kth.sePublications
12345 1 - 50 of 246
rss atomLink to result list
Permanent link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
  • Public defence: 2024-03-15 10:00 D3, Stockholm
    Aguiar, Miguel
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Decision and Control Systems (Automatic Control).
    Learning flow functions: architectures, universal approximation and applications to spiking systems2024Licentiate thesis, monograph (Other academic)
    Abstract [en]

    Learning flow functions of continuous-time control systems is considered in this thesis. The flow function is the operator mapping initial states and control inputs to the state trajectories, and the problem is to find a suitable neural network architecture to learn this infinite-dimensional operator from measurements of state trajectories. The main motivation is the construction of continuous-time simulation models for such systems. The contribution is threefold.

    We first study the design of neural network architectures for this problem, when the control inputs have a certain discrete-time structure, inspired by the classes of control inputs commonly used in applications. We provide a mathematical formulation of the problem and show that, under the considered input class, the flow function can be represented exactly in discrete time. Based on this representation, we propose a discrete-time recurrent neural network architecture. We evaluate the architecture experimentally on data from models of two nonlinear oscillators, namely the Van der Pol oscillator and the FitzHugh-Nagumo oscillator. In both cases, we show that we can train models which closely reproduce the trajectories of the two systems.

    Secondly, we consider an application to spiking systems. Conductance-based models of biological neurons are the prototypical examples of this type of system. Because of their multi-timescale dynamics and high-frequency response, continuous-time representations which are efficient to simulate are desirable. We formulate a framework for surrogate modelling of spiking systems from trajectory data, based on learning the flow function of the system. The framework is demonstrated on data from models of a single biological neuron and of the interconnection of two neurons. The results show that we are able to accurately replicate the spiking behaviour.

    Finally, we prove an universal approximation theorem for the proposed recurrent neural network architecture. First, general conditions are given on the flow function and the control inputs which guarantee that the architecture is able to approximate the flow function of any control system with arbitrary accuracy. Then, we specialise to systems with dynamics given by a controlled ordinary differential equation, showing that the conditions are satisfied whenever the equation has a continuously differentiable right-hand side, for the control input classes of interest.

    Download full text (pdf)
    fulltext
  • Uzunel, Sinem
    et al.
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Biomedical Engineering and Health Systems, Health Informatics and Logistics.
    Xu, Joanna
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Biomedical Engineering and Health Systems, Health Informatics and Logistics.
    Prestandajämförelse mellan krypterade och okrypterade tidsseriedatabaser med IoT-baserad temperatur- och geopositionsdata2024Independent thesis Basic level (university diploma), 10 credits / 15 HE creditsStudent thesis
    Abstract [en]

    The Internet of Things (IoT) is a growing technology that plays an increasingly significant role in society. It encompasses a network of internet-connected devices that collect and exchange data. As IoT continues to expand, challenges arise regarding the management of large volumes of data and security aspects. The company Softhouse faces the challenge of choosing an efficient time-series database for handling temperature and geoposition data from heating systems in homes, where both performance and data integrity through encryption are of great importance.

    Therefore, this thesis has conducted a performance comparison between AWS Timestream and InfluxDB, using various tests to measure the execution times for data ingestion of sensor data and database queries. The comparison includes AWS Timestream in encrypted form versus InfluxDB in its AWS cloud version in encrypted form, as well as InfluxDB AWS in encrypted form versus InfluxDB in unencrypted form. The aim of the study was to provide guidelines for the selection of time-series databases with a focus on performance and security aspects, including encryption. The study also explored how the choice of the right database affects companies like Softhouse, both in terms of quantitative and qualitative benefits, and provided an assessment of costs.

    The results showed that InfluxDB in its AWS cloud version generally outperformed AWS Timestream and InfluxDB in its standard version. There were clear performance differences between AWS Timestream and InfluxDB in its AWS cloud version, but not as pronounced differences in performance between InfluxDB in itsAWS cloud version and the standard version. Considering both performance and security, InfluxDB in its AWS cloud version appears to be the most suitable option. However, it is crucial to consider the cost aspect, as AWS Timestream proves to be significantly more cost-effective than InfluxDB.

    Download full text (pdf)
    fulltext
  • Gårdestam, Sofie
    et al.
    KTH, School of Industrial Engineering and Management (ITM), Engineering Design, Integrated Product Development and Design.
    Kronér, Amalia
    KTH, School of Industrial Engineering and Management (ITM), Engineering Design, Integrated Product Development and Design.
    Robotic Automation of Mechanical Verification2024Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    This master thesis explores the implementation of Collaborative robots into Elekta's mechanical product verification process. Elekta is a pioneer in precision radiation therapy solutions. Ensuring the compliance of Elekta's products with performance, safety, and quality criteria is of great importance. The existing verification process faces challenges regarding resource management, ergonomics, efficiency, and test accuracy.This thesis delves into three main research areas, the study begins with theoretical research investigating Industry 4.0, automation, and human-robot collaboration, focusing on the ABB GoFa Cobot. The following research area analyzes the existing market, regarding the exploration of different robotic grippers. Lastly, the third research area focuses on the user. Including interviews and observations with Elekta testing employees and industry experts from ABB and Atlas Copco.This thesis examines insights from theoretical, market, and user research to identify key factors influencing the development of concepts, prototypes, and models. By this, strengths, weaknesses,and a criteria value matrix were created for various gripper options for Elekta's tests. Furthermore, test parameters were defined to make the test suitable for automated Cobot verification. Lastly, the conclusion from the user research was examined in several parts, including, efficiency, safety, ergonomics, limitations, robotics, and industry. The thesis continued with a creative process that contains the development and iterations of concepts by CAD modeling, 3D printing, and Cobot programming for a specific verification test. The results of this thesis include Cobot implementation guidelines, covering the selection of suitable tests for automation, gripper selection, and finger development. The report then delves into the sustainable regards related to Cobot implementation, studying social, economic, and environmental sustainability. Furthermore, decisive factors that could influence the project were discussed. In conclusion, this master's thesis successfully addresses Elekta's mechanical product verification challenges through the strategic implementation of Cobots. It is expected to increase efficiency, reduce ergonomic risks, improve resource management, and increase test precision. This research provides a blueprint for Cobot implementation at Elekta and positions the company at the forefront of modern industrial technology, emphasizing the potential for further development in this field.

    Download full text (pdf)
    fulltext
  • Barsoumi, Rabi
    et al.
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Biomedical Engineering and Health Systems.
    Odowa, Mohammed
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Biomedical Engineering and Health Systems.
    Utvärdering av luftrenare som använder sig av centrifugalteknologi: På uppdrag av Airission i samarbete med Karolinska Universitetssjukhuset i Huddinge2023Independent thesis Basic level (university diploma), 10 credits / 15 HE creditsStudent thesis
    Abstract [en]

    This report presents an evaluation of a centrifugal air purifier from the company Airission used in an infectious ward at Karolinska University Hospital. Airission's air purifier removes particles and aerosols using centrifugal technology, a relatively untested technology for air purification. The goal was to investigate and attempt to verify the functionality and performance of the air purifier and compare it to a conventional air purifier that uses traditional two-stage filters for air purification.To conduct the study, a bioaerosol measuring instrument was used to measure real-time particle levels in the room. The tests were performed under different operating conditions and time intervals with the air purifier turned on and off. Data collection and analysis included calculating the mean values, comparing the particle levels between different test cases, and calculating the standard deviation.The results showed that Airission's centrifugal air purifier effectively purifies air from particles and aerosols. The purification efficiency was comparable to a conventional air purifier. It was more effective than a conventional air purifier without the use of highefficiency air filters, commonly known as HEPA filters. A significant reduction in the number of particles in the air was observed while the air purifier was in operation. However, some complications arose during the application of the HEPA filter, which could have had a negative impact on both air purifiers.In summary, the analysis demonstrates that the applied centrifugal technology in Airission's air purifier works well. The comparison with the conventional air purifier shows certain advantages of an air purifier that uses centrifugal technology - in terms of both efficiency and quality.This report contributes to the knowledge of air purification solutions to improve air quality and reduce the spread of airborne diseases, especially in hospital environments. The results can be useful for further research and development of more effective air purifiers, which in turn provide better protection for patients and hospital staff exposed to airborne pathogens.

    Download full text (pdf)
    Utvärdering av luftrenare som använder sig av centrifugalteknologi
  • Amirapu, Lalitha Swetha
    et al.
    KTH, School of Industrial Engineering and Management (ITM), Engineering Design.
    Yalamanchili, Haswanth
    KTH, School of Industrial Engineering and Management (ITM), Engineering Design.
    Efficient FE Modeling of Large Casted Parts2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The design and analysis of large casted parts present significant challenges due to their complex geometry. Finite Element (FE) modeling is a vital tool for understanding the performance of casted components. However, the computational requirements associated with these parts often lead to excessive processing times and resource utilization. This thesis aims to enhance the efficiency of the mid-surface model creation by developing an FE modeling approach suited explicitly for large casted components. The study begins by exploring the background of casted parts and their applications. A comprehensive analysis of modeling and meshing techniques is conducted, emphasizing their application to large casted components. Building upon this knowledge, different ideas are examined, leading to the proposal of a methodology combining CAD strategies for design features, hybrid meshing techniques, and approaches aimed at reducing FE modeling time to streamline the overall process.To validate the proposed approach, a series of case studies involving casted parts with varying levels of complexity are undertaken. Real-world casting process parameters are considered, highlighting the advantages and limitations in each ideation phase. The proposed methodology is tested and show cased to expert engineers to evaluate its efficiency and feasibility. Furthermore, the efficiency of the new approach is quantitatively evaluated in terms of processing time. The developed methodology offers engineers and researchers a powerful tool to accelerate the design process and optimize FE modeling time while managing computational costs. As industries continue to push the boundaries of size and complexity in casted part design, the insights and techniques presented in this thesis offer a valuable resource for addressing the various engineering challenges inherent in future endeavors.

    Download full text (pdf)
    fulltext
  • Naserallah, Dina
    KTH, School of Architecture and the Built Environment (ABE), Civil and Architectural Engineering, Building Technology and Design.
    Brandskyddsegenskaper, miljöpåverkan och hållbarhet med högtrycks vattendimsprinklers jämfört med traditionella sprinklers: En undersökning av högtrycks vattendimsprinklers jämfört traditionella sprinklers från brandsäkerhets-, miljö- och hållbarhetsperspektiv2023Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
    Abstract [en]

    This study explores differences between two types of fire suppression systems: traditional watersprinkler system and high-pressure water mist system from the perspective of fire protection aswell as environmental properties. The purpose of this study is to gain a deeper understanding ofthese two systems and their use for fire protection.Traditional water sprinkler systems systems showed to offer effective fire detection and quickresponse thanks to heat-sensitive glass bulbs. It was also particiularly more effective in largeopen spaces. Traditional water sprinklers have also appeared to be effective for enfironmentswith high fibrosity, such as sawmills.High-pressure water has shown several advantages, such as early fire detection and responsetime, as well as a more efficient fire extinguishing capacity. The system uses small waterdroplets discharged at high pressure, which creates a larger coverage area and leads to up to90% less water usage. This reduces the risk of water damage and makes the system particularlysuitable for use in confined spaces and for protecting objects like lithium-ion batteries, as wellas in most other environments except those with high fibrosity.From an environmental perspective, water mist proved to be superior from several aspects. Thisincludes reducing water usage, which leads to less, environmentally hazardous firewater, thatneeds to be managed after extinguishing a fire. Higher energy efficiency is also achieved,resulting in lower carbon dioxide emissions because the pump does not need to handle as muchwater as a traditional sprinkler system. High-pressure water mist also contributes to improvedair quality by suppressing hazardous fire gases.In conclusion this study has shown important insights and conclusions, offering a betterunderstanding of the compared fire protection systems. The analyses indicate that both high-pressure water mist and traditional water sprinkler systems have their specific advantagesdepending on their usage areas. The purpose of the study was to investigate and compare thesetwo systems with regard to their fire-fighting capabilities, contributing to a more efficient firesuppression. There has been a lack of information from previous studies, making the subjectvery interesting to explore more in future studies.

    Download full text (pdf)
    fulltext
  • Al-Bazarkan, Abdulla
    et al.
    KTH, School of Architecture and the Built Environment (ABE), Civil and Architectural Engineering, Concrete Structures.
    Alahmad, Hussain
    KTH, School of Architecture and the Built Environment (ABE), Civil and Architectural Engineering, Concrete Structures.
    Loading tests on composite concrete columns: A comparison between composite columns made of permanent 3DPC formwork filled with SCC and columns made of homogenous SCC2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    In order to achieve increased sustainability when casting concrete structures this thesis investigates the possibility of utilizing 3D-printing to construct formwork and digitalize construction. By digitizing construction less workers are needed and therefore less capital. The main aim of this thesis is to examine the load-carrying capacity of composite columns made of a permanent 3D-printed concrete formwork filled with SCC and comparing their strength with that of homogenous columns made of SCC. Another aim of this thesis is to examine the formwork pressure and lastly to examine the bond strength between the two concrete materials. The cylindrical columns were reinforced and cast in Tumba, Stockholm at ConcretePrint. Another rectangular temporary formwork was also constructed to measure the formwork pressure. The measured formwork pressure is compared to a calculated theoretical pressure. The load-carrying capacity was examined with the help of a hydraulic press at RISE in Borås. Bond strength was also tested at RISE in Borås by drilling out cores of the composite columns. Results prove that formwork pressure is not an issue for these types of columns. The results also show that the composite columns are stronger when wholly loaded over their cross section while they perform similarly when a concentrated centrally placed load is applied. When subjected to a concentrated centrally placed load it is thought that an eccentricity occurs and therefore lowers the load-carrying capacity in both the homogeneous and composite columns. Further, bond strength between the composite materials had such values that the bond is not thought to be a limiting factor either. While no conclusive conclusion can be drawn due to the limited number of columns, the results are promising and further research is recommended. Such research could for example be durability studies which will be treated in a special report by KTH and possibly a study regarding the economic benefit.

     

    Download full text (pdf)
    fulltext
  • Public defence: 2024-03-12 13:15 Ka-301, Stockholm
    Khorsandmanesh, Yasaman
    KTH, School of Electrical Engineering and Computer Science (EECS), Computer Science, Communication Systems, CoS.
    Hardware Distortion-Aware Beamforming for MIMO Systems2024Licentiate thesis, comprehensive summary (Other academic)
    Abstract [en]

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

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

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

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

    Download full text (pdf)
    fulltext
  • Kruse, Josefin
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Biomedical Engineering and Health Systems, Ergonomics.
    Product development of hand-held scooping utensil for the pharmaceutical industry2024Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Musculoskeletal disorders (MSDs) are one of the most common groups of work-related illnesses. At PET Dispensing, AstraZeneca Södertälje, several MSDs had been linked to a specific scooping utensil made of stainless steel. This master thesis aimed to, using a user-centered design approach, identify the MSD risks associated with this utensil and then develop, test, and suggest a re-design to reduce these risks. PET Dispensing was a pre-manufacturing department with the main task of re-packaging supplier raw material, such as cellulose and color pigment, before entering AstraZeneca’s factories where it needed to fit their industrial processes for producing medicines. The repackaging was sometimes done using a lifting aid but more often using a hand-held scooping utensil in stainless steel. The packaging varied in size, material, and weight depending on the type of raw material and supplier and the work tasks were therefore hard to standardize. Interviews with operators and observation of the scooping tasks with the utensil showed that a full scoop could weigh up to 2.5 kg per scoop and the scooping motion was often done about 56 times per 30 min. The utensil handle was so positioned that downward bending of the wrist occurred during most of the scooping task.

    The result of the thesis project was two-fold: identification of risk factors and a requirement specification with a final design concept for a new scooping utensil. The most prominent MSD risks were repetitive motions, especially of the wrist, high loads using one arm, and awkward body postures. For the risks to lessen one or more of these parameters must be addressed. Organizational factors such as rotation of work tasks, training of employees, and customized work material and tools such as lifting aids and height adjustable surfaces were also recommended measures. The product development of a new scooping utensil concept was done using ideation, prototyping, and testing. Prototypes of increasingly higher levels of detail were used to visualize ideas and were brought to the operators for feedback. The final design concept is based on literature studies on ergonomic handle design and requirements collected from operators and process responsible at AstraZeneca. It is visualized using 3D modeling software and has a new handle placement with a grip optimized for force transmission. The proposed scooping utensil concept includes using stainless steel as a main material because of hygiene requirements and regulations at PET Dispensing. Other materials are encouraged to be further evaluated.

    Download full text (pdf)
    fulltext
  • Public defence: 2024-03-11 15:00 F3, Stockholm
    Taghavian, Liam Hamed
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Decision and Control Systems (Automatic Control).
    Externally positive systems: Analysis and control based on combinatorial polynomials2024Doctoral thesis, monograph (Other academic)
    Abstract [en]

    Monotonic tracking is required in many control systems, including those that cannot tolerate any overshoots and undershoots in their closed-loop responses. Classical examples are found in vehicle cruise control and liquid tank level control. In the former, an overshoot happens when the speed of the vehicle goes beyond the set value violating the safety measures, and in the latter an overshoot is considered as filling up the tank with an excessive amount of liquid which leads to a waste of resources. Controllers that eliminate overshoots are undeniably more desirable in these examples. In fact, the same requirement is in place for many more engineering applications, including biological systems, robotics and process control, indicating widespread benefits of controllers which can guarantee monotonicity in the system response. Formally, linear systems that exhibit monotonically increasing step responses are called externally positive. Designing controllers that render the closed-loop system externally positive requires a thorough understanding of this property in linear systems. In this thesis, we leverage combinatorial polynomials and their properties to study external positivity in both discrete-time and continuous-time linear systems modelled by transfer functions or impulse responses. Several conditions are provided that are either necessary, sucientor both necessary and sufficient for a linear system to be externally positive. These conditions are then used to synthesize controllers that ensure external positivity in closed-loop systems and hence, eliminate both overshoots and undershoots in the system response. In particular, we provide synthesis techniques based on convex optimization that ensure stability, robustness and offset-free monotonic tracking in the closed-loop system and improve its decay rate and sensitivity. We compare the results with the state-of-the-art in the literature and demonstrate the efficacy of the proposed controller synthesis methods through several numerical examples.

    Download full text (pdf)
    Thesis
  • Tengana Hurtado, Lizzy
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Computation Offloading for Real-Time Applications: Server Time Reservation for Periodic Tasks2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Edge computing is a distributed computing paradigm where computing resources are located physically closer to the data source compared to the traditional cloud computing paradigm. Edge computing enables computation offloading from resource-constrained devices to more powerful servers in the edge and cloud. To offer edge and cloud support to real-time industrial applications, the communication to the servers and the server-side computation needs to be predictable. However, the predictability of offloading cannot be guaranteed in an environment where multiple devices are competing for the same edge and cloud resources due to potential server-side scheduling conflicts. To the best or our knowledge, no offloading scheme has been proposed that provides a highly predictable real-time task scheduling in the face of multiple devices offloading to a set of heterogeneous edge/cloud servers. Hence, this thesis approaches the problem of predictable offloading in real-time environments by proposing a centralized server time reservation system to schedule the offloading of real-time tasks to edge and cloud servers. Our reservation system allows end-devices to request external execution time in advance for real-time tasks that will be generated in the future, therefore when such a task is created, it already has a designated offloading server that guarantees its timely execution. Furthermore, this centralized reservation system is capable of optimizing the reservation scheduling strategy with the goal of minimizing energy consumption of edge servers while meeting the stringent deadline constraints of real-time applications.

    Download full text (pdf)
    fulltext
  • Shen, Xiner
    KTH, School of Electrical Engineering and Computer Science (EECS).
    3D Coating of Interface Materials for High-Performance RF Passive Devices2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The demand for high-performance Radio Frequency (RF) passive devices has been steadily increasing due to the growing complexity and sophistication of wireless communication systems. The Quality factor (Q-factor) is a key parameter for describing the signal losses and the energy efficiency of resonators. Previous studies have been done on the spin coating technique of intermediate coating, which presented some limitations in terms of 3D resonators. In this master thesis, we investigate the development of a intermediate layer using dip coating to enhance the Q-factor, i.e., the performance of RF passive devices. The dip coating method is applied to add a nano ceramic coating to the 3D structure as the intermediate layer between the resonator ceramic substrate and the conductive silver coating. After the fabrication process, the samples are observed under Scanning Electron Microscope (SEM) and Atomic Force Microscope (AFM) and tested with Vector Network Analysis (VNA). Analysis and calculations are mainly conducted with the software Matlab and Gwyddion. The proposed technique improves the smoothness of the samples by 78.95%, and the Q-factor is tested to have a 20.87% enhancement using VNA. The results demonstrate that the intermediate layer with the dip coating technique significantly improves the performance of RF passive devices by reducing the roughness of the resonator surface. These findings open up new opportunities for the design and development of high-performance RF passive devices in various applications, including wireless communication systems, radar systems, and satellite communication. Further studies can be carried out to reduce defects during fabrication and to stabilize the performance of the silver coating.

    Download full text (pdf)
    fulltext
  • Salvador Lopez, Eduardo
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Design of a grating lobe mitigated antenna array architecture integrated with low loss PCB filtering structures2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Massive multiple input multiple output - MIMO systems are a reality and modern communication systems rely upon this technology to cope with the increasing need for capacity and network usage. Antenna arrays are at the heart of the of the massive-MIMO system and are the enabling technology. The defining cost of such a system is the number of transmit receive ports TRx as they dictate the number of control points and the associated digital control computational capacity. Typically users are spread along the azimuth and there is limited angular user spread along elevation. This enables us to group the elements in elevation which of course limits the elevation scanning performance. The element grouping result in grating lobes when we do elevation scanning. In the newly introduced frequency range 3 - FR3 in the envisioned 6G communication systems that is from 6-20 GHz it will not be allowed to transmit power above the horizon and the resulting grating lobes from the standard grouping should be mitigated. This project is structured into two parts. In the first part a grating lobe mitigation technique based on irregular subarray grouping utilizing the wellknown Penrose irregular tessellation is developed. This tessellation is based into two geometrical shapes where when put together they can fully tile the space aperiodically. Introducing this apperiodicity the grating or quantization lobes of the array are mitigated. In addition, in the first part a beam forming algorithm is developed based on particle swarm optimization that is able to produce the optimal weights for the array steering as well as optimize some of the embedded patterns of the irregular grouping. The last optimization step of the irregular subarray patterns is utilized only when the grouping results in a narrow pattern in azimuth and as a result we have static single port beamforming networks. This of course is a trade off between the broadside gain and the azimuth steerability of the array. In the second part of this thesis two low loss band pass filters have been developed with a PCB integrated suspended stripline techology. The filters were optimised for the frequencies within FR3. The resulted filtering structures can further be integrated at the input port of the proposed feeding network with the same technology. The two parts of this thesis target to introduce on one hand a antenna array architecture with subarray groupings that produce no grating lobes and on the other hand the proposed filtering structures have small enough dimensions to fit within the subarray footprint.

    Download full text (pdf)
    fulltext
  • Koutlis, Dimitrios
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Estimation of Voltage Drop in Power Circuits using Machine Learning Algorithms: Investigating potential applications of machine learning methods in power circuits design2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Accurate estimation of voltage drop (IR drop), in Application-Specific Integrated Circuits (ASICs) is a critical challenge, which impacts their performance and power consumption. As technology advances and die sizes shrink, predicting IR drop fast and accurate becomes increasingly challenging. This thesis focuses on exploring the application of Machine Learning (ML) algorithms, including Extreme Gradient Boosting (XGBoost), Convolutional Neural Network (CNN) and Graph Neural Network (GNN), to address this problem. Traditional methods of estimating IR drop using commercial tools are time consuming, especially for complex designs with millions of transistors. To overcome that, ML algorithms are investigated for their ability to provide fast and accurate IR drop estimation. This thesis utilizes electrical, timing and physical features of the ASIC design as input to train the ML models. The scalability of the selected features allows for their effective application across various ASIC designs with very few adjustments. Experimental results demonstrate the advantages of ML models over commercial tools, offering significant improvements in prediction speed. Notably, GNNs, such as Graph Convolutional Network (GCN) models showed promising performance with low prediction errors in voltage drop estimation. The incorporation of graph-structures models opens new fields of research for accurate IR drop prediction. The conclusions drawn emphasize the effectiveness of ML algorithms in accurately estimating IR drop, thereby optimizing ASIC design efficiency. The application of ML models enables faster predictions and noticeably reducing calculation time. This contributes to enhancing energy efficiency and minimizing environmental impact through optimised power circuits. Future work can focus on exploring the scalability of the models by training on a smaller portion of the circuit and extrapolating predictions to the entire design seems promising for more efficient and accurate IR drop estimation in complex ASIC designs. These advantages present new opportunities in the field and extend the capabilities of ML algorithms in the task of IR drop prediction.

    Download full text (pdf)
    fulltext
  • Dikonimaki, Chrysoula
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Edge Compute Offloading Strategies using Heuristic and Reinforcement Learning Techniques.2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The emergence of 5G alongside the distributed computing paradigm called Edge computing has prompted a tremendous change in the industry through the opportunity for reducing network latency and energy consumption and providing scalability. Edge computing extends the capabilities of users’ resource-constrained devices by placing data centers at the edge of the network. Computation offloading enables edge computing by allowing the migration of users’ tasks to edge servers. Deciding whether it is beneficial for a mobile device to offload a task and on which server to offload, while environmental variables, such as availability, load, network quality, etc., are changing dynamically, is a challenging problem that requires careful consideration to achieve better performance. This project focuses on proposing lightweight and efficient algorithms to take offloading decisions from the mobile device perspective to benefit the user. Subsequently, heuristic techniques have been examined as a way to find quick but sub-optimal solutions. These techniques have been combined with a Multi-Armed Bandit algorithm, called Discounted Upper Confidence Bound (DUCB) to take optimal decisions quickly. The findings indicate that these heuristic approaches cannot handle the dynamicity of the problem and the DUCB provides the ability to adapt to changing circumstances without having to keep adding extra parameters. Overall, the DUCB algorithm performs better in terms of local energy consumption and can improve service time most of the times.

    Download full text (pdf)
    fulltext
  • Zheng, Zhuo
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Direct writing metal-freebio-organic piezoelectricenergyharvester2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The project is about piezoelectric energy harvesters and piezoelectric bio-organic materials.Nowadays, various kinds of energy harvesters based on micro or nano materials are appliedinmanyelectronic applications, such as wearable devices and electricity generators. Amongthem, thepiezoelectric effect-based energy harvesters are more attractive in research and industryfields. Inrecent years, piezoelectric biomaterials become a popular option because they are availabletocouple electrical and mechanical energy in a biological, ecofriendly systemto generate electricityinreal time. Among them, γ- glycine crystals have been recently synthesized in wafer scale throughasimple polyvinyl alcohol (PVA)-assisted evaporation process exhibiting good piezoelectricperformance. However, so far there are no metal-free energy-harvesting devices basedonPVA-glycine film to enable green manufacturing. In this project, we proposed the direct inkwritingorganic PEDOT:PSS electrodes and PVA-glycine-PVA piezoelectric crystals to fabricate metal-freeenergy harvesters. The output voltage reaches 1.5 V at a load resistance of 500 MΩandunderaforce of 10 N. The performance is comparable to other glycine-based devices in recent literature.Our scalable, sustainable and low-cost printing process is expected to greatly contribute tothefieldof biomaterials-based piezoelectric energy harvesting.

    Download full text (pdf)
    fulltext
  • Thorell, Anton
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Sporadic-E layers in the polar cap ionosphere: A review on Es occurrence, dynamics and formation theory2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Sporadic-E layers (Es) are layers of metallic ions that appear in the lower E-region of the ionosphere and can last from under one hour to several hours. Es are found at all latitudes, but polar cap Es, and specifically over Svalbard at a latitude of around 78◦ , are the focus of this study. Data is provided from several instruments: the EISCAT Svalbard Dynasonde, the EISCAT Svalbard Incoherent Scatter Radar (ESR), SuperDARN, and modelled data from the latest Horizontal Wind Model (HWM14). Data on the Interplanetary Magnetic Field (IMF) is acquired from the NASA OMNIWeb data base. It is found that polar cap Es are a summer phenomenon, confined to the later afternoon to a couple of hours after midnight in universal time (UT), with a peak occurrence between 18-21 UT. The layer heights are mostly confined to ∼92-120 km, although there is a discrepancy between Es found in ESR data and Dynasonde data, with ESR events being confined to 92-110 km, and Dynasonde events to 95-120 km. It is also found that Es occurrence is dependent on Interplanetary Magnetic Field (IMF) direction, with a higher occurrence during a southward and eastward IMF. The Dynasonde automatic signal processing of echoes is found to be unreliable at times with intense E-region density enhancements, such as Es. From a Superposed Epoch Analysis (SEA) on Es found in ESR data, it is found that there is a density buildup from the start of the events, peaking by 30 % of the Es lifetime, that is followed by fading at a slower rate until the end of the layer lifetime. Layer thickness is found to be largely confined to <9 km. From SuperDARN E-field data it is found that layers can form and migrate downwards to low altitudes for a strictly northward E-field. ’Flat’, low altitude layers are found during E-field directions in both the southwest and northwest quadrants. Some cases of Es formation and migration fit current theories and whilst some does not. Indications of particle precipitation that induce layer formation is found.

    Download full text (pdf)
    fulltext
  • Gajanan Naik, Harshavardhan
    KTH, School of Industrial Engineering and Management (ITM), Production engineering.
    Effective Digitization in Brownfield Factories: A conceptualized model for technology transfer to brownfield production factories through smart factory lab2024Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The exploration of Smart Factories and Industry 4.0 technologies has indeed sparked curiosity and interest in the industrial world. The potential of these advancements to revolutionize manufacturing processes, enhance efficiency, and drive innovation is immense. However, there is a gap in research when it comes to the practical implementation of these advanced technologies in real-world production settings, especially in already established factories so-called Brownfield Factories.

    This thesis work was conducted within one such brownfield factory to comprehend the tangible challenges associated with transferring smart technologies. Within this specific company, a laboratory had already been established for testing novel smart technologies in the context of production and logistics. The aim in companies is to test smart technologies in a controlled environment without causing any disruption to the ongoing profit-generating production processes. This laboratory setup also serves the additional purpose of educating the personnel within traditional production facilities about the upcoming smart technologies in the market. The Lab showcases the potential of new and emerging technologies in addressing long-standing issues with a fresh perspective, thereby inspiring innovation. The central approach of this thesis revolves around the establishment of a standardized laboratory work process through which smart technology can be tested in a structured way.

    In this context, an illustrative example of a technology, namely "Virtual Training for Assembly Operators," was chosen as a case study to explore and comprehend the challenges associated with technology transfer. This case study also played a pivotal role in assessing the credibility of the standard technology transfer model formulated within the company. Notably, it was deduced that knowledge and competence are two key obstacles impeding the smooth transfer of technology. Building upon the insights garnered from the case study on virtual training technology and drawing from interviews with engineers and managers employed at the case company, a refined technology transfer process named the "Smart Factory Lab Process" was developed. This process aims to enable the effective transfer of smart technologies, informed by the lessons learned from the practical application of technology in real-world scenarios.

    Download full text (pdf)
    fulltext
  • Chatelais, Léa
    KTH, School of Engineering Sciences (SCI), Engineering Mechanics, Vehicle Engineering and Solid Mechanics, Väg- och spårfordon samt konceptuell fordonsdesign.
    Vehicle dynamics modelling of electromagnetic suspensions for MAGLEV applications2024Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    MAGnetic LEVitation Guidance System (MAGLEV) technology was commercially introduced relatively recently in the guided transport field. It is based on removing the wheels and rails of classic railway systems and supporting and guiding the train with magnets and magnetic forces instead. But, as for conventional railways, those trains need to fulfil dynamic requirements in order to make trains safe and comfortable. The dynamics of a train being mainly influenced by its suspensions characteristics, the magnetic forces generated in MAGLEV systems are of prime importance. Having a model of those systems allows to check the requirements of a certain design, and to consider the influence of different parameters on their fulfilment. This thesis leans on research work on MAGLEV vehicle modelling to model and implement magnetic levitation components in a quarter-car model in order to study the fulfilment of vehicle dynamics requirements. Specifically, the modelled vehicles are based on Electro- Magnetic Suspension (EMS) and Electro-Dynamic Suspension (EDS) (Inductrack) technologies, for which the modelling equations are analysed to study the magnetic force dependencies with physical and operational parameters. Finally, the dynamic requirements are checked in response to a set of track irregularities amplitudes, anda parametric study is carried out to verify the fulfilment of those requirements for other design cases. The results show that it is possible to model and implement simple MAGLEV MBS models for dynamic studies, although it is challenging to model and simulate specific MAGLEV components because of the lack of component specifications or experimental data on track irregularities.

    Download full text (pdf)
    fulltext
  • Roudiere, Elie
    KTH, School of Engineering Sciences (SCI), Engineering Mechanics, Vehicle Engineering and Solid Mechanics, Väg- och spårfordon samt konceptuell fordonsdesign.
    Data analytics and machine learning for railway track degradation: Using Bothnia Line track measurements for maintenance forecasting2024Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    In this paper, a statistical method is developed to improve predictive maintenance on railway track. The problem tackled is being able to predict when the next maintenance event should take place to guarantee a certain track quality class. To solve the problem, The prediction is made using track measurement data exclusively, with no maintenance history to support the data analysis. The dataset consists of track measurements taken over eleven years and 170 kilometres on the Bothnia line in Northern Sweden. Different track degradation models and machine learning approaches are discussed and implemented. In the end, the tool developed was able to predict track degradation with an error within reasonable bounds of the typical maintenance limit. This will allow an operator to predict the recommended date for the next maintenance event at all locations using only historical track measurements as an input and little to no user intervention on the programme.

    Download full text (pdf)
    fulltext
  • Wollberg, Daniel
    KTH, School of Architecture and the Built Environment (ABE), Real Estate and Construction Management, Surveying – Geodesy, Land Law and Real Estate Planning.
    Enhancement of a Power Line Information System by Combining BIM and LiDAR Data2024Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    With the great ongoing energy transition in Sweden, Svenska Kraftnät (SVK) sees a huge need for investment in the Swedish transmission network and supporting IT- systems.  SVK has a great amount of collected laser data over the electric power transmission network however this data does not contain any semantic attribution that can be analyzed on broader information systems. The overall focus of this thesis is to investigate ways to enhance an information system.

    This thesis focuses particularly on the ability to combine information from pylon-3D-BIM models with point clouds of pylons gathered via an airborne laser scanner. Point cloud data from a 200-kilometer-long power line corridor between Djurmo and Lindbacka, including pylons, was provided by SVK.

    Two different methods to compile different data types to a single information system are investigated in this thesis. The first method is established by matching different types of pylon 3D-models to the respective reference point clouds using the Iterative Closest Point algorithm (ICP). The pylon models used for this method were S1J, B1J and SV2J pylons.The second method is based on segmentation of meaningful pylon parts from the point cloud data using predefined information about the shape of the pylons. The pylon models used for this method were S1J and B1J pylons.

    The goal was to automate the process and extract information as well as perform computations dynamically. This has been done using Feature Manipulation Engine (FME). The results are evaluated by comparing the two methods based on performance, reliability, and purpose. 93 % of the ICP comparisons showed that the best match between a point cloud model and a 3D-model was achieved when comparing models of the same pylon type. The highest accuracy was achieved when comparing an S1J pylon point cloud to an S1J pylon 3D-model.The segmentation method was used to successfully segment the beam, insulators and legs from the pylon point cloud data. A small sample size of pylon point clouds as well as a low number of different pylon 3D-models were used but both methods can be seen as a proof of concept that could be further evaluated in the future. In conclusion both methods used in this project were used successfully in order to enhance a power line information system.

    Download full text (pdf)
    fulltext
  • Ya Ting, Hu
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Features as Indicators for Delirium: An Application on Single Wrist-Worn Accelerometer Data from Adult Intensive Care Unit Patients2022Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Objective: The diagnosis of delirium in intensive care unit patients is frequently missed. Key symptoms to identify delirium are motoric alterations, changes in activity level, and delirium-specific movements. This study aimed to explore features collected by a single wrist-worn accelerometer as indicators of delirium. Methods: The study included twenty-two patients in the intensive care unit. The data was collected with the GENEActiv accelerometer device and the activity level was calculated. Differences between the delirious and nondelirious patients were tested. Results: Differences in activity level and rest-activity patterns were noticed between the delirious and non-delirious patients. However, the differences were not found to be significant. Conclusion: Activity patterns revealed differences between delirious and non‐delirious patients. Further study is required to confirm the potential of actigraphy in the early detection of delirium in the intensive care unit.

    Download full text (pdf)
    fulltext
  • El Azrak, Fatiha
    KTH, School of Architecture and the Built Environment (ABE), Civil and Architectural Engineering, Structural Engineering and Bridges.
    CFD Analyses for Wind Load AssessmentCFD Analyses for Wind Load Assessment: Wind-induced vibrations in the Bomarsunds Bridge hangers2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The trend in recent years of building longer and slender bridge components has introduced new challenges to ensure their stability and strength. This master thesis focuses on the effects of wind-induced vibrations in the context of long, slender arch bridge components, particularly the recently constructed Bomarsunds Bridge in Åland, Finland.

    The primary goal of this study is to comprehensively analyze the dynamic wind effects on the hangers due to the vortex shedding phenomenon, as the resulting vibrations pose potential risks to its safety and structural integrity. The slenderhangers of the bridge, close to the centre of the span, have exhibited significant vibrations, necessitating an in-depth investigation to understand the bridge’s response to wind forces. Computational fluid dynamics (CFD) simulations wereperformed using ANSYS Fluent to estimate more accurate aerodynamic quantities. Using CFD analysis, the behaviour of a given hanger section subjected to wind flow can be described. In this way, it was possible to calculate the aerodynamic coefficients that characterize that given section (i.e. Strouhal number (St), drag coefficient (CD), etc). By integrating advanced computer simulations and CFD analysis, the research addresses the complex challenges of investigating the vortex-induced vibration (VIV) phenomena at different wind speeds.

    The results showed an inconsistent trend for drag coefficients at varying wind speeds and lower drag for geometries with rounded edges, with an average value of drag coefficient of 1.60. The study highlighted the significant dependence of theStrouhal number on wind speed, varying from 0.129 for a wind speed of 2.5 m/s to 0.063 for a wind speed of 30 m/s, challenging traditional geometry-based estimations for this parameter. The drag frequency for each wind speed investigated is twice as high as the lift frequency, showing that at wind speeds of 7.5 m/s a drag frequency close to the fundamental transversal frequency of 6.6 Hz of the longest hanger is reached. This leads to the conclusion that for this particular case study, the headwind response is much more critical than the crosswind response. These findings can be used to implement effective measures to mitigate wind-induced vibrations in the studied hanger along the critical direction.

    By analyzing the complex vortex shedding phenomenon, the study contributes valuable insights into the field of wind engineering. This research plays a key role in ongoing efforts to design robust, safe, and resilient bridge structures.

    Download full text (pdf)
    fulltext
  • Das, Sandipan
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Information Science and Engineering. Scania, Sweden.
    Boberg, Bengt
    Scania, Sweden.
    Fallon, Maurice
    ORI, University of Oxford, UK.
    Chatterjee, Saikat
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Information Science and Engineering.
    IMU-based Online Multi-lidar Calibration2024Manuscript (preprint) (Other academic)
    Abstract [en]

    Modern autonomous systems typically use several sensors for perception. For best performance, accurate and reliable extrinsic calibration is necessary. In this research, we proposea reliable technique for the extrinsic calibration of several lidars on a vehicle without the need for odometry estimation or fiducial markers. First, our method generates an initial guess of the extrinsics by matching the raw signals of IMUs co-located with each lidar. This initial guess is then used in ICP and point cloud feature matching which refines and verifies this estimate. Furthermore, we can use observability criteria to choose a subset of the IMU measurements that have the highest mutual information — rather than comparing all the readings. We have successfully validated our methodology using data gathered from Scania test vehicles.

    Download full text (pdf)
    fulltext
  • Jabeli, Habib
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Biomedical Engineering and Health Systems, Health Informatics and Logistics.
    Optimizing Flight Ranking:A Machine Learning Approach: Applying Machine Learning to Upgrade Flight Sorting and User Experience2024Independent thesis Basic level (university diploma), 10 credits / 15 HE creditsStudent thesis
    Abstract [en]

    Flygresor.se, a leading flight comparison platform, uses machine learning to rankflights based on their likelihood of being clicked. The main goal of this project was toimprove this flight sorting to obtain a better user experience. The platform's existingmodel is based on a neural network approach and a limited set of features. The solution involved developing and comparing two machine learning models, Random Forest and XGBoost besides using a set of existing and newly created features. TheXGBoost model demonstrated superior performance by significantly improving theprediction of clicked flights by 4.18% while also achieving a remarkable increase inefficiency by being 125 times faster than the existing model.

    Download full text (pdf)
    fulltext
  • Das, Sandipan
    et al.
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Information Science and Engineering. Scania, Sweden.
    Mahabadi, Navid
    Scania, Sweden.
    Chatterjee, Saikat
    KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Information Science and Engineering.
    Fallon, Maurice
    Oxford Robotics Institute, UK.
    Multi-modal curb detection and filtering2022Conference paper (Other academic)
    Abstract [en]

    Reliable knowledge of road boundaries is critical for autonomous vehicle navigation. We propose a robust curb detection and filtering technique based on the fusion of camera semantics and dense lidar point clouds. The lidar point clouds are collected by fusing multiple lidars for robust feature detection. The camera semantics are based on a modified EfficientNet architecture which is trained with labeled data collected from onboard fisheye cameras. The point clouds are associated with the closest curb segment with L2-norm analysis after projecting into the image space with the fisheye model projection. Next, the selected points are clustered using unsupervised density-based spatial clustering to detect different curb regions. As new curb points are detected in consecutive frames they are associated with the existing curb clusters using temporal reachability constraints. If no reachability constraints are found a new curb cluster is formed from these new points. This ensures we can detect multiple curbs present in road segments consisting of multiple lanes if they are in the sensors' field of view. Finally, Delaunay filtering is applied for outlier removal and its performance is compared to traditional RANSAC-based filtering. An objective evaluation of the proposed solution is done using a high-definition map containing ground truth curb points obtained from a commercial map supplier. The proposed system has proven capable of detecting curbs of any orientation in complex urban road scenarios comprising straight roads, curved roads, and intersections with traffic isles. 

    Download full text (pdf)
    fulltext
  • Larsson, Axel
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Image-Guided Zero-Shot Object Detection in Video Games: Using Images as Prompts for Detection of Unseen 2D Icons2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Object detection deals with localization and classification of objects in images, where the task is to propose bounding boxes and predict their respective classes. Challenges in object detection include large-scale annotated datasets and re-training of models for specific tasks. Motivated by these problems, we propose a zero-shot object detection (ZSD) model in the setting of user interface icons in video games. Allowing to quickly and accurately analyze the state of a game, with potentially millions of people watching, would greatly benefit the large and fast-growing video game sector. Our resulting model is a modification of YOLOv8, which, at inference time, is prompted with the specific object to detect in an image. Many existing ZSD models exploit semantic embeddings and high-dimensional word vectors to generalize to novel classes. We hypothesize that using only visual representations is sufficient for the detection of unseen classes. To train and evaluate our model, we create synthetic data to reflect the nature of video game icons and in-game frames. Our method achieves similar performance as YOLOv8 on bounding box prediction and detection of seen classes while retaining the same average precision and recall for unseen classes, where the number of unseen classes is in the order of thousands.

    Download full text (pdf)
    fulltext
  • Hellberg, Ebba
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Exploring GPT models as biomedical knowledge bases: By evaluating prompt methods for extracting information from language models pre-trained on scientific articles2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Scientific findings recorded in literature help continuously guide scientific advancements, but manual approaches to accessing that knowledge are insufficient due to the sheer quantity of information and data available. Although pre-trained language models are being explored for their utility as knowledge bases and structured data repositories, there is a lack of research for this application in the biomedical domain. Therefore, the aim in this project was to determine how Generative Pre-trained Transformer models pre-trained on articles in the biomedical domain can be used to make relevant information more accessible. Several models (BioGPT, BioGPT-Large, and BioMedLM) were evaluated on the task of extracting chemical-protein relations between entities directly from the models through prompting. Prompts were formulated as a natural language text or an ordered triple, and provided in different settings (few-shot, one-shot, or zero-shot). Model-predictions were evaluated quantitatively as a multiclass classification task using a macro-averaged F1-score. The result showed that out of the explored methods, the best performance for extracting chemical-protein relations from article-abstracts was obtained using a triple-based text prompt on the largest model, BioMedLM, in the few-shot setting, albeit with low improvements from the baseline (+0.019 F1). There was no clear pattern for which prompt setting was favourable in terms of task performance, however, the triple based prompt was generally more robust than the natural language formulation. The task performance of the two smaller models underperformed the random baseline (by at best -0.026 and -0.001 F1). The impact of the prompt method was minimal in the smallest model, and the one-shot setting was the least sensitive to the prompt formulation in all models. However, there were more pronounced differences between the prompt methods in the few-shot setting of the larger models (+0.021-0.038 F1). The results suggested that the method of prompting and the size of the model impact the knowledge eliciting performance of a language model. Admittedly, the models mostly underperformed the baseline and future work needs to look into how to adapt generative language models to solve this task. Future research could investigate what impact automatic prompt-design methods and larger in-domain models have on the model performance.

    Download full text (pdf)
    fulltext
  • Feng, Jiayi
    KTH, School of Electrical Engineering and Computer Science (EECS).
    EMONAS: Evolutionary Multi-objective Neuron Architecture Search of Deep Neural Network2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Customized Deep Neural Network (DNN) accelerators have been increasingly popular in various applications, from autonomous driving and natural language processing to healthcare and finance, etc. However, deploying them directly on embedded system peripherals within real-time operating systems (RTOS) is not easy due to the paradox of the complexity of DNNs and the simplicity of embedded system devices. As a result, DNN implementation on embedded system devices requires customized accelerators with tailored hardware due to their numerous computations, latency, power consumption, etc. Moreover, the computational capacity, provided by potent microprocessors or graphics processing units (GPUs), is necessary to unleash the full potential of DNN, but these computational resources are often not easily available in embedded system devices. In this thesis, we propose an innovative method to evaluate and improve the efficiency of DNN implementation within the constraints of resourcelimited embedded system devices. The Evolutionary Multi-Objective Neuron Architecture Search-Binary One Optimization (EMONAS-BOO) optimizes both the image classification accuracy and the innovative Binary One Optimization (BOO) objectives, with Multiple Objective Optimization (MOO) methods. The EMONAS-BOO automates neural network searching and training, and the neural network architectures’ diversity is also guaranteed with the help of an evolutionary algorithm that consists of tournament selection, polynomial mutation, and point crossover mechanisms. Binary One Optimization (BOO) is used to evaluate the difficulty in implementing DNNs on resource-limited embedded system peripherals, employing a binary format for DNN weights. A deeper implementation of the innovative Binary One Optimization will significantly boost not only computation efficiency but also memory storage, power dissipation, etc. It is based on the reduction of weights binary 1’s that need to be computed and stored, where the reduction of binary 1 brings reduced arithmetic operations and thus simplified neural network structures. In addition, analyzed from a digital circuit waveform perspective, the embedded system, in interpreting the neural network, will register an increase in zero weights leading to a reduction in voltage transition frequency, which, in turn, benefits power efficiency improvement. The proposed EMONAS employs the MOO method which optimizes two objectives. The first objective is image classification accuracy, and the second objective is Binary One Optimization (BOO). This approach enables EMONAS to outperform manually constructed and randomly searched DNNs. Notably, 12 out of 100 distinct DNNs maintained their image classification accuracy. At the same time, they also exhibit superior BOO performance. Additionally, the proposed EMONAS ensures automated searching and training of DNNs. It achieved significant reductions in key performance metrics: Compared with random search, evolutionary-searched BOO was lowered by up to 85.1%, parameter size by 85.3%, and FLOPs by 83.3%. These improvements were accomplished without sacrificing the image classification accuracy, which saw an increase of 8.0%. These results demonstrate that the EMONAS is an excellent choice for optimizing innovative objects that did not exist before, and greater multi-objective optimization performance can be guaranteed simultaneously if computational resources are adequate.

    Download full text (pdf)
    fulltext
  • Guo, Xiaolinglong
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Comparative Analysis and Development of Receivers for Drone Remote Identification2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Similar to a car’s license plate, the European Union Aviation Safety Agency (EASA) has released new regulations imposing the registration of drone operators and the broadcasting of the drone’s “digital license plate”, i.e., the drone Remote Identification (RID) in flight, which gives a new opportunity in drone surveillance. Thus, it is meaningful to test the performance of Direct Remote Identification (DRI). To evaluate whether DRI can further improve the performance of the current Counter-Unmanned Aerial System (CUAS), it is essential to understand the performance of DRI, e.g., the effective region (maximal range). In this project, the field test of DRI reception with two broadcast protocols, Wireless Fidelity (Wi‑Fi) and Bluetooth, is carried out. As a result, with a 10dB high-gain receiving antenna and Mavic 3 as the transmitter in suburban areas, the maximal range can reach 1300 meters and still performs well in urban areas with a 700-meter maximal range when predicted with empirical propagation models. The field test results regard DRI signal as a helpful assistant in drone detection and surveillance. Therefore, a DRI receiver is developed in this project. All the basic functions like signal receiving and processing, Internet connection, and data transmission are successfully implemented. With further developments, the receiver could become a product that provides drone detection and tracking services.

    Download full text (pdf)
    fulltext
  • Hao, Kun
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Counterfactual and Causal Analysis for AI-based Modulation and Coding Scheme Selection2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Artificial Intelligence (AI) has emerged as a transformative force in wireless communications, driving innovation to address the complex challenges faced by communication systems. In this context, the optimization of limited radio resources plays a crucial role, and one important aspect is the Modulation and Coding Scheme (MCS) selection. AI solutions for MCS selection have been predominantly characterized as black-box models, which suffer from limited explainability and consequently hinder trust in these algorithms. Moreover, the majority of existing research primarily emphasizes enhancing explainability without concurrently improving the model’s performance which makes performance and explainability a trade-off. This work aims to address these issues by employing eXplainable AI (XAI), particularly counterfactual and causal analysis, to increase the explainability and trustworthiness of black-box models. We propose CounterFactual Retrain (CF-Retrain), the first method that utilizes counterfactual explanations to improve model performance and make the process of performance enhancement more explainable. Additionally, we conduct a causal analysis and compare the results with those obtained from an analysis based on the SHapley Additive exPlanations (SHAP) value feature importance. This comparison leads to the proposal of novel hypotheses and insights for model optimization in future research. Our results show that employing CF-Retrain can reduce the Mean Absolute Error (MAE) of the black-box model by 4% while utilizing only 14% of the training data. Moreover, increasing the amount of training data yields even more pronounced improvements in MAE, providing a certain level of explainability. This performance enhancement is comparable to or even superior to using a more complex model. Furthermore, by introducing causal analysis to the mainstream SHAP value feature importance, we provide a novel hypothesis and explanation of feature importance based on causal analysis. This approach can serve as an evaluation criterion for assessing the model’s performance.

    Download full text (pdf)
    fulltext
  • Qian, Qiran
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Register Caching for Energy Efficient GPGPU Tensor Core Computing2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The General-Purpose GPU (GPGPU) has emerged as the predominant computing device for extensive parallel workloads in the fields of Artificial Intelligence (AI) and Scientific Computing, primarily owing to its adoption of the Single Instruction Multiple Thread architecture, which not only provides a wealth of thread context but also effectively hide the latencies exposed in the single threads executions. As computational demands have evolved, modern GPGPUs have incorporated specialized matrix engines, e.g., NVIDIA’s Tensor Core (TC), in order to deliver substantially higher throughput for dense matrix computations compared with traditional scalar or vector architectures. Beyond mere throughput, energy efficiency is a pivotal concern in GPGPU computing. The register file is the largest memory structure on the GPGPU die and typically accounts for over 20% of the dynamic power consumption. To enhance energy efficiency, GPGPUs incorporate a technique named register caching borrowed from the realm of CPUs. Register caching captures temporal locality among register operands to reduce energy consumption within a 2- level register file structure. The presence of TC raises new challenges for Register Cache (RC) design, as each matrix instruction applies intensive operand delivering traffic on the register file banks. In this study, we delve into the RC design trade-offs in GPGPUs. We undertake a comprehensive exploration of the design space, encompassing a range of workloads. Our experiments not only reveal the basic design considerations of RC but also clarify that conventional caching strategies underperform, particularly when dealing with TC computations, primarily due to poor temporal locality and the substantial register operand traffic involved. Based on these findings, we propose an enhanced caching strategy featuring a look-ahead allocation policy to minimize unnecessary cache allocations for the destination register operands. Furthermore, to leverage the energy efficiency of Tensor Core computing, we highlight an alternative instruction scheduling framework for Tensor Core instructions that collaborates with a specialized caching policy, resulting in a remarkable reduction of up to 50% in dynamic energy consumption within the register file during Tensor Core GEMM computations.

    Download full text (pdf)
    fulltext
  • Shen, Xuying
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Algorithmic Multi-Ported Memories Enabled Power-Efficient Pre-Distorter Design in ASIC2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The transition from the 5G to the 6G era is a pivotal juncture in contemporary wireless communication. Under such a circumstance, Digital Pre-Distortion (DPD) technology has established its significance as an effective method to linearize Power Amplifiers. However, DPD is facing a series of challenges, notably the increased bandwidth which necessitates more complex modeling techniques. This thesis focuses on the fact that the DPD requires multi-ported memories for the Look-Up-Tables to store correction coefficients, where two research questions are identified. Firstly, this thesis analyses the power, area, and delay-performance trade-offs with an increase in the number of read and write ports of Flip-Flop (FF)-based memories. Secondly, this thesis evaluates and compares the performance of the conventional FF-based multi-ported memories and algorithmic FF-based multi-ported memories. As a Master’s thesis project, this research utilizes the knowledge and practice skills expected of a Master’s student specializing in Embedded Systems. In this thesis, conventional and algorithmic multi-ported memories are implemented and evaluated after studying related works. Subsequently, an industrial Application-Specific Integrated Circuit (ASIC) design flow is executed, undergoing iterative refinements. And in the end, the conclusions are drawn based on an analysis of the software reports. The results underscore that area and power consumption exhibit linear growth alongside increased port numbers within conventional multi-ported memories. Also, the algorithmic multi-ported memory presents a promising alternative, engendering improvements across all three dimensions of delay, area, and power consumption. The implemented memories can be integrated into DPD forward path with customized port numbers in the future, offering adaptability in terms of port configuration and better performance in terms of timing, area and power. Additionally, these implemented memories stand as a valuable point of reference for engineers engaged in the development of FF-based multi-ported memories within the context of ASIC.

    Download full text (pdf)
    fulltext
  • Ihling, Nina
    et al.
    AVT – Biochemical Engineering RWTH Aachen University Aachen Germany.
    Berg, Christoph
    AVT – Biochemical Engineering RWTH Aachen University Aachen Germany.
    Paul, Richard
    AVT – Biochemical Engineering RWTH Aachen University Aachen Germany.
    Munkler, Lara Pauline
    AVT – Biochemical Engineering RWTH Aachen University Aachen Germany.
    Mäkinen, Meeri
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Industrial Biotechnology.
    Chotteau, Veronique
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Industrial Biotechnology. KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Centres, Centre for Advanced BioProduction by Continuous Processing, AdBIOPRO.
    Büchs, Jochen
    AVT – Biochemical Engineering RWTH Aachen University Aachen Germany.
    Scale‐down of CHO cell cultivation from shake flasks based on oxygen mass transfer allows application of parallelized, non‐invasive, and time‐resolved monitoring of the oxygen transfer rate in 48‐well microtiter plates2023In: Biotechnology Journal, ISSN 1860-6768, E-ISSN 1860-7314, Vol. 18, no 11, article id 2300053Article in journal (Refereed)
    Abstract [en]

    Cultivating Chinese hamster ovary (CHO) cells in microtiter plates (MTPs) with time-resolved monitoring of the oxygen transfer rate (OTR) is highly desirable to provide process insights at increased throughput. However, monitoring of the OTR in MTPs has not been demonstrated for CHO cells, yet. Hence, a CHO cultivation process was transferred from shake flasks to MTPs to enable monitoring of the OTR in each individual well of a 48-well MTP. For this, the cultivation of an industrially relevant, antibody-producing cell line was transferred from shake flask to MTP based on the volumetric oxygen mass transfer coefficient (kLa). Culture behavior was well comparable (deviation of the final IgG titer less than 10%). Monitoring of the OTR in 48-well MTPs was then used to derive the cytotoxicity of dimethyl sulfoxide (DMSO) based on a dose–response curve in a single experiment using a second CHO cell line. Logistic fitting of the dose–response curve determined after 100 h was used to determine the DMSO concentration that resulted in a cytotoxicity of 50% (IC50). A DMSO concentration of 2.70% ± 0.25% was determined, which agrees with the IC50 previously determined in shake flasks (2.39% ± 0.1%). Non-invasive, parallelized, and time-resolved monitoring of the OTR of CHO cells in MTPs was demonstrated and offers excellent potential to speed up process development and assess cytotoxicity.

    Download full text (pdf)
    fulltext
  • Rauch, Reinhard
    et al.
    Engler-Bunte-Insitut, Karlsruhe Institute of Technology (KIT), Engler-Bunte-Ring 1, 76131 Karlsruhe, Germany.
    Kiros, Yohannes
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Chemical Engineering, Process Technology.
    Engvall, Klas
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Chemical Engineering, Process Technology.
    Kantarelis, Efthymios
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Chemical Engineering, Process Technology.
    Brito, Paulo
    VALORIZA—Research Center for Endogenous Resource Valorization, Campus Politécnico, 10, 7300-555 Portalegre, Portugal.
    Nobre, Catarina
    VALORIZA—Research Center for Endogenous Resource Valorization, Campus Politécnico, 10, 7300-555 Portalegre, Portugal.
    Santos, Santa Margarida
    VALORIZA—Research Center for Endogenous Resource Valorization, Campus Politécnico, 10, 7300-555 Portalegre, Portugal.
    Graefe, Philipp A.
    Engler-Bunte-Insitut, Karlsruhe Institute of Technology (KIT), Engler-Bunte-Ring 1, 76131 Karlsruhe, Germany.
    Hydrogen from Waste Gasification2024In: Hydrogen, E-ISSN 2673-4141, Vol. 5, no 1, p. 70-101Article in journal (Refereed)
    Abstract [en]

    Hydrogen is a versatile energy vector for a plethora of applications; nevertheless, itsproduction from waste/residues is often overlooked. Gasification and subsequent conversion ofthe raw synthesis gas to hydrogen are an attractive alternative to produce renewable hydrogen. Inthis paper, recent developments in R&D on waste gasification (municipal solid waste, tires, plasticwaste) are summarised, and an overview about suitable gasification processes is given. A literaturesurvey indicated that a broad span of hydrogen relates to productivity depending on the feedstock,ranging from 15 to 300 g H2/kg of feedstock. Suitable gas treatment (upgrading and separation) isalso covered, presenting both direct and indirect (chemical looping) concepts. Hydrogen productionvia gasification offers a high productivity potential. However, regulations, like frame conditions orsubsidies, are necessary to bring the technology into the market.

  • Hu, Yu
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Using a Deep Generative Model to Generate and Manipulate 3D Object Representation2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The increasing importance of 3D data in various domains, such as computer vision, robotics, medical analysis, augmented reality, and virtual reality, has gained giant research interest in generating 3D data using deep generative models. The challenging problem is how to build generative models to synthesize diverse and realistic 3D objects representations, while having controllability for manipulating the shape attributes of 3D objects. This thesis explores the use of 3D Generative Adversarial Networks (GANs) for generation of 3D indoor objects shapes represented by point clouds, with a focus on shape editing tasks. Leveraging insights from 2D semantic face editing, the thesis proposes extending the InterFaceGAN framework to 3D GAN model for discovering the relationship between latent codes and semantic attributes of generated shapes. In the end, we successfully perform controllable shape editing by manipulating the latent code of GAN.

    Download full text (pdf)
    fulltext
  • Valentin Maza, Axel
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Analysis of speaking time and content of the various debates of the presidential campaign: Automated AI analysis of speech time and content of presidential debates based on the audio using speaker detection and topic detection2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The field of artificial intelligence (AI) has grown rapidly in recent years and its applications are becoming more widespread in various fields, including politics. In particular, presidential debates have become a crucial aspect of election campaigns and it is important to analyze the information exchanged in these debates in an objective way to let voters choose without being influenced by biased data. The objective of this project was to create an automatic analysis tool for presidential debates using AI. The main challenge of the final system was to determine the speaking time of each candidate and to analyze what each candidate said, to detect the topics discussed and to calculate the time spent on each topic. This thesis focus mainly on the speaker detection part of this system. In addition, the high overlap rate in the debates, where candidates cut each other off, posed a significant challenge for speaker diarization, which aims to determine who speaks when. This problem was considered appropriate for a Master’s thesis project, as it involves a combination of advanced techniques in AI and speech processing, making it an important and difficult task. The application to political debates and the accompanying overlapping pathways makes this task both challenging and innovative. There are several ways to solve the problem of speaker detection. We have implemented classical approaches that involve segmentation techniques, speaker representation using embeddings such as i-vectors or x-vectors, and clustering. Yet, due to speech overlaps, the End-to-end solution was implemented using pyannote-audio (an open-source toolkit written in Python for speaker diarization) and the diarization error rate was significantly reduced after refining the model using our own labeled data. The results of this project showed that it was possible to create an automated presidential debate analysis tool using AI. Specifically, this thesis has established a state of the art of speaker detection taking into account the particularities of the politics such as the high speaker overlap rate.

    Download full text (pdf)
    fulltext
  • Bhatnagar, Kunal
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Modelling synaptic rewiring in brain-like neural networks for representation learning2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    This research investigated the concept of a sparsity method inspired by the principles of structural plasticity in the brain in order to create a sparse model of the Bayesian Confidence Propagation Neural Networks (BCPNN) during the training phase. This was done by extending the structural plasticity in the implementation of the BCPNN. While the initial algorithm presented two synaptic states (Active and Silent), this research extended it to three synaptic states (Active, Silent and Absent) with the aim to enhance sparsity configurability and emulate a more brain-like algorithm, drawing parallels with synaptic states observed in the brain. Benchmarking was conducted using the MNIST and Fashion-MNIST dataset, where the proposed threestate model was compared against the previous two-state model in terms of representational learning. The findings suggest that the three-state model not only provides added configurability but also, in certain low-sparsity settings, showcases similar representational learning abilities as the two-state model. Moreover, in high-sparsity settings, the three-state model demonstrates a commendable balance between accuracy and sparsity trade-off.

    Download full text (pdf)
    fulltext
  • Kolanowski, Mikael
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Enhancing an Existing Attack Projection System with Deep Learning2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    As organizations and critical infrastructure increasingly rely on computer networks for their function, cyber defense becomes more and more important. A recent trend is to employ predictive methods in cybersecurity. Attack projection attempts to predict the next step in an ongoing attack. Previous research has attempted to solve attack projection using deep learning relying solely on LSTM networks. In this work, by contrast, we solved the attack projection problem using three different neural network architectures: an LSTM, a Transformer, and a hybrid LSTM­Transformer model. We then proposed a way to integrate our neural models into an existing software framework that relies on sequential rule mining to predict future security alerts. The models were trained and evaluated on a publicly available dataset of network security alerts and evaluated with respect to precision and recall of alert predictions. We found that the Transformer architecture had the best overall performance in all but one experiment and that the LSTM architecture performed the worst across all experiments.

    Download full text (pdf)
    fulltext
  • Ekman von Huth, Simon
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Multi-Scale Task Dynamics in Transfer and Multi-Task Learning: Towards Efficient Perception for Autonomous Driving2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Autonomous driving technology has the potential to revolutionize the way we think about transportation and its impact on society. Perceiving the environment is a key aspect of autonomous driving, which involves multiple computer vision tasks. Multi-scale deep learning has dramatically improved the performance on many computer vision tasks, but its practical use in autonomous driving is limited by the available resources in embedded systems. Multi-task learning offers a solution to this problem by allowing more compact deep learning models that share parameters between tasks. However, not all tasks benefit from being learned together. One way of avoiding task interference during training is to learn tasks in sequence, with each task providing useful information for the next – a scheme which builds on transfer learning. Multi-task and transfer dynamics are both concerned with the relationships between tasks, but have previously only been studied separately. This Master’s thesis investigates how different computer vision tasks relate to each other in the context of multi-task and transfer learning, using a state-ofthe-art efficient multi-scale deep learning model. Through an experimental research methodology, the performance on semantic segmentation, depth estimation, and object detection were evaluated on the Virtual KITTI 2 dataset in a multi-task and transfer learning setting. In addition, transfer learning with a frozen encoder was compared to constrained encoder fine tuning, to uncover the effects of fine-tuning on task dynamics. The results suggest that findings from previous work regarding semantic segmentation and depth estimation in multi-task learning generalize to multi-scale learning on autonomous driving data. Further, no statistically significant correlation was found between multitask learning dynamics and transfer learning dynamics. An analysis of the results from transfer learning indicate that some tasks might be more sensitive to fine-tuning than others, suggesting that transferring with a frozen encoder only captures a subset of the complexities involved in transfer relationships. Regarding object detection, it is observed to negatively impact the performance on other tasks during multi-task learning, but might be a valuable task to transfer from due to lower annotation costs. Possible avenues for future work include applying the used methodology to real-world datasets and exploring ways of utilizing the presented findings for more efficient perception algorithms.

    Download full text (pdf)
    fulltext
  • Popova, Victoria
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Co-creating Futures for Integrating Generative AI into the Designers’ Workflow2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    In recent years Generative AI tools have become increasingly ubiquitous and have given rise to much discussion concerning their impact on jobs, both in personal use and corporate settings. Despite Generative AI being a rapidly growing field, there is currently an existing research gap regarding the adoption of these tools across different domains. This study aims to fill this gap by contributing with knowledge on how designers might integrate Generative AI into their workflows. By adopting a Research through design (RtD) approach, three workshops were held where designers used generative AI tools to co-create design fictions envisioning how AI tools would permeate their future workflows. Thematic analysis of the workshop data revealed both desirable and undesirable futures from the designers’ perspectives situating AI at various stages of design – from assisting designers with mundane tasks to helping with ideation and testing. The futures brought up reflections on designers in control of the workflow, the dynamics of human-AI collaboration and the evolving role of the designer. The study contributes knowledge about different forms human-AI interactions could take in the near future, and highlights the importance of careful consideration when deploying these tools in a human-centric manner.

    Download full text (pdf)
    fulltext
  • Wagner, Jannik
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Improving Behavior Trees that Use Reinforcement Learning with Control Barrier Functions: Modular, Learned, and Converging Control through Constraining a Learning Agent to Uphold Previously Achieved Sub Goals2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    This thesis investigates combining learning action nodes in behavior trees with control barrier functions based on the extended active constraint conditions of the nodes and whether the approach improves the performance, in terms of training time and policy quality, compared to a purely learning-based approach. Behavior trees combine several behaviors, called action nodes, into one behavior by switching between them based on the current state. Those behaviors can be hand-coded or learned in so-called learning action nodes. In these nodes, the behavior is a reinforcement learning agent. Behavior trees can be constructed in a process called backward chaining. In order to ensure the success of a backward-chained behavior tree, each action node must uphold previously achieved subgoals. So-called extended active constraint conditions formalize this notion as conditions that must stay true for the action node to continue execution. In order to incentivize upholding extended active constraint conditions in learning action nodes, a negative reward can be given to the agent upon violating extended active constraint conditions. However, this approach does not guarantee not violating the extended active constraint conditions since it is purely learning-based. Control barrier functions can be used to restrict the actions available to an agent so that it stays within a safe subset of the state space. By defining the safe subset of the state space as the set in which the extended active constraint conditions are satisfied, control barrier functions can be employed to, ideally, guarantee that the extended active constraint conditions will not be violated. The results show that significantly less training is needed to get comparable, or slightly better, results, when compared to not using control barrier functions. Furthermore, extended active constraint conditions are considerably less frequently violated and the overall performance is slightly improved.

    Download full text (pdf)
    fulltext
  • Basu, Sampriti
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Phase Noise Performance of a PLL Frequency Synthesizer when Powered by Silent Switchers2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    In use today are ‘normal’ DC-DC switching regulators with considerable switching noise and ringing, which is bad for noise-sensitive applications. This project involves a solution based on ‘Silent Switchers’ to prove its effectiveness in reducing noise. This idea is then coupled into identifying the susceptibility of a PLL synthesizer to ensure we understand the sensitivity of this type of component and if this can be used with a silent switcher. A particular PLL synthesizer evaluation board is currently powered by linear regulators, which is used as a basis for comparing results. A new board involving silent switchers is designed and manufactured. Phase noise measurements are done to evaluate if silent switchers are a suitable alternative for power supply.

    Download full text (pdf)
    fulltext
  • Skarf, Frida
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Uncertainty Estimation in Radiation Dose Prediction U-Net2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The ability to quantify uncertainties associated with neural network predictions is crucial when they are relied upon in decision-making processes, especially in safety-critical applications like radiation therapy. In this paper, a single-model estimator of both epistemic and aleatoric uncertainties in a regression 3D U-net used for radiation dose prediction is presented. To capture epistemic uncertainty, Monte Carlo Dropout is employed, leveraging dropout during test-time inference to obtain a distribution of predictions. The variability among these predictions is used to estimate the model’s epistemic uncertainty. For quantifying aleatoric uncertainty quantile regression, which models conditional quantiles of the output distribution, is used. The method enables the estimation of prediction intervals of a user-specified significance level, where the difference between the upper and lower bound of the interval quantifies the aleatoric uncertainty. The proposed approach is evaluated on two datasets of prostate and breast cancer patient geometries and corresponding radiation doses. Results demonstrate that the quantile regression method provides well-calibrated prediction intervals, allowing for reliable aleatoric uncertainty estimation. Furthermore, the epistemic uncertainty obtained through Monte Carlo Dropout proves effective in identifying out-of-distribution examples, highlighting its usefulness for detecting anomalous cases where the model makes uncertain predictions.

    Download full text (pdf)
    fulltext
  • Closa, Oriol
    KTH, School of Electrical Engineering and Computer Science (EECS).
    LSTM-attack on polyalphabetic cyphers with known plaintext: Case study on the Hagelin C-38 and Siemens and Halske T522023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Polyalphabetic cyphers have been used for centuries and well into the 1970s to transmit all kinds of messages. Since then, computers and modern cryptography have taken over making bruteforce attacks unfeasible when designed properly. However, there was a time where mechanical machines, built to operate in harsh conditions and sometimes even without power, were the state of the art for keeping secrets secret. During World War II both Axis and Allied powers used different machines to ensure their dominance. To communicate with occupied territories, Germany installed the Siemens & Halske T52, also known as Geheimschreiber (the secret teleprinter), on telex lines running through Sweden. On the allied side, the United States of America adopted the Hagelin C-38 (known as the M-209) of Swedish invention for field operation. While both machines are inherently different as to what tasks they perform and how, they can both be considered complex polyalphabetic cyphers. Many different methods and attacks on these kind of cyphers have been developed, some relying on inner knowledge and some on mechanical devices. However, the application of Machine Learning to extract key information from intercepts is not a well researched area yet. This thesis aims to demonstrate the potential of LSTM networks on known plaintext attacks against different classical as well as stream cyphers. The techniques used have proven to be effective on the Vigenère as well as with the Hagelin C-38 while being partially successful on the Geheimschreiber with crib lengths of only 15 characters.

    Download full text (pdf)
    fulltext
  • Lassarat, Côme
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Self-learning for 3D segmentation of medical images from single and few-slice annotation2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Training deep-learning networks to segment a particular region of interest (ROI) in 3D medical acquisitions (also called volumes) usually requires annotating a lot of data upstream because of the predominant fully supervised nature of the existing stateof-the-art models. To alleviate this annotation burden for medical experts and the associated cost, leveraging self-learning models, whose strength lies in their ability to be trained with unlabeled data, is a natural and straightforward approach. This work thus investigates a self-supervised model (called “self-learning” in this study) to segment the liver as a whole in medical acquisitions, which is very valuable for doctors as it provides insights for improved patient care. The self-learning pipeline utilizes only a single-slice (or a few-slice) groundtruth annotation to propagate the annotation iteratively in 3D and predict the complete segmentation mask for the entire volume. The segmentation accuracy of the tested models is evaluated using the Dice score, a metric commonly employed for this task. Conducting this study on Computed Tomography (CT) acquisitions to annotate the liver, the initial implementation of the self-learning framework achieved a segmentation accuracy of 0.86 Dice score. Improvements were explored to address the drifting of the mask propagation, which eventually proved to be of limited benefits. The proposed method was then compared to the fully supervised nnU-Net baseline, the state-of-the-art deep-learning model for medical image segmentation, using fully 3D ground-truth (Dice score ∼ 0.96). The final framework was assessed as an annotation tool. This was done by evaluating the segmentation accuracy of the state-of-the-art nnU-Net trained with annotation predicted by the self-learning pipeline for a given expert annotation budget. While the self-learning framework did not generate accurate enough annotation from a single slice annotation yielding an average Dice score of ∼ 0.85, it demonstrated encouraging results when two ground-truth slice annotations per volume were provided for the same annotation budget (Dice score of ∼ 0.90).

    Download full text (pdf)
    fulltext
  • Allard, Nathan
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Introducing GA-SSNN: A Method for Optimizing Stochastic Spiking Neural Networks: Scaling the Edge User Allocation Constraint Satisfaction Problem with Enhanced Energy and Time Efficiency2023Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    As progress within Von Neumann-based computer architecture is being limited by the physical limits of transistor size, neuromorphic comuting has emerged as a promising area of research. Neuromorphic hardware tends to be substantially more power efficient by imitating the aspects of computations in networks of neurons in the brain. It features massive parallelism, colocation of processing and memory at the neurons and synapses, inherent scalability, temporally sparse event-driven computation, and stochasticity. This thesis explores the application of neuromorphic comuting, specifically Stochastic Spiking Neural Networks (SSNNs), to large-scale edge user allocation constraint satisfaction problems (CSPs). These problems are central in the era of 5G networks, augmented reality and computational offloading, yet existing solutions struggle with scalability and stability. The thesis introduces the Genetic Algorithm for Stochastic Spiking Neural Networks (GA-SSNN), an algorithm designed to optimize complex and stochastic objective functions. The GA-SSNN algorithm leverages adaptive mutation, simulation time management, constraint approximation, and specialized tournament selection to efficiently traverse the search space and achieves better performance than the current state of the art (NSGA-II). Furthermore, the thesis elaborates on designing an SSNN structure to efficiently solve a complex CSP. The outcome of this thesis represents a significant step towards applying neuromorphic computing to real-world scenarios, with the potential to greatly enhance solution speed and energy efficiency.

    Download full text (pdf)
    fulltext
  • Engholm, Albin
    et al.
    KTH, School of Industrial Engineering and Management (ITM), Centres, Integrated Transport Research Lab, ITRL.
    Kristoffersson, Ida
    VTI Statens väg- och transportforskningsinstitut.
    Frölander, Simon (Contributor)
    KTH, School of Industrial Engineering and Management (ITM), Centres, Integrated Transport Research Lab, ITRL.
    Brunner, Sabrina (Contributor)
    VTI Statens väg- och transportforskningsinstitut.
    MUST Managing Deep Uncertainty in Planning for Sustainable Transport: Project report: phase 12024Report (Other academic)
    Abstract [en]

    There is a growing recognition that traditional forecasting and decision-making approaches might fall short considering the many uncertainties and complexities facing the development of the transport system. The project Managing deep Uncertainty in planning for Sustainable Transport (MUST), funded by Trafikverket and conducted by KTH ITRL and VTI, aims to explore emerging methods for improving the handling of deep uncertainty in the long-term planning of future transport systems. The core of MUST is to explore, develop, and demonstrate tools and methods grounded in Decision Making under Deep Uncertainty (DMDU) and Exploratory Modeling and Analysis (EMA). These approaches are intended to support a shift towards more robust and adaptable planning methodologies.

    The project is performed in two phases, with the first phase dedicated to laying a foundational understanding of deep uncertainty in transport planning. This report covers the first phase which has included the following tasks: 

    • A literature review on deep uncertainty and existing decision-making and system analysis methods under such conditions, with a focus on transportation. 
    • A workshop series with Trafikverket identifying transport planning challenges marked by deep uncertainty.
    • A case study of applying DMDU through a case study on climate policy robustness (primarily reported in other deliverables).

    The literature review covers how the nature of uncertainty in socio-technical systems can be understood, classified, and analyzed. For policy analysis and decision making, the literature underscores the importance of considering multiple futures in model-based analysis when faced with deep uncertainties. DMDU and EMA methods are reviewed and summarized, and their application to transport are discussed. The literature also summarizes studies on uncertainty in model-based transport planning and policy analysis and concludes that the primary location of deep uncertainty is in the model inputs in the form of “scenario uncertainty”. In the workshop series, uncertainty related to producing the base forecast (Swe: basprognos) and policy analysis for domestic transport climate policy was analyzed. This analysis suggested that scenario uncertainty is a main source of deep uncertainty, but also uncertainty related to the system boundaries where highlighted. Furthermore, potential benefits and drawbacks of EMA and DMDU were discussed. In the case study, it is explored how the Scenario tool can be further leveraged by DMDU. More specifically, MORDM (see Section 2.2.3) is applied to assess to what extent it may allow a broader set of policy options to be explored, and how it can provide a better understanding of the robustness and vulnerabilities of different types of policies. 

    A key takeaway from MUST phase 1 is that DMDU and EMA could provide several potential benefits and that methods and tools for applying them are maturing. However, it is possibly a long way to go before DMDU and EMA can be integrated as a regularly used method during the planning process. This is due to organization and process-related issues, as well as technical issues on how to effectively apply DMDU and EMA to Trafikverket’s national transport models. These technical issues will partly be explored in MUST phase 2. 

    Download full text (pdf)
    fulltext
  • Ambhore, Dhairysheel Shivaji
    KTH, School of Industrial Engineering and Management (ITM), Energy Technology, Energy Systems.
    Navigating Industry 4.0 to Industry 5.0: Challenges and Strategies for Workforce Transition and its Relation to SDGs2024Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The Fourth Industrial Revolution, known as Industry 4.0, has ushered in a new era of technological advancement and disruption across Manufacturing Industries. As organizations embrace digital technologies, artificial intelligence, and automation, the workforce faces profound changes in job roles, skill requirements, and training needs. Several challenges are faced by industries during this Industrial transformation. This research begins by focusing on Industry 4.0 and Industry 5.0, understanding there evolving concept. Through comprehensive literature review, the challenges and practices faced during this Industrial transformation and SDGs are discussed. Following the literature review, survey and interview questions were drafted trying to dug deeper into aspects that the literature could not fully capture. Furthermore, the relation between this Industrial Transformation and SDGs were established. This research contributes to a deeper understanding of the dynamic relationship between technology, the workforce, and sustainable development. This thesis report serves as a valuable resource for policymakers, business leaders, educators, and researchers seeking to navigate the transformative landscape of Industrial Transformation and its implications for a rapidly evolving workforce and sustainabledevelopment.

    Download full text (pdf)
    fulltext
  • Zawahri, Aya
    et al.
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Biomedical Engineering and Health Systems, Health Informatics and Logistics.
    Ibrahim, Nanci
    KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Biomedical Engineering and Health Systems, Health Informatics and Logistics.
    Dynamik och tillförlighet i finansiell prognostisering: En analys av djupinlärningsmodeller och deras reaktion på marknadsmanipulation2024Independent thesis Basic level (university diploma), 10 credits / 15 HE creditsStudent thesis
    Abstract [en]

    Over the years, intensive research has been conducted to enhance the capability of machine learning models to predict market movements. Despite this, during financial history, several events, such as the "Flash-crash," have impacted the market and had dramatic consequences for price movements. Therefore, it is crucial to examine how the models are affected by manipulative actions in the financial market to ensure their robustness and reliability in such situations.

    To carry out this work, the process has been divided into three steps. Firstly, a review of previous studies was conducted to identify the most robust models in the field. This was achieved by training the models on the FI-2010 dataset, which is a publicly available dataset for high-frequency trading of stocks on the NASDAQ Nordic stock exchange. The examined models included DeepLOB, DeepLOB-Attention, DeepLOB-seq2seq, DTNN, and TCN. The second step involved acquiring the Swedish dataset from Nasdaq Nordic, providing data on Swedish stock Limit Order Books (LOB). The two models that demonstrated the best results in the first step were then trained with this dataset. Finally, a manipulation was performed on the Swedish order books to investigate how these models would be affected.

    The result constituted a clear assessment of the models' robustness and reliability in predicting market movements through a comprehensive comparison and analysis of all tests and their results. The work also highlights how the models' outcomes are affected by manipulative actions. Furthermore, it becomes evident how the choice of normalization method affects the models' results.

    Download full text (pdf)
    Examensarbete