https://www.mdu.se/

Imagination Studios is a company specialized in motion capturing and animation. Part of their daily business is working at trade shows where they have a booth to keep close contact with existing customers and also to find new ones. However, usually only two to three people will be working at the booth, and frequently, these people will be in meetings with potential customers. During a time like this, nobody is free to attend to other people checking out the booth. This can result in a potential loss of a new customer. This project seeks a way to alleviate that problem.The idea behind this project was to create an application that trade show visitors can interact with in a playful and innovative way while also giving them a feel of what Imagination Studios is all about while looking for information about the company. To do this it was decided to let users interact with the system by using a Microsoft Kinect. The Kinect allows for easy implementation of a user interface based on motion capturing while also being very cost effective. A new user interface was to be designed as well, without copying already existing solutions and without simply expanding a traditional UI with new elements. To achieve this several design sketches were made, and the most interesting ones were then turned into storyboards. These were then used to decide on the final design, which was then elaborated on by use of video sketches and a collage in Adobe Photoshop.Several tools were used during the actual implementation. For the actual visualization and graphical design, the Unreal Engine 3 in combination with UDK was decided upon. To connect Kinect and Unreal Engine 3, a third party addon called NIUI which makes use of the open source SDK OpenNI was used. For ease of debugging and programming in Unrealscript, the programming language used by the Unreal Engine 3, an addon for Microsoft Visual Studio 2010 called nFringe (Pixel Mine, Inc., 2010) was used.

The edge-cloud computing continuum effectively uses fog and cloud servers to meet the quality of service (QoS) requirements of tasks when edge devices cannot meet those requirements. This paper focuses on the workflow offloading problem in edge-cloud computing and formulates this problem as a nonlinear mathematical programming model. The objective function is to minimize the monetary cost of executing a workflow while satisfying constraints related to data dependency among tasks and QoS requirements, including security and deadlines. Additionally, it presents a genetic algorithm for the workflow offloading problem to find near-optimal solutions with the cost minimization objective. The performance of the proposed mathematical model and genetic algorithm is evaluated on several real-world workflows. Experimental results demonstrate that the proposed genetic algorithm can find admissible solutions comparable to the mathematical model and outperforms particle swarm optimization, bee life algorithm, and a hybrid heuristic-genetic algorithm in terms of workflow execution costs.

A hybrid cloud is an efficient solution to deal with the problem of insufficient resources of a private cloud when computing demands increase beyond its resource capacities. Cost-efficient workflow scheduling, considering security requirements and data dependency among tasks, is a prominent issue in the hybrid cloud. To address this problem, we propose a mathematical model that minimizes the monetary cost of executing a workflow and satisfies the security requirements of tasks under a deadline. The proposed model fulfills data dependency among tasks, and data transmission time is formulated with exact mathematical expressions. The derived model is a Mixed-integer linear programming problem. We evaluate the proposed model with real-world workflows over changes in the input variables of the model, such as the deadline and security requirements. This paper also presents a post-optimality analysis that investigates the stability of the assignment problem. The experimental results show that the proposed model minimizes the cost by decreasing inter-cloud communications for dependent tasks. However, the optimal solutions are affected by the limitations that are imposed by the problem constraints. 

Recent advancements of multiprocessor architectures have led to increasing use of multiprocessors in real-time embedded systems. The two most popular real-time scheduling approaches in multiprocessors are global and partitioned scheduling. Cluster based multiprocessor scheduling can be seen as a hybrid approach combining benefits of both partitioned and global scheduling. Virtual clustering further enhances it by providing dynamic cluster resource allocation duringrun-time and applying hierarchical scheduling to ensure temporal isolation between different software components. Over the years, the study of virtual clustered-based multiprocessor scheduling has been limited to theoretical analysis. In this thesis, we implemented a Virtual-Clustered Hierarchical Scheduling Framework (VC-HSF) in Linux without modifying the base Linux kernel. This work includes complete design, implementation and experimentation of this framework in a multiprocessor platform. Our main contributions are twofold: (i) to the best of our knowledge, our work is the first implementation of any virtual-clustered real-time multiprocessor scheduling in an operating system, (ii) our design and implementation gives practical insights about challenges of implementing any virtual-clustered algorithms for real-time scheduling.

Most of the hardware products today, especially those that people interact with, are controlled by software. There are many devices with inbuilt software which many people do not bother to notice. Software may be critical in terms of strict quality requirements due to failures which enable risks of endangering the production and more importantly the lives of people. The testing team of Westermo Network Technologies AB faced a challenge with not being able to identify the coverage of tested cases in software. Identifying the coverage of total executed test cases enables the possibility of knowing the potential future quality of a software. By reaching such a stage a software will likely not suffer from failures due to higher quality, therefore the lives of people and the production at stake will not get harmed. The essential problem is that test cases are both tested and skipped, which makes it more difficult for the company to identify what has been actually tested and skipped. The purpose of this thesis is to identify and calculate the coverage of test cases, the process was mainly to understand the essence in the identification of executed test cases. The ethical aspect of putting people's lives at stake is what inspired us to investigate software failures. Software is typically tested before a release, therefore our investigation was to research testing process of software. When investigating test result over many days and test systems we developed a coverage calculator system which helps Westermo to decide and determine the release of tested software, either the software test result is acceptable for a release or not.

The massive increase in 5G usage for video streaming can cause short but intense spikes in data transfer rates, potentially causing network congestion. Congestion detection and prediction are essential for ensuring user satisfaction and enhancing the performance of real-time applications. By leveraging historical network measurement data such as Round Trip Time (RTT), Jitter, and packet loss ratio, machine learning algorithms can analyze traffic patterns, predict congestion based on the attributes, and provide adaptive solutions. We have focused primarily on ML models due to their adaptability to learn from historical patterns, a capability where traditional rule-based methods may fall short. Furthermore, past researchers faced difficulties in obtaining measurement data for fully comprehensive real-world networks. We have collected real-world network measurement datasets over a six-month period for a complete network. Using the datasets and topology information, we trained Support Vector Regression (SVR), Long Short-TermMemory (LSTM), and XGBRegressor (Extreme Gradient Boosting) to forecast the RTT values of each router. To predict network anomalies based on packet loss and high RTT values, Random Forest and Support Vector Classifier (SVC) models have been developed. By analyzing the outputs of these two models, a method is proposed to predict network congestion. While this study is limited by data quantity and the use of simpler ML models, it offers a comprehensive solution and a solid foundation for predicting network congestion symptoms. Network congestion prediction can be further improved by exploring areas such as model ensembling, explainability in deep learning models, transfer learning approaches, and integrating multi-source data.

Tests are important artifacts in the software development process. Testing activities such as test automation, test maintenance, and test suite optimization mainly rely on an in-depth understanding of test specifications. The manual process of writing testspecifications in natural language can create many different quality issues such as ambiguous, incomplete, redundant, or inconsistent test cases. Nowadays, the concept of test smells is proposed by several researchers to be used as indicators of low-qualityattributes in test specifications. Quality assurance processes for test specifications often rely on manual reviews to detect these smells. The manual process of detecting these smells is considered time consuming and costly. However, there is currently no work that implements a comprehensive quality model that classifies and identifies these smells by using a systematic strategy. As a result, there is a need for machine-supported analytical measures that decrease the time and effort needed to detect these smells manually, especially when it comes to reviewing and validating large test specifications.This study aims to investigate which natural language smell metrics implemented in NALABS can be found in test specifications and to measure the sufficiency of those smellmetrics. It also aims to extend these smell metrics by exploring, proposing, or combining with new bad smell metrics to cover more aspects of natural language test quality. The results of the study show that the smell metrics exists in real-world test specifications and can uncover many potential quality issues by assisting test designers in identifying certain types of quality issues pertaining to for example the understandability and complexity of test specifications. Moreover, the results show thatthe list of smell metrics implemented in NALABS is incomplete and can be extended to cover more aspects of test quality.

Development in the field of software architecture, from the early days in the mid-80’s, has been significant. From purely technical descriptions to decision based architectural knowledge, software architecture has seen fundamental changes to its methodologies and techniques. Architectural knowledge is a resource that is managed and stored by companies, this resource is valuable because it can be reused and analysed to improve future development. Companies today are interested in the reasoning behind the software architecture. This reasoning is mainly formulated through the architectural decisions made during development. For architectural decisions to be easier to analyse they need to be stored in a way that enables use of common analytical tools so that comparisons between decisions are consistent and relevant. Additionally, it is also important to have enough data, which leads us to the problem that, preferably, all the individual architectural knowledge cases must be structured and stored. To do this we present a tool that uses graph generation algorithms to generate architectural knowledge as graphs based on an architectural decision canvas called GRADE. This enables multiple decision cases to be encoded through graphs that can be used to analyse relationships and balances between different architectural knowledge elements represented through nodes and edges within a graph.

Safety and security ontologies quickly become essential support for integrating heterogeneous knowledge from various sources. Today, there is little standardization of ontologies and almost no discussion of how to compare concepts and their relationships, establish a general approach to create relationships or model them in general. However, concepts with similar names are not semantically similar or compatible in some cases. In this case, the problem of correspondence arises among the concepts and relationships found in the ontologies. To solve this problem, a comparison between the Hazard Ontology (HO) and the Combined Security Ontology (CSO) is proposed, in which the value of equivalence between their concepts and their relationships was extracted and analyzed. Although the HO covers the concepts related to the safety domain and the CSO includes securityrelated concepts, both are based on the Unified Foundational Ontology (UFO). For this study, HO and CSO were compared, and the results were summarized in the form of comparison tables. Our main contribution involves the comparisons among the concepts in HO and CSO to identify equivalences and differences between the two. Due to the increasing number of ontologies, their mapping, merging, and alignment are primary challenges in bridging the gaps that exist between the safety and security domains. 

Security ontologies have been developed to facilitate the organization and management of security knowledge. A comparison and evaluation of how these ontologies relate to one another is challenging due to their structure, size, complexity, and level of expressiveness. Differences between ontologies can be found on both the ontological and linguistic levels, resulting in errors and inconsistencies (i.e., different concept hierarchies, types of concepts, definitions) when comparing and aligning them. Moreover, many concepts related to security ontologies have not been thoroughly explored and do not fully meet security standards. By using standards, we can ensure that concepts and definitions are unified and coherent. In this study, we address these deficiencies by reviewing existing security ontologies to identify core concepts and relationships. The primary objective of the systematic literature review is to identify core concepts and relationships that are used to describe security issues. We further analyse and map these core concepts and relationships to five security standards (i.e., NIST SP 800-160, NIST SP 800-30 rev.1, NIST SP 800-27 rev.A, ISO/IEC 27001 and NISTIR 8053). As a contribution, this paper provides a set of core concepts and relationships that comply with the standards mentioned above and allow for a new security ontology to be developed.

Embedded systems are typically resource constrained, i.e., resources such as processors, I/O devices, shared buffers or shared memory can be limited for tasks in the system. Therefore, techniques that enable an efficient usage of such resources are of great importance.

In the industry, typically large and complex software systems are divided into smaller parts (applications) where each part is developed independently. Migration towards multiprocessor platforms has become inevitable from an industrial perspective. Due to such migration and to efficient use of system resources, these applications eventually may be integrated on a shared multiprocessor platform. In order to facilitate the integration phase of the applications on a shared platform, the timing and resource requirements of each application can be provided in an interface when the application is developed. The system integrator can benefit from such provided information in the interface of each application to ease the integration process. In this thesis, we have provided the resource and timing requirements of each application in their interfaces for applications that may need several processors to be allocated on when they are developed.

Although many scheduling techniques have been studied for multiprocessor systems, these techniques are usually based on the assumption that tasks are independent, i.e. do not share resources other than the processors. This assumption is typically not true. In this thesis, we provide an extension to such systems to handle sharing of resources other than processor among tasks. Two traditional approaches exist for multiprocessor systems to schedule tasks on processors. A recent scheduling approach for multiprocessors has combined the two traditional approaches and achieved a hybrid more efficient approach compared to the two previous one. Due to the complex nature of this scheduling approach the conventional approaches for resource sharing could not be used straight forwardly. In this thesis, we have modified resource sharing approaches such that they can be used in such hybrid scheduling systems. A second concern is that enabling resource sharing in the systems can cause unpredictable delays and variations in response time of tasks which can degrade system performance. Therefore, it is of great significance to improve the resource handling techniques to reduce the effect of imposed delays caused by resource sharing in a multiprocessor platform. In this thesis we have proposed alternative techniques for resource handling that can improve system performance for special setups.

Two traditional approaches exist for a task that is blocked on a global resource; a task either performs a non-preemptive busy wait, i.e., spins, or suspends and releases the processor. Previously, we have shown that both approaches can be viewed as spinning either at the highest priority HP or at the lowest priority on the processor LP, respectively. Based on this view, previously we have generalized a task's blocking behavioral model, as spinning at any arbitrary priority level. In this paper, we focus on a particular class of spin-lock protocols from the introduced flexible spin-lock model where spinning is performed at a priority equal to or higher than the highest local ceiling of the global resources accessed on a processor referred to as CP spin-lock approach. In this paper, we assume that all tasks of a specific processor are spinning on the same priority level. Given this class and assumption, we show that there exists a spin-lock protocol in this range that dominates the classic spin-lock protocol which tasks spin on highest priority level (HP). However we show that this new approach is incomparable with the CP spin-lock approach. Moreover, we show that there may exist an intermediate spin-lock approach between the priority used by CP spin-lock approach and the new introduced spin-lock approach that can make a task set schedulable when those two cannot. We provide an extensive evaluation results comparing the HP, CP and the new proposed approach.

Two traditional approaches exist for a task that is blocked on a global resource; a task either performs a non-preemptive busy wait, i.e., spins, or suspends and releases the processor. Previously, we have shown that both approaches can be viewed as spinning either at the highest priority HP or at the lowest priority on the processor LP, respectively. Based on this view, previously we have generalized a task's blocking behavioral model, as spinning at any arbitrary priority level. In this paper, we focus on a particular class of spin-lock protocols from the introduced flexible spin-lock model where spinning is performed at a priority equal to or higher than the highest local ceiling of the global resources accessed on a processor referred to as CP spin-lock approach. In this paper, we assume that all tasks of a specific processor are spinning on the same priority level. Given this class and assumption, we show that there exists a spin-lock protocol in this range that dominates the classic spin-lock protocol which tasks spin on highest priority level (HP). However we show that this new approach is incomparable with the CP spin-lock approach. Moreover, we show that there may exist an intermediate spin-lock approach between the priority used by CP spin-lock approach and the new introduced spin-lock approach that can make a task set schedulable when those two cannot. We provide an extensive evaluation results comparing the HP, CP and the new proposed approach.

Utvecklingsmetoder är olika arbetsprocesser som företag kan använda för att förbättra och effektivisera sitt arbete. Idag är de agila utvecklingsmetoderna Scrum och Kanban två av de mer använda metoderna inom företag som utvecklar mjukvara. I mitt arbete undersöks det hur nära små utvecklingsteam inom mindre företag förhåller sig till Scrum eller Kanban i verkligheten. Undersökningen görs för att se om det finns förbättringar som mindre företag med små utvecklingsteam kan ta till sig från litteratur för att efterlikna hur teorin är beskriven, samt kunna diskutera kring delar relevans som inte efterlevs.

Syftet med detta arbete är att identifiera hur nära agila utvecklingsmetoder som Scrum eller Kanban efterlevs i praktiken, hos mindre företag med små utvecklingsteam, enligt hur metoderna är teoretiskt beskrivna. Målet i detta arbete är se om det finns förbättringar som små utvecklingsteam kan ta till sig från litteratur om utvecklingsmetoderna, Scrum eller Kanban, för att effektivisera sin arbetsprocess genom att efterleva mer enligt hur dessa utvecklingsmetoder beskrivs i litteraturen.

För att uppnå arbetets syfte och mål undersökas tre frågeställningar. Ena frågeställningen är om hur litteratur beskriver metoderna för agil mjukvaruutveckling för små utvecklingsteam i mindre företag, vilket besvarades genom en litteraturstudie. Genom intervjuer besvarades frågeställningen om hur utvecklare i små utvecklingsteam inom mindre företag beskriver sin utvecklingsmetod. Tillsammans med intervjuerna och en observationsstudie besvarades den sista frågeställningen angående hur små utvecklingsteam i mindre företag arbetar med utveckling av mjukvara i praktiken.

Resultatet visade att utvecklarna från det medverkande företaget i praktiken endast följde vissa delar från Scrum men inte alla centrala delar, trots att Scrum enligt litteratur är anpassad till att användas inom små utvecklingsteam. Genom att inte följa Scrum centrala delar går det inte att säga att de använder Scrum, och därmed riskerar att inte leva upp till alla de fördelar Scrum kan bidra med.

Datorstött kooperativt arbete (hädanefter förkortat CSCW efter engelskans Computer-Supported Cooperative Work) är när ett system underlättar datorbaserat arbete i grupp på distans och söker lämplig teknik för att uppnå optimal effektivitet i gruppens arbetsprocess. Ett exempel på ett sådant system kan vara skissamarbete på digitala plattformar. Inom interaktionsdesign finns ett intresse för skissamarbete på digitala plattformar. Vi har därför valt att inom ramen för vårt arbete undersöka om skissamarbete på digitala plattformar är ett effektivt arbetssätt för designers och vilka funktioner en sådan plattform bör ha. För att få svar på våra frågor har vi valt att skapa en Android samt en Webbapplikation för skissamarbete. I våra applikationer har vi beaktat designers önskemål gällande funktioner. Applikationerna användes i en undersökning för att kunna grundligt analysera hur designers skissamarbetar på digitala plattformar. Resultatet har visat att skissamarbete på digitala plattformar är ett smidigt och mobilt arbetssätt som medför aktiva diskussioner, skapar nya idéer och bättre lösningar, bidrar till kunskapsöverföring, bättre arbetsmiljö och samspel mellan designers.

The superior performance of Deep Neural Networks (DNNs) has led to their application in various aspects of human life. Safety-critical applications are no exception and impose rigorous reliability requirements on DNNs. Quantized Neural Networks (QNNs) have emerged to tackle the complexity of DNN accelerators, however, they are more prone to reliability issues.In this paper, a recent analytical resilience assessment method is adapted for QNNs to identify critical neurons based on a Neuron Vulnerability Factor (NVF). Thereafter, a novel method for splitting the critical neurons is proposed that enables the design of a Lightweight Correction Unit (LCU) in the accelerator without redesigning its computational part.The method is validated by experiments on different QNNs and datasets. The results demonstrate that the proposed method for correcting the faults has a twice smaller overhead than a selective Triple Modular Redundancy (TMR) while achieving a similar level of fault resiliency. 

Deep Neural Networks (DNNs) and their accelerators are being deployed ever more frequently in safety-critical applications leading to increasing reliability concerns. A traditional and accurate method for assessing DNNs' reliability has been resorting to fault injection, which, however, suffers from prohibitive time complexity. While analytical and hybrid fault injection-/analyticalbased methods have been proposed, they are either inaccurate or specific to particular accelerator architectures. In this work, we propose a novel accurate, fine-grain, metric-oriented, and accelerator-agnostic method called DeepVigor that provides vulnerability value ranges for DNN neurons' outputs. An outcome of DeepVigor is an analytical model representing vulnerable and non-vulnerable ranges for each neuron that can be exploited to develop different techniques for improving DNNs' reliability. Moreover, DeepVigor provides reliability assessment metrics based on vulnerability factors for bits, neurons, and layers using the vulnerability ranges. The proposed method is not only faster than fault injection but also provides extensive and accurate information about the reliability of DNNs, independent from the accelerator. The experimental evaluations in the paper indicate that the proposed vulnerability ranges are 99.9% to 100% accurate even when evaluated on previously unseen test data. Also, it is shown that the obtained vulnerability factors represent the criticality of bits, neurons, and layers proficiently. DeepVigor is implemented in the PyTorch framework and validated on complex DNN benchmarks.

Trigger-Action Platforms (TAPs) play a vital role in fulfilling the promise of the Internet of Things (IoT) by seamlessly connecting otherwise unconnected devices and services. While enabling novel and exciting applications across a variety of services, security and privacy issues must be taken into consideration because TAPs essentially act as persons-in-the-middle between trigger and action services. The issue is further aggravated since the triggers and actions on TAPs are mostly provided by third parties extending the trust beyond the platform providers. Node-RED, an open-source JavaScript-driven TAP, provides the opportunity for users to effortlessly employ and link nodes via a graphical user interface. Being built upon Node.js, third-party developers can extend the platform’s functionality through publishing nodes and their wirings, known as flows. This paper proposes an essential model for Node-RED, suitable to reason about nodes and flows, be they benign, vulnerable, or malicious. We expand on attacks discovered in recent work, ranging from exfiltrating data from unsuspecting users to taking over the entire platform by misusing sensitive APIs within nodes. We present a formalization of a runtime monitoring framework for a core language that soundly and transparently enforces fine-grained allowlist policies at module-, API-, value-, and context-level. We introduce the monitoring framework for Node-RED that isolates nodes while permitting them to communicate via well-defined API calls complying with the policy specified for each node. 

Trigger-Action Platforms (TAPs) enable applications to integrate various devices and services otherwise unconnected. Recent features of TAPs introduce additional sources of data such as queries in IFTTT. The current TAPs, like IFTTT, demand that trigger and query services transmit excessive amounts of user data to the TAP. To limit the data to what is actually necessary for the execution to comply with the principle of data minimization, input services should send no more than the necessary data. LazyTAP proposes a new paradigm of data minimization by construction in TAPs, introducing a novel perspective for data collection from input services. While the existing push-all approach of TAPs entails coarse-grained data over-approximation, LazyTAP pulls input data on-demand at the level of attributes, once accessed by the app execution. Thanks to the fine granularity provided by LazyTAP, multiple trigger and query services can be naturally minimized while the behavior of app executions is preserved. In addition, a great benefit of LazyTAP is being seamless for third-party app developers. By leveraging laziness, LazyTAP defers computation and proxies objects to load necessary remote data behind the scenes. Our evaluation study on app benchmarks shows that on average LazyTAP improves minimization by 95% over IFTTT and by 38% over minTAP, with a tolerable performance overhead. This poster goes into further details about LazyTAP and elaborates on its prototype implementation. 

A difficult issue in stress management is to use biomedical sensor signal in the diagnosis and treatment of stress. Clinicians often make their diagnosis and decision based on manual inspection of physiological signals such as, ECG, heart rate, finger temperature etc. However, the complexity associated with manual analysis and interpretation of the signals makes it difficult even for experienced clinicians. Today the diagnosis and decision is largely dependent on how experienced the clinician is interpreting the measurements.  A computer-aided decision support system for diagnosis and treatment of stress would enable a more objective and consistent diagnosis and decisions.

A challenge in the field of medicine is the accuracy of the system, it is essential that the clinician is able to judge the accuracy of the suggested solutions. Case-based reasoning systems for medical applications are increasingly multi-purpose and multi-modal, using a variety of different methods and techniques to meet the challenges of the medical domain. This research work covers the development of an intelligent clinical decision support system for diagnosis, classification and treatment in stress management. The system uses a finger temperature sensor and the variation in the finger temperature is one of the key features in the system. Several artificial intelligence techniques have been investigated to enable a more reliable and efficient diagnosis and treatment of stress such as case-based reasoning, textual information retrieval, rule-based reasoning, and fuzzy logic. Functionalities and the performance of the system have been validated by implementing a research prototype based on close collaboration with an expert in stress. The case base of the implemented system has been initiated with 53 reference cases classified by an experienced clinician. A case study also shows that the system provides results close to a human expert. The experimental results suggest that such a system is valuable both for less experienced clinicians and for experts where the system may function as a second option.

Navigation and orientation in an indoor environment are a challenging task for visually impaired people. This paper proposes a portable vision-based system to provide support for visually impaired persons in their daily activities. Here, machine learning algorithms are used for obstacle avoidance and object recognition. The system is intended to be used independently, easily and comfortably without taking human help. The system assists in obstacle avoidance using cameras and gives voice message feedback by using a pre-trained YOLO Neural Network for object recognition. In other parts of the system, a floor plane estimation algorithm is proposed for obstacle avoidance and fuzzy logic is used to prioritize the detected objects in a frame and generate alert to the user about possible risks. The system is implemented using the Robot Operating System (ROS) for communication on a Nvidia Jetson TX2 with a ZED stereo camera for depth calculations and headphones for user feedback, with the capability to accommodate different setup of hardware components. The parts of the system give varying results when evaluated and thus in future a large-scale evaluation is needed to implement the system and get it as a commercialized product in this area.

AI methods referred to as interpretable are often discredited as inaccurate by supporters of the existence of a trade-off between interpretability and accuracy. In many problem contexts however this trade-off does not hold. This paper discusses a regression problem context to predict flight take-off delays where the most accurate data regression model was trained via the XGBoost implementation of gradient boosted decision trees. While building an XGB-CBR Twin and converting the XGBoost feature importance into global weights in the CBR model, the resultant CBR model alone provides the most accurate local prediction, maintains the global importance to provide a global explanation of the model, and offers the most interpretable representation for local explanations. This resultant CBR model becomes a benchmark of accuracy and interpretability for this problem context, and hence it is used to evaluate the two additive feature attribute methods SHAP and LIME to explain the XGBoost regression model. The results with respect to local accuracy and feature attribution lead to potentially valuable future work. © 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). CEUR Workshop Proceedings (CEUR-WS.org)

A key factor of software architecting is the decision-making process. All phases of software development contain some kind of decision-making activities. However, the software architecture decision process is the most challenging part. To support the decision-making process, a research project named ORION provided a knowledge repository that contains a collection of decision cases. To utilize the collected data in an efficient way, eliciting correlations between decision cases needs to be automated. 

The objective of this thesis is to select appropriate method(s) for automatically detecting correlations between decision cases. To do this, an experiment was conducted using a dataset of collected decision cases that are based on a taxonomy called GRADE. The dataset is stored in the Neo4j graph database. The Neo4j platform provides a library of graph algorithms which allow to analyse a number of relationships between connected data. In this experiment, five Similarity algorithms are used to find correlated decisions, then the algorithms are analysed to determine whether the they would help improve decision-making. 

From the results, it was concluded that three of the algorithms can be used as a source of support for decision-making processes, while the other two need further analyses to determine if they provide any support. 

Understanding the behavior of pedestrians in road transportation is critical to maintain a safe en- vironment. Accidents on road transportation are one of the most common causes of death today. As autonomous vehicles start to become a standard in our society, safety on road transportation becomes increasingly important. Road transportation is a complex system with a lot of dierent factors. Identifying risky behaviors and preventing accidents from occurring requires better under- standing of the behaviors of the dierent persons involved. In this thesis the activities and behavior of a pedestrian is analyzed. Using sensor data from phones, eight dierent events of a pedestrian are classied using machine learning algorithms. Features extracted from phone sensors that can be used to model dierent pedestrian activities are identied. Current state of the art literature is researched to nd relevant machine learning algorithms for a classication model. Two models are implemented using two dierent machine learning algorithms: Articial Neural Network and Hid- den Markov Model. Two dierent experiments are conducted where phone sensor data is collected and classied using the models, achieving a classication accuracy of up to 93%.

Robot-assisted training, where social robots can be used as motivational coaches, provides an interesting application area. This paper examines how feedback given by a robot agent influences the various facets of participant experience in robot-assisted training. Specifically, we investigated the effects of feedback type on robot acceptance, sense of safety and security, attitude towards robots and task performance. In the experiment, 23 older participants performed basic arm exercises with a social robot as a guide and received feedback. Different feedback conditions were administered, such as flattering, positive and negative feedback. Our results suggest that the robot with flattering and positive feedback was appreciated by older people in general, even if the feedback did not necessarily correspond to objective measures such as performance. Participants in these groups felt better about the interaction and the robot.

Over the past few decades the use of industrial robots has increased the efficiency as well as the competitiveness of several sectors. Despite this fact, in many cases robot automation investments are considered to be technically challenging. In addition, for most small and medium-sized enterprises (SMEs) this process is associated with high costs. Due to their continuously changing product lines, reprogramming costs are likely to exceed installation costs by a large margin. Furthermore, traditional programming methods of industrial robots are too complex for most technicians or manufacturing engineers, and thus assistance from a robot programming expert is often needed. The hypothesis is that in order to make the use of industrial robots more common within the SME sector, the robots should be reprogrammable by technicians or manufacturing engineers rather than robot programming experts. In this thesis, a novel system for task-level programming is proposed. The user interacts with an industrial robot by giving instructions in a structured natural language and by selecting objects through an augmented reality interface. The proposed system consists of two parts: (i) a multimodal framework that provides a natural language interface for the user to interact in which the framework performs modality fusion and semantic analysis, (ii) a symbolic planner, POPStar, to create a time-efficient plan based on the user's instructions. The ultimate goal of this work in this thesis is to bring robot programming to a stage where it is as easy as working together with a colleague.This thesis mainly addresses two issues. The first issue is a general framework for designing and developing multimodal interfaces. The general framework proposed in this thesis is designed to perform natural language understanding, multimodal integration and semantic analysis with an incremental pipeline. The framework also includes a novel multimodal grammar language, which is used for multimodal presentation and semantic meaning generation. Such a framework helps us to make interaction with a robot easier and more natural. The proposed language architecture makes it possible to manipulate, pick or place objects in a scene through high-level commands. Interaction with simple voice commands and gestures enables the manufacturing engineer to focus on the task itself, rather than the programming issues of the robot. The second issue addressed is due to inherent characteristics of communication with the use of natural language; instructions given by a user are often vague and may require other actions to be taken before the conditions for applying the user's instructions are met. In order to solve this problem a symbolic planner, POPStar, based on a partial order planner (POP) is proposed. The system takes landmarks extracted from user instructions as input, and creates a sequence of actions to operate the robotic cell with minimal makespan. The proposed planner takes advantage of the partial order capabilities of POP to execute actions in parallel and employs a best-first search algorithm to seek the series of actions that lead to a minimal makespan. The proposed planner can also handle robots with multiple grippers, parallel machines as well as scheduling for multiple product types.

Organizations gather enormous amounts of data and analyze these data to extract insights that can be useful for them and help them to make better decisions. Predictive data analytics is a crucial subfield within data analytics that make accurate predictions. Predictive data analytics extracts insights from data by using machine learning algorithms. This thesis presents the supervised learning algorithm to perform predicative data analytics in Embedded Test System at the Nordic Engineering Partner company. Predictive Maintenance is a concept that is often used in manufacturing industries which refers to predicting asset failures before they occur. The machine learning algorithms used in this thesis are support vector machines, multi-layer perceptrons, random forests, and gradient boosting. Both binary and multi-class classifier have been provided to fit the models, and cross-validation, sampling techniques, and a confusion matrix have been provided to accurately measure their performance. In addition to accuracy, recall, precision, f1, kappa, mcc, and roc auc measurements are used as well. The prediction models that are fitted achieve high accuracy.

Efficient production scheduling is important for both maximizing productivity and minimizing costs in manufacturing environments. This thesis presents an approach to optimizing production scheduling using Artificial Intelligence (AI) and Genetic Algorithms (GAs). The primary goal is to develop a generalized solution that can be modified and adapted to the varying needs that different production- or manufacturing lines may have. This research has two main research questions that address the problem at hand. (1) Can Genetic Algorithms be used to optimize a sequence of products in a production line? and (2) How effectively can Genetic Algorithms optimize the sequencing of production tasks in diverse production lines to minimize total order processing times? Through experimentation with various GA configurations the results achieved suggested that GAs were appropriate for sequence optimization. The study demonstrates that GAs can optimize a production line up to almost 42% , which significantly reduced the total processing time. The thesis also highlights the importance of the representation of data, which is essential for the optimization of the sequence. 

In the past, industrial control systems (ICS) and supervisory control and data acquisition (SCADA) systems were planned to run as isolated networks, and not interconnect with other networks e.g., the internet or other parts of a corporate’s network. Because of the isolation, no cybersecurity mechanism was required. In the modern society, ICS/SCADA systems has evolved to communicate over public IP networks and has been incorporated in a company’s intranet or directly to the internet. This integration opens up for threats that were not envisioned at the time when the system was created. When ICS/SCADA systems get exposed to the internet, there is a risk that vulnerabilities in the systems get exploited by a malicious force. This can lead to data loss, destruction of data and devices, damage to infrastructure, financial losses for the company, and even loss of human life could occur. To mitigate and prevent attacks it is crucial to understand the attacks and the behaviour of the attacker. One way to achieve this is setting up a system that mimics the real system. This fake system is separated from the production network and closely monitored. The data collected can be analysed and used to gain knowledge about the attacks.

This thesis will present a possible way to study attacks on an ICS/SCADA system using a honeypot designed for this purpose. To do this, a suitable honeypot had to be found that could collect relevant data regarding what kind of attacks that may be used against an ICS/SCADA system. This was achieved by experimenting with different set ups, and the collected data was analysed. This led to the use of T-pot as the chosen honeypot and the collected data showed that a lot of the traffic were directed towards the ICS/SCADA communication protocols Modbus and s7comm. To secure an ICS/SCADA system, it is important to gain knowledge about attacks and attack vectors. A honeypot can be a useful tool that provide information regarding attacks and attackers and can be a help in setting up a defence-in-depth strategy to improve the security in an ICS/SCADA network.

This thesis explores the integration of machine learning and computer vision technologies to automate calorie intake tracking, aiming to revolutionize traditional methods by leveraging object detection techniques. The study focuses on training and evaluating different versions of YOLO object detection model (YOLOv5, YOLOv6, YOLOv7 and YOLOv8) on a custom food dataset. Through experiments and simulated testing, the performance of these models is analyzed in terms of accuracy, speed, and practical usability. Additionally, a FastApi web application is developed to provide a user-friendly interface for capturing and uploading food images, processing them with the object detection model, and storing the daily calorie intake. The results demonstrate the potential of machine learning-driven solutions in addressing the challenges of manual calorie tracking, paving the way for more efficient and accurate dietary management systems. Results indicate that YOLOv5 and YOLOv6 exhibited similar performance, offering a balance between accuracy and speed, with YOLOv6 demonstrating slightly better accuracy. In contrast, YOLOv7-tiny, trained with a smaller image size, showed a significant drop in recall and F1 scores compared to its predecessors. Conversely, YOLOv8 achieved the highest accuracy among the tested models due to architecture improvements and optimizations.

The paper presents an approach to solve the software and hardware related failures in edge-cloud environments, more precisely, in cloud manufacturing environments. The proposed approach, called TOLERANCER, is composed of distributed components that continuously interact in a peer to peer fashion. Such interaction aims to detect stress situations or node failures, and accordingly, TOLERANCER makes decisions to avoid or solve any potential system failures. The efficacy of the proposed approach is validated through a set of experiments, and the performance evaluation shows that it responds effectively to different faults scenarios.

Inom byggindustrin uppstår många olika situationer som kan leda till olyckor där en person kan skadas eller avlida. Att utveckla metoder för att förhindra olyckor är därför mycket angeläget. I denna studie undersöker jag möjligheterna att skapa en applikation, anpassad för autonoma displaydatorer, för att öka säkerheten runt arbetsmaskiner genom att med objektdetektering och kameraövervakning skapa situationsmedvetenhet runt maskinen. Objektdetekteringen används för att hitta alla personer och koner som är med på bilden från kameran. Genom att detektera utställda trafikkoner byggs ett virtuellt stängsel upp som avgränsar det farliga området runt maskinen. För att sedan bedöma om personer befinner sig innanför området har jag tagit fram och utvärderat tre olika algoritmer. Jag har utvärderat algoritmerna med avseende på om exekveringstiden påverkar möjligheten att analysera 30 bilder per sekund och att de bedömer personens position korrekt i förhållande till det virtuella stängslet i olika scenarion.  Tester har också genomförts för att verifiera att algoritmerna fungerar i praktiken integrerade i realtidsapplikationen. Resultaten visar att inte någon av algoritmerna har exekveringstid som kommer att påverka analysen av 30 bilder per sekund. Vidare bedömer två av algoritmerna alla testade situationer korrekt förutom i vissa specifika fall när en person befinner sig exakt på en avgränsningslinje eller på en kon. Felbedömningen har dock ingen praktisk betydelse i denna tillämpning. Testerna av realtidsapplikationen visar att algoritmerna hanterar att området ändras dynamiskt och att det larmar när en person går in i det farliga området. Utifrån utvärderingen kan jag konstatera att det är möjligt att skapa en applikation för en autonom displaydator för att skapa situationsmedvetenhet för tunga arbetsmaskiner på byggarbetsplatser.

As IT-systems become more complex they become more susceptible to suffering of security threats and vulnerabilities. Accordingly, security has also become an increasingly growing concern that must be considered before the system has already been designed and put into operation. In particular, at the requirements phase of the Software Development Life Cycle. However, obtaining security requirements is non-trivial in view of requirements engineers’ insufficient knowledge of security terminology. There exist many methods that help to elicit security requirements by using various security concepts. Among the possible methods, this thesis compares Abuse Frame, Misuse Case and Common Criteria methods to identify the common security concepts they use to elicit security requirements and how they link security concepts to the security requirements. The research findings show that the most common security concepts that have been used by the above-mentioned methods are: threat, asset, and countermeasure. These concepts are directly related to security requirements in such manner where security requirements protect assets from harm caused by threats through describing the countermeasures that mitigate the threats and achieve the security objectives. It is also found that two out of three methods used the countermeasure concept as a synonym to a security requirementrather than a security architecture mechanism. A countermeasure is viewed as an abstract statement that describe how to mitigate a threat or eliminate a vulnerability without specifying any technical details or architecture mechanisms.

We live in a world that is constantly evolving where new technologies and innovations are being introduced. This progress partly results in developing new technologies and also in the improvement of the current ones. Docker containers are a virtualization method that is one of these new technologies that has become a hot topic around the world as it is said to be a better alternative to today's current virtual machines. One of the aspects that has contributed to this statement is the difference from virtual machines where containers isolate processes from each other and not the entire operating system. The company Saab AB wants to be at the forefront of today's technology and is interested in investigating the possibilities with container technology. The purpose with this thesis work is partly to investigate whether the container solution is in fact an alternative to traditional VMs and what differences there are between these methods. This will be done with the help of an in-depth literature study of comperative studies between containers and VMs. The results of the comparative studies showed that containers are in fact a better alternative than VMs in certain aspects such as performance and scalability and are worthy for the company. Thus, in the second part of this thesis work, a proof of concept implementation was made, by recreating a part of the company’s subsystem TactiCall into containers, to ensure that this transition is possible for the concrete use-case and that the container solution works as intended. This task has succeeded in highlighting the benefits of containers and showing through a proof of concept that there is an opportunity for the company to transition from VMs into containers. 

This thesis explores a novel approach to the computer vision eld in the form of lowcost computer vision intended for industrial use. The system proposed in this thesis, calledSimpleEye, is implemented and tested against an existing system. Dierent approachesto object detection and data extraction from a scene, as well as common applications ofcomputer vision in the industry, are examined. Three algorithms are implemented, aimedat dierent industrial applications. These are two types of object recognition, using CannyEdge detection and connected-component labeling, as well as barcode scanning. The tests,each targeting one of the implemented approaches, show promising results for low costcomputer vision. While the system is expectedly lacking in speed, it has no diculties inachieving good result in applications which are not highly time critical. SimpleEye yieldedaccuracy and precision comparable to commercial systems, with parts costing approximately100 USD. The tests show that the system is able to function in several computer visionapplications used today, including visual servoing, blob detection, blob tracking, and barcodescanning.

Globally, population numbers are growing, and the lifespan of the elderly is increasing. Therefore, this phenomenon requires that society commit more money, facilities, and staff for health care. The Internet of Things (IoT) can be employed to cover the financial shortfall in healthcare resources by using sensors for remote health monitoring such as the Shimmer device. The Shimmer physiological sensor is a Bluetooth-enabled radio device designed and used for monitoring various human health conditions. Using a sufficient number of the Shimmer devices with proper sampling rate can affect the provision of health and related information in realtime. Moreover, this type of sensor can sometimes be attached to the human body, which can create an inference between the sensors and the human body. The high-noise environment may also impact the sensor. This thesis reviews and analyses several scenarios in which Shimmer devices can be used by medical practitioners to offer reliable physiological measurements, such as ECG and movement. This study found that the Shimmers device can provide reliable data by using a specific configuration when the maximum number of sensors participate in a piconet network.

Software-defined networking (SDN) is a technology that was developed a decade ago and has become apole of attraction for researchers due to its unique ability to distinguish the control plane from the data planeon network devices. With SDN, network security has the potential to evolve to a great extent. Manycompanies, large and small, are threatened by attacks with the ultimate goal of the attacker’s personal gain.One way for outsiders to enter is through Rogue Access Points (RAPs). RAPs are malicious devices thatare connected to the network and provide wireless connection to unauthorized devices. The presented thesislooks into different methods for detecting and isolating a RAP from the network with SDN architectures.The implementation of this system is developed in a platform called Mininet-WiFi that simulatesheterogeneous types of networks. A topology is created on Mininet-WiFi that includes ten hosts, anOpenFlow switch, a ryuretic controller and a RAP. The thesis aims to detect and block a rogue device inan SDN network, and three algorithms are developed to achieve the main goal of this thesis. The resultsshow that the detection and isolation algorithms work successfully. This can be understood becauseillegitimate users cannot ping the legitimate network users. In addition, the scalability and reliability of thenetwork is being tested by increasing the number of users. 

Testing is an important activity in software development used to ensure the quality of a product. In industrial practice when developing control software, such as Programmable Logic Controllers (PLCs), software testing is, in some cases, neglected during the development process. There is little support for automated testing on PLC software and the creation of test cases is hence mainly a manual activity that can be both costly and time-consuming.

In this thesis, we propose a solution for an integrated tool chain for automating the testing process for IEC 61131-3 PLC software in the integrated development environment (IDE) of CODESYS (Controller Development System), using a combinatorial testing tool (i.e., SEAFOX) for test case generation. Further, we measure the applicability and usefulness of the integrated tool chain in terms of the decision coverage (i.e, Observable Decision Coverage) achieved by the generated test cases. For this purpose, several available tools for testing in CODESYS IDE have been examined as well as solutions on how to integrate the chosen tool (i.e., CfUnit) with SEAFOX. After a solution for the tool chain was implemented and integrated, an experiment was conducted to measure decision coverage for the generated test cases on nine industrial programs using Pairwise, Base choice and Random combinatorial testing techniques.

The result of our thesis is a fully integrated tool chain consisting of CODESYS, CfUnit, SEAFOX, and a python script used to combine these, and where the SEAFOX tool was further extended to support additional standard data types. This tool chain can be used to create test cases, generate new input values for these using SEAFOX, automate the additional test case creation and integration with CfUnit in CODESYS IDE, as well as for test execution. The results of the experiment show that test cases generated by this tool chain achieved on average 90% decision coverage or higher regardless of the combinatorial technique used, with the overall average level being 94%. Interestingly enough, Random testing scored higher than both Pairwise and Base choice testing.

We present our integrated tool chain as a contribution to the automation of test creation and execution for industrial control software in CODESYS IDE. However, we identified several limitations with testing CODESYS timer blocks and when executing larger test suites that can make it problematic to fully test PLC programs in a real-time simulation environment. These challenges should be investigated further by both researchers and practitioners.

Identifying intrusions in computer networks is important to be able to protect the network. The network is the entry point that attackers use in an attempt to gain access to valuable information from a company or organization or to simply destroy digital property. There exist many good methods already but there is always room for improvement. This thesis proposes to use reservoir computing as a feature extractor on network traffic data as a time series to train machine learning models for anomaly detection. The models used in this thesis are neural network, support vector machine, and linear discriminant analysis. The performance is measured in terms of detection rate, false alarm rate, and overall accuracy of the identification of attacks in the test data. The results show that the neural network generally improved with the use of a reservoir network. Support vector machine wasn't hugely affected by the reservoir. Linear discriminant analysis always got worse performance. Overall, the time aspect of the reservoir didn't have a huge effect. The performance of my experiments is inferior to those of previous works, but it might perform better if a separate feature selection or extraction is done first. Extracting a sequence to a single vector and determining if it contained any attacks worked very well when the sequences contained several attacks, otherwise not so well. 

Designing cutting tools for the turning industry, providing optimal cutting parameters is of importance for both the client, and for the company's own research. By examining the metal chips that form in the turning process, operators can recommend optimal cutting parameters. Instead of doing manual classification of metal chips that come from the turning process, an automated approach of detecting chips and classification is preferred. This thesis aims to evaluate if such an approach is possible using either a Convolutional Neural Network (CNN) or a CNN feature extraction coupled with machine learning (ML).

The thesis started with a research phase where we reviewed existing state of the art CNNs, image processing and ML algorithms. From the research, we implemented our own object detection algorithm, and we chose to implement two CNNs, AlexNet and VGG16. A third CNN was designed and implemented with our specific task in mind. The three models were tested against each other, both as standalone image classifiers and as a feature extractor coupled with a ML algorithm. Because the chips were inside a machine, different angles and light setup had to be tested to evaluate which setup provided the optimal image for classification.

A top view of the cutting area was found to be the optimal angle with light focused on both below the cutting area, and in the chip disposal tray. The smaller proposed CNN with three convolutional layers, three pooling layers and two dense layers was found to rival both AlexNet and VGG16 in terms of both as a standalone classifier, and as a feature extractor. The proposed model was designed with a limited system in mind and is therefore more suited for those systems while still having a high accuracy. The classification accuracy of the proposed model as a standalone classifier was 92.03%. Compared to the state of the art classifier AlexNet which had an accuracy of 92.20%, and VGG16 which had an accuracy of 91.88%. When used as a feature extractor, all three models paired best with the Random Forest algorithm, but the accuracy between the feature extractors is not that significant. The proposed feature extractor combined with Random Forest had an accuracy of 82.56%, compared to AlexNet with an accuracy of 81.93%, and VGG16 with 79.14% accuracy.

När olika datatyper överförs mellan två platser har, historiskt sett, detta skett på skildinfrastruktur; analoga signaler för sig, och digitala data för sig. Inom flygindustrin hanteras båda dessa datatyper än idag, då övergång från analoga röstsamtal till digitala röstsamtal tar tid. För att förenkla infrastrukturen är det önskvärt att skicka all data på samma medium. Time Division Multiplexing (TDM) möjliggör detta och har studerats med fokus på robusthet och prestanda. Studien har fokuserat på hur TDM-enheter ska konfigureras med avseende på Ethernet-frame-storlek för att till fullo utnyttja länkkapaciteten. Teoretiska resonemang har validerats av experimentella data som utförts på TDM-enheter sammankopplade med en fiberlänk. Antalet förlorade frames för systemet har mätts med varierande datatillförselshastighet. Detta upprepades för ett flertal Ethernet-frame-storlekar. Mindre frame-storlekar gav en högre möjlig överföringshastighet innan data förloras, och uppvisade koherens mellan teori och experiment. Stora frame-storlekar fyllde i snitt TDM-tidsluckorna sämre, och gjorde att databuffert på enheten fylldes snabbare än för mindre framstorlekar. Detta ledde till dataförluster i systemet. Andra faktorer, som overhead och payload, gjorde att framestorleken borde väljas efter applikation. Mindre datamängder som skickas ofta kan med fördel använda mindre frame-storlekar, medan stora datamängder som behöveröverföras snabbt effektivare överförs med större frame-storlekar.

An important step in developing software is to test the system thoroughly. Testing software requires a generation of test cases that can reach large numbers and is important to be performed in the correct order. Certain information is critical to know to schedule the test cases incorrectly order and isn’t always available. This leads to a lot of required manual work and valuable resources to get correct. By instead analyzing their test specification it could be possible to detect the functional dependencies between test cases. This study presents a natural language processing (NLP) based approach and performs cluster analysis on a set of test cases to evaluate the correlation between test case dependencies and their semantic similarities. After an initial feature selection, the test cases’ similarities are calculated through the Cosine distance function. The result of the similarity calculation is then clustered using the HDBSCAN clustering algorithm. The clusters would represent test cases’ relations where test cases with close similarities are put in the same cluster as they were expected to share dependencies. The clusters are then validated with a Ground Truth containing the correct dependencies. The result is an F-Score of 0.7741. The approach in this study is used on an industrial testing project at Bombardier Transportation in Sweden. 

The rapid growth of the mobile consumer market has influenced companies to make their workplace digital and mobile, with the goal to remove ties to a physical workplace, increase productivity, and trivialize tedious tasks. The construction industry is in need of modernization; it is a competitive market and to get a contract it is vital to provide a reasonable price quote fast.This bachelor thesis was performed together with Byggwalle AB in Västerås with the objective to create a mobile solution for price quotes, based on an earlier experimental prototype.

This thesis will look at how the manual price quote creation process can be sped up and made more transparent to build up trust between the companies and the customers. It will also look at the choices made when developing a mobile application for the construction industry, how to make it flexible enough to fit different kinds of businesses, and look at how the data is administrated using a web based administration system.

The result of this thesis is a promotional prototype, which sped up the delivery of price quotes from 3 weeks to 3 hours, and will be distributed to selected companies for further testing in the near future. We hope that this product can change the way business is done in construction industry.

This thesis investigates the implementation of an Retrieval-Augmented Generation (RAG) Teamschat bot to enhance the efficiency of a service organization, utilizing Microsoft Azure’s AI services.The project combines the retrieval capabilities of Azure AI Search with OpenAI’s GPT-3.5 Turboand Meta’s Llama 3 70B-instruct. The aim is to develop a chat bot capable of handling bothstructured and unstructured data. The motivation for this work comes from the limitations ofstandalone Large Language Models (LLMs) which often fail to provide accurate and contextuallyrelevant answers without external knowledge. The project uses the retriever and two languagemodels and evaluates them using F1 scoring. The retriever performs well, but the RAG modelproduces wrong or too long answers. Metrics other than F1 scoring could be used, and future workin prompt engineering as well as larger test datasets could improve model performance.

I dagens samhälle finns det flera typer av missbruk, ett av det vanligaste är alkoholmissbruk. Alkoholmissbruk har stora negativa effekter för individen men även för samhället. Det är därför viktigt att hitta effektiva metoder för att behandla alkoholmissbruk. Två metoder som används för att skapa en beteendeförändring hos missbrukarna är medicin och konsultation. På senare tid har ett nytt verktyg inom området börjat användas, nämligen mobilapplikationer. De mobilapplikationer vars mål är att skapa en beteendeförändring kan använda sig av designprinciper från ramverket Persuasive systemdesign (PSD). I den här studien har gränssnittet för applikationen Previct Task varit i fokus. Applikationen används som ett verktyg i kombination med att patienten har kontakt med en vårdgivare. Syftet med studien var att få fram ett designförslag som kan öka motivationen hos användarna jämfört med det nuvarande gränssnittet. Resultatet kan användas som riktlinjer för hur gränssnitt för applikationer som ska behandla alkoholmissbruk kan designas. För att få ett resultat har det nuvarande gränssnittet utvärderats genom en heuristisk utvärdering. De viktigaste designprinciperna från PSD och funktionerna i gränssnittet har även identifierats av användarna och vårdgivarna. Resultatet visar att designprinciperna Self-monitoring, Praise, Rewards och Trustworthiness var viktigast för att användarna skulle bli motiverade. Det nuvarande gränssnittet uppfyllde delvis designprinciperna, men jag kunde se att designprinciperna kunde förbättras. Slutsatsen av undersökningen var att applikationens idé var bra men att applikationens struktur och utförande behöver förbättras för att användarna ska känna sig mer motiverade.

With the simulations tools available today, simulation can be utilised as a platform for more advanced software testing. By introducing simulations to software testing of robot controllers, the motion performance testing phase can begin at an earlier stage of development. This would benefit all parties involved with the robot controller. Testers at ABB would be able to include more motion performance tests to the regression tests. Also, ABB could save money by adapting to simulated robot tests and customers would be provided with more reliable software updates. In this thesis, a method is developed utilising simulations to create a test set for detecting motion anomalies in new robot controller versions. With auto-generated test cases and a similarity analysis that calculates the Hausdorff distance for a test case executed on controller versions with an induced artificial bug. A test set has been created with the ability to detect anomalies in a robot controller with a bug.

With the growing popularity of the web, companies are starting to extend current development to reflect this. When extending desktop applications to the web, it can be difficult to choose what techniques and technologies to use when solving a problem, as current solutions might not be directly applicable. Rendering high-performance interactive 2D graphics on the web can be achieved in multiple ways. The rise in open standards such as the Canvas API allows the client to render natively in the browser, provided they can receive the full object state. There are some cases where this is simply not possible, where the object state is too large, or the client is lacking sufficient hardware. A possible solution is migrating the rendering of the graphic from the client to the server. However, remote rendering comes with new sets of issues as it often lacks high interaction capabilities, and would theoretically require more resources with multiple connections. This thesis will evaluate the performance differences and individual capabilities of remote and local rendering in terms of scalability and Quality of Experience using ASP.NET Core Blazor. The evaluation is done through the implementation of the four different solutions for the scenario. These implementations are based on Canvas and SVG using remote and local rendering. Different configurations of the performed tests, such as how much data should be rendered and how many clients are connected, were used to see how they affect response time and interaction latency. The results show that remote rendering performed better in all scalability tests, with remote SVG being the recommended approach. Due to implementation issues and lack of a proper testing environment, the number of concurrent clients was downsized. This caused problems when analyzing the results, and drawing concrete conclusions were difficult. In tests with increasing image size, the client solution suffered memory exceptions, preventing the local versions to be tested further. When testing interaction capabilities by measuring interaction latency, the SVG technology significantly outperformed Canvas, since SVG does not require a full re-render of the elements.

This paper presents an interpretable machinelearning model for anomaly detection in door locks using torque data. The model aims to replace the human tactile sense in the quality control process, reducing repetitive tasks and improving reliability. The model achieved an accuracy of 96%, however, to gain social acceptance and operators' trust, interpretability of the model is crucial. The purpose of this study was to evaluate anapproach that can improve interpretability of anomalousclassifications obtained from an anomaly detection model. Weevaluate four instance-based counterfactual explanators, three of which, employ optimization techniques and one uses, a less complex, weighted nearest neighbor approach, which serve as ourbaseline. The former approaches, leverage a latent representation of the data, using a weighted principal component analysis, improving plausibility of the counter factual explanations andreduces computational cost. The explanations are presentedtogether with the 5-50-95th percentile range of the training data, acting as a frame of reference to improve interpretability. All approaches successfully presented valid and plausible counterfactual explanations. However, instance-based approachesemploying optimization techniques yielded explanations withgreater similarity to the observations and was therefore concluded to be preferable despite the higher execution times (4-16s) compared to the baseline approach (0.1s). The findings of this study hold significant value for the lock industry and can potentially be extended to other industrial settings using timeseries data, serving as a valuable point of departure for further research.