system was developed for autonomous paratransit vehicles. The proposedsystem integrates physical accessibility support, environmental guidance and orientation,multimodal communication, and access to remote human assistance to support usersduring both normal travel and critical events such as sudden braking and collisions.Prototype evaluation indicated that combining accessibility, orientation and guidance,communication, and reassurance mechanisms can improve user understanding, perceivedsafety, and trust when interacting with autonomous transport systems.The study contributes design knowledge on translating driver support functions into autonomousvehicle interiors and highlights the importance of addressing both functionaland emotional user needs when designing future mobility solutions for an ageing population.

Background/purpose: The digitalization of the public sector is characterized by increased demands for collaboration and efficiency, while decentralized organizational structures create new challenges in the coordination of digital initiatives. This study aims to explore how IT development strategies are motivated and communicated between strategic and operational levels in relation to their value-creating potential within the context of decentralized collaboration.  

Method: The study has been conducted as a qualitative case study of the development initiative Joint Information Provision for Financial Assistance (GIF-EKBI), which is coordinated by the Swedish Association of Local Authorities and Regions (SKR). Data has been collected through public documents and semi-structured interviews. The collected data has been used to produce a case description and thematic analysis of respondents' statements.  

Results: The results show that collective prospective sensemaking in decentralized digital development initiatives is an iterative process where values and goals are renegotiated based on experiences, institutional differences and technological frames of reference. Four central dimensions were identified in this process: Constellations; core values; performances, and discrepancies. These concepts are illustrated in the study through a dynamic theoretical model of how friction, willingness to collaborate, and means of communication drive the need for interpretation, renegotiation and new understanding of the implied value of the development initiative.  

Conclusions: Value propositions in decentralized digital development initiatives are not translated linearly from strategic to operational level. Collective prospective sensemaking is formed for this context through temporary organization, where accessibility and immediacy are limited by the forms of communication that are applied. Centralization and active mediation of forms of communication can enable better conditions for sensemaking. 

This thesis aims to design a fail-safe locking device for a hydraulic lifting mechanism that moves a pivoting structure between a lowered and raised position. Hydraulic systems are common in many industries due to their ability to generate large forces while maintaining a compact design. The systems are generally reliable, but there are cases in which the hydraulics fail, which often lead to pressure loss in the system. This type of failure causes the piston rod to retract unintentionally, which in turn may cause equipment damage and system failure. 

The project originates from a request for a safety measure from the market even though the current system is yet to fail, forwarding the question of perceived reliability. This is investigated in relevant literature along with design parameters for mechanical reliability, specifically focusing on geometric design, loads, failure modes and the effects of weather and wear. The impact of these parameters were explored in the development of this project's locking mechanism. 

The product development process started with multiple idea generation methods to create a broad base of ideas, which are then evaluated, with the most promising ones selected for continued development and refinement through calculations and simulations. A risk assessment is made for the final concept, identifying potential risks with the concept as well as recommended actions for future development of the product.

The project results in a final concept that consists of three parts; automatic locking and unlocking, manual unlocking and safe lowering. The locking mechanism engages automatically if pressure in the hydraulic system is lost, and is capable of holding both the weight of the structure and applied external forces. Three concepts for manual unlocking are presented, as well as one for secure lowering of the structure. In the end, conclusions are drawn as to which design parameters are most important when designing a fail-safe system for a hydraulic product, as well as for perceived reliability.

Polychlorinated n-alkanes (PCAs), the main component of chlorinated paraffin mixtures (CPs), are complex industrial chemicals widely used as plasticizers, flame retardants, and additives in numerous commercial products and building materials. Due to their persistence, bioaccumulative potential, and widespread use, PCAs are ubiquitous in environmental and biological matrices. However, their extreme compositional complexity hinders harmonized analytical methodologies and toxicological information, challenging exposure assessment and risk evaluation.

The overall objective of this thesis was to improve the understanding of human exposure to PCAs. To contribute to this goal, the thesis aimed to develop analytical workflows, investigate PCA contamination in indoor environments, assess human exposure, and identify important exposure sources and pathways.

To address analytical challenges, CPxplorer, a harmonized high-resolution mass spectrometry-based data analysis workflow, was developed for the identification and quantification of PCAs and related compounds. The workflow was successfully applied across different instrumental platforms and matrices, including indoor dust, indoor organic films (IOFs), silicone wristbands (SWBs), food, and human serum, improving analytical consistency and comparability.

Indoor dust and IOFs collected from Swedish homes, schools, offices, and sport halls revealed widespread PCA contamination. Medium-chain PCAs (ΣPCAs-C_14–17) dominated most indoor matrices, while differences between dust and IOFs reflected matrix-specific partitioning and accumulation processes. Ventilation, cleaning practices, and indoor activities influenced PCA concentrations and patterns, although no single factor fully explained the observed variability.

Human exposure assessment demonstrated that PCA exposure is highly multipathway. Estimated dietary intake exceeded exposure estimates derived from indoor matrices, indicating that food likely represents the major contributor to total external exposure. Nevertheless, chemical forensic fingerprinting analyses showed that PCA homologue patterns measured in serum were more similar to indoor dust and IOFs than to food. Likewise, PCA profiles measured in SWBs were well reconstructed using PCA patterns in indoor environmental matrices, highlighting indoor environments, particularly homes, as important exposure scenarios. These findings suggest that the pathways contributing most to total exposure are not necessarily those most strongly reflected in internal exposure profiles, likely due to variation in homologue-specific bioavailability and toxicokinetics.

Overall, this thesis demonstrates that understanding human exposure to PCAs requires integrated analytical, environmental, and biomonitoring approaches. The results provide new insights into PCA occurrence, partitioning behavior, exposure pathways, and internal exposure patterns, while also contributing methodological tools to support future monitoring, exposure assessment, and regulatory efforts aimed at reducing human exposure to these persistent contaminants.

Distributed stream-processing pipelines deployed on heterogeneous systems can stop

producing useful output when a worker node fails. The same problem can occur when

one pipeline task is affected by a partial node failure while the node remains reachable

to the orchestrator. SkePU-Streaming currently leaves such failures to the surrounding

platform or application code. That leaves a gap for long-running pipelines that need the

runtime itself to detect a failure, remap the affected pipeline tasks, and reconnect the

stream.

The prototype adds framework-level recovery support to SkePU-Streaming. It coor-

dinates recovery inside the runtime and restores the stream path after a failure trigger.

The evaluation uses controlled fault injection and repeated timing measurements on

distributed pipelines. For recovery scope, the evaluation tests six fault cases that expose

the core one-to-one, one-to-many, and many-to-one local recovery patterns, and checks

the same patterns inside a larger motion-detection pipeline. In these cases, the prototype

detects the injected failures, remaps the affected tasks, restores the expected stream path,

and passes the recovery checks. For the evaluated producer–consumer workload, enabling

recovery support leaves the mean per-run median latency in the same range (0.97 ms

vs. 1.09 ms), with no latency penalty distinguishable from run-to-run variation. Warm

standby reduces the mean post-detection recovery interval from 24.03 s under cold failover

to 54.69 ms.

The results show that failure detection, task remapping, and bounded recovery sup-

port can be integrated into SkePU-Streaming. Within the evaluated cases, skeleton ap-

plications resume streaming after the runtime detects a failure and repairs the affected

path.

Generativ AI (GAI) är en teknik som snabbt har utvecklats och fått genomslag i både människors vardag och arbetsliv. Tidigare forskning belyser hur GAI kan användas inom olika områden, särskilt inom affärssystem,vilket denna studie fokuserar på. Däremot saknas forskning som undersöker hur tekniken faktiskt upplevs av adoptörer i praktiken. Denna studie analyserar hur GAI har och kan påverka adoptörers arbetsroll inom affärssystemsbranschen. Vidare undersöker studien vad respondenterna uppfattar att deras kunder har förförväntningar på teknikens kapacitet och vad GAI kan leverera i praktiken. Studien utgår från teori om teknologiska ramar för att belysa hur uppfattningar formas inom en specifik roll med komplement av teori om adoptörer för att beskriva respondenterna. I studien genomfördes semistrukturerade intervjuer med adoptörer och dokumentstudier från leverantörer och konsultföretag inom affärssystem. Resultatet av denna studie visar att adoptörer använder GAI som ett verktyg i sitt arbete idag. Samtidigt framkommer det att GAI-användningen upplevs som överdriven enligt studiens respondenter. Utifrån respondenterna använder deras kunder GAI begränsat, vilket visar på att teknologin inte har fått ett starkt genomslag inom denna bransch. 

Den ökade digitaliseringen har gjort informationssäkerhet till en central utmaning för organisationer, särskilt inom offentlig sektor där komplexa strukturer och ett växande hotlandskap ställer höga krav. Mot denna bakgrund och i ljuset av NIS2-direktivet syftar studien till att undersöka hur chefer inom en svensk kommun upplever arbetet med informationssäkerhet, med fokus på deras medvetenhet om NIS2-direktivet, förståelse av informationssäkerhetsansvar samt hur säkerhetskrav påverkar deras dagliga arbete. Studien har en kvalitativ ansats och baseras på semistrukturerade intervjuer som analyserades tematiskt. Resultatet visar att medvetenheten om NIS2-direktivet är låg och att förståelsen av dess innebörd är begränsad, där många chefer förlitar sig på satta rutiner eller inväntar tydlig styrning från högsta ledningen. Samtidigt uppfattas ansvaret för informationssäkerhet som formellt tydligt men i praktiken komplext och stundvis svårhanterligt, till följd av oklara gränser och organisatoriska förutsättningar. Informationssäkerhetskrav påverkar det dagliga arbetet genom ökad administration, nya arbetsmoment och behov av prioriteringar mellan konkurrerande mål. Studien bidrar med kunskap om hur chefer i kommunal kontext förhåller sig till NIS2-direktivet och visar tydligt att ansvar inte nödvändigtvis innebär praktisk hanterbarhet, samt att organisatoriska och mänskliga faktorer utgör centrala utmaningar i informationssäkerhetsarbetet.

This study compares two laboratory approaches for embedded systems education on the STM32L562E-DK platform: one where participants complete preparatory laboratory exercises before a project, and one where participants proceed directly to the project. The study investigates how these approaches affect student motivation, engagement, learning outcomes and efficiency. Four participants were divided into a laboratory group and a project group, each given two weeks to complete a Black Jack implementation using RTOS, UART communication, and LCD graphics. Data was collected through pre- and post-study surveys using a six-point Likert scale, a multiple-choice knowledge test, and the evaluation of a artifact based on predefined criteria. Results indicate that both groups achieved similar learning outcomes and comparable efficiency. Motivation and engagement tended to favour the project-only approach. Due to the small sample size, no definitive conclusions can be drawn regarding the two approaches, although a project based laboratory series showed positive results for learning, motivation and engagement. For more definitive results, future work should replicate the study with a larger and more experience-balanced sample.

 This thesis investigates how a temporal memory layer can be added to an LLM-based robotic system to support reasoning over past observations. The work was carried out on the Spot robot at Linköping University, where the robot coordinator could interpret natural-language instructions but lacked persistent memory. Two alternative memory representations were implemented and compared: an event-based temporal knowledge graph using Graphiti and an RDF-star representation stored in Oxigraph. Both were exposed through the same query interface to enable a controlled comparison. The evaluation consisted of ingest benchmarking, offline query testing, and an exploratory live integration on the physical Spot robot. During the live session, 15 observations were ingested into a fresh Graphiti graph. The system successfully retrieved object information from the live graph and applied a temporal filter to observations recorded after a specific timestamp. A directional query also caused the robot to rotate towards a remembered object when its directional metadata had been ingested correctly. Other directional attempts were affected by inconsistent mapping of the location and camera metadata. JSON-based ingest reduced average time by 60.2% compared with text-based ingest, although most of the runtime was still spent inside Graphiti’s internal processing. The query evaluation used 83 natural-language questions covering retrieval, temporal filtering, pose lookup, directional reasoning, and naming variation. Four approaches were compared: raw Graphiti, Graphiti with an LLM-assisted query pipeline, raw RDF-star, and RDF-star with an LLM-assisted query pipeline. The results show that the Graphiti LLM pipeline performed best overall and was the only approach that handled the full range of query types, providing high level results. The raw Graphiti baseline performed substantially worse, showing that the application-layer query pipeline was essential. Raw RDF-star performed better than RDF-star with LLM-generated SPARQL, indicating that symbolic querying was effective when queries were formalised but difficult to use reliably through natural language. The thesis concludes that a practical temporal memory layer for LLM-based robots depends on the combination of memory representation, retrieval method, and specialised query-handling logic.

This thesis examines Financing Locally Led Climate Adaptation (FLLoCA) within the Embu County organisation in Kenya. It does so from the perspective of a theory of the translation between policy and practice, grounded in organisational learning. The thesis is limited to the field of monitoring and evaluation, as this can be used as a tool to assess how policy implementation is progressing. The problem formulation highlights that Locally Led Adaptation (LLA) is fundamental to ensuring that policies do not risk maladaptation. FLLoCA is built on LLA, yet the theory also indicates that there is often a gap between policy and practice, as these two concepts fail to reflect one another. Consequently, there is an area within the translation between FLLoCA policy and practice for this thesis to analyse. Data collection was carried out through a thematic analysis of documents and interviews. This resulted in three themes on which the findings are based. The analysis of the findings discusses these themes in relation to the three phases of translation; i) policy ii) translation iii) practice. The main conclusions of this thesis are that the translation of FLLoCA policy into practice within the Embu County organisation has both positive aspects and challenges. Challenges include political interference in decisions that should be made at local level, as well as capacity issues relating to funding and systems. Furthermore, FLLoCA has positive aspects in term of training of skills and resources, it is also highly regarded for its bottom-up approach.

A major challenge of developing tools for the environment of a host language isthat external library modules can add new syntax elements to the host language,making its environment difficult to develop and maintain using tooling based on afixed set of grammar rules. Storm is another example of this kind of extended hostlanguage; i.e., all extensions are defined in the same source file as the hostlanguage. Therefore, each developer has to figure out at least once how theirpieces of code are being parsed.This thesis introduces a new, cursor-centered, filtered syntax-tree visualization forStorm as a side panel in the popular Emacs editor. The visualization is centeredaround the developer’s cursor position, limited to the smallest enclosing multi-lineconstruct, and rendered via a grammar-agnostic renderer that works directly withthe structural characteristics of the syntactic parse trees instead of working withthe names of individual grammatical rules. The editor renders the resulting syntaxtree as a tree view with bidirectional links between source code and syntax treenodes, as well as a direct link to the Storm documentation buffer.

Bayesian neural networks (BNNs) have been widely applied to solve inverse problems (e.g., deconvolution), offering flexibility between prediction fidelity and uncertainty quantification. In this work, we focus on BNNs with heavy-tailed distributed weights, which aim to preserve the sharpness of the underlying unknown state.

Numerically, we consider a one-dimensional deconvolution problem to illustrate the effectiveness of heavy-tailed BNNs. Gaussian and Laplace distributed priors are also studied to show the limitations of smooth and sparse promoting priors. Across all considered priors, we evaluated three posterior approximation methods: deep ensembles, Markov Chain Monte Carlo (MCMC) sampling and, our primary interest, the mean-field variational approximation with Cauchy-distributed factors.

The results suggest that the Cauchy mean-field approximation introduces a sharp-promoting bias, making it a suitable modeling choice for signals with discontinuous structure. By explicitly imposing a heavy-tailed structure on the posterior distribution, the recovered signal exhibits sharp transitions that closely resemble the edges of the underlying hidden state.

Furthermore, we find that the Cauchy mean-field approximation achieves competitive edge-preserving reconstructions even with a significantly reduced network capacity, provided the network is large enough to sample a sufficient number of extreme weights. This makes it a computationally appealing alternative to full Bayesian inference methods such as MCMC.

Pain assessment relies heavily on a patient describing how they feel. Because this ishighly subjective, there is a need of objective, biological way to measure pain using proteinbiomarkers. A laboratory technique called 2D-GE can map out these proteins visually, buthaving humans analyze these complex image maps is slow and inconsistent. To solve this,this thesis evaluates how deep learning can automatically scan 2D-GE images of differentbodily fluids to find objective, physical evidence of pain.To achieve this, the study systematically evaluated three distinct neural network architectures:convolutional neural networks (CNNs), specifically ResNet18 and ResNet34, and avision transformer (ViT). The models were trained and validated on a proprietary datasetof 2D-GE images derived from heterogeneous bodily fluids (saliva, cerebrospinal fluid[CSF], and blood plasma). The methodology evaluated both global (mixed-fluid) andfluid-specific classification tasks. Monte Carlo cross-validation across 50 iterations was employedto ensure evaluation, comparing metrics such as accuracy, balanced accuracy, andMatthews Correlation Coefficient (MCC). Furthermore, the necessity of transfer learningwas tested against training from scratch. To confirm the significance of the observed performancedisparities, two-sided pairwise permutation tests (10,000 resamples) and Wilcoxonsigned-rank tests (α = 0.05) were executed. Finally, visual explainability algorithms specially,GradCAM++ for CNNs and attention heatmap for the ViT were utilized to interpretthe spatial reasoning behind the models’ predictions.The study established that pain classification is feasible but fluid-dependent. Modelstrained on heterogeneous global data struggled to surpass a random-chance baseline dueto confounding factor. However, for fluid-specific classification, saliva emerged as an significantdiagnostic medium, achieving a good predictive accuracy of 82.49%˘5.63%. Conversely,blood plasma proved unsuitable for this task, largely due to high variations insample quality and background noise.Computationally, transfer learning was proven necessary; without pre-trained weights, thedeeper ResNet34 model suffered considerable performance degradation by performing at50%. With transfer learning, the CNN architectures demonstrated a significant statisticalsuperiority over the ViT. For the overall binary classification task, ResNet34 achievedthe highest overall accuracy of 65.2% ˘ 5.8%. Whereas, ResNet18 achieved an accuracy of62.5%˘4.4% and ViT achieved an accuracy of 59.4%˘5.2%. For the complex 6-class multiclasstask, ResNet34 achieved 63.0% ˘ 4.0% accuracy against a 16.67% baseline, againestablishing significant superiority over the ViT (p = 0.0002). Despite the predictiveshortcomings of the ViT, explainability analyses revealed a crucial trade-off: GradCAM++showed that ResNet models relied on broad, diffuse regional patterns, while the ViT attentionheatmaps acted as a more localized focus, successfully pinpointing discrete, individualprotein spots.Ultimately, this research demonstrates that while deeper CNN (ResNet34) are the superiordiagnostic classifiers for evaluating complex 2D-GE images, the ViT provides the precisehigher localized focus required for downstream biological identification. This thesis laysa foundation for the algorithmic pain-related protein biomarkers detection. Future effortsmust focus on mitigating data scarcity within the underrepresented fluid cohorts and exploringmultimodal data fusion to further enhance diagnostic capabilities.

The Semantic Web aims to make web data machine-readable by embedding semantics into the data. As key technologies in the Semantic Web, ontologies and knowledge graphs (KGs) are used to formalize and represent knowledge in a structured way. They are core components in application scenarios involving searching, integrating, managing, and extracting value from heterogeneous data sources at scale, and are widely used by major data and database providers and consumers, such as Google, Amazon, and Netflix. As a result, their quality is critical for developing high-quality semantically enabled applications. However, in practice, many ontologies and KGs, especially large-scale ones, contain defects. Using such defective resources can lead to incorrect conclusions or missed valid conclusions. Ensuring their quality, particularly in terms of correctness and completeness, is a major challenge. 

In this thesis, we focus on repairing ontologies and KGs represented by description logic knowledge bases. We first discuss the kind of defects concerning correctness and completeness that can occur in ℰℒℋ⟂ ontologies and KGs. We then propose an interactive repairing approach that combines five basic quality-improving operations: debugging, removing, weakening, strengthening, and adding, to address the problem of repairing ℰℒℋ⟂ ontologies and KGs with defects. Traditional debugging approaches repair ontologies and KGs by simply removing incorrect information, which may result in the loss of correct information that was previously derivable from the removed information. In contrast, our approach aims to preserve as much correct knowledge as possible while removing incorrect information by, in addition to removing the incorrect information, also adding correct knowledge, thereby mitigating the negative effects and improving the quality of repaired ontologies and KGs. 

Furthermore, we show that there are also choices to be made when using and combining these operations such as how to combine them and when to apply them, as well as the different autonomy levels of different components in a KG. We define combination operators that reflect these different choices, and discuss how these choices influence the quality of the repaired ontologies and KGs in terms of correctness and completeness. Comparing these different combinations provides insights into how to make choices based on the desired level of correctness and completeness of repaired ontologies and KGs. Finally, we conduct experiments and present several use cases to validate our approach and demonstrate the framework. Overall, this framework serves as a template for developing repair systems with specific properties regarding correctness and completeness of the repaired ontologies and KGs.  

 In today’s tech-driven society, we increasingly rely on the security of internet commu-

nication to handle personal data, financial transactions and digital infrastructure. TLS

1.3 is one of the most used cryptographic communications protocols for securing internet

communication, ensuring confidentiality, authenticity, and integrity. However, imple-

mentations remain vulnerable to critical software defects: Heartbleed and Apple’s ”goto

fail;” are two of the major discovered security bugs which risked data leakage and enabled

authentication-bypasses, respectively. For this reason systematic methods are needed to

find bugs before they are exploited. This thesis presents a KLEE-based Differential Sym-

bolic Execution framework to test the parsing of the ServerHello message in OpenSSL

and wolfSSL implementations. The framework progressively introduces message fields as

symbolic variables in order to detect end-to-end divergences between implementations. It

is developed and guided by a proof of concept (PoC) implementation, in which it discovers

9 out of 12 injected bugs. Applied onto the real TLS implementations, two unique diver-

gences are found through four runs of Differential symbolic execution for which OpenSSL

is classified as non-conformant to the RFC 8446 (TLS 1.3). Furthermore, through the

runs discrepancies in build configurations was uncovered, assisting in ensuring a correct

setup. The results demonstrate that Differential Symbolic Execution is a viable method

for finding parsing level divergences, but is prone to state-space explosion and hence faces

difficulties in making larger fields of the message symbolic, and finding deeper vulnera-

bilities. It is also solely capable of finding divergences between implementations, possibly

vulnerabilities, and is no guarantee of proving absence of bugs.

Electrification of societies and the rapid development of artificial intelligence are predicted to significantly increase global energy demand in the coming years. Meeting this demand sustainably requires renewable energy sources. Organic solar cells (OSCs) have emerged as a promising candidate, achieving power conversion efficiencies exceeding 20% in laboratory settings through the development of non-fullerene acceptors (NFAs). However, limited long-term stability remains a major challenge for commercialization.

In this work, the photodegradation of the NFAs Y6 and IT4F (outdoor photovoltaics), GS-ISO and FTCC-Br (indoor photovoltaics), and the donor polymer PTQ10, was investigated in neat films, bulk heterojunction (BHJ) blends and solar cell devices under standardized AM1.5G LED illumination without UV in air. A combination of UV-Vis-NIR absorption spectroscopy, photoluminescence (PL), transient absorption spectroscopy (TAS) and device characterization was used to correlate material degradation with device performance.

The results showed wavelength dependent degradation for IT4F, GS-ISO and FTCC-Br, with higher energy photons causing more severe damage, while Y6 exhibited high photostability. Although the absorption of PTQ10 remained stable after light soaking, a reduced photoluminescence quantum yield (PLQY) and shortened lifetimes indicated increased non-radiative recombination. TAS revealed accelerated formation of bound polaron pairs in light soaked IT4F, GS-ISO and FTCC-Br, while no such features were observed for PTQ10. Despite efficient charge transfer in BHJ blends, the devices showed a dramatic loss of performance and external quantum efficiency after 12 hours of light soaking. This loss in performance could not be explained solely by the photodegradation of the active layer, and indicated additional problems within the device stack, including increased series resistance and reduced shunt resistance. Overall, the results demonstrated that a dominant degradation pathway is the formation of non-radiative recombination channels driven by photoinduced degradation of the NFAs.

Memory safety vulnerabilities remain a prevalent security challenge, accounting for a significant majority of critical software defects. This paper investigates how programming language standards influence memory safety by analyzing three distinct approaches: the manual memory management of C++, the ownership-based compile-time model of Rust, and the automatic garbage collection of Python. These languages are evaluated by reproducing a suite of Common Weakness Enumeration (CWE) vulnerabilities, including buffer overflows, use-after-free, and null pointer dereference conditions to determine the stage at which faults are detected: compile-time, runtime, or silent execution.The study highlights the fundamental trade-offs in language design: Rust shifts detection to the compilation phase by imposing strict compile-time discipline on the developer. C++ prioritizes runtime performance and control, often relying on warnings rather than errors at the risk of silent failures, and Python ensures safety through significant runtime overhead. These findings quantify the cost of safety, offering a comparative view of the memory overhead and safety guarantees inherent to each language's design.

Den här uppsatsen utgörs av en komparativ analys av Margarets Atwoods roman Tjänarinnans berättelse och P.D. James roman Människors barn utifrån aspekterna religion, infertilitet, det kvinnliga könet och bokförbud samt ett avslutande avsnitt om verkens reella politiska omvärld. Det som undersöks är två likartade romaner men där enbart den förstnämnda har blivit förbjuden. Syftet med denna analys är därför att undersöka både likheter och avgörande skillnader för att ge en rimlig förklaring till varför Tjänarinnans berättelse är förbjuden medan Människors barn inte är det trots deras många likheter. Teorin för uppsatsen har sin grund i Gregory Claeys definition av begreppet dystopi samt Andreu Domingos underkategori demodystopi, tillsammans med Yvonne Hirdmans genusteorier om det kvinnliga könet och det stereotypiska genuskontraktet. Tidigare forskning tar upp olika perspektiv på aspekterna ovan, där bland annat religionens roll och kvinnors villkor är några framträdande teman. Till själva bokförbudsaspekten tar uppsatsen också hjälp av Svenska PEN och deras gemensamma databas om verk och deras bokförbud. Metoden för den här uppsatsen är närläsning av de båda verken för att kunna genomföra den komparativa analysen. I det avslutande kapitlet av analysen (Det politiska klimatet) skiljer sig metoden från de resterande delarna då denna del istället utgörs av en undersökning av romanernas verkliga omgivande samhälle. Denna förändring av metod tillämpas för att skapa en djupare förståelse och en mer nyanserad bild över vad som kan tänkas ha orsakat de olika bokförbuden. Resultatet av analysen synliggör hur Atwood och James olika användning av kristendomen utgör en viktig roll för mottagandet av boken, där Tjänarinnans berättelse har anklagats för exempelvis blasfemi medan Människors barn istället kan läsas som en kristen fabel. Detta tillsammans med hur Atwoods roman börjat figurera i dagens politik drar den här uppsatsen som slutsats till att utgöra de största skälen till verkets förbud.

Accurate dynamic models of industrial robot manipulators are essential for achieving high-performance control and for realizing accurate physics simulation. However, improving model fidelity typically requires extensive experimental data and time-consuming identification procedures. This thesis addresses this trade-off by advancing system identification methods aimed at improving the accuracy of physically parameterized robot models. The advancements are focused on resource- efficiency in terms of shortening experiment duration for data collection, decreasing data volume and reducing need for manual effort related to experiment design and tuning of the identification algorithm.

The work first investigates high-fidelity modeling approaches, including multibody formulations with flexible links derived from Finite Element data. These models enable accurate prediction of robot dynamics early in the development process without requiring experimentally identified stiffness parameters. To further enhance model realism, extended joint models incorporating gear, bearing, and link flexibility are developed, allowing separation and identification of transmission- and arm-side dynamics using dual encoder measurements.

The second central contribution of the thesis is the development of improved experiment design strategies for frequency-domain identification. By formulating and solving optimization problems based on information criteria, optimal robot configurations for data collection are identified, significantly reducing experimental effort while maintaining or improving parameter estimation accuracy. These methods are validated both in simulation and on real industrial manipulators, demonstrating substantial reductions in experiment time. The approach is further extended to systems incorporating nonlinear transmission stiffness through the use of statistical linearization.

Third, the thesis introduces advanced estimation techniques for extracting more information from limited data. In particular, local parametric methods for frequency response function estimation are adapted to nonlinear, closed-loop, multi-input multi-output robotic systems, reducing the number of required experiments and improving estimation accuracy compared to classical approaches.

A key contribution of the thesis is the experimental validation of the proposed methods and models, ensuring their practical relevance. Overall, the results demonstrate that substantial improvements in model fidelity can be achieved without increasing experimental resources. By combining physically parameterized models, optimized experiment design, and efficient estimation methods, the proposed framework has significant practical and industrial impact, supporting the development of accurate and computationally efficient dynamic models for industrial robot manipulators.

Drinking water can become contaminated through several mediums, the distribution network being a major one. Sewage water may enter the pipe network through pipe breaks or leaks, and there are many documented cases historically of this causing people to fall ill. The drinking water in the distribution network is currently monitored through manual sampling, which is time-consuming and infrequent. This means contaminations in the network are usually not discovered until after people have already fallen ill. There is therefore a need for detecting contaminations in real time in the distribution network before people become sick.

In this project methods for detecting sewage contamination in the drinking water distribution network are explored and evaluated. It investigates the feasibility of using a sensor called the electronic tongue alongside one of five different algorithms to detect contaminations in real time. The most important factors include being able to detect low concentrations of sewage as well as minimizing false alarms, and a delay in the response is also introduced to reduce the number of false alarms. The algorithms are evaluated with regard to the trade-off between contamination detection, false alarm rate and response delay, and the results show that out of the five algorithms, two are viable for detecting sewage contaminations using the electronic tongue. The thesis highlights the decisions that need to be made when tackling this problem and provides some alternatives for solving it. To evaluate the real efficacy of the algorithms, further testing with more nuanced data is required, especially in real scenarios.

Background: The number of individuals dependent on financial assistance due to illness and poor health is increasing. Long-term dependence on financial assistance is associated with an increased risk of social exclusion and ill health. Within municipal labor market units, individuals may be offered community-based vocational rehabilitation. Occupational therapists working in these units often use the Assessment of Work Performance (AWP) to assess work ability, and in some units the informant report data collection method (AWP-IR) is applied.

Purpose: The aim of the study was to investigate aspects of usability of the data collection method AWP-IR in work ability assessments within community-based vocational rehabilitation.

Method: The study had a qualitative design with a deductive approach. Data were collected through focus group interviews with two groups: occupational therapists in municipal labor market units with experience of AWP and AWP-IR, and supervisors with experience of using AWP-IR in collaboration with occupational therapists. The data were analyzed using manifest content analysis based on an analysis framework derived from usability concepts in relevant literature.

Results: The results show that AWP-IR achieves usability in terms of appropriateness, effectiveness, satisfaction, perceived usability, as well as situational and organizational usability. AWP-IR appears to be usable within the context of community-based vocational rehabilitation. Further studies are needed focusing on the application of the informant report data collection method within AWP, where occupational therapists as assessors and supervisors as informants conduct systematic data collection for work ability assessment.

Conclusion: Increased knowledge about work ability assessments using AWP-IR within municipal community-based vocational rehabilitation may contribute to improved quality of work ability assessments for individuals receiving financial assistance due to illness or ill health. Improved assessments may provide a basis for appropriate interventions and increased opportunities for equality in activity, participation, and self-sufficiency through work.

Bakgrund: Personer med substansbrukssyndrom (SUD) har komplexavårdbehov men möter ofta stigma och negativa attityder i samhället och ihälso- och sjukvården.Syfte: Att belysa hur patienter med SUD upplever mötet med vårdpersonalinom den somatiska hälso- och sjukvården.Metod: En allmän litteraturstudie med systematisk ansats. Datainsamlinggenomfördes i CINAHL och PubMed. Elva kvalitativa artiklar och enmixad-metod artikel (2016–2026) inkluderades efter urval,kvalitetsgranskning och analys enligt kvalitativ innehållsanalys.Resultat: Patienter beskrev avhumanisering, misstro och skuldbeläggning,vilket påverkade tillit och vårdsökande. De upplevde även att de inte blevlyssnade på eller inkluderade i vårdbeslut. Somatiska symtom och smärta togsibland inte på allvar och behandling kunde vara bristfällig.Konklusion: Patienter med SUD upplever ojämlikt och stigmatiserandebemötande inom somatisk vård. Ett ökat fokus på personcentrering, kunskapoch förbättrade attityder krävs för att säkerställa jämlik och kvalitativ vård.

This thesis investigates how the history of mathematics can be integrated into contemporary mathematics education through the design, implementation, and evaluation of a teaching resource on Maya mathematics for upper secondary school students. The study is grounded in research highlighting the need for more varied and engaging mathematics instruction, as well as curriculum guidelines that emphasize the historical and cultural dimensions of mathematics. The aim was to design, implement, and evaluate a mathematics-history-based teaching unit in mathematics education, as well as to identify areas for improvement in its design. Furthermore, the study examined students' perceptions of the teaching unit and the potential effects of incorporating elements from the history of mathematics on student learning.

The teaching resource was developed iteratively through three design phases and was also inspired by principles of Slow Design to promote reflection, curiosity, and cultural understanding. The material consisted of instructional videos, worksheets, and grids with items through which students performed the four basic arithmetic operations using methods derived from Maya mathematics, adapted to base ten to facilitate comparisons with the decimal number system. The study was conducted in an upper secondary school class studying Mathematics 2c and was evaluated through classroom observations and a questionnaire completed by 25 students.

The results indicate that the lesson was generally well received. Most students found the material clear, appropriately challenging, and engaging. For example, 76% of the students appreciated the opportunity to work with an alternative calculation method, 48% experienced a higher level of engagement than during a regular math class and 60% considered the history of mathematics to be an important component of mathematics education. At the same time, several pedagogical challenges were identified, particularly regarding the format of instruction and some students’ tendency to rely on their usual calculation methods rather than the historical methods presented.

The study indicates that teaching activities based on the history of mathematics can contribute to increased interest, engagement, variation, and understanding in mathematics education. However, it also shows that their design requires careful adaptation to students’ prior knowledge, working methods, and educational context.

Over the past decades, the circular economy has developed around products with either high material value or large, uniform waste volumes. However, many other products are still locked in to the linear economy of take–make–dispose. Few products illustrate the logic of the linear economy as clearly as product category of bonded, short-lifecycle products with examples such as comfort air filters. Comfort air filters are manufactured to improve air quality, protect human health and machines, and they require frequent replacement as their efficiency declines when they become saturated with dust and particles. This frequent replacement generates substantial end-of-life waste, which is typically sent to incineration. Climbing higher in the waste hierarchy, from incineration with energy recovery to material recycling, would allow the embedded material value to circulate rather than be lost, easing pressure on virgin resource extraction and aligning the product category with the broader transition toward a circular economy. This master’s thesis investigates why incineration persists as the main end-of-life pathway, how a manufacturer of bonded, short-lifecycle products can implement a take-back system, and what value arises from closing a material loop. This is done through a case study of the company Camfil AB, a market-leading manufacturer of comfort air filters, with the main empirical collection constituting of semi-structured interviews, primarily with employees and customers.  The thesis highlights that incineration persists for several mutually reinforcing reasons. Sweden has developed one of the most extensive waste-to-energy systems with incineration being deeply integrated with the district heating networks. Regarding the material recycling, infrastructure is built around either high-volume single-material products or high value materials. Large recycling companies deem the volumes of these products as insufficient for a dedicated processing line to be economically viable to operate. The material value is also too low to motivate them to acquire the necessary machinery for separation of the bonded materials. Customers handling spent products are influenced by several types of operational frictions – all leading the filter handling towards the path of least resistance, which is incineration. Additionally, the study concludes that an internal take-back system is preferable to outsourcing for a market-leading manufacturer of bonded, short-lifecycle products. An internal system secures control over material, the economic value of the recovered material, and the communicative benefits of a closed material loop. Companies should leverage existing local presence and prioritize locations where customer and product-density is sufficient. Implementation should proceed incrementally, starting where conditions are most favorable, rather than as a full-scale rollout. Finally, the findings show that the economic viability of a take-back system need not rest on material value alone. Value is realized indirectly through won procurements where circularity has become a qualifying criterion, through strengthened relationships with actors throughout the value chain, through a strengthened sustainability position that limits competitors' ability to compete on environmental claims, and through first-mover advantages such as customer lock-in and influence over emerging regulation. For a market-leading manufacturer, the greatest value lies in maintaining its market position and failing to act puts this position at risk. The risk from inaction is greater than the risk from early investments, as brand position and customer trust are difficult to rebuild once lost.  

This thesis aims to design, implement, and cross-compare software capable of updating parts of a running system, without requiring a restart of the host application. Dynamic Software Updating (DSU) enables such runtime modifications. High-availability systems aim to remain continuously in service, which makes dynamic updates particularly valuable for applying changes with minimal service interruption in such systems. In academia, DSU has been explored in general-purpose computing where prior work largely focuses on abstract mechanisms rather than application-specific ones. In this work, we take an application-specific perspective, using DSU to update program functions within a running Radio Access Network (RAN) application.

The scientific novelty of this work lies in applying established DSU techniques to a uniquely constrained RAN application. Restructuring the application to support DSU introduced some pros and cons, prompting an investigation into how different DSU strategies affect downtime, performance, memory usage and system stability. By implementing four update solutions and evaluating them under conditions that resemble a worst-case packet transmission workload, the thesis quantifies the costs of enabling runtime updates in a high-availability RAN system.

The mining industry has a central but ambivalent role in the sustainability transition, as it both supplies critical raw materials for low-carbon technologies and needs to reduce its own emissions. Electrification of mobile machinery is a important step, but it requires more than replacing diesel equipment with electric alternatives. It also involves changes in infrastructure, mine planning, supplier relations, routines and decision-making, while several technologies remain under development. Therefore, electrification in mining can be understood as a socio-technicaltransition rather than a purely technical change.

Combining transition studies with perspectives from organizational change, this thesis analyzes how perceived organizational barriers and technological uncertainty shape how an incumbent mining organization handles electrification. While transition studies provide a lens for understanding electrification as part of a broader socio-technical transition, organizational change perspectives are used to examine how this transition is interpreted, negotiated and enacted insidethe organization. Previous transitions research has often treated incumbent firms as relatively coherent regime actors, leaving limited understanding of how internal organizational conditions shape transition processes. Empirically, the study is based on a qualitative embedded singlecase study, including semi-structured interviews with internal and external actors, as well as observations from two mine sites.

The findings show that electrification is widely perceived as a necessary and desirable long-term direction within the organization, driven by environmental goals, improved work environment, long-term competitiveness and potential cost efficiency. At the same time, uncertainty regarding technology maturity, supplier capacity, infrastructure requirements, geopolitics, future costs and operational consequences makes the organization hesitant to test, invest in and commit to specific electrification path. These uncertainties are interpreted through existing responsibilities, priorities and routines, meaning that support for electrification remains conditional andvaries across organizational roles. The study identifies this as an uncertainty paradox: because uncertainty limits experimentation and commitment, opportunities for learning are reduced, which in turn reproduces the uncertainty that slows the transition. As a result, electrification remains exploratory and fragmented. Pilot projects and supplier collaborations generate valuable knowledge, but this learning is not always sufficiently shared or translated into coordinated organizational action.

The study further conceptualizes the incumbent firm as a micro-regime, where internal routines, production-oriented priorities, decentralized decision-making, local autonomy and role-specific interpretations reproduce stabilizing mechanisms usually associated with broader socio-technical regimes. This shows that electrification is constrained not only by external technology and market conditions, but also by internal organizational structures that shape how new technologies are evaluated and acted upon.

By opening the incumbent “black box”, this thesis contributes to a more nuanced understanding of how established industrial firms handle sustainability transitions. The findings show that incumbent firms are neither simply resisting nor straightforwardly driving change. Instead, they navigate transitions through internal negotiations between long-term sustainability ambitions, technological uncertainty and established operational logics. The study therefore highlights the importance of shared learning, clearer strategic translation and stronger coordination across organizational levels in order to move from fragmented exploration toward more deliberate transition management.

Growing energy demands, electrification, and geopolitical tensions have been increasing the urgency of the Swedish energy transition in recent years. The past decade of Swedish energy policy has seen drastic changes, going from the goal of 100% renewable energy to 100% fossil-free energy. This is not just a change in words or technologies; it is a change in the futures that are imagined, valued, and enacted. This thesis focuses on how Swedish energy futures are imagined, valued, and enacted in different contexts: a national policy and strategy context, a regional energy context, and the context of municipal energy and climate advisors. With an approach that is centred on sociotechnical imaginaries and values, this thesis studies how energy future imaginaries are enacted and embedded in society. The thesis has its starting point in four sociotechnical imaginaries analysed in a document published by the Swedish Energy Agency in 2016. These imaginaries serve as an analytical point of reference in studying how national policy, regional energy work, and municipal energy and climate advisors value and enact energy futures. The analysis shows how different contexts allow different energy future imaginaries to manifest. The thesis points out that while energy policy has changed drastically in the last decade, Sweden has been locked into an imaginary focused on large-scale energy production, a strong and competitive industry, job creation, and minor changes in households’ energy use patterns. In this imaginary, households are valued for their contribution to stabilising the grid through flexibility and demand response, and the regional context is valued for its contributions to grid capacity management. The thesis identifies tensions with this dominant imaginary. In the regional and local context, obstacles to the embedding of the dominant imaginary are identified, such as limited institutional and financial resources. Additionally, it points out that the regional and local contexts value and enact different energy futures, but do not have the means to embed those in society.

In Mary Wollstonecraft Shelley's (1797-1851) seminal novel Frankenstein, or, The Modern Prometheus (1818), the body and the monster are revealed to be closely connected as Frankenstein's creature comes to embody the ultimate Other, equally repellent, frightening, and fascinating in his artificial corporeality. In this study I examine three modern novels – Poor Things (1992) by Alasdair Gray (1934-2019), Bunny (2019) by Mona Awad (b. 1978) and Our Hideous Progeny (2023) by C.E. McGill (b. 2000) – which, I argue, may all be linked to Shelley's novel through their shared usage of the Frankenstein motif: the creation or construction of artificial bodies. Focusing especially on the female protagonists of the novels, this study aims to illustrate how the novels use, subvert, and recontextualize the Frankenstein motif to explore the notion of the monstrous body, particularly the relationship between monstrosity, femininity, and queer existence. Combining feminist and posthumanist theory, I examine the novels' portrayals of female agency, reproduction, and sexuality as well as the seemingly contradictory term monstrous cuteness, showcasing the various ways in which the body may be used as a tool for subversion and reclaiming agency. The findings of this study highlight the figure of the monster as simultaneously ostracized object and powerful subject, revealing the implicit power dichotomies inherent in Cartesian humanism as well as advocating for a broader, more inclusive understanding of the human.

Att arbeta utifrån agila arbetssätt har blivit allt vanligare inom systemutveckling. I takt med denna utveckling har frågan om hur användarupplevelse (UX) hanteras blivit mer aktuell. Tidigare forskning  har identifierat flera utmaningar kopplade till integrationen av UX i agila arbetssätt, såsom Scrum. Exempelvis har det visats att agila arbetssätt inte utgör processer med starkt användarfokus, samt att UX-arbete kommer in för sent i processen, vilket riskerar att slutprodukten inte motsvarar användarens behov. Samtidigt finns det begränsad kunskap om hur UX integreras i storskaliga agila sammanhang ur ett kognitionsvetenskapligt perspektiv. Därför undersöker denna studie hur användarupplevelse kan integreras i storskaliga sammanhang. Studien genomfördes i samarbete med Trafikverket och använde en fallstudiemetodik för att undersöka deras nuvarande arbetssätt. Data samlades in genom bland annat intervjuer och observationer samt analyserades med distribuerad kognition som analyslins. Resultatet visade att kommunikation mellan aktörer, samarbete och den fysiska/digitala miljön stödjer hanteringen av användarinformation, medan begränsningar i artefakter, otydligt ansvar för UX och hur information representeras kan försvåra integreringen av UX. Utifrån de identifierade framgångsfaktorerna och hinder bidrar studien med rekommendationer för hur UX kan integreras mer kontinuerligt i storskaliga agila arbetssätt.

Denna studie syftade till att undersöka hur vårdpersonal inom reumatologin i Region Östergötland upplever användbarheten i behandlingsprogrammet för reumatoid artrit på plattformen Stöd och behandling via 1177. Målet var att identifiera användbarhetsproblem och förbättringsområden som kunde realiseras i konkreta designförslag. Studien följde perspektivet Forskning genom design och använde sig av ett användarcentrerat perspektiv där semistrukturerade intervjuer med inslag av användartest utfördes för att samla in kvalitativa data om programmet. Den kvalitativa data analyserades utifrån en tematisk analys där fyra huvudteman identifierades, vilka var: Patientinformation, Navigering och struktur, Feedback, samt Terminologi och begreppsförståelse. Insikterna från den tematiska analysen lade grunden för de designförslag som utvecklades och den slutgiltiga Hi-Fi prototypen. Designförslagen genomgick en iterativ process där olika förslag skapades, testades och utvärderades för att slutresultatet skulle motsvara personalens förväntningar och önskningar så pass bra som möjligt. De slutgiltiga Hi-Fi prototyperna presenterades för verksamheten som bidrog med insikter och områden som borde vidareutvecklas. Insikter från intervjuer, användbarhetstester och utvärderingar av designförslagen användes för att besvara studiens frågeställningar och hur området bör utforskas vidare. Studien är menad att fungera som en grund för vidare arbete kring hur Stöd och behandling för dig med RA ska fungera så pass bra som möjligt i praktiken. Studien utgår ifrån ett distribuerat perspektiv på kognition, vilket innebär att kognition förstås som något som uppstår i samspelet mellan människor, artefakter och den omgivande miljön. 

This literature review investigates which problem-solving methods and strategies students can use when working with mathematical word problems, as well as which teaching strategies teachers can use to support students in this process. The study is based on a systematic literature study in which articles were identified through database searches in ERIC and UniSearch. The selected articles were then analyzed through a thematic analysis. The results identified four overarching themes: language, the use of digital tools, mathematical representations and teachers' teaching strategies. The result shows that students can use strategies such as retelling the word problem in their own words, creating personal och mathematical connections, asking questions, dividing the word problem into smaller parts, visualizing the word problem by creating pictorial and schematic representations and using digital tools. Students can also use the C-C model or a problem-solving model that involves dividing problem solving into two phases. Teachers can use strategies such as introducing the task without numbers and the question itself and asking students questions. Furthermore, teachers should teach about mathematical language and key concepts, work with language and reading comprehension and focus on practicing students' deep reading skills. In teaching, teachers can use different colors, diagrams and different forms of representation that can support students. Finally, teachers should teach strategies and teach students about the C-C model and about the two phases of problem solving.

Background

Spontaneous coronary artery dissection (SCAD) is an acute, non-atheromatous, non-inflammatory, non-iatrogenic cause of myocardial infarction (MI) characterized by a spontaneous separation of the coronary vessel wall layers due to either vessel wall hematoma or an initial rift in the intimal-medial vessel wall. Little is known on its pathogenesis and, until a few years past, interventional cardiologists were little aware of the variations of its radiographic appearance. Specifically in Sweden, little research has been done on the demographics (incidence, prevalence, treatment and prognosis) of this disease and its diagnostic patterns.

Method

We utilised data from the SWEDEHEART registry to describe a Swedish SCAD population. We recruited a national cohort of acute SCAD patients to compare coronary computed tomography angiography (CCTA) as a non-invasive diagnostic modality against invasive coronary angiography (ICA) and further describe our angiographic findings. We investigated SCAD prevalence, diagnostic accuracy and sensitivity in women with MI <50 years between 2015-2023 from 3 Swedish hospitals, by comparing SWEDEHEART registry data and angiographic evaluations by two independent interventional cardiologists.

Results

We found an initial low prevalence of SCAD in Sweden. Compared to atherosclerotic (type 1) MI, SCAD patients were treated almost the same. SCAD patients had lower mortality but comparable morbidity and a higher incidence of unplanned repeat angiography than atherosclerotic MI patients. CCTA was less effective than ICA in identifying and locating acute SCAD, especially if distally located. There was significant underdiagnosing of SCAD in SWEDEHEART with an improving trend in accuracy during the reviewed period.

Conclusion

SCAD was initially underdiagnosed in Sweden, likely leading to an underestimation of its true prevalence. ICA remains the most reliable method for diagnosing SCAD since non-invasive CCTA could not demonstrate comparable precision. Since implementation of a SCAD variable in 2015, SCAD diagnosis in Sweden has improved.