No description available.

The aim of this thesis is to study problems of deep convolutional neural networks and the connected classification of images and to experiment with the architecture of particular network with the aim to get the most accurate results on the selected dataset. The thesis is divided into two parts, the first part theoretically outlines the properties and structure of neural networks and briefly introduces selected networks. The second part deals with experiments with this network, such as the impact of data augmentation, batch size and the impact of dropout layers on the accuracy of the network. Subsequently, all results are compared and discussed with the best result achieved an accuracy of 86, 44% on test data.

The main benefit of the work is the optimization of the hardware configuration for the calculation of neural networks. The theoretical part describes neural networks, deep learning frameworks and hardware options. The next part of the thesis deals with implementation of performance tests, which include application of Inception V3 and ResNet models. Network models are applied to various graphics cards and computing hardware. The output of the thesis is the implemented model of the network Inception V3, which examines the graphics cards and their performance, time-consuming calculations and their efficiency. The ResNet model is applied to a section that examines other impacts on neural network computing such as used disk, operating memory, and so on. Each practical part contains a discussion where the knowledge of the given part is explained. In the case of consumption measurement, a mismatch between the declaration by the manufacturer and the measured values was identified.

The stable operating range of axial compressors is limited by the onset of rotating stall and surge. Mass injection upstream of the tip of an axial compressor rotor is a stability enhancement approach which can be effective in suppressing stall in tip-critical rotors, and thus increasing the operating range of compressors. In this article, investigations on active flow control related to the rotor tip gap sensitivity are discussed. The experiments were performed in a 1.5-stage low-speed research compressor. Measurements at part speed (80 per cent) and full speed (100 per cent) with varying injection rates are discussed. These tests were performed for two rotor tip clearances of 1.3 per cent and 4.3 per cent of rotor blade tip chord. Results on the compressor map, the flow field as well as transient measurements to identify the stall inception are discussed. Supplementary, the numerical results are compared to the experiments based on the configuration with the greatest benefit in operating range enhancement.

Convolutional neural networks have been successfully applied to both object classification and audio classification. The aim of this thesis is to evaluate the degree of how well transfer learning of convolutional neural networks, trained in the object classification domain on large datasets (such as CIFAR-10, and ImageNet), can be applied to the audio classification domain when only a small dataset is available. In this work, four different convolutional neural networks are tested with three configurations of transfer learning against a configuration without transfer learning. This allows for testing how transfer learning and the architectural complexity of the networks affects the performance. Two of the models developed by Google (Inception-V3, Inception-ResNet-V2), are used. These models are implemented using the Keras API where they are pre-trained on the ImageNet dataset. This paper also introduces two new architectures which are developed by the authors of this thesis. These are Mini-Inception, and Mini-Inception-ResNet, and are inspired by Inception-V3 and Inception-ResNet-V2, but with a significantly lower complexity. The audio classification dataset consists of audio from RC-boats which are transformed into mel-spectrogram images. For transfer learning to be possible, Mini-Inception, and Mini-Inception-ResNet are pre-trained on the dataset CIFAR-10. The results show that transfer learning is not able to increase the performance. However, transfer learning does in some cases enable models to obtain higher performance in the earlier stages of training.

An increased understanding of how inlet flow distortion affects transonic fans enables improved fan design and performance prediction. Inlet distortion refers to non-uniformities in the incoming flow properties. Complex inlet ducts in high performance aircraft result in distorted flow at the fan inlet. In this thesis, two studies were performed using Unsteady Reynolds-Averaged Navier Stokes (URANS) simulations. The first study focused on understanding how the transition abruptness between the clean and distorted sector in the inlet Pt profile as well as the circumferential extent of the distorted sector affect distortion transfer and generation through a transonic fan. Simulations on two main distortion sector sizes were carried out. For each sector size, variants with decreasing levels of transition abruptness were applied to the inlet of fan. Simulations were conducted at various operating points, ranging from choke to near-stall. Fourier-based distortion descriptors were used to quantify levels of distortion transfer and generation at various axial locations. It is shown that variations in rotor incidence occur as a result of the applied Pt distortion at the inlet. A less abrupt transition diminishes the local extrema in rotor incidence, which in turn reduces the amount of distortion transfer and generation through the rotor. The near-stall condition is affected most of all operating points considered, with a 23.4% average reduction in the amount of distortion transfer at any span. The size the inlet distorted sector affects the amount of distortion transfer and generation, particularly at the near-stall operating point. This is shown to be due to the dynamic response of the fan. The second study compared the mechanisms of stall inception for cases of both clean and distorted inlet flow. In each instance, the mechanism of stall inception is shown to be interactions between the detached bow shock and the tip clearance vortex. These interactions result in the formation of two vortices within the blade passage. The location and strength of these vortices affect the LE spillage in the adjacent blade rows. Stall inception occurs when the bow shock has moved far enough upstream to allow the resultant vortices from shock/tip clearance vortex interaction to pass in front of the leading edge. When inlet distortion is present, mass redistribution upstream of the fan results in variations in rotor incidence. Within the high incidence region, the bow shock is detached 3.9%-8.1% chord more than the clean inlet case, making LE spillage more severe. The rotating stall cell grows out of the stalled passages present at the near-stall operating point and ultimately extends 180° circumferentially and 18.7% span radially. Understanding the effects of distortion on the mechanisms of stall inception will allow appropriate steps to be taken to extend the stable operating range of modern commercial and high performance fans.

This thesis is missing page 168. The other copies do not have this page. -Digitization Centre / The Canadian Health Insurance program has been in place for quite some time now and it has always been said that Canadians have some of the best healthcare in the world. Canadian healthcare is very well known throughout the world because every Canadian citizen has the right to healthcare without having to pay for it. The benefits of this program are quite well known but some benefits one might think would result, just might not be there. This paper looks at whether the inception of Canadian Health Insurance has had an effect on the mortality rates of Canadians. Through a statistical analysis, this paper shows that there is no evidence that the Canadian Health Insurance program has had an effect on Canadian aggregate mortality rates. This paper shows that Canadian mortality rates follow a trend to that of the United States. To say Canadians have a similar trend in mortality rate to the United States is perhaps surprising since Canadians are supposed to have a far superior healthcare system. / Thesis / Master of Science (MS)

Image classification is the process of identifying to which class a previously unobserved object belongs to. Classifying images is a commonly occurring task in companies. Currently many of these companies perform this classification manually. Automated classification however, has a lower expected accuracy. This thesis examines how automated classification could be improved by the addition of augmented data into the learning process of the classifier. We conduct a quantitative empirical study on the effects of two image augmentations, random horizontal/vertical flips and random rotations (<180◦). The data set that is used is from an auction house search engine under the commercial name of Barnebys. The data sets contain 700 000, 50 000 and 28 000 images with each set containing 28 classes. In this bachelor’s thesis, we re-trained a convolutional neural network model called the Inception-v3 model with the two larger data sets. The remaining set is used to get more class specific accuracies. In order to get a more accurate value of the effects we used a tenfold cross-validation method. Results of our quantitative study shows that the Inception-v3 model can reach a base line mean accuracy of 64.5% (700 000 data set) and a mean accuracy of 51.1% (50 000 data set). The overall accuracy decreased with augmentations on our data sets. However, our results display an increase in accuracy for some classes. The highest flat accuracy increase observed is in the class "Whine & Spirits" in the small data set where it went from 42.3% correctly classified images to 72.7% correctly classified images of the specific class. / Bildklassificering är uppgiften att identifiera vilken klass ett tidigare osett objekt tillhör. Att klassificera bilder är en vanligt förekommande uppgift hos företag. För närvarande utför många av dessa företag klassificering manuellt. Automatiserade klassificerare har en lägre förväntad nogrannhet. I detta examensarbete studeradas hur en maskinklassificerar kan förbättras genom att lägga till ytterligare förändrad data i inlärningsprocessen av klassificeraren. Vi genomför en kvantitativ empirisk studie om effekterna av två bildförändringar, slumpmässiga horisontella/vertikala speglingar och slumpmässiga rotationer (<180◦). Bilddatasetet som används är från ett auktionshus sökmotor under det kommersiella namnet Barnebys. De dataseten som används består av tre separata dataset, 700 000, 50 000 och 28 000 bilder. Var och en av dataseten innehåller 28 klasser vilka mappas till verksamheten. I det här examensarbetet har vi tränat Inception-v3-modellen med dataset av storlek 700 000 och 50 000. Vi utvärderade sedan noggrannhet av de tränade modellerna genom att klassificera 28 000-datasetet. För att få ett mer exakt värde av effekterna använde vi en tiofaldig korsvalideringsmetod. Resultatet av vår kvantitativa studie visar att Inceptionv3-modellen kan nå en genomsnittlig noggrannhet på 64,5% (700 000 dataset) och en genomsnittlig noggrannhet på 51,1% (50 000 dataset). Den övergripande noggrannheten minskade med förändringar på vårat dataset. Dock visar våra resultat en ökad noggrannhet i vissa klasser. Den observerade högsta noggrannhetsökningen var i klassen Åhine & Spirits", där vi gick från 42,3 % korrekt klassificerade bilder till 72,7 % korrekt klassificerade bilder i det lilla datasetet med förändringar.

Tracking algorithms provide the model to recognize objects’ motion in the past. Moreover, applied to an artificial intelligence algorithm, these algorithms allow, to some degree, the capacity to forecast the future position of an object. This thesis uses deep learning algorithms to predict the ball’s position in the three-dimensional (3D) Cartesian space given the players’ motion and referees on the 2D space. The algorithms implemented are the encoder-decoder attention-based Transformer and the Inception Time, which is comprised of an ensemble of Convolutional Neural Networks. They are compared to each other under different parametrizations to understand their ability to capture temporal and spatial aspects of the tracking data on the ball prediction. The Inception Time proved to be more inconsistent on different areas of the pitches, especially on the end-lines and corners, motivating the decision to choose the Transformer network as the optimal algorithm to predict the ball position since it achieved less volatile errors on the pitch. / Spårningsalgoritmer möjliggör för modellen att känna igen objekts tidigare rörelser. Dessutom om tillämpad till en Artificiell intelligensalgoritm, de tillåter till viss mån att prognostisera ett objekts framtida position. Detta examensarbete använder djupinlärningsalgoritmer för att förutsäga bollens position i det tredimensionella (3D) kartesiska utrymmet baserat på spelarnas och domarnas rörelse i 2D-rymden. De implementerade algoritmerna är den kodare-avkodare-uppmärksamhetsbaserade Transformer och Inception Time, som består av en sammansättning faltningsnätverk (CNN). De jämförs med varandra med olika parametriseringar för att se deras förmåga att fånga upp tidsmässiga och rumsliga aspekter av spårningsdata för att förutsäga bollens rörelse. Inception Time visade sig vara mer inkonsekvent på olika områden på planen. Det var extra tydligt på slutlinjerna och i hörnen. Det motiverade beslutet att välja Transformer-nätverket som den optimala algoritmen för att förutsäga bollpositionen, eftersom den resulterade i färre ojämna fel på planen.

The widespread application of pressurized SF6 gas and its mixtures as insulating medium in many electric power applications is the result of recent advances in technologies. The likelihood of failure for a Gas Insulated Substation or Transmission Line (GIS/GITL) is primarily due to the presence of defects inside the equipment. Defects can be introduced into the GIS/GITL system for various reasons. Partial discharge (PD) is a natural phenomenon occurring in the GIS/GITL systems, which invariably contains defects. During commissioning or routine tests PD measurements serve to identify the type and status of a defect. Of particular interest for this research work will be the critical review of PD measurement for different types of free conducting particles in the gas using the UHF method due to its superiority among others. The work highlights the integrity of the method as a tool for both commissioning and routine tests and its alignment with the high voltage SF6 test standards is reviewed. 80/20 N2/SF6 mixture is used to reduce the surface roughness effect in pure SF6, as well as for the reduction of economical and environmental risks.