Luleå University of Technology Publications
Not a member yet
39356 research outputs found
Sort by
Leveraging Uncensored, Self-hosted Large Language Models for Interactive Penetration Testing
No full textGenerative AI, especially large language models (LLMs), have quickly advanced and transformed cybersecurity in the recent years, bringing powerful new capabilities, but at the same time with unprecedented security risks and potential for misuse. With uncensored LLMs now widely available, there is a growing concern how these powerful models can be utilized and exploited in the context of penetration testing. Understanding how they behave in realistic and interactive environments is essential for assessing their potential and risks. Much of the identified existing research had focused and was limited mostly to comparisons of chat prompts and the differences in the responses between publicly available LLMs and uncensored ones. This leaves a major gap of not assessing how the uncensored models can perform in practical penetration testing scenarios and dynamic real-world like settings. Our thesis aims to address that by practically evaluating the integration of uncensored, self-hosted Large Language Models (LLMs) into penetration testing workflows. The goal was to understand their operational capabilities, ethical risks, and potential value for cybersecurity practitioners in offensive security contexts. We deployed multiple uncensored LLMs on cloud-hosted infrastructure and integrated them with an interface capable of querying the LLMs for valid commands and executing them as well as using the LLMs as a live chatbot for assistance during the different scenarios when needed. We ran a series of penetration testing scenarios, like vulnerability scanning, exploit generation and phishing simulation, against vulnerable machines from ethical hacking platforms. Outputs, command accuracy, and quality of interaction were recorded and later analysed. Uncensored LLMs consistently outperformed censored ones in response flexibility. They generated more detailed payloads, offering accurate command-line suggestions, and adapted well to iterative prompts. However, we noticed that they also introduced risks such as hallucinating commands, suggesting unsafe actions, and lacked ethical constraints. While we did not specifically benchmark it, task completion and automation success were significantly higher in uncensored models across all tested categories. We concluded that uncensored LLMs could significantly enhance penetration testing when used responsibly, serving as powerful co-pilots for security professionals. But their lack of internal guardrails means they must be handled with extra caution. The findings highlight both the upsides and risks of these tools, and point to the need for ethical governance, access controls, and improved safeguards. For cybersecurity practitioners and researchers, this work adds practical insight into how LLMs might reshape offensive security work in the near future
Konceptframtagning av infångningsystem till fastvingad drönare
No full textDetta examensarbete har utförts i samarbete med Sjöräddningssällskapet (SSRS) och behandlar utvecklingen av ett koncept för infångning av fastvingade drönare. SSRS är en ideell organisation som arbetar med att rädda liv till sjöss. Inom projektet Eyes On Scene används kameraförsedda drönare för att snabbt ge räddningsledningen en tydlig lägesbild vid olyckor. Idag saknas dock en effektiv metod för landning, vilket innebär att drönaren måste plockas upp manuellt. Detta är tidskrävande och kan begränsa tillgängligheten för framtida uppdrag. Syftet har därför varit att utveckla ett system som möjliggör en säker och automatiserad infångning. Arbetet har följt Ulrich och Eppingers modell för produktutveckling med fokus på konceptframtagning. Processen omfattade funktionsnedbrytning, behovsanalys, kravspecifikation och konceptgenerering genom brainstorming och morfologisk matris. Koncepten utvärderades med elimineringsmatris, screening, scoring och delar av konstruktionen verifierades med FEM-analyser. Resultatet blev ett koncept där drönaren fångas av en lina som hålls av infångningsarmar på teleskopiska stolpar. Infångningen dämpas med fjäder och spolsystem och hela konstruktionen är monterad på en svängkransaxel som möjliggör 360° rotation för att följa inflygningen. Armarnas rörelser anpassas efter drönarens position, vilket gör att systemet klarar viss felmarginal vid landning. Konstruktionen är kompakt, ryms i en takbox och är enkel att transportera. Lösningen inkluderar även ett planerat styrsystem samt utrymme för sensorer som kamera eller radar, vilka måste utvecklas i framtida arbete
Mitigation of the Pressure Pulsations in a Hydraulic Axial Turbine with Asynchronous Guide Vanes
No full textHydraulic turbines are increasingly used for power grid regulation as intermittent energy resources, such as wind and solar power, gain prominence. The power output from these renewable sources fluctuates over short and long periods depending on weather conditions. Therefore, hydraulic turbines often operate away from their design point to mitigate grid imbalances, such as low-load operation, which presents challenges. The guide vanes control the flow, and their opening angle is limited during low-load conditions to restrict the flow rate and reduce power output, creating a high swirl flow condition. During part load (PL) operation, the residual swirl entering the draft tube can initiate a rotating-vortex-rope (RVR) that wraps around a stagnant region, introducing severe pressure pulsations. Some turbines are expected to provide a spinning reserve, enabling rapid responses to grid power shortages. One method to achieve a spinning reserve is to allow the turbine to operate under speed-no-load (SNL) conditions, where the turbine rotates at synchronous speed without generating electrical power. Since the turbine does not extract any power, the energy must be dissipated through the flow field. The resulting chaotic flow field features sometimes rotating vortices attached to the head cover in the vaneless space, extending into the draft tube. The axial flow primarily occurs in a thin region near the outer wall, while a recirculating region exists at the center of the draft tube, potentially extending upstream of the runner. Similar to the RVR, the rotating vortices induce harmful pressure pulsations throughout the turbine, jeopardizing safe operation and shortening the turbine's lifespan due to an increased risk of material fatigue. This thesis aims to study the flow under low-load operating conditions for a Kaplan model turbine, specifically the Porjus U9 model, using computational fluid dynamics (CFD), and to explore a mitigation strategy that reduces pressure pulsations during these low-load operating conditions. The idea is to limit swirl and, consequently, the pressure pulsations caused by the flow structures by employing asynchronous guide vanes. This involves adjusting some guide vanes to a larger opening angle, while keeping the others closed and maintaining the same power output. Unlike other mitigation techniques, no additional installations are needed aside from an option to control some of the guide vanes asynchronously. CFD is combined with machine learning to explore various guide vane configurations efficiently. Results indicate that vortices in the vaneless space during SNL operation can be mitigated, significantly reducing the associated pressure pulsations by opening one consecutive section of guide vanes. However, the jet like flow from the opened guide vane section generates a significant radial force on the runner due to the asymmetric flow field and pressure pulsations on the runner blades, oscillating at the runner's rotational frequency. Both issues can be addressed by opening two sections of guide vanes on opposite sides of the runner axis while maintaining most of the mitigation effect. Furthermore, the flow field is predictable, and most stochastic pressure pulsations are reduced, positively impacting the turbine's lifespan. One section with open guide vanes during PL operation can decrease the amplitude of the pressure pulsations related to the RVR rotating mode (RM) and plunging mode (PM) in the draft tube. Conversely, the stochastic pressure variations increase, and the runner is subjected to an asymmetric radial force and pressure pulsations on the blades that oscillate at the runner's rotational frequency, much like for SNL operation. Additionally, efficiency decreases, and the torque on the runnervaries more. The mitigation effect diminishes when opening two guide vane sections, as the RVR reduces in size, but is not completely mitigated. The asynchronous guide vanes primarily affect the flow upstream of the runner, making the technique suitable for SNL operation, as the vortices originate upstream of the runner. The mitigation strategy is less effective during PL because the RVR originates in the draft tube. While the pressure pulsations related to the RVR can be reduced, significant stochastic variations persist. Furthermore, since the flow rate is higher during PL than SNL, the runner is subjected to higher-amplitude pressure pulsations and a more excessive radial force. Ultimately, implementing asynchronous guide vanes balances increased life expectancy and the cost of turbine operation. Experimental investigations are necessary to validate the above findings and clarify the actual effect on material fatigue and other parts of the turbine not included in the simulations
Digitala tvillingar för deltagande design : Ett nytt verktyg för stadsutveckling
No full textThe urban development process faces challenges due to the lack of citizen participation for a more democratic development. This issue results in several challenges, necessitating more active involvement from citizens to incorporate their ideas and opinions into urban development. One of the primary challenges is creating opportunities for more citizens to engage in the urban development process. Much of the current visual material provided to citizens often lacks real-time attributes and shows a deficient understanding of the development area. To address these issues, Digital Twins, usually described as a live coupling of the state of a physical asset or process to a virtual representation with a functional output, have recently begun to be utilised in the urban development process to function in real-time by connecting the urban landscape, stakeholders, and citizens. Compared to commonly used 3D visualisation technology, digital twins offer a more transparent and rapid participation process. The concept behind such a support system is to provide citizens with greater opportunities to express themselves and comprehend the urban development process. The objective of this research is to explore the design of digital twins and expand the knowledge of how digital twins can be integrated into urban development by connecting stakeholders and citizens through these technologies. Accordingly, the following research question has been defined: RQ. How can digital twins support citizen participation in urban developments? This thesis has evolved from broad to more specific aspects to deepen the understanding of digital twin implementations in urban development. The process of creating Urban Digital Twin Framework has revealed a lack of knowledge regarding how different digital twins are interconnected and what alternatives exist within urban digital twins. This framework was subsequently utilised to determine the visualisation accuracy required for digital twins across various urban development processes. It emphasised the significance of unique digital twins and their interaction with the development process. This thesis has also analysed and compared three different datasets: LiDAR, Photogrammetry, and Remote sensing based on visualisation attributes. This led to a new metric called Visualisation Accuracy, which describes the datasets' capability to render urban environments in ways that are visually realistic and experientially meaningful when deployed in visualisation platforms such as virtual or augmented reality. Employing diverse urban spaces and digital methodologies to enhance citizens' opportunities to participate in the architectural design process presents an innovative approach, ensuring that every citizen can be involved in the process.
20 Years of Hosting Capacity Studies, 2004–2024
No full textThe term “hosting capacity”, in the context of power systems, was first introduced in March 2004 and has since resulted in a range of applications and research. Network operators commonly use the concept to quantify the ability of their networks to accept new production and consumption. Academic research on hosting capacity took off seriously after 2010 and has resulted in thousands of publications. This paper presents a brief history of the early years of hosting capacity studies, gives an overview of the status of both applications and research, summarises the different methods and types of studies, and correlates all that with the underlying fundamental principles of the hosting capacity concept, as it was introduced in 2004. The main focus of this paper is to review and relate these methods and studies to the fundamental principles. Having a clear understanding of these fundamental principles enables a wide range of applications for hosting capacity studies with detailed methods and models. However, it also requires transparency to ensure a coherent analysis, correct interpretation of the results, and an open discussion between stakeholders. With research on hosting capacity, it is important to refer to the fundamental principles; with applications, it is important to maintain transparency and objectivity. Validerad;2025;Nivå 2;2025-10-09 (u8);Full text license: CC BY-NC-ND;This article has previously appeared as a manuscript in a thesis.</p
How to Measure Negative Effects of Psychological Interventions
No full textNegative effects of psychological interventions, such as deterioration, non-response, and novel symptoms, remain underexamined despite their ethical and clinical significance. This commentary reviews current methods for assessing such effects, highlighting the limitations of relying solely on symptom deterioration and the need for more direct patient-reported measures. Eight commonly used instruments are described and compared in terms of content, coverage, and psychometric properties. Methodological challenges, such as timing of assessment, measurement reactivity, and patients' reluctance to disclose harm, are also discussed. Recommendations are provided for improving the monitoring and reporting of negative effects in both clinical and research settings, with an emphasis on transparency, patient safety, and the inclusion of diverse populations and treatment modalities.Validerad;2025;Nivå 2;2025-08-14 (u1);Full text license: CC BY 4.0Research of Excellence on Digital Technologies and Wellbein
Statistical Reliability Assessment : Integrating Predictive Analytics for Maintenance Strategies
No full textThis thesis presents a data-driven framework for enhancing preventive maintenance strategies by integrating statistical reliability analysis and predictive analytics. The study, conducted in collaboration with Volvo CE, addresses the underutilization of maintenance data in current industrial applications. By analyzing failure data extracted from the maintenance system Avantis of Volvo CE, this research applies statistical methods including Trend Testing, Goodness of Fit Testing and fitting Parametric Distributions. These methods are used in order to identify potential failure patterns, which is a strong tool to apply when optimizing preventive maintenance timing. The results inform a proactive approach to maintenance that when applied effectively helps reduce system downtime, increasing its reliability. This approach is presented in a framework that is both scalable and adaptable for broader industrial applications aiming to bridge the gap between maintenance data and its practical application, which transitions into a greater degree of predictive maintenance.
Designing Pre-Ingest Systems for Digital Preservation
No full textOur society is experiencing a rapid digital transformation, producing vast amounts of digital information at an unprecedented rate. As technology advances, data stored in legacy systems risks becoming inaccessible. For government agencies, however, legal requirements for records management, public access, confidentiality, and data protection mean that preservation is not optional but mandatory. These requirements demand verifiable authenticity, integrity, provenance, and controlled access over long periods that often exceed the lifespan of the originating systems. To maintain the integrity and authenticity of digital information over time, proper techniques and systematic processes are essential for transferring digital assets from producers’ systems to long-term preservation environments, such as digital archives. This transfer process, known as pre-ingest, is complex and resource-intensive, creating a need for solutions that simplify workflows, ensure authenticity, and enable the transfer of digital objects into preservation environments designed for long-term accessibility. This dissertation explores how to design and govern preservation workflows, with a focus on the pre-ingest stage, the transfer from operational systems to archival systems, where legal and policy obligations are operationalized through metadata handling, machine-actionable checks, and the creation of verifiable SIPs aligned with the OAIS reference model. To address the need for practically usable pre-ingest systems, this dissertation contributes a set of six novel design principles that cover collaborative engagement, adaptive SIP generation, quality assessment, purpose-driven configuration, workflow orchestration, and submission-agreement flexibility. Drawing on empirical evidence from the Test Platform (TP), ENSURE, and ForgetIT projects, the study assesses approaches and technologies that help organizations prepare and transfer digital collections for long-term preservation. Collectively, these principles establish pre-ingest as a strategic, configurable subprocess, demonstrating how design-driven development can enable adaptive, automated, and policy-aware pre-ingest across diverse archival contexts. This work provides a foundation rather than a final achievement; future research should validate and extend these principles in operational settings, advance compliance-as-code approaches, and strengthen interoperability across organizational and archival environments
I just want to sleep : An interview study on nurse anesthetists’ experiences of patients who wish to switch from regional to general anesthesia
No full textIntroduktion: Regional anestesi har många fördelar för patienters säkerhet men en kontraindikation till regional anestesi kan vara att patienten inte vill göra ingreppet med anestesimetoden. Det kan hända att patientens åsikter om byte av anestesimetod framkommer på själva operationsdagen när anestesisjuksköterskan träffar patienten för första gången. Denna studie syftar till att beskriva anestesisjuksköterskors upplevelser av patienter som vill byta från regional till generell anestesi. Metod: Studien är en kvalitativ intervjustudie med manifest ansats där sju anestesisjuksköterskor intervjuades. Datan analyserades med kvalitativ innehållsanalys. Resultat: Resultatet från intervjuerna genererade kategorierna; “Att bemöta behov hos oroliga patienter”, “Att informera och motivera” och “Att hantera utmaningar vid förändrad anestesimetod”. Slutsats: Resultatet visar att patienters önskemål om generell anestesi ofta bottnar i oro, ångest eller tidigare negativa erfarenheter. Anestesisjuksköterskan har en central roll i att minska denna oro genom individanpassad information, närvaro och avledande åtgärder, vilket stärker personcentrerad vård. Samtidigt utgör organisatoriska hinder som tidsbrist och hög arbetsbelastning en risk för att patientmöten inte blir optimala
Experimental characterization of additively manufactured material using Split-Hopkinson Pressure Bar for material modeling
No full textThis thesis explores the mechanical response of precipitation-hardened stainless steel, PH 17-4, across different strain rates and temperatures. The experimental part combined high strain rate compression tests using Split-Hopkinson Pressure Bar (SHPB) with quasi-static compression tests on a Gleeble testing system. Tests were performed at 20°C, 200°C and 400°C, and the resulting data were carefully processed to obtain true stress–strain curves, yield points, and strain rate dependencies. The constitutive behavior was modeled using the Johnson–Cook (JC) model which accounts for strain hardening, strain rate sensitivity, and thermal softening. Fitting the model to both SHPB and Gleeble datasets gave unsatisfactory results, particularly at high strain rates, due to limiations of the JC model. Therefore, the model was also calibrated using only the SHPB data for comparison, which showed a significant improvement in prediction. Overall, the study provides a set of calibrated JC parameters for PH 17-4 stainless steel that can be implemented in numerical simulations of high strain rate scenarios. The work highlights both the strengths and the limitations of the JC approach when applied to PH 17-4 which offers a foundation for future research into more advanced constitutive models. The results are particularly relevant for applications in aerospace and additive manufacturing, where PH 17-4 is frequently subjected to high strain rates and elevated temperatures.