Gaze estimation, a vital component in numerous applications such as humancomputer interaction, virtual reality, and driver monitoring systems, is the process of predicting the direction of an individual’s gaze. The predominant methods for gaze estimation can be broadly classified into intrusive and nonintrusive approaches. Intrusive methods necessitate the use of specialized hardware, such as eye trackers, while non-intrusive methods leverage images or recordings obtained from cameras to make gaze predictions. This thesis concentrates on appearance-based gaze estimation, specifically within the non-intrusive domain, employing various deep learning models. The primary focus of this study is to compare the efficacy of Vision Transformers (ViTs), a recently introduced architecture, with Convolutional Neural Networks (CNNs) for gaze estimation on RGB images. Performance evaluations of the models are conducted based on metrics such as the angular gaze error, stimulus distance error, and model size. Within the realm of ViTs, two variants are explored: pure ViTs and hybrid ViTs, which combine both CNN and ViT architectures. Throughout the project, both variants are examined in different sizes. Experimental results demonstrate that all pure ViTs underperform in comparison to the baseline ResNet-18 model. However, the hybrid ViT consistently emerges as the best-performing model across all evaluation datasets. Nonetheless, the discussion regarding whether to deploy the hybrid ViT or stick with the baseline model remains unresolved. This uncertainty arises because utilizing an exceedingly large and slow model, albeit highly accurate, may not be the optimal solution. Hence, the selection of an appropriate model may vary depending on the specific use case. / Ögonblicksbedömning, en avgörande komponent inom flera tillämpningar såsom människa-datorinteraktion, virtuell verklighet och övervakningssystem för förare, är processen att förutsäga riktningen för en individs blick. De dominerande metoderna för ögonblicksbedömning kan i stort sett indelas i påträngande och icke-påträngande tillvägagångssätt. Påträngande metoder kräver användning av specialiserad hårdvara, såsom ögonspårare, medan ickepåträngande metoder utnyttjar bilder eller inspelningar som erhållits från kameror för att göra bedömningar av blicken. Denna avhandling fokuserar på utseendebaserad ögonblicksbedömning, specifikt inom det icke-påträngande området, genom att använda olika djupinlärningsmodeller. Studiens huvudsakliga fokus är att jämföra effektiviteten hos Vision Transformers (ViTs), en nyligen introducerad arkitektur, med Convolutional Neural Networks (CNNs) för ögonblicksbedömning på RGB-bilder. Prestandautvärderingar av modellerna utförs baserat på metriker som den vinkelmässiga felbedömningen av blicken, felbedömning av stimulusavstånd och modellstorlek. Inom ViTs-området utforskas två varianter: rena ViTs och hybrid-ViT, som kombinerar både CNN- och ViT-arkitekturer. Under projektet undersöks båda varianterna i olika storlekar. Experimentella resultat visar att alla rena ViTs presterar sämre jämfört med basmodellen ResNet-18. Hybrid-ViT framstår dock konsekvent som den bäst presterande modellen över alla utvärderingsdatauppsättningar. Diskussionen om huruvida hybrid-ViT ska användas eller om man ska hålla sig till basmodellen förblir dock olöst. Denna osäkerhet uppstår eftersom användning av en extremt stor och långsam modell, även om den är mycket exakt, kanske inte är den optimala lösningen. Valet av en lämplig modell kan därför variera beroende på det specifika användningsområdet.

The recent development of massive multilingual transformer networks has resulted in drastic improvements in model performance. These models, however, are so large they suffer from large inference latency and consume vast computing resources. Such features hinder widespread adoption of the models in industry and some academic settings. Thus there is growing research into reducing their parameter count and increasing their inference speed, with significant interest in the use of knowledge distillation techniques. This thesis uses the existing approach of deep self-attention distillation to develop a task-agnostic distillation of the language agnostic BERT sentence embedding model. It also explores the use of the Switch Transformer architecture in distillation contexts. The result is DistilLaBSE, a task-agnostic distillation of LaBSE used to create a 10 times faster version of LaBSE, whilst retaining over 99% cosine similarity of its sentence embeddings on a holdout test from the same domain as the training samples, namely the OpenSubtitles dataset. It is also shown that DistilLaBSE achieves similar scores when embedding data from two other domains, namely English tweets and customer support banking data. This faster version of LaBSE allows industry practitioners and resourcelimited academic groups to apply a more convenient version of LaBSE to their various applications and research tasks. / Den senaste utvecklingen av massiva flerspråkiga transformatornätverk har resulterat i drastiska förbättringar av modellprestanda. Dessa modeller är emellertid så stora att de lider av stor inferenslatens och förbrukar stora datorresurser. Sådana funktioner hindrar bred spridning av modeller i branschen och vissa akademiska miljöer. Således växer det forskning om att minska deras parametrar och öka deras inferenshastighet, med stort intresse för användningen av kunskapsdestillationstekniker. Denna avhandling använder det befintliga tillvägagångssättet med djup uppmärksamhetsdestillation för att utveckla en uppgiftsagnostisk destillation av språket agnostisk BERT- innebördmodell. Den utforskar också användningen av Switch Transformerarkitekturen i destillationskontexter. Resultatet är DistilLaBSE, en uppgiftsagnostisk destillation av LaBSE som används för att skapa en 10x snabbare version av LaBSE, samtidigt som man bibehåller mer än 99 % cosinuslikhet i sina meningsinbäddningar på ett uthållstest från samma domän som träningsproverna, nämligen OpenSubtitles dataset. Det visas också att DistilLaBSE uppnår liknande poäng när man bäddar in data från två andra domäner, nämligen engelska tweets och kundsupportbankdata. Denna snabbare version av LaBSE tillåter branschutövare och resursbegränsade akademiska grupper

Vikten av hållbarhetsredovisning kan ses genom den uppmärksamhet ämnet har från företag, media, myndigheter och den ökande regleringen genom införandet av nya direktiv och lagstiftning. Att manuellt analysera företags hållbarhetsredovisningar är en tidskrävande process. En automatiserad analys av hållbarhetsredovisningar skulle innebära ekonomiska och tidsmässiga vinster när viktiga insikter tas fram relaterat till större företags påverkan på sin miljö och omgivning. Denna studie syftar till att utforska möjligheterna till en automatisering av en befintlig manuell arbetsmetod. Prototypen som utvecklats tillämpar moderna språkbehandlingsmetoder, ett område inom maskininlärning, för att realisera denna vision. Studiens implementation uppnår för de utvärderade språkmodellerna upp till 96% precision för majoritetsklassen vid bearbetning av grunddatat respektive 55% precision för minoritetsdataklassen vid bearbetning av grunddata jämfört resultat från den manuellt genomförda metoden. Slutsatsen är att en automatiserad version av den befintliga manuella analysmetoden kan konstrueras och även förbättras med den snabba utveckling som sker inom teknologi och språkmodeller, om ytterligare resurser avsätts. Resultaten visar hopp om potentialen för en metodik som utvecklas i vidare arbeten. / The importance of sustainability reporting can be observed by the attention directed towards the subject from companies, media and authorities’ continuous new directives and laws. To manually analyze companies’ sustainability reports is a time-consuming process. An automated approach analyzing sustainability reports would give advantages regarding both time and economics when important insights related to companies’ operations are brought into light. This study aims to explore possibilities in automating an existing manual method related to analyzing sustainability reports. The developed prototype applies modern language models and methods related to machine learning to realize this vision. For the evaluated language models, the study’s implementation achieves up to 96% precision for the majority class, while the minority class achieves up to 55% precision in processing of data, when compared to reference results from the manual evaluation method. The work’s conclusion indicates that an automated version of the existing manual method for analysis can be constructed with sufficient resources, and even further improved as the area of technology further advances. The results are positive for the potential for a more sophisticated method that can be developed in further work.

Röststyrning som användargränssnitt kan erbjuda flera fördelar jämfört med mer traditionella styrmetoder. Det saknas dock färdiga lösningar för specifika industriella miljöer, vilka ställer särskilda krav på att korta kommandon tolkas korrekt i olika grad av buller och med begränsad eller ingen internetuppkoppling. Detta arbete ämnade undersöka potentialen för röststyrning i industriella miljöer. Ett koncepttest genomfördes där ordfelsfrekvens (på engelska Word Error Rate eller kortare WER) användes för att utvärdera träffsäkerheten för olika kombinationer av taligenkänningsarkitekturer, brusreduceringstekniker samt kommandolängder i verkliga bullriga miljöer. Undersökningen tog dessutom hänsyn till Lombard-effekten.  Resultaten visar att det för samtliga testade miljöer finns god potential för röststyrning med avseende på träffsäkerheten. Framför allt visade DeepSpeech, en djupinlärd taligenkänningsmodell med rekurrent lagerstruktur, kompletterad med domänspecifika språkmodeller och en riktad kardioid-mikrofon en ordfelsfrekvens på noll procent i vissa scenarier och sällan över fem procent. Resultaten visar även att utformningen av kommandon påverkar ordfelsfrekvensen.  För en verklig implementation i industriell miljö behövs ytterligare studier om säkerhetslösningar, inkluderande autentisering och hantering av risker med falskt positivt tolkade kommandon. / Voice command as a user interface can offer several advantages over more traditional control methods. However, there is a lack of ready-made solutions for specific industrial environments, which place particular demands on short commands being interpreted correctly in varying degrees of noise and with limited or no internet connection. This work aimed to investigate the potential for voice command in industrial environments. A proof of concept was conducted where Word Error Rate (WER) was used to evaluate the accuracy of various combinations of speech recognition architectures, noise reduction techniques, and command lengths in authentic noisy environments. The investigation also took into account the Lombard effect.  The results indicate that for all tested environments there is good potential for voice command with regard to accuracy. In particular, DeepSpeech, a deep-learned speech recognition model with recurrent layer structure, complemented with domain-specific language models and a directional cardioid microphone, showed WER values of zero percent in certain scenarios and rarely above five percent. The results also demonstrate that the design of commands influences WER. For a real implementation in an industrial environment, further studies are needed on security solutions, including authentication and management of risks with false positive interpreted commands.

Thesis (D. Tech. Engineering : Electrical ; Dept. Applied Electronics and Electronic Communications) -- Vaal University of Technology / Two methods to convert an AC current source to an AC voltage source are presented. Both methods make use of a current transformer to provide energy extraction from the main system while maintaining galvanic isolation between the main system and the output system. Control is via a pulse width modulation scheme in both instances. The ftrst method uses a storage element to provide a DC voltage, which feeds an inverter that supplies the AC load with a voltage source. The second method does not use a storage element but is a direct AC current source to AC voltage source converter employing a current transformer. This has not been done before. A possible application of this study is in the extraction of power from high voltage transmission lines without the conventional substation and rural grid to supply small users such as the telecommunication industry.

A dissertation submitted to the Faculty of Engineering and the Built Environment, University of the Witwatersrand, in fulfilment of the requirements for the degree of Master of Science in Engineering Date submitted: 28 August 2015 / Many generator transformers were installed many years ago during the initial commissioning of Eskom’s power stations. Many of these transformers have started showing signs of significant ageing of the paper insulation and hence require regular monitoring. There are two methods that are currently being employed to assess the degree of ageing of the paper insulation in a generator transformer, which are paper sampling and furan level measurement. This dissertation investigates an alternative method of predicting the degree of ageing of the paper insulation instead of what is used currently. This method uses the Arrhenius equation that relates time and temperature to determine the degree of degradation of organic materials. The reliability of the Arrhenius estimation method is assessed by comparing the predicted DP (Degree of polymerisation) values with the measured DP values of the same transformer paper insulation. The results obtained showed that there is reasonable correlation between the DP values estimated from the Arrhenius equation and the DP values estimated from the measured furan levels. The accuracy of the prediction method is reduced when the oil temperature greatly differs from the paper insulation temperature. The application of the Arrhenius equation to estimate the ageing of paper insulation is a great milestone in the quest to predict the remaining life of a transformer. It is the only method available to do this prediction and using online temperature measurement on transformers makes the method more reliable. / MT 2017

Este trabalho tem por meta focalizar a adaptação necessária na metodologia de planejamento atual, norteada segundo uma ótica determinística e conservadora, afim de permitir a elaboração de Programas de Obras de Transmissão/Subtransmissão, condicionados por limitações de teto no montante de investimento a curto e médio prazos. Para tanto, como tema central, busca-se desenvolver uma base conceitual e metodológica, para permitir uma exploração maximizada das possibilidades de carregamento de equipamentos e instalações do sistema de transmissão, contemplando a especificação das ferramentas computacionais necessárias para dar consequência prática à proposta. Considerando o potencial de análise proporcionado por métodos e técnicas de natureza probabilística, aplicados ao planejamento de Sistemas Elétricos, a espinha dorsal da evolução metodológica delineada no texto repousa sobre a avaliação de confiabilidade global e quantificação de riscos operativos. O texto foi estruturado em 8 capítulos onde, a partir de um capítulo introdutório, enfoca-se sequencialmente os temas: * Conceitos e Metodologias para Maximização do Carregamento de Transformadores de Potência. * Conceitos e Metodologias para Maximização do Carregamento de Linhas aéreas de Transmissão. * Impacto das Restrições Financeiras no planejamento e possibilidade de atuação a nível técnico para administrar o conflito entre qualidade de serviço e recursos limitados. * Conceitos e Metodologias para relaxação dos Critérios limitantes para o carregamento máximo de componentes do sistema, em ambiente de severas restrições financeiras, no horizonte de curto prazo. * Resultados obtidos em aplicações concretas a situações do Setor Elétrico, visando caracterizar o alcance do ferramental preconizado nos capítulos anteriores. * Critérios e Procedimentos para Padronização de Subestações e Linhas de Transmissão , coerentes com os conceitos de maximização de carregamento introduzidos nos dois primeiros capítulos. Ao final do texto, incluiu-se um capítulo de \"Conclusões\", com o objetivo de sumarizar os aspectos mais relevantes abordados no texto, assim como indicar linhas de ação e novos desenvolvimentos, necessários para permitir a efetivação prática da metodologia proposta. / An important problem posed to power system planners in developing countries is now becoming commonplace, even in developed countries, and concerns on how to fit an expansion plan into a tight budget, with the least reduction in reliability levels. This problem is known as power system planning under financial constraints and a series of methods is now available to deal with it. So, this work addresses the necessary adjustments in order to permit the Transmission / Sub-transmission Expansion Program establishment when there are investments constraints in a medium/short term range. The main emphasis is on the development of a methodological and conceptual framework, aiming at a maximizes exploration of the loading capability of transformers and transmission lines of the power network. In this context, a new methodological framework for transmission system planning in a financial constrained environment is presented. The proposed approach combines transmission projects priority evaluation and ranking as well as criteria flexibilization, being particularly suitable for short-term planning. The problem to be faced is to operate the system after a project postponement, while the service quality is kept so adequate as it is possible. One of the main issues to be focused is the transmission system uprating since a maximized equipment utilization can contribute to avoid a more significant restriction to the costumers. Therefore, a detailed description of the models for transmission lines and transformers loading limits assessment is presented, with emphasis on the main concepts and the application of probabilistic tools to circumvent the inherent conservatism of deterministic methods. A summary of the contents is presented in what follows: * Concepts and Methodology for Transformers loading maximization. * Concepts and Methodology for Transmission Lines loading maximization. * Financial Restrictions impacts on Expansion Planning Process and the possibilities to manager the conflict between service quality and investment constraints. * Concepts and Methodologies to the relaxation of Criteria that limits the loading capability of transmission lines, aiming at short term Planning under severe financial restrictions, * Obtained results in concrete applications of the proposed methodology and computational tools to Brazilian Electric Power System. * Methods and Criteria to standardization of Substations and Transmission Lines, according the main issues formerly introduced. * Conclusion and Future Research.

O objetivo deste trabalho de pesquisa foi de fazer a modelagem por meio de sistemas de inferência Fuzzy da análise da concentração dos gases dissolvidos no óleo mineral isolante e desta forma diagnosticar defeitos elétricos internos em transformadores de Potência. Assim sendo o sistema proposto deve fornecer respostas que auxiliem no diagnóstico de falhas e avarias nos interiores dos transformadores e no processo de tomada de decisões no acompanhamento da evolução destas falhas de forma a aumentar a confiabilidade em relação à utilização dos métodos individualmente. O sistema desenvolvido baseou-se na pesquisa acadêmica de normas e técnicas mais utilizadas na literatura que relacionam o gás dissolvido no óleo mineral isolante com a falha. O sistema proposto foi validado por meio de dados reais de dois transformadores pilotos e vários sistemas Fuzzy foram construídos, cada um especialista em um determinado método. Os resultados encontrados mostraram-se compatíveis com aqueles obtidos pelos métodos convencionais comprovando que a ferramenta pode ser utilizada como suporte para uma análise rápida e confiável do estado do óleo isolante dos transformadores. / The objective of this research was to model by analyzing the fuzzy inference systems the concentration of gases dissolved in the insulating oil and diagnosing thereby internal defects in electrical power transformers. Therefore the proposed system should provide answers that help in the diagnosis of faults and malfunctions in the interiors of the transformers and in the decision-making process in monitoring the evolution of these failures in order to increase the reliability regarding the use of the methods individually. The system developed was based on academic research standards and techniques commonly used in the literature relating the gas dissolved in the insulating oil with failure. The proposed system was validated by real data of two pilots transformers and various Fuzzy systems were built, each an expert in a particular method. The results were compatible with those obtained by conventional methods proving that the tool can be used as support for quick and reliable analysis of the insulating oil condition of transformers.

Este trabalho apresenta o projeto de um amplificador de potência (PA) de rádio-frequência (RF) linear em tecnologia complementar metal-oxido silício (CMOS). Nele são analisados os desafios encontrados no projeto de PAs CMOS assim como soluções encontradas no estado-da-arte. Um destes desafios apresentados pela tecnologia é a baixa tensão de alimentação e passivos com alta perda, o que limita a potência de saída e a eficiência possível de ser atingida com métodos tradicionais de projeto de PA e suas redes de transformação de impedância. Este problema é solucionado através do uso de redes de combinação de impedância integradas, como a usada neste trabalho chamada transformador combinador em série (SCT). Os problemas com o uso de tecnologia CMOS se tornam ainda mais críticos para padrões de comunicação que requerem alta linearidade como os usados para redes sem-fio locais (WLAN) ou padrões de telefonia móvel 3G e 4G. Tais protocolos requerem que o PA opere em uma potência menor do que seu ponto de operação ótimo, degradando sua eficiência. Técnicas de linearização como pré-distorção digital são usadas para aumentar a potência média transmitida. Uma ténica analógica de compensação de distorção AM-PM através da linearização da capacitância de porta dos transistores é usada neste trabalho. O processo de projeto é detalhado e evidencia as relações de compromisso em cada passo, particularmente o impacto da terminação de harmônicos e a qualidade dos passivos na rede de transformação de carga. O projeto do SCT é otimizado para sintonia da impedância de modo comum que é usada para terminar o segundo harmonico de tensão do amplificador. O amplificador projetado tem um único estágio devido a área do chip ser limitada a 1:57 x 1:57 mm2, fato que impacta seu desempenho. O PA foi analisado através de simulação numérica sob várias métricas. Ele atinge uma potência máxima de saída de 24:4 dBm com uma eficiência de dreno de 24:53% e Eficiência em adição de potência (PAE) de 22%. O PA possui uma curva de ganho plana em toda faixa ISM de 2.4 GHz, com magnitude de 15:8 0:1dB. O PA tem um ponto de compressão de OP1dB = 20:03 dBm e o sinal tem um defasamento não-linear de = 1:2o até esta potência de saída. Um teste de intermodulação de dois tons com potência 3dB abaixo do OP1dB tem como resultado uma relação entre intermodulação de terceira ordem e fundamental de IMD3 = 24:22 dB, e de quinta ordem inferior e superior e fundamental de IMD5Inferior = 48:16 dB e IMD5Superior = 49:8 dB. Por fim, mostra-se que o PA satisfaz os requerimentos para operar no padrão IEEE 802.11g. Ele atinge uma potência média de saída de 15:4 dBm apresentando uma magnitude do vetor erro (EVM) de 5:43%, ou 25:3 dB e satisfazendo a máscara de saída para todos os canais. / This work presents the design of a fully integrated Radio-frequency (RF) linear Power Amplifier( PA) in complementary metal-oxide silicon (CMOS) technology. In this work we analyse the challenges in CMOS PA design as well as the state-of-the-art solutions. One such challenge presented by this technology is the low supply voltage and high-loss passives, which pose severe limits on the output power and efficiency achieved with traditional PA design methods and load impedance transformation networks. This issue is addressed by the use of on-chip, highly efficient power combining networks such as the one in this work: A series combining transformer (SCT). The problem of using CMOS becomes even more critical for recent communications standards that require high transmitter linearity such as the ones used for wireless local area network (WLAN) or 3G and 4G mobile communications. This requirement is such that the PA operate at a high power back-off from its optimum operating point, degrading efficiency. To address this problem linearization techniques such as digital pre-distortion can be used in order to decrease the necessary power back-off. In this work an analog technique of AM-PM distortion compensation is used to linearize the capacitance at the input of the amplifier’s transistors and reduce this type of distortion that severely impacts the error vector magnitude (EVM) of the signal. The design process is detailed and aims to make evident the trade-offs of PA design and particularly the impact of harmonic termination and the quality of passives on the load transformation network, the series combining transformer design is optimized for common-mode impedance tuning used for 2nd harmonic termination. The circuit has only a single amplifying stage due to its area being limited to 1:57 x 1:57 mm2 and the design is very constrained by this fact. The PA simulated performance is analyzed under various metrics. It achieves a simulated maximum output power of 24:4 dBm with a drain efficiency of 24:53% and power added efficiency (PAE) of 22%. The PA has a very flat power gain of 15:8 0:1 dB throughout the 2.4 GHz industrial, scientific and medical (ISM) band and is unconditionally stable with 4:9. The PA has a compression point of OP1dB = 20:03 dBm and the signal has a non-linear phase shift of = 1:2o up to this output power. A two-tone intermodulation test with 3dB back-off from OP1dB has a ratio of third-order intermodulation to fundamental of IMD3 = 24:22 dB, and lower and upper fifth order intermodulation to fundamental of IMD5Lower = 48:16 dB and IMD5Upper = 49:8 dB. Finally the PA is shown to satisfy the requirements for operation within the institute of electrical and electronic engineers (IEEE) 802.11g standard. It achieves an average output power of 15:4 dBm while having an EVM of 5:43% or 25:3 dB while satisfying the output spectrum mask for all channels.

Transformadores de potência representam um grande volume de investimento em sistemas de transmissão e distribuição. Falhas em transformadores têm um alto impacto econômico na exploração de uma rede. O parâmetro mais importante e que define a vida útil de isolação de um transformador é a temperatura do ponto mais quente de seus enrolamentos. O trabalho analisa dois modelos térmicos de transformadores de potência. Eles nos permitem calcular as temperaturas do óleo no topo e na parte inferior e a temperatura do ponto quente no enrolamento. Ambos os modelos são baseados em circuitos elétricos a parâmetros concentrados. O primeiro é uma proposta de Susa, que envolve a solução de dois circuitos elétricos acoplados. Já Tang sugere um segundo modelo, que consiste em um circuito elétrico de três nós. O principal objetivo deste trabalho é a obtenção dos parâmetros térmicos de ambos os modelos baseados em temperaturas medidas para uma determinada carga. Duas abordagens diferentes de otimização serão avaliadas: uma mono objetivo e outra multi objetivo. A segunda oferece-nos uma análise de sensibilidade de parâmetros do modelo. Ambos os métodos de otimização foram utilizadas para obter os parâmetros térmicos de um transformador de potência 250 MVA resfriamento ONAN (Óleo Natural, Ar Natural). Os parâmetros obtidos do modelo térmico são então utilizadas para outras curvas de carga e também são comparadas com a norma IEEE C57.91 1995 com boa precisão. / Power transformers represent a large amount of investment in transmission and distribution systems. Faults in transformers have a high economic impact in the operation of a network. One of the most important parameters that govern the life of transformers is its hotspot temperature, which defines the loss of life of the equipament. The work analyzes two thermal models for a power transformer. They allow us to calculate the top oil temperature, the bottom oil temperature and the hotspot temperature of power transformers. Both models are based on lumped parameters. The first one is proposed by Susa, and involves the solution of two coupled electrical circuits. Tang suggests the second one, which consists on an electric circuit of three nodes to be solved. The main objective of this work is to obtain the thermal lumped parameters of both models based on measured temperatures for a given load curve. Two different approaches will be adopted: a mono-objective optimization and a multiobjective optimization. The second one provides us a sensitivity analysis of the model parameters. Both optimization methods were used to obtain the thermal lumped parameters in a 250 MVA ONAN power transformer. The obtained parameters of the thermal model are then used for other load curves and compared with the guide IEEE C57.91 1995 with good accuracy.