Energy-Based Models (EBMs) are a family of unsupervised machine learning models that associate each point in the input space with an energy value in which low energy indicates a high likelihood. Specifically, an EBM can be viewed as an unnormalized probabilistic model, which upon normalization, gives rise to the probability density function of data. The main difficulty in learning EBMs lies in the computation of the normalization constant, or the partition function, a task known to be intractable in general. Several approaches have been proposed to avoid or overcome this difficulty, including Maximum Likelihood Estimation (MLE) with Markov chain Monte Carlo (MCMC), Score Matching (SM), Noise Contrastive Estimation (NCE), and so on. This thesis studies the learning of EBMs using NCE. Briefly, in NCE, the EBM learning problem is converted to learning a binary classifier, which aims to distinguish the real data from fake data drawn from a noise distribution. This process allows the learning of the energy function in the EBM to bypass a direct estimation of the partition function and a certain theoretical guarantee is available under some assumptions in some asymptotic limit. Despite the nice theoretical properties of NCE, in this work, we show that learning EBMs using NCE entails significant practical limitations. Specifically, there appears a tension between the quality of the learned model and the computational efficiency, due to which we must sacrifice one to achieve the other. We establish these limitations via empirical studies as well as a theoretical analysis based on a simple “Gaussian data learning problem”. Our analysis inspires a revised NCE scheme, Adaptive Noise Contrastive Estimation (ANCE), to overcome these limitations. Empirically, we show that ANCE achieves a better quality-efficiency trade-off than the standard NCE in some regimes.

Le procédé de fabrication par fusion de poudre à l'aide d'un faisceau d'électrons est appelé procédé Electron Beam Melting (EBM). Il permet la fabrication de pièces métalliques à partir de poudres. Grâce au niveau de qualité (géométrique et mécanique) des pièces produites, le procédé peut être utilisé afin de produire des pièces fonctionnelles et non plus uniquement des prototypes. Ce procédé, ainsi que les autres procédés additifs métalliques, permettent d'envisager le passage de l'impression 3D à la fabrication additive métallique.L'utilisation de la fabrication additive dans un contexte industriel impose le respect de critères en termes de qualité, coût et délai des pièces produites. L'ensemble des étapes numériques de mise en production d'une pièce constitue la chaîne numérique. Cette dernière a un impact fort sur l'ensemble de ces trois critères. Ainsi, cette thèse apporte une réponse à la question suivante :Comment la fabrication assistée par ordinateur peut-elle améliorer le triptyque qualité, coût, délai du procédé de fabrication EBM?Le problème est abordé par la question sous-jacente suivante :Quelles caractéristiques doit posséder un environnement de fabrication assistée par ordinateur adapté au procédé EBM ?Pour répondre à cette question, la chaîne numérique actuelle est analysée . Les principales limites identifiées sont :- l'utilisation de fichiers au format STL- l’impossibilité d’optimiser le procédé à différentes échelles- l’impossibilité de simuler le procédé EBMAfin de résoudre l'ensemble des problèmes énoncés, un environnement de FAO est proposé. Celui-ci permet de centraliser l'ensemble des opérations de mise en production au sein d'un environnement unique. Il autorise le travail avec l'ensemble des formats de fichiers reconnus comme les formats natifs des logiciels de CAO ou le format STEP. Des développements informatiques permettent de concrétiser l’environnement proposé.L'implémentation de l'environnement de FAO a mis en évidence le rôle fondamental de la simulation au sein de celui-ci. Il a donc fallu répondre à la question :Comment obtenir une simulation du procédé EBM permettant sa mise au point hors ligne en temps raisonnable ?Bien que la simulation du procédé EBM est largement traitée dans la littérature scientifique, les études proposées reposent sur la méthode des éléments finis et le temps de calcul nécessaire n'est pas compatible avec une utilisation au sein d'un environnement de FAO. Un type de simulation alternatif a donc été créé : une simulation par abaques. Elle est constituée d’une simulation par la méthode des éléments finis qui permet d'obtenir des cartes de températures pour des cas de chauffes et de refroidissements standards. Ces cartes de températures sont ensuite transformées en abaques. La simulation par abaques est vue comme la succession d'une multitude de cas standards. Ainsi l'algorithme de simulation par abaques cherche l'abaque le plus proche de la situation simulée, afin d’estimer les températures au pas de temps suivant.Cette méthode de simulation a permis une réduction des temps de calcul tout en gardant une précision suffisante pour pouvoir être utilisée pour optimiser les paramètres de fabrication.Grâce à une telle simulation, un outil d'optimisation des stratégies de fusion a pu être créé. Il permet d’améliorer la qualité des pièces produites en calculant des stratégies de fusion respectant certains critères thermiques.Les apports majeurs de ces travaux de thèses sont :- l'établissement d'un cahier des charges pour une chaîne numérique performante en EBM- le développement d'un environnement de FAO adapté au procédé EBM- la mise au point d'une simulation rapide du procédé EBM basée sur des abaques- la création d'un outil d'optimisation des stratégies de fusion / The Electron Beam Melting (EBM) process allows to build metallic parts from powder. Thanks to the geometric and mechanical quality of the parts produced, the EBM process can be used to manufacture functional parts and not only prototypes. This process, with other additive metallic processes, make it possible to consider a transition from 3D printing to metallic additive manufacturing.The use of additive manufacturing in an industrial environment requires compliance with quality, cost and time criteria for the parts produced. The production of manufactured parts involves a series of numerical stages which is called the numerical chain. The numerical chain has a significant impact on the three criteria mentioned above. Thus, this thesis provides an answer to the following question:How Computer Aided Manufacturing can improve the quality, cost and time of the EBM manufacturing process?This problem is addressed through the following underlying question:What are the required characteristics for a Computer Aided Manufacturing system adapted to the EBM process?In order to answer this question, the current numerical chain is analyzed. Three main limitations are found:- the use of STL files format- the process cannot be optimized at different scales- the process cannot be simulatedTo solve these issues, a CAM environment is proposed. It allows the centralization of all numerical operations in a single environment. All supported formats can be used within this environment, such as native CAD file formats or STEP format. Software developments are done to prove the feasibility of such an environment.The CAM environment implementation reveals the crucial role of simulation in this system. It is therefore necessary to answer this second question:How to obtain an EBM process simulation allowing the development of parameters, virtually?Although EBM simulation is a recurrent subject in scientific literature, existing studies are based on the finite elements method but the calculation time needed is too important to be used in an CAM environment. Thus, an alternative type of simulation is created in this thesis: a simulation based on abacus. It is composed of a finite elements model, that allows heat maps generation for standards cases of heating and cooling. These heat maps are then transformed in abacus. The simulation algorithm based on abacus search the nearest abacus from the simulated situation in order to estimate the temperatures at the next time step.This simulation method was used to reduce the calculation time while keeping a sufficient precision to optimize process parameters.With the simulation based on abacus, a tool for the optimization of melting strategies is developed. This tool allows quality improvement for the produced parts through the calculation of melting strategies according to thermic criteria.To summarize, the main contributions of this work are:- the definition of requirements specifications of a powerful numerical chain for the EBM process- the development of a CAM environment adapted to the EBM process- the proposal of a fast simulation for the EBM process, based on abacus- the development of a tool for the optimization of melting strategies

Simple one-dimensional heat balance equations have been used to understand climate concepts since the 1960s, when a class of models was developed known as energy balance models (EBMs). EBMs use the growth or loss of polar surface ice as a climatic feedback, giving rise to surprisingly complex non-linear behaviours. One aspect of EBMs that has been relatively poorly examined is the effects of feedbacks caused by the other two phases of water in Earth’s climate other than ice: water clouds and water vapour. Cloud and water vapour play a critical role in the energy balance of Earth’s climate, and yet are some of the least well understood elements of the global climate system. This thesis explores the behaviour and interrelationships of climatic feedbacks caused by water in all three phases as it exists in the climate: surface ice caps, water vapour, and liquid water clouds. A two-layered EBM was modified with parameterizations of water vapour and liquid water clouds in order to conduct experiments. Three variants of the model were produced, each with progressively more water feedbacks than the last: a 1 phase model (with only surface ice feedback), a 2 phase model (with surface ice and water vapour) and a 3 phase model (with surface ice, water vapour, and cloud). The models were found to give generally realistic results, but with an underestimation of water vapour density, which in turn reduced the generated cloud fraction in the 3 phase model. Thus, the impacts of these extra feedbacks were likely to be underestimated in the analysis in general. The sensitivity of the model to several prognostic variables was studied by observing the changes in the model to a range of each variable. The 3 phase model was less sensitive to changes to the solar constant, S0, which measures incoming solar radiation, than the 1 phase model. This was probably caused by cloud reflecting and absorbing some radiation from the sun that would have otherwise reached the surface, changing the ratio of atmospheric heat transport to surface heat transport from 2.4953 for the 1 phase model to 2.0626 for the 3 phase model. Changing surface and ice albedo values resulted in changes in the model’s stability. The model was found to be insensitive to changes in surface humidity that drives the amount of water vapour the system has available, due to underestimation of water vapour in the model. The stability of the model was examined, and the 1 phase model was found to respond faster to changes in S0 than the 3 phase model. The model was tested for hysteresis, which was confirmed for all three model variants. The 1 phase model showed less stability then the 3 phase model as S0 was increased, but both models were similarly stable as S0 was decreased.

This thesis explores the possibilities with additive manufacturing, applied to the engineering of high performance bike parts. The goal was to study the technique and materials that make up the additive manufacturing in order to apply this to bike stems. Also, the goal was to develop both a physical and a virtual model to further evaluate the possibilities with additive manufacturing. Knowledge of the additive manufacturing processes was gained through an extensive information study. After that, a focus on a particular method, EBM, electron beam melting, was made. The process of development of a bike part, the steering stem, was documented to identify the strength and weaknesses of the technique and to finally evaluate it to a traditionally manufactured reference stem. The design process started with a state-of-the-art study and concept stage to later be followed with an iterative process of modelling and simulating. The concepts were created with accordance of the constraints set upon them from our technical specification and the European standards regarding bike parts which limits the maximum allowed deformation and fatigue. One concept was selected to be modelled as a physical component. Finally, the developed component is discussed and compared to a chosen reference stem. The final concept did not outperform the reference in weight, but valuable insight and knowledge were gained along the way. The main conclusion of the thesis is that additive manufacturing is a suitable tool for manufacturing of high performing bike parts. Suggestions for future work are also given, where a further analysis with other materials using the EBM-technique is suggested. / Detta kandidatsexamensarbete utforskar möjligheterna med additiv tillverkning applicerat på högpresterande cykelkomponenter. Målet med arbetet var att utforska möjligheterna att utnyttja additiv tillverkning för styrstammar, samt ta fram en virtuell och en fysisk modell av en styrstam anpassad för detta. Genom en omfattande infosökning hämtades kunskap in om de tekniker och material som utgör den additiva tillverkningsgruppen. Därefter gjordes en fördjupning i en specifik teknik, nämligen EBM, electron beam melting. Produktutvecklingsprocessen för cykelkomponenten, en styrstam, dokumenterades för att identifiera styrkorna och svagheterna hos tekniken och för att i slutet utvärderas mot en traditionellt tillverkad referenskomponent. Designprocessen inleddes med infosökning och konceptgenerering för att sedan, med hjälp av digital mjukvara såsom Solid Edge och ANSYS, övergå till en iterativ process av modellering och simulering. Koncepten modellerades efter önskade egenskaper definierat i en kravspecifikation samt rådande EU-standard för styrstammar i tillåten deformation och utmattningsbrott. Ett slutgiltigt koncept valdes och sedan tillverkades en 3D-utskriven fysisk prototyp. I den avslutande delen diskuteras den utvecklade komponenten och jämförs med en vald referenskomponent. Det slutgiltiga konceptet lyckades inte prestera bättre än referenskomponenten i vikt. Men däremot erhölls värdefull insikt och kunskap angående den additiva processen. Slutsatsen som drogs var att additiv tillverkning är en legitim tillverkningsmetod för konstruktion av högpresterande cykelkomponenter. Förslag ges även för framtida arbete där framförallt en vidare analys med andra material som utnyttjar EBM-tekniken föreslås.

Background Evidence-based medicine (EBM) involves the care of patients using the best available evidence from the results of good quality clinical research to guide clinical decision making 1 – 3. By incorporating the principles of Evidence-based Medicine (EBM), the family practitioner would be able to treat a patient according to the best clinical research available. This principle is implemented widely in the USA, Canada, the United Kingdom and Europe. In South Africa, however, EBM is not yet as widely incorporated into family practice. This is so despite the plethora of websites available to practitioners and the relative ease with which applicable research evidence can be found. Very few published studies are available regarding EBM or Evidence–based Practice (EBP) in the South African context. The findings of this study would thus highlight reasons and/ or barriers preventing family practitioners from implementing EBM in their respective practices. This could also lead to further research into possible methods of implementation of EBM into South African family practices. Aim: The aim of the study was to describe the perceptions, knowledge and attitudes of private practitioners regarding evidence based practice and to identify the barriers encountered in evidence based practice. Methods A questionnaire survey of general practitioners in Gauteng, South Africa, was conducted. Questionnaires were distributed to a random sample of practitioners in the Gauteng region. Two hundred and twenty one (221) practitioners participated in the survey and responded to questionnaires mailed to them. The questionnaire was mailed, faxed or e-mailed to the practitioners, which they then completed and returned for statistical analysis. Study design The study design is that of quantitative, statistical analysis (descriptive cross-sectional survey). Setting General practitioners were randomly selected from a list of practitioners in the Gauteng Province. Doing a nationwide survey would have been a mammoth undertaking. It was therefore decided to limit the research to one province and therefore it was only concentrated on practitioners practicing in the Gauteng area. Results It is interesting to note that of the two hundred and twenty one participants in this study; only 10% of the practitioners were against using EBM in their practices. This, however, stands in stark contrast to the 56% of practitioners who do not implement EBM in their practices or make use of the EBM principle at all. The major barriers preventing practitioners from implementing EBM is depicted in the following graph: Lack of time and the training in aspects of Evidence-based medicine were the main barriers preventing the full scale implementation of EBM in family practices in Gauteng. Conclusion Participating Gauteng doctors were in principle, very positive towards the implementation of EBM in their respective practices. Most of the participants agreed that EBM would benefit their patients’ care and treatment. Very few of the participants, however, make use of EBM in practice. A lack of training and time constraints were the main barriers with regards to the implementation of EBM. Proper training of medical students at undergraduate level at faculties of health sciences, would go a long way assisting prospective doctors in mastering the concept of EBM and increasing their overall awareness of EBM. Further definitive research would assist in establishing whether such awareness would be associated with improved implementation of evidence in the form of evidence based guidelines in practice.

Vid additiv tillverkning med EBM-metoden har hanteringen och återcirkulationen av titanpulver en betydande roll för effektivitet, kvalitet och spårbarhet i tillverkningsprocessen. I takt med att additivt tillverkade produkter fått större marknadsandelar och används i alltmer ingenjörsmässiga applikationer ställs högre krav på tillverkningsprocesserna och därmed pulverhanteringen.   Detta projekt initierades då företaget AIM Sweden såg behovet av att införa ny utrustning i form av en pulverhanteringsstation, med syftet att erhålla en mer effektiv och spårbar pulverhantering vid deras EBM tillverkning med maskinerna Arcam EBM Q10- och Q20plus.   Projektet har genomförts som ett examensarbete för maskiningenjörsutbildningen vid Mittuniversitetet. Målet var att ta fram ett konstruktionsunderlag för en utrusning som uppfyllde identifierade krav och önskemål från uppdragsgivaren AIM Sweden och de närmsta intressenterna för projektet.   Utrustningen togs fram med produktutvecklingsmetodik i fem projektfaser där framstegen från varje milstolpe redovisades för uppdragsgivaren under projektets gång.   Resultatet av projektet bestod i ett konstruktionsunderlag av en pulverhanteringsstation med funktioner för att väga, transportera och homogenisera titanpulver. Stationen skapar förutsättningar för en mer effektiv tillverkningsprocess med förenklad spårbarhetsdokumentation. Detta kan bidra till möjlighet att leverera produkter till fler typer av kunder och en ökad lönsamhet för AIM Sweden.   Innan utrustningen kan tillverkas krävs tester och vidare arbete i form av val av standardkomponenter, konstruktion av drivningssystem samt en mer omfattande ekonomisk kalkyl. / <p>Betyg 2020-07-14</p>

The beekeepers need to inspect their hives regularly in order to protect them from various stressors. Manual inspection of hives require a lot of time and effort. Hence, many researchers have started using electronic beehive monitoring (EBM) systems to collect critical information from beehives, so as to alert the beekeepers of possible threats to the hive. EBM collects information by applying multiple sensors into the hive. The sensors collect information in the form of video, audio or temperature data from the hives. This thesis involves the automatic classification of audio samples from a beehive into bee buzzing, cricket chirping and ambient noise, using machine learning models. The classification of samples in these three categories will help the beekeepers to determine the health of beehives by analyzing the sound patterns in a typical audio sample from beehive. Abnormalities in the classification pattern over a period of time can notify the beekeepers about potential risk to the hives such as attack by foreign bodies (Varroa mites or wing virus), climate changes and other stressors.

No description available.

The work presented in this dissertation is concerned with the microstructures andsurface roughness of test slabs of Ti-6Al-4V produced by one of powder based AdditiveManufacturing (AM) technique namely Electron Beam melting (EBM). The effects of processparameters of a EBM system and geometry factors of a EBM build such as slabs’ thickness andheight etc. on the microstructure and the surface roughness of the EBM produced Ti-6Al-4Vhave being investigated. The processing parameters of the EBM system involved in the presentwork include beam current, scan speed, offset focus and scanning length etc. In this study threedifferent batches of samples were prepared. Microstructures of EBM built Ti-6Al-4V werestudied using Optical Microscopy (OM), Scanning Electron Microscopy (SEM), TransmissionElectron Microscopy (TEM), and X-ray diffraction (XRD). Confocal microscopy and imageanalysis codes ImageJ’s routine SurfCharJ were used to quantify the surface roughness of the testslabs. The microstructures of EBM built Ti-6Al-4V in general consist of columnar grains of priorβ phase. Inside the columnar grains there is a typical (α+β) microstructure of titanium alloyscontaining Widmanstätten α platelets and rod-shaped β phase. Grain boundary α layer has alsobeen observed on the grain boundaries of prior b columnar grains. By using TEM, the β phase inEBM built Ti-6Al-4V has been identified as a rod-like structure located on the grain boundariesof the fine a grains and often grows up along the build direction. The size of the β rods is about200nm and the distance between the β rods is ranged between 0.5 -2μm. Chemical compositionsfor different phases have been measured by TEM/EDX and volume fraction of the β phase in theEBM Ti-6Al-4V has been determined to be 2.7%. The phase transformation sequence in EBMbuilt Ti-6Al-4V has been discussed according to processing history and microstructuresobserved. It has been observed that, the size, number and geometry of prior β columnar graindepend on the sample thickness and other process parameters setting. The diameter of columnargrain varies between 2-70μm. The increase in size, number and regularity of columnar grains hasbeen observed with increase in sample thickness, beam energy density and scanning length.While with increase in height of the build it decreases. The length and smoothness of α plateletsincreases with increase in diameter of prior β columnar grain. The β phase rods are unaffected bysample thickness and process parameter settings. The sample thickness and beam energy densityhas a strong effect on the surface roughness of the test slabs. The value of surface roughnesscoefficient Ra for different test slabs varies between 1-20μm. It is observed that the surfaceroughness increases of the test slab increases with increase in thickness of the slab and beamenergy density. The possible reasons for these variations in microstructures and surfaceroughness have been discussed. / <p>All papers in dissertation as manuscripts.</p>

I detta   projekt har en studie av krympning vid additiv tillverkning i materialet   Ti6Al4V gjorts, detta för att analysera hur utskrifter i olika storlekar och   riktningar påverkas. Arbetet har genomförts av två studenter vid   Mittuniversitetet i Östersund och görs på uppdrag av universitetets   forskningssida. I projektet har förstudier gjorts för att lägga en grund för   arbetet. Under förstudierna hittades information om att krympningen kunde   minskas med hjälp av att använda stödmaterial i utskrifterna. Design för de   detaljer som ska skrivas ut har valts ut med hjälp av tidigare forskning där   trappor varit rekommenderat för att lätt kunna avgöra vad som händer i olika   delar av en utskrift. Även detaljer för mätning av vassa kanter och   avrundningar skrevs ut för att kunna kontrollera dessa om tid blev över men   dessa mättes aldrig då fokus under hela detta arbetet legat på trapporna.   Trapporna skrevs ut i tre olika storlekar. De skrevs ut både med och utan   stöd och i både liggande och stående led i en EBM-maskin, Arcam S12. När de   blivit utskrivna mättes de med hjälp av mikroskopet “Dino-Light edge Digital   Microscope”. Informationen som kommit fram ur dessa mätningar har sedan lagts   in i ett Excel dokument där tabeller och grafer tagits fram för att göra det   lätt att uppfatta tendenser. De utmärkande tendenser som upptäcktes var att   en större krympning alltid inträffade vid det första och tredje steget i en   utskrift. Detta inträffade oavsett vilken storlek av trappor som   kontrollerades och kunde undvikas med hjälp av att använda stödmaterial. I   modellerna med stödmaterial kunde inte speciella tendenser synas och den   krympning som syns i dessa mätningar kan bero på fel från den mänskliga   faktorn vid mätningen. Trenderna är tydliga i de detaljer som skrivits ut   utan stödmaterial. Detta fenomen kan bero på att materialet i skrivplattan   som finns i skrivaren inte är detsamma som Ti6Al4V, vilket kan ha en inverkan   på det utskrivna materialet i de lagren som är närmast skrivplattan. Krymp i   detaljerna beror också på att pulvret i skrivaren blir mindre då det smälter   vilket gör att varje lager inte får förväntad/ önskad tjocklek. Detta är ett   problem som följer med genom hela utskriften oavsett om stödmaterial används   eller ej. Arbetet som gjorts anses lyckat. / This project is a about studying and anylazing shrinkage during the   use of a EBM-machine in the material Ti6A14V. The study is done to anaylze   how prints of diffrent size and printing direction are affected by the   shrinkage. The project is done by two students at Mid Sweden University in   Östersund and the project is requested by the research department of the   university. Research has been made before the start of the project to give   the students a solid understanding in the matter. During the research for   information it was found that the shrink could be minimized by using   supportmaterial in the print. The design for the parts that was going to be   printed have been chosen through earlier studies of the matter where stairs   was recommended to easy be able to see what goes on in the different parts of   a print. Some details was also created so that measurements considering sharp   edges and round details could be checked for deformations if there was enough   time. Those were never measured since all the time of this project was   focused around the stairs. The stairs were printed in three different sizes.   They were also printed with and without supportmaterial and both standing up   and lying down in the printer EBM Arcam S12. When the parts had been printed   they were measured with a microscope called “Dino-Light edge Digital   Microscope”. The information gotten from these measurements were put into a   Microsoft Excel document where tables and graphs were created to make it easy   to spot the tendencies of the prints. The tendencies that were noticeable was   that the first and third step of every print had more shrink than the others.   This occurred no matter the size the stairs were but did not occur in the   parts printed with supportmaterial. No tendencies were noticed in the prints   with supportmaterial and the shrinkage that was noted in these prints might   also be because of the human factor. The trends are clearly visible in the   parts printed without supportmaterial. This might be because of the different   material in the printing plate that is stainless steel instead of Ti6Al4V.   This might cause problem for the layers that are printed closest to the   plate. The reason for shrinking in the details is also because of the fact   that the powder that is being printed gets melted and therefor shrinks. This   results in a thicker layer of powder everytime the powder is applied to the   printing surface. This is a problem that goes on throughout the whole print   no matter if supportmaterial is used or not. The work completed is considered to be well done. / <p>Betyg: 180809</p>