Dissertations / Theses on the topic 'Deep structures'
To see the other types of publications on this topic, follow the link: Deep structures.
Create a spot-on reference in APA, MLA, Chicago, Harvard, and other styles
Consult the top 50 dissertations / theses for your research on the topic 'Deep structures.'
Next to every source in the list of references, there is an 'Add to bibliography' button. Press on it, and we will generate automatically the bibliographic reference to the chosen work in the citation style you need: APA, MLA, Harvard, Chicago, Vancouver, etc.
You can also download the full text of the academic publication as pdf and read online its abstract whenever available in the metadata.
Browse dissertations / theses on a wide variety of disciplines and organise your bibliography correctly.
1
Lambert, C. P. "Multimodal segmentation of deep cortical structures." Thesis, University College London (University of London), 2012. http://discovery.ucl.ac.uk/1344055/.
Full textAbstract:
The organisation of the human cortex is characterised by macroscopically defined areas consisting of functionally distinct subunits, each connected to an array of local and distant targets forming distinctive networks. Classically, these structures have been parcellated according to ex vivo cytochemical and connectivity properties. However, the emergent flaw with this approach is the presence of significant inter-hemispheric and inter-individual anatomical variability. By exploiting several MRI modalities, a similar approach to sub-regional parcellation can be applied in vivo across large numbers of individuals. Using diffusion tensor imaging (DTI), probabilistic tractography can be used to generate a representation of the white matter pathways originating from or passing through a single voxel. By quantifying the degree of similarity between different tract distributions, regional parcellation can be achieved through several algorithms. These have previously been used on regions such as the thalamus and basal ganglia. However, due to computational limitations, it is normal practice to apply dimension reduction tactics prior to parcellation, thereby generating an upper bound on the degree of accuracy that can be achieved. I have set out to further this pre-existing framework by developing methods to analyse and cluster massive matrices without down-sampling data, thereby generating a prior free, bottom-up approach to regional parcellation based on regional connectivity. I have applied this approach to several areas including the sub-thalamic nucleus, amygdala and human brainstem. Several fundamental properties and limitations of the technique are revealed, and additional methods developed to further improve the white matter parcellation. This includes a novel method of multichannel segmentation, which was applied to the human brainstem and cortex. The new tissue classes were used both for quantitative analysis, and also to improve DTI based segmentation. Throughout, the findings are extrapolated to examine a variety of neuropathological scenarios, including symptom networks, pre-clinical diagnosis and therapeutic interventions such as deep brain stimulation.
2
Xu, Yuan. "Statistical shape analysis for deep brain structures." Diss., Restricted to subscribing institutions, 2008. http://proquest.umi.com/pqdweb?did=1581917061&sid=11&Fmt=2&clientId=1564&RQT=309&VName=PQD.
Full text
3
Billingsley, Richard John. "Deep Learning for Semantic and Syntactic Structures." Thesis, The University of Sydney, 2014. http://hdl.handle.net/2123/12825.
Full textAbstract:
Deep machine learning has enjoyed recent success in vision and speech-to-text tasks, using deep multi-layered neural networks. They have obtained remarkable results particularly where the internal representation of the task is unclear. In parsing, where the structure of syntax is well studied and understood from linguistics, neural networks have so far not performed so well. State-of-the-art parsers use a tree-based graphical model that requires a large number of equivalent classes to represent each parse node and its phrase label. A recursive neural network (RNN) parser has been developed that works well on short sentences, but falls short of the state-of-the-art results on longer sentences. This thesis aims to investigate deep learning and improve parsing by examining how neural networks could perform state-of-the-art parsing by comparison with PCFG parsers. We hypothesize that a neural network could be configured to implement an algorithm parallel to PCFG parsers, and examine their suitability to this task from an analytic perspective. This highlights a missing term that the RNN parser is unable to model, and we identify the role of this missing term in parsing. We finally present two methods to improve the RNN parser by building upon the analysis in earlier chapters, one using an iterative process similar to belief propagation that yields a 0.38% improvement and another replacing the scoring method with a deeper neural model yielding a 0.83% improvement. By examining an RNN parser as an exemplar of a deep neural network, we gain insights to deep machine learning and some of the approximations it must make by comparing it with well studied non-neural parsers that achieve state-of-the-art results. In this way, our research provides a better understanding of deep machine learning and a step towards improvements in parsing that will lead to smarter algorithms that can learn more accurate representations of information and the syntax and semantics of text.
4
Olowe, Adedayo Christianah. "Corrosion assessment and cathodic protection design parameters for steel structures in deep and ultra deep offshore waters." Thesis, University of Aberdeen, 2013. http://digitool.abdn.ac.uk:80/webclient/DeliveryManager?pid=201965.
Full textAbstract:
Carbon steel finds much application for use in industries including civil; manufacturing; oil and gas; as well as, renewable energy. Common examples for usage of steel include water pipelines; oil pipelines; bridges; etc. The main advantages of steel over other engineering materials are its strength and affordability. However, steel undergoes corrosion which is a degradation mechanism that occurs as a result of the electrochemical interaction between steel and its environment. There are two main options to control corrosion, aside from material selection techniques, namely, the use of protective coating systems to isolate the steel from the environment; or the use of cathodic protection. Cathodic protection involves the use of galvanic anodes or impressed current system to prevent steel corrosion. Currently the oil and gas industry accounts for the major share of consumption of galvanic anodes for the protection of steel in engineering applications. Recent incursions into deep water depths by the Oil and Gas industry in the last decade or so has brought to the fore the need to understand better the performance of steel at deep and ultra deep water depths; as well as to develop an understanding of how cathodic protection works at these water depths. So far, the bulk of industry experience lies in shallow waters and current international cathodic protection design guidelines are based on data collated at these shallow water depths. It is the objective of this research work to assess the corrosion properties of steel with deep seawater parameters and determine design current density requirements for effective cathodic protection of steel at deep and ultra deep water depths offshore.
5
Grice, James Robert. "Prediction of extreme wave-structure interactions for multi-columned structures in deep water." Thesis, University of Oxford, 2013. http://ora.ox.ac.uk/objects/uuid:dd7320c1-7121-4ea7-827f-527af9405e9a.
Full textAbstract:
With a continuing and rising demand for hydrocarbons, the energy companies are installing infrastructure ever further offshore, where such infrastructure is often exposed to extreme waves. This thesis explores some aspects of wave-structure interaction, particularly the maximum water surface elevation increase in severe storms due to these local interactions. The effects on wave-structure interactions of column cross-sectional shape are investigated using linear and second-order wave diffraction theory. For multi-column structures, the excitation of locally resonant wave modes (near-trapping) is studied for several column cross-sectional shapes, and a simple method for estimating the surface elevation mode shape is given. The structure of the quadratic transfer functions for second-order sum wave elevation is investigated and an approximation assuming these QTFs are flat perpendicular to the leading diagonal is shown to be adequate for the first few lowest frequency modes. NewWave-type focused wave groups can be used as a more realistic model of extreme ocean waves. A Net Amplification Factor based on the NewWave model is given as an efficient tool for finding the incident frequencies most likely to cause a violent wave-structure interaction and where these violent responses are likely to occur. Statistics are collected from Monte Carlo type simulations of random waves to verify the use of the Net Amplification Factor. Going beyond linear calculations, surface elevation statistics are collected to second-order and a `designer' wave is found to model the most extreme surface elevation responses. A `designer' wave can be identified at required levels of return period to help to understand the relative size of harmonic components in extreme waves. The methods developed with a fixed body are then applied to an identical hull which is freely floating, and the responses between the fixed and moving cases are compared. The vertical heave motion of a semi-submersible in-phase with the incident wave crests is shown to lead to a much lower probability of water-deck impact for the same hull shape restrained vertically. The signal processing methods developed are also applied to a single column to allow comparison with experimental results. Individual harmonic components of the hydrodynamic force are identified up to at least the fifth harmonic. Stokes scaling is shown to hold even for the most violent interactions. It is also shown that the higher harmonic components of the hydrodynamic force can be reconstructed from just the fundamental force time history, and a transfer function in the form of a single phase and an amplitude for each harmonic. The force is also reconstructed well to second-order from the surface elevation time history using diffraction transfer functions. Finally, possible causes of damage to a platform high above mean water level in the North Sea are investigated.
6
Dikdogmus, Halil. "RISER CONCEPTS FOR DEEP WATERS." Thesis, Norges teknisk-naturvitenskapelige universitet, Institutt for marin teknikk, 2012. http://urn.kb.se/resolve?urn=urn:nbn:no:ntnu:diva-18528.
Full textAbstract:
Oil and gas exploration and production activities in deep and ultra deep waters in hostile environments necessitates the need to develop innovative riser systems capable of ensuring transfer of fluids from the seabed to a floating vessel and vice versa, with little or no issues with respect to influences of environmental loads and vessel motions.The design of the riser system must focus on different types of loading and load effects than for traditional water-depth. A variety of different riser concepts are proposed, both with respect to geometric shape and selection of materials.In the last few years, steel catenary risers have been a preferred riser solution for deep-water field developments due to its simple engineering concept, cost effective, flexibility in using different host platform and flexibility in geographical and environmental conditions. In this report, a case study considering a steel catenary riser operating in 1000 m water depth was conducted. The riser was subjected to extreme environmental conditions and static and dynamic response analyses were performed by the computer program RIFLEX.Last, parametric study is carried out to investigate the effects of parameter variation based on some parameters like current profiles, mesh density, wall thickness and so on. These parameters have significant effect on the structural response, especially in the touch down region.
7
Romagna, Pinter Patricia. "Reappraising the Numidian system (Miocene, southern Italy) deep-water sandstone fairways confined by tectonised substrate." Thesis, University of Aberdeen, 2017. http://digitool.abdn.ac.uk:80/webclient/DeliveryManager?pid=238534.
Full text
8
Oyallon, Edouard. "Analyzing and introducing structures in deep convolutional neural networks." Thesis, Paris Sciences et Lettres (ComUE), 2017. http://www.theses.fr/2017PSLEE060.
Full textAbstract:
Cette thèse étudie des propriétés empiriques des réseaux de neurones convolutifs profonds, et en particulier de la transformée en Scattering. En effet, l’analyse théorique de ces derniers est difficile et représente jusqu’à ce jour un défi : les couches successives de neurones ont la capacité de réaliser des opérations complexes, dont la nature est encore inconnue, via des algorithmes d’apprentissages dont les garanties de convergences ne sont pas bien comprises. Pourtant, ces réseaux de neurones sont de formidables outils pour s’attaquer à une grande variété de tâches difficiles telles la classification d’images, ou plus simplement effectuer des prédictions. La transformée de Scattering est un opérateur mathématique, non-linéaire dont les spécifications sont inspirées par les réseaux convolutifs. Dans ce travail, elle est appliquée sur des images naturelles et obtient des résultats compétitifs avec les architectures non-supervisées. En plaçant un réseau de neurones convolutifs supervisés à la suite du Scattering, on obtient des performances compétitives sur ImageNet2012, qui est le plus grand jeux de donnée d’images étiquetées accessibles aux chercheurs. Cela nécessite d’implémenter un algorithme efficace sur carte graphique. Dans un second temps, cette thèse s’intéresse aux propriétés des couches à différentes profondeurs. On montre qu’un phénomène de réduction de dimensionnalité progressif à lieu et on s’intéresse aux propriétés de classifications supervisées lorsqu’on varie des hyper paramètres de ces réseaux. Finalement, on introduit une nouvelle classe de réseaux convolutifs, dont les opérateurs sont structurés par des groupes de symétries du problème de classificationThis thesis studies empirical properties of deep convolutional neural networks, and in particular the Scattering Transform. Indeed, the theoretical analysis of the latter is hard and until now remains a challenge: successive layers of neurons have the ability to produce complex computations, whose nature is still unknown, thanks to learning algorithms whose convergence guarantees are not well understood. However, those neural networks are outstanding tools to tackle a wide variety of difficult tasks, like image classification or more formally statistical prediction. The Scattering Transform is a non-linear mathematical operator whose properties are inspired by convolutional networks. In this work, we apply it to natural images, and obtain competitive accuracies with unsupervised architectures. Cascading a supervised neural networks after the Scattering permits to compete on ImageNet2012, which is the largest dataset of labeled images available. An efficient GPU implementation is provided. Then, this thesis focuses on the properties of layers of neurons at various depths. We show that a progressive dimensionality reduction occurs and we study the numerical properties of the supervised classification when we vary the hyper parameters of the network. Finally, we introduce a new class of convolutional networks, whose linear operators are structured by the symmetry groups of the classification task
9
Astolfi, Pietro. "Toward the "Deep Learning" of Brain White Matter Structures." Doctoral thesis, Università degli studi di Trento, 2022. http://hdl.handle.net/11572/337629.
Full textAbstract:
In the brain, neuronal cells located in different functional regions communicate through a dense structural network of axons known as the white matter (WM) tissue. Bundles of axons that share similar pathways characterize the WM anatomy, which can be investigated in-vivo thanks to the recent advances of magnetic resonance (MR) techniques. Diffusion MR imaging combined with tractography pipelines allows for a virtual reconstruction of the whole WM anatomy of in-vivo brains, namely the tractogram. It consists of millions of WM fibers as 3D polylines, each approximating thousands of axons. From the analysis of a tractogram, neuroanatomists can characterize well-known white matter structures and detect anatomically non-plausible fibers, which are artifacts of the tractography and often constitute a large portion of it. The accurate characterization of tractograms is pivotal for several clinical and neuroscientific applications. However, such characterization is a complex and time-consuming process that is difficult to be automatized as it requires properly encoding well-known anatomical priors. In this thesis, we propose to investigate the encoding of anatomical priors with a supervised deep learning framework. The ultimate goal is to reduce the presence of artifactual fibers to enable a more accurate automatic process of WM characterization. We devise the problem by distinguishing between volumetric and non-volumetric representations of white matter structures. In the first case, we learn the segmentation of the WM regions that represent relevant anatomical waypoints not yet classified by WM atlases. We investigate using Convolutional Neural Networks (CNNs) to exploit the volumetric representation of such priors. In the second case, the goal is to learn from the 3D polyline representation of fibers where the typical CNN models are not suitable. We introduce the novelty of using Geometric Deep Learning (GDL) models designed to process data having an irregular representation. The working assumption is that the geometrical properties of fibers are informative for the detection of tractogram artifacts. As a first contribution, we present StemSeg that extends the use of CNNs to detect the WM portion representing the waypoints of all the fibers for a specific bundle. This anatomical landmark, called stem, can be critical for extracting that bundle. We provide the results of an empirical analysis focused on the Inferior Fronto-Occipital Fasciculus (IFOF). The effective segmentation of the stem improves the final segmentation of the IFOF, outperforming with a significant gap the reference state of the art. As a second and major contribution, we present Verifyber, a supervised tractogram filtering approach based on GDL, distinguishing between anatomically plausible and non-plausible fibers. The proposed model is designed to learn anatomical features directly from the fiber represented as a 3D points sequence. The extended empirical analysis on healthy and clinical subjects reveals multiple benefits of Verifyber: high filtering accuracy, low inference time, flexibility to different plausibility definitions, and good generalization. Overall, this thesis constitutes a step toward characterizing white matter using deep learning. It provides effective ways of encoding anatomical priors and an original deep learning model designed for fiber.
10
Yang, Yuzhe S. M. Massachusetts Institute of Technology. "On exploiting structures for deep learning algorithms with matrix estimation." Thesis, Massachusetts Institute of Technology, 2020. https://hdl.handle.net/1721.1/127319.
Full textAbstract:
Thesis: S.M., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, May, 2020Cataloged from the official PDF of thesis.
Includes bibliographical references (pages 113-118).
Despite recent breakthroughs of deep learning, the intrinsic structures within tasks have not yet been fully explored and exploited for better performance. This thesis proposes to harness the structured properties of deep learning tasks using matrix estimation (ME). Motivated by the theoretical guarantees and appealing results, we apply ME to study the following two important learning problems: 1. Adversarial robustness. Deep neural networks are vulnerable to adversarial attacks. This thesis proposes ME-Net, a defense method that leverages ME. In ME-Net, images are preprocessed using two steps: first pixels are randomly dropped from the image; then, the image is reconstructed using ME. We show that this process destroys the adversarial structure of the noise, while re-enforcing the global structure in the original image. Comparing ME-Net with state-of-the-art defense mechanisms shows that ME-Net consistently outperforms prior techniques, improving robustness against both black-box and white-box attacks. 2. Value-based planning and deep reinforcement learning (RL). This thesis proposes to exploit the underlying low-rank structures of the state-action value function, i.e., Q function. We verify empirically the existence of low-rank Q functions in the context of control and deep RL tasks. As our key contribution, by leveraging ME, we propose a generic framework to exploit the underlying low-rank structure in Q functions. This leads to a more efficient planning procedure for classical control, and additionally, a simple scheme that can be applied to any value-based RL techniques to consistently achieve better performance on "low-rank" tasks. The results of this thesis demonstrate the value of using matrix estimation to capture the internal structures of deep learning tasks, and highlight the benefits of leveraging structure for analyzing and improving modern learning algorithms.
by Yuzhe Yang.
S.M.
S.M. Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science
11
Diczfalusy, Elin. "Modeling and Simulation of Microdialysis in the Deep Brain Structures." Licentiate thesis, Linköpings universitet, Biomedicinsk instrumentteknik, 2012. http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-84277.
Full textAbstract:
Microdialysis is a method for monitoring of the local biochemical environment in a region of interest. The method uses a catheter, mimicking the function of a blood capillary, to sample substances from the surrounding medium through diffusion. A recent application for microdialysis is the sampling of neuroactive substances in the deep brain, or basal ganglia, during deep brain stimulation (DBS) for patients with Parkinson’s disease. The basal ganglia consist of nuclei interconnected by chemical synapses, and it is hypothesized that the levels of neurotransmitter substances around the synapses are affected by DBS treatment. In order to relate the microdialysis data to their anatomical origin and to the effects of DBS, it is suitable to estimate the tissue volume which is sampled during a microdialysis experiment. In this thesis, the maximum tissue volume of influence (TVImax) for a microdialysis catheter was simulated and evaluated using the finite element method (FEM), to allow interpretation of biochemical data in relation to anatomical structures. A FEM model for simulation of the TVImax for a microdialysis catheter placed in grey brain matter was set up, using Fick’s law of diffusion. The model was used to investigate the impact of the analyte diffusion coefficient (D), the tissue tortuosity (λ) and the loss rate constant (k) on the size of the TVImax by regression analysis. Using relevant parameter intervals, the radius of the TVImax of a neurotransmitter was estimated to 0.85 ± 0.25 mm. A microdialysis experiment on calf brain tissue showed agreement with the regression model. A heterogeneous anisotropic FEM model based on diffusion tensor imaging (DTI) showed that the radius of the TVImax may vary by up to 0.5 mm as a consequence of local tissue properties, which was reasonable in relation to the 95% confidence interval from the regression estimation. The TVImax was simulated and patient-specifically visualized in relation to MRI images for four patients undergoing microdialysis in parallel to DBS. The size of the TVImax showed to be relevant in relation to the basal ganglia nuclei, and the obtained microdialysis data indicated that the biochemical response to DBS depends on the catheter position. The simulations of the TVImax were combined with patient-specific DBS electric field simulations, for further interpretation of the results in relation to the effects of DBS. In conclusion, simulations and visualizations of the TVImax allowed relating microdialysis data to its anatomical origin. Detailed knowledge about the parameters affecting the microdialysis sampling volume is valuable for the current application as well as other applications related to the migration of analytes in tissue.Mikrodialys är en metod som används för studera lokala nivåer av biokemiska substanser i ett specifict organ eller struktur. Metoden använder sig av en kateter med ett semipermeabelt membran, över vilket utbyte av substanser sker genom diffusion. Mikrodialys har nyligen använts för att studera nivåer av neurotransmittorer i de djupa hjärnstrukturerna, ävan kallade basala ganglierna, under djup hjärnstimulering (DBS) för patienter med Parkinsons sjukdom. De basala ganglierna består av ett antal millimeterstora hjärnstrukturer, sammankopplade via biokemiska synapser, och nivåerna av signalsubstanser runt dessa synapser tros påverkas av DBS. För att relatera mikrodialysmätningarna till dess anatomiska ursprung, och till effekterna av DBS, är det önskvärt att få en uppskattning av den vävnadsvolym som påverkar mätningen från en mikrodialyskateter. Målet med denna licentiatavhandling har varit att simulera och utvärdera den maximala påverkansvolymen (TVImax) för en mikrodialyskateter med hjälp av finita element-metoden (FEM), för att underlätta tolkningen av de biokemiska data som samlats in. En FEM-modell sattes upp för att simulera TVImax för en kateter placerad i grå hjärnvävnad, baserat på Ficks diffusionslag och lämpliga rand- och initialvillkor. Modellen användes för att göra en regressionsanalys av hur TVImax påverkades av analytens diffusionskoefficient (D), hjärnvävnadens tortuositet (λ) och analytens nedbrytningshastighet (k), och radien för TVImax för en neurotransmitter uppskattades till 0.85 ± 0.25 mm då fysiologiskt relevanta parameterintervall användes. En experimentell studie av mikrodialys på hjärnvävnad från kalv gav god överensstämmelse med simuleringsresultaten. En heterogen och anisotrop FEM-modell sattes upp med hjälp av diffusionstensordata (DTI), vilket visade att lokala vävnadsegenskaper påverkar diffusionen av analyter i de basala ganglierna med upp till 0.5 mm i enighet med den regressionsmodell som tagits fram. TVImax simulerades och visualiserades sedan i relation till MRI-bilder för fyra patienter som genomgått mikrodialys parallellt med DBS. Målområdena för mikrodialysmätningarna visade sig skilja mellan patienterna, och den insamlade mikrodialysdatan indikerade att den biokemiska responsen på DBS berodde på kateterns position. För att ytterligare underlätta tolkningen av resultatet i relation till effekterna av DBS, kombinerades TVImax-simuleringarna med simuleringar av det elektriska fältet runt DBS-elektroderna. Sammanfattningsvis kan simuleringar av TVImax vara en hjälp vid den fysiologiska tolkningen av insamlad mikrodialysdata, vilket underlättar jämförelser mellan patienter. Detaljerad kunskap om de parametrar som påverkar samplingsvolymen för en mikrodialyskateter är värdefulla både för den aktuella applikationen, och övriga applikationer relaterade till diffusion av substanser i vävnad.
12
Zhang, Zhe. "Probing Secondary Structures of Self-cleaving Ribozymes by Deep Mutational Scanning." Thesis, Griffith University, 2021. http://hdl.handle.net/10072/402262.
Full textAbstract:
Self-cleaving ribozymes are the smallest catalytic RNAs found in nature, and are believed to have played an important role in the origin of life. The evidence that these natural RNAs can catalyze site-specific scission of their phosphodiester backbone demonstrates that RNAs can also fold into intricate functionally specific structures. Despite the biological importance of ribozymes, their structural characterization remains challenging because of the difficulty to crystallize. Moreover, their sizes are often too large for NMR structure determination. This thesis seeks to infer the base-pairing information of self-cleaving ribozymes by deep mutational scanning.
Chapter 1 of this thesis gives an overview of self-cleaving ribozymes and current RNA structural biology techniques.
In Chapters 2, 3 and 4, we performed a large-scale mutational analysis to gain structural information which is important for understanding the function of ribozymes. Generally, we constructed the mutant library of three ribozymes (CPEB3, LINE-1, OR4K15 ribozyme) from the human genome by error-prone PCR or using doped synthesis. These variants of ribozymes were assayed for their self-cleaving activity by exploiting deep sequencing for every randomized variant. A complete activity profile of each variant was acquired based on this large-scale mutational analysis.
To better predict the structural information, we developed a method called covariation-induced deviation of activity (CODA). When in combination with Monte Carlo simulated annealing, it provides an accurate inference of noncanonical and all canonical Watson-Crick base pairs at 100% precision for two self-cleaving ribozymes studied (CPEB3 and twister ribozyme). By extending this method to two unknown ribozymes (LINE-1, OR4K15 ribozyme), we were able to identify the core elements and their secondary structure. According to the secondary structure information, we identified homologs of these ribozymes by using a secondary structure-based similarity search.
In summary, our results show that combining deep mutational scanning and CODA analysis provides a highly accurate secondary-structure characterization of RNAs for the discovery of additional homologous sequences.Thesis (PhD Doctorate)
Doctor of Philosophy (PhD)
Institute for Glycomics
Griffith Health
Full Text
13
Sedhain, Ashok. "Optical properties of ALN and deep UV photonic structures studied by photoluminescence." Diss., Kansas State University, 2011. http://hdl.handle.net/2097/8522.
Full textAbstract:
Doctor of PhilosophyDepartment of Physics
Jingyu Lin
Time-resolved deep ultraviolet (DUV) Photoluminescence (PL) spectroscopy system has been employed to systematically monitor crystalline quality, identify the defects and impurities, and investigate the light emission mechanism in III-nitride semiconducting materials and photonic structures. A time correlated single photon counting system and streak camera with corresponding time resolutions of 20 and 2 ps, respectively, were utilized to study the carrier excitation and recombination dynamics. A closed cycle He-flow cryogenic system was employed for temperature dependent measurements. This system is able to handle sample temperatures in a wide range (from 10 to 900 K). Structural, electrical, and morphological properties of the material were monitored by x-ray diffraction (XRD), Hall-effect measurement, and atomic force microscopy (AFM), respectively. Most of the samples studied here were synthesized in our laboratory by metal organic chemical vapor deposition (MOCVD). Some samples were bulk AlN synthesized by our collaborators, which were also employed as substrates for homoepilayer growth. High quality AlN epilayers with (0002) XRD linewidth as narrow as 50 arcsec and screw type dislocation density as low as 5x10[superscript]6 cm[superscript]-2 were grown on sapphire substrates. Free exciton transitions related to all valence bands (A, B, and C) were observed in AlN directly by PL, which allowed the evaluation of crystal field (Δ[subscript]CF) and spin-orbit (Δ[subscript]SO) splitting parameters exerimentally. Large negative Δ[subscript]CF and, consequently, the difficulties of light extraction from AlN and Al-rich AlGaN based emitters due to their unique optical polarization properties have been further confirmed with these new experimental data. Due to the ionic nature of III-nitrides, exciton-LO phonon Frohlich interaction is strong in these materials, which is manifested by the appearance of phonon replicas accompanying the excitonic emission lines in their PL spectra. The strength of the exciton-phonon interactions in AlN has been investigated by measuring the Huang-Rhys factor. It compares the intensity of the zero phonon (exciton emission) line relative to its phonon replica. AlN bulk single crystals, being promising native substrate for growing nitride based high quality device structures with much lower dislocation densities (<10[superscript]4 cm[superscript]-2), are also expected to be transparent in visible to UV region. However, available bulk AlN crystals always appear with an undesirable yellow or dark color. The mechanism of such undesired coloration has been investigated. MOCVD was utilized to deposit ~0.5 μm thick AlN layer on top of bulk crystal. The band gap of strain free AlN homoepilayers was 6.100 eV, which is ~30 meV lower compared to hetero-epitaxial layers on sapphire possessing compressive strain. Impurity incorporation was much lower in non-polar m-plane growth mode and the detected PL signal at 10 K was about an order of magnitude higher from a-plane homo-epilayers compared to that from polar c-plane epilayers. The feasibility of using Be as an alternate p-type dopant in AlN has been studied. Preliminary studies indicate that the Be acceptor level in AlN is ~330 meV, which is about 200 meV shallower than the Mg level in AlN. Understanding the optical and electronic properties of native point defects is the key to achieving good quality material and improving overall device performance. A more complete picture of optical transitions in AlN and GaN has been reported, which supplements the understanding of impurity transitions in AlGaN alloys described in previous reports.
14
Poulenard, Adrien. "Structures for deep learning and topology optimization of functions on 3D shapes." Thesis, Institut polytechnique de Paris, 2020. http://www.theses.fr/2020IPPAX007.
Full textAbstract:
Le domaine du traitement de la géométrie suit un cheminement similaire à celui de l'analyse d'images avec l'explosion des publications consacrées à l'apprentissage profond ces dernières années. Un important effort de recherche est en cours pour reproduire les succès de l'apprentissage profond dans le domaine de la vision par ordinateur dans le contexte de l'analyse de formes 3D. Contrairement aux images, les formes 3D peuvent peuvent être représentées de différentes manières comme des maillages ou des nuages de points souvent dépourvus d'une structure canonique. Les algorithmes d'apprentissage profond traditionnels tels que les réseaux neuronaux convolutifs (CNN) ne sont donc pas faciles à appliquer aux formes 3D. Dans cette thèse, nous proposons trois contributions principales : premièrement, nous introduisons une méthode permettant de comparer des fonctions sur des domaines différents sans correspondances et de les déformer afin de rendre la topologie de leur ensemble de niveaux similaires. Nous appliquons notre méthode au problème classique de la correspondance de formes dans le contexte des applications fonctionnelles (functional maps) afin de produire des correspondances plus lisses et plus précises. Par ailleurs notre méthode reposant sur l'optimisation continue d'une énergie différentiable par rapport aux fonctions comparées elle est applicable à l'apprentissage profond. Nous apportons deux contributions directes à l'apprentissage profond des données 3D. Nous introduisons un nouvel opérateur de convolution sur des maillages triangulaires basés sur des coordonnées polaires locales et l'appliquons à l'apprentissage profond sur les maillages. Contrairement aux travaux précédents, notre opérateur prend en compte tous les choix de coordonnées polaires sans perte d'information directionnelle. Enfin, nous introduisons un nouveau module de convolution invariant par rotation sur les nuages de points et montrons que les CNN basés sur ce dernier peuvent surpasser l'état de l'art pour des tâches standard sur des ensembles de données non alignés même avec augmentation des donnéesThe field of geometry processing is following a similar path as image analysis with the explosion of publications dedicated to deep learning in recent years. An important research effort is being made to reproduce the successes of deep learning 2D computer vision in the context of 3D shape analysis. Unlike images shapes comes in various representations like meshes or point clouds which often lack canonical structure. This makes traditional deep learning algorithms like Convolutional Neural Networks (CNN) non straightforward to apply to 3D data. In this thesis we propose three main contributions:First, we introduce a method to compare functions on different domains without correspondences and to deform them to make the topology of their set of levels more alike. We apply our method to the classical problem of shape matching in the context of functional maps to produce smoother and more accurate correspondences. Furthermore, our method is based on the continuous optimization of a differentiable energy with respect to the compared functions and is applicable to deep learning. We make two direct contributions to deep learning on 3D data. We introduce a new convolution operator over triangles meshes based on local polar coordinates and apply it to deep learning on meshes. Unlike previous works our operator takes all choices of polar coordinates into account without loss of directional information. Lastly we introduce a new rotation invariant convolution layer over point clouds and show that CNNs based on this layer can outperform state of the art methods in standard tasks on un-alligned datasets even with data augmentation
15
Yosoi, Masaru. "Structures and fragmentations of the deep-hole states in 11B and 15N." 京都大学 (Kyoto University), 2003. http://hdl.handle.net/2433/149153.
Full text
16
Gane, Georgiana Andreea. "Building generative models over discrete structures : from graphical models to deep learning." Thesis, Massachusetts Institute of Technology, 2019. https://hdl.handle.net/1721.1/121611.
Full textAbstract:
Thesis: Ph. D., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2019Cataloged from PDF version of thesis. Page 173 blank.
Includes bibliographical references (pages 159-172).
The goal of this thesis is to investigate generative models over discrete structures, such as binary grids, alignments or arbitrary graphs. We focused on developing models easy to sample from, and we approached the task from two broad perspectives: defining models via structured potential functions, and via neural network based decoders. In the first case, we investigated Perturbation Models, a family of implicit distributions where samples emerge through optimization of randomized potential functions. Designed explicitly for efficient sampling, Perturbation Models are strong candidates for building generative models over structures, and the leading open questions pertain to understanding the properties of the induced models and developing practical learning algorithms.
In this thesis, we present theoretical results showing that, in contrast to the more established Gibbs models, low-order potential functions, after undergoing randomization and maximization, lead to high-order dependencies in the induced distributions. Furthermore, while conditioning in Gibbs' distributions is straightforward, conditioning in Perturbation Models is typically not, but we theoretically characterize cases where the straightforward approach produces the correct results. Finally, we introduce a new Perturbation Models learning algorithm based on Inverse Combinatorial Optimization. We illustrate empirically both the induced dependencies and the inverse optimization approach, in learning tasks inspired by computer vision problems. In the second case, we sequentialize the structures, converting structure generation into a sequence of discrete decisions, to enable the use of sequential models.
We explore maximum likelihood training with step-wise supervision and continuous relaxations of the intermediate decisions. With respect to intermediate discrete representations, the main directions consist of using gradient estimators or designing continuous relaxations. We discuss these solutions in the context of unsupervised scene understanding with generative models. In particular, we asked whether a continuous relaxation of the counting problem also discovers the objects in an unsupervised fashion (given the increased training stability that continuous relaxations provide) and we proposed an approach based on Adaptive Computation Time (ACT) which achieves the desired result. Finally, we investigated the task of iterative graph generation. We proposed a variational lower-bound to the maximum likelihood objective, where the approximate posterior distribution renormalizes the prior distribution over local predictions which are plausible for the target graph.
For instance, the local predictions may be binary values indicating the presence or absence of an edge indexed by the given time step, for a canonical edge indexing chosen a-priori. The plausibility of each local prediction is assessed by solving a combinatorial optimization problem, and we discuss relevant approaches, including an induced sub-graph isomorphism-based algorithm for the generic graph generation case, and a polynomial algorithm for the special case of graph generation resulting from solving graph clustering tasks. In this thesis, we focused on the generic case, and we investigated the approximate posterior's relevance on synthetic graph datasets.
by Georgiana Andreea Gane.
Ph. D.
Ph.D. Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science
17
FERRONE, LORENZO. "On interpretable information in deep learning: encoding and decoding of distributed structures." Doctoral thesis, Università degli Studi di Roma "Tor Vergata", 2016. http://hdl.handle.net/2108/202245.
Full text
18
Liedtke, Eric Arthur. "Effects from uncertainties in bathymetric measurements and variability in topography on computed stability of offshore slopes in deep water /." Full text (PDF) from UMI/Dissertation Abstracts International, 2001. http://wwwlib.umi.com/cr/utexas/fullcit?p3008380.
Full text
19
Boonkongkird, Chotipan. "Deep learning for Lyman-alpha based cosmology." Electronic Thesis or Diss., Sorbonne université, 2023. https://accesdistant.sorbonne-universite.fr/login?url=https://theses-intra.sorbonne-universite.fr/2023SORUS733.pdf.
Full textAbstract:
Au fur et à mesure que les relevés cosmologiques progressent et deviennent plus sophistiquées, ils fournissent des données de meilleure résolution, et de plus grand volume. La forêt Lyman-α est apparue comme une sonde puissante pour étudier les propriétés du milieu intergalactique (MIG) jusqu’à des redshift très élevés. L’analyse de ces données massives nécessite des simulations hydrodynamiques avancées capables d’atteindre une résolution comparable à celles des observations, ce qui exige des ordinateurs puissants et une quantité considérable de temps de calcul. Les développements récents dans le domaine de l’apprentissage automatique, notamment les réseaux de neurones, offrent de potentielles alternatives. Avec leur capacité à fonctionner comme des mécanismes d’ajustement universels, les réseaux de neurones gagnent du terrain dans diverses disciplines, y compris l’astrophysique et la cosmologie. Dans cette thèse de doctorat, nous explorons un cadre d’apprentissage automatique, plus précisément un réseau de neurones artificiels qui émule des simulations hydrodynamiques à partir de simulations N-corps de matière noire. Le principe fondamental de ce travail est basé sur l’approximation fluctuante de Gunn-Peterson (AFGP), un cadre couramment utilisé pour émuler la forêt Lyman-α à partir de la matière noire. Bien qu’utile pour la compréhension physique, l’AFGP ne parvient pas à prédire correctement l’absorption en négligeant la non-localité dans la construction du MIG. Au lieu de cela, notre méthode prend en compte la diversité du MIG, ce qui ne profite pas exclusivement à la forêt Lyman-α et s’étend à d’autres applications, tout en étant transparente dans son fonctionnement. Elle offre également une solution plus efficace pour générer des simulations, réduisant considérablement le temps de calcul par rapport aux simulations hydrodynamiques standard. Nous testons également la résilience du modèle en l’entraînant sur des données produites avec différentes hypothèses concernant la physique du MIG, via une méthode d’apprentissage par transfert. Nous comparons nos résultats à ceux d’autres méthodes existantes. Enfin, les simulateurs Lyman-α standards construisent généralement le volume d’observation en utilisant une seule époque des simulations cosmologiques. Cela implique un environnement astrophysique identique partout, ce qui ne reflète pas l’univers réel. Nous explorons la possibilité d’aller au-delà de cette limitation en prenant en compte dans notre émulateur des effets baryoniques variables le long de la ligne de visée. Bien que préliminaire, cette méthode pourrait servir à la construction de cônes de lumière cohérents. Afin de fournir des observables simulées plus réalistes, ce qui nous permettrait de mieux comprendre la nature du MIG et de contraindre les paramètres du modèle ΛCDM, nous envisageons d’utiliser des réseaux de neurones pour interpoler la rétroaction astrophysique à travers différentes cellules dans les simulationsAs cosmological surveys advance and become more sophisticated, they provide data with increasing resolution and volume. The Lyman-α forest has emerged as a powerful probe to study the intergalactic medium (IGM) properties up to a very high redshift. Analysing this extensive data requires advanced hydrodynamical simulations capable of resolving the observational data, which demands robust hardware and a considerable amount of computational time. Recent developments in machine learning, particularly neural networks, offer potential solutions. With their ability to function as universal fitting mechanisms, neural networks are gaining traction in various disciplines, including astrophysics and cosmology. In this doctoral thesis, we explore a machine learning framework, specifically, an artificial neural network to emulate hydrodynamical simulations from N-body simulations of dark matter. The core principle of this work is based on the fluctuating Gunn-Peterson approximation (FGPA), a framework commonly used to emulate the Lyman-α forest from dark matter. While useful for physical understanding, the FGPA misses to properly predict the absorption by neglecting non-locality in the construction of the IGM. Instead, our method includes the diversity of the IGM while being interpretable, which does not exclusively benefit the Lyman-α forest and extends to other applications. It also provides a more efficient solution to generate simulations, significantly reducing time compared to standard hydrodynamical simulations. We also test its resilience and explore the potential of using this framework to generalise to various astrophysical hypotheses of the IGM physics using a transfer learning method. We discuss how the results relate to other existing methods. Finally, the Lyman-α simulator typically constructs the observational volume using a single timestep of the cosmological simulations. This implies an identical astrophysical environment everywhere, which does not reflect the real universe. We explore and experiment to go beyond this limitation with our emulator, accounting for variable baryonic effects along the line of sight. While this is still preliminary, it could become a framework for constructing consistent light-cones. We apply neural networks to interpolate astrophysical feedback across different cells in simulations to provide mock observables more realistic to the real universe, which would allow us to understand the nature of IGM better and to constrain the ΛCDM model
20
Kushibar, Kaisar. "Automatic segmentation of brain structures in magnetic resonance images using deep learning techniques." Doctoral thesis, Universitat de Girona, 2020. http://hdl.handle.net/10803/670766.
Full textAbstract:
This PhD thesis focuses on the development of deep learning based methods for
accurate segmentation of the sub-cortical brain structures from MRI. First,
we have proposed a 2.5D CNN architecture that combines convolutional and
2/2
spatial features. Second, we proposed a supervised domain adaptation
technique to improve the robustness and consistency of deep learning model.
Third, an unsupervised domain adaptation method was proposed to eliminate the
requirement of manual intervention to train a deep learning model that is
robust to differences in the MRI images from multi-centre and multi-scanner
datasets. The experimental results for all the proposals demonstrated the
effectiveness of our approaches in accurately segmenting the sub-cortical
brain structures and has shown state-of-the-art performance on well-known
publicly available datasetsEsta tesis doctoral se centra en el desarrollo de métodos basados en el aprendizaje profundo para la segmentación precisa de las estructuras cerebrales subcorticales a partir de la resonancia magnética. En primer lugar, hemos propuesto una arquitectura 2.5D CNN que combina características convolucionales y espaciales. En segundo lugar, hemos propuesto una técnica de adaptación de dominio supervisada para mejorar la robustez y la consistencia del modelo de aprendizaje profundo. En tercer lugar, hemos propuesto un método de adaptación de dominio no supervisado para eliminar el requisito de intervención manual para entrenar un modelo de aprendizaje profundo que sea robusto a las diferencias en las imágenes de la resonancia magnética de los conjuntos de datos multicéntricos y multiescáner. Los resultados experimentales de todas las propuestas demostraron la eficacia de nuestros enfoques para segmentar con precisión las estructuras cerebrales subcorticales y han mostrado un rendimiento de vanguardia en los conocidos conjuntos de datos de acceso público
21
Ramadi, Khalil B. (Khalil Basil). "A chronically implantable neural device for on-demand microdosing of deep brain structures." Thesis, Massachusetts Institute of Technology, 2016. http://hdl.handle.net/1721.1/107067.
Full textAbstract:
Thesis: S.M., Massachusetts Institute of Technology, Department of Mechanical Engineering, 2016.Cataloged from PDF version of thesis.
Includes bibliographical references (pages 54-59).
Chronic neuropsychiatric diseases are increasingly consuming a larger portion of healthcare costs, in part due to a lack of effective treatment techniques. Through research into the pathology of these diseases we now know that most of these disorders are due to a loss in synchrony in a specific neural network. Effective treatments must seek to attenuate these network dynamics to establish normal neural communication. However, current treatments lack the spatiotemporal resolution to target networks with such specificity. The 'Injectrode' device developed here is a dual-lumen brain probe that is chronically implanted with wirelessly programmable micropumps for drug delivery on-demand. We establish the functionality of the system for repeated delivery of down to a few nanoliters of drug on-demand in vitro and in vivo, and show its biocompatibility over a 2-month implantation. This provides the foundation for testing of the system in a disease model, as well as the incorporation of additional features such as a recording or stimulating electrode. Combined with these tools, the injectrode system could serve as a closed loop device, delivering drug only when needed, ultimately allowing for efficacious independent disease management for chronic disorders.
by Khalil B. Ramadi.
S.M.
22
Kayesh, Humayun. "Deep Learning for Causal Discovery in Texts." Thesis, Griffith University, 2022. http://hdl.handle.net/10072/415822.
Full textAbstract:
Causality detection in text data is a challenging natural language processing task. This is a trivial task for human beings as they acquire vast background knowledge throughout their lifetime. For example, a human knows from their experience that heavy rain may cause flood or plane accidents may cause death. However, it is challenging to automatically detect such causal relationships in texts due to the availability of limited contextual information and the unstructured nature of texts. The task is even more challenging for social media short texts such as Tweets as often they are informal, short, and grammatically incorrect. Generating hand-crafted linguistic rules is an option but is not always effective to detect causal relationships in text because they are rigid and require grammatically correct sentences. Also, the rules are often domain-specific and not always portable to another domain. Therefore, supervised learning techniques are more appropriate in the above scenario. Traditional machine learning-based model also suffers from the high dimensional features of texts. This is why deep learning-based approaches are becoming increasingly popular for natural language processing tasks such as causality detection. However, deep learning models often require large datasets with high-quality features to perform well. Extracting deeply-learnable causal features and applying them to a carefully designed deep learning model is important. Also, preparing a large human-labeled training dataset is expensive and time-consuming. Even if a large training dataset is available, it is computationally expensive to train a deep learning model due to the complex structure of neural networks. We focus on addressing the following challenges: (i) extracting highquality causal features, (ii) designing an effective deep learning model to learn from the causal features, and (iii) reducing the dependency on large training datasets. Our main goals in this thesis are as follows: (i) we aim to study the different aspects of causality and causal discovery in text in depth. (ii) We aim to develop strategies to model causality in text, (iii) and finally, we aim to develop frameworks to design effective and efficient deep neural network structures to discover causality in texts.Thesis (PhD Doctorate)
Doctor of Philosophy (PhD)
School of Info & Comm Tech
Science, Environment, Engineering and Technology
Full Text
23
Brothers, Richard John. "The mechanical formation of vein structures as fluid flow pathways in Peru margin sediments and the Monterey formation, California." Thesis, University of Southampton, 1995. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.262451.
Full text
24
Johansson, Johannes. "Thermocoagulation in Deep Brain Structures : Modelling, simulation and experimental study of radio-frequency lesioning." Licentiate thesis, Linköping : Linköping University, Department of Biomedical Engineering, 2006. http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-7406.
Full text
25
Brünnler, Kai. "Deep Inference and Symmetry in Classical Proofs." Doctoral thesis, Saechsische Landesbibliothek- Staats- und Universitaetsbibliothek Dresden, 2003. http://nbn-resolving.de/urn:nbn:de:swb:14-1064911987703-38192.
Full textAbstract:
In this thesis we see deductive systems for classical propositional and predicate logic which use deep inference, i.e. inference rules apply arbitrarily deep inside formulas, and a certain symmetry, which provides an involution on derivations. Like sequent systems, they have a cut rule which is admissible. Unlike sequent systems, they enjoy various new interesting properties. Not only the identity axiom, but also cut, weakening and even contraction are reducible to atomic form. This leads to inference rules that are local, meaning that the effort of applying them is bounded, and finitary, meaning that, given a conclusion, there is only a finite number of premises to choose from. The systems also enjoy new normal forms for derivations and, in the propositional case, a cut elimination procedure that is drastically simpler than the ones for sequent systems.
26
Phan, Van Trung. "Modelling of the in service behaviour of passive insulated structures for deep sea offshore applications." Thesis, Brest, 2012. http://www.theses.fr/2012BRES0098/document.
Full textAbstract:
L’étude se situe dans le cadre de la recherche de gains de performance de structures d’isolations passives pour l’offshore profond. Le travail proposé a pour support des analyses expérimentales et numériques de tubes revêtus par des matériaux isolants utilisés en eau profonde pour transporter du fluide chaud. Le raboutage des tubes en acier, préalablement revêtus en atelier, nécessite un dégagement du revêtement aux extrémités pour réaliser l'opération d'assemblage (généralement par soudure). La partie dégagée est ensuite recouverte par un nouveau matériau pouvant être appliqué sur site. Ainsi l’isolation de cette partie du tube (Field Joint), qui est soumise à des chargements thermomécaniques en service, doit être optimisée pour assurer une durée de vie compatible avec les contraintes de l’exploitation offshore en eau profonde. Le travail comporte principalement quatre parties : - la modélisation du comportement thermique pour analyser l’évolution en temps et en espace de la température du matériau au cours de la fabrication, de la pose et en service sachant que pour les matériaux d’isolation le comportement mécanique est fortement dépendant de la température,- une partie expérimentale pour l’analyse du comportement des matériaux isolants en fonction de la température et en fonction de la pression hydrostatique qui est le principal chargement mécanique de ces structures en service,- la modélisation du comportement mécanique des isolants,- et une partie modélisation et simulation du comportement en service d'assemblages multi-matériaux de type industriel, avec prise en compte du comportement non-linéaire des constituantsUltra deep offshore oil exploitation presents new challenges to offshore engineering and operating companies. Such applications require the use of pipelines with an efficient thermal protection. Passive insulation materials are commonly used to guarantee the thermal performance of the pipes, and syntactic foams are now the preferred material for this application. The mechanical behaviour of such insulation materials is quite complex, associating time-dependent behaviour of polymers with damage behaviour of glass microspheres. In order to allow an optimisation of such systems, while ensuring in-service durability, accurate numerical models of insulation materials are thus required. During the service life in deep water, hydrostatic pressure is the most important mechanical loading of the pipeline, so this study aims to describe the mechanical behaviour of the material under such loading. Using a hyperbaric chamber, the analysis of the evolution of the volumetric strain with time, with respect to the temperature, under different time-evolutions of the applied hydrostatic pressure is presented in this paper. Such experimental results associated with the mechanical response of the material under uniaxial tensile creep tests, allow the development of a thermo-mechanical model, so that representative loadings can be analysed
27
Dahl, Kristian. "Hybrid model testing of deep-water moored structures by active control of simulated line forces." Thesis, Norges teknisk-naturvitenskapelige universitet, Institutt for marin teknikk, 2010. http://urn.kb.se/resolve?urn=urn:nbn:no:ntnu:diva-11468.
Full textAbstract:
Increased activity in deeper water areas, in connection with oil and gas exploration, has resulted in a steadily increased demand for model testing technology to predict reliable behavior of floating offshore structures at deep water. Common model testing technology, developed for more shallow water, must be modified to meet the challenges that deep water model testing poses. The approach using active control of line forces to replace the mooring and riser system seems promising. Actuators should induce the same resulting forces as calculated by numerical methods on to the vessel in real time. This master’s thesis has its focus on the development of an experimental setup to investigate the performance of the control system inducing desired forces on to a simplified vessel model. To simplify the complexity with an experiment using a vessel moving in 6 degrees of freedom (DOF), a simplified vessel model with motion only in the heave direction is considered. Active control of the line force connected to the vessel, moving only in heave direction, is launched to investigate if the actuator or motor-winch system, mange to induce the correct calculated mooring force for a measured heave motion. Necessary procurement of parts to the experimental setup has been obtained. Reasonable values for the maximum and minimum top tension for one mooring line has been used to find an appropriate actuator. A Faulhaber electrical motor was chosen to meet the specifications. The drive electronics controlling the electrical motor has a wide variety of control options, and control of the motor shaft torques, necessary to obtain desired forces, is possible. A mathematical model describing the experimental setup was developed. A model of the DC motor dynamics and the mass dynamics are mathematically derived. The unknown parameters in the model have been found using measurement data from experiment. The measurement data are established using a pretensioned spring connected to the DC motor shaft, as well as a freely running shaft. In the experiments two test cases are considered. In test case one a spring and line, representing the active mooring line, is connected in-between the vessel model, or oscillating actuator, and a motor-winch system. The motor-winch, or DC motor, will then try to pull the line and spring till it reaches the desired reference. The measured heave motion to the oscillating system is utilized to calculate the mooring force references, the active mooring line will then try to induce the same forces as calculated. The calculated forces are only available for a given heave motion. A second test case is performed with a vessel model represented as a mass. The mass, or vessel, is connected to the oscillating system actuator with a spring. The oscillating system actuator will represent the wave motion, while the spring should represent the water plane stiffness. A rope, or line, connected in-between the mass and the motor-winch system will represent the active mooring line. Different sinusoidal references, as well as piecewise constant signals are used to test if the active mooring line is able to induce the desired forces on to the mass real time. The results show that for the given heave motion the motor-winch setup will be able to actuate the desired mooring line forces on to the oscillating body. Sinusoidal force references are also possible to follow. There is however a constraint in the frequency, and a high oscillating system frequency will be a limitation. By decreasing the oscillating frequency less noise in the measured signal will occur. To avoid a lot of noise in the measured forces filtering the signal was necessary. An observer has been developed to remove this noise. The filtered signal (estimated force) can then be used as feedback to the controller. A vessel model considered in three DOF is used to find an appropriate sampling frequency. The real time response measurements to the vessel need to be sampled in order to calculate desired mooring forces. A study is therefore conducted to find the bandwidth, or highest frequency, in the system to conclude which sampling frequency that is needed to recreate the response signals without loss of information. The results show that 50 Hz sampling frequency will be necessary to recreate the continuous time signals for the specific case considered.
28
Ran, Peipei. "Imaging and diagnostic of sub-wavelength micro-structures, from closed-form algorithms to deep learning." Electronic Thesis or Diss., université Paris-Saclay, 2020. http://www.theses.fr/2020UPASG061.
Full textAbstract:
Le test électromagnétique d’un ensemble fini en forme de grille de tiges diélectriques cylindriques circulaires infiniment longues dont certaines manquent est investigué à partir de données fréquence simple et multiple et en régime temporel. Les distances sous-longueur d’onde entre tiges adjacentes et des diamètres de tige de sous-longueur d’onde sont considérées sur toute la bande de fréquences d’opération et cela conduit à un défi majeur en raison du besoin de super-résolution dans la microstructure, bien au-delà du critère de Rayleigh. Tout un ensemble de méthodes de résolution est étudié et des simulations numériques systématiques illustrent avantages et inconvénients, complétées par le traitement de données expérimentales en laboratoire acquises sur un prototype de micro-structure en chambre anéchoïque micro-onde. Ces méthodes, qui diffèrent selon les informations a priori prises en compte et la polyvalence qui en résulte, comprennent retournement temporel, inversions de source de contraste, binaires ou parcimonieuses, ainsi que réseaux de neurones convolutifs éventuellement combinés avec des réseaux récurrentsElectromagnetic probing of a gridlike, finite set of infinitely long circular cylindrical dielectric rods affected by missing ones is investigated from time-harmonic single and multiple frequency data. Sub-wavelength distances between adjacent rods and sub-wavelength rod diameters are assumed throughout the frequency band of operation and this leads to a severe challenge due to need of super-resolution within the present micro-structure, well beyond the Rayleigh criterion. A wealth of solution methods is investigated and comprehensive numerical simulations illustrate pros and cons, completed by processing laboratory-controlled experimental data acquired on a micro-structure prototype in a microwave anechoic chamber. These methods, which differ per a priori information accounted for and consequent versatility, include time-reversal, binary-specialized contrast-source and sparsity-constrained inversions, and convolutional neural networks possibly combined with recurrent ones
29
Gerard, Sarah E. "Multi-scale convolutional neural networks for segmentation of pulmonary structures in computed tomography." Diss., University of Iowa, 2018. https://ir.uiowa.edu/etd/6578.
Full textAbstract:
Computed tomography (CT) is routinely used for diagnosing lung disease and developing treatment plans using images of intricate lung structure with submillimeter resolution. Automated segmentation of anatomical structures in such images is important to enable efficient processing in clinical and research settings. Convolution neural networks (ConvNets) are largely successful at performing image segmentation with the ability to learn discriminative abstract features that yield generalizable predictions. However, constraints in hardware memory do not allow deep networks to be trained with high-resolution volumetric CT images. Restricted by memory constraints, current applications of ConvNets on volumetric medical images use a subset of the full image; limiting the capacity of the network to learn informative global patterns. Local patterns, such as edges, are necessary for precise boundary localization, however, they suffer from low specificity. Global information can disambiguate structures that are locally similar.
The central thesis of this doctoral work is that both local and global information is important for segmentation of anatomical structures in medical images. A novel multi-scale ConvNet is proposed that divides the learning task across multiple networks; each network learns features over different ranges of scales. It is hypothesized that multi-scale ConvNets will lead to improved segmentation performance, as no compromise needs to be made between image resolution, image extent, and network depth. Three multi-scale models were designed to specifically target segmentation of three pulmonary structures: lungs, fissures, and lobes.
The proposed models were evaluated on a diverse datasets and compared to architectures that do not use both local and global features. The lung model was evaluated on humans and three animal species; the results demonstrated the multi-scale model outperformed single scale models at different resolutions. The fissure model showed superior performance compared to both a traditional Hessian filter and a standard U-Net architecture that is limited in global extent.
The results demonstrated that multi-scale ConvNets improved pulmonary CT segmentation by incorporating both local and global features using multiple ConvNets within a constrained-memory system. Overall, the proposed pipeline achieved high accuracy and was robust to variations resulting from different imaging protocols, reconstruction kernels, scanners, lung volumes, and pathological alterations; demonstrating its potential for enabling high-throughput image analysis in clinical and research settings.
30
Von, Maltitz Kosma. "The deep optical ZoA galaxy catalogue in Vela first indications of previously hidden large-scale structures." Master's thesis, University of Cape Town, 2012. http://hdl.handle.net/11427/12065.
Full textAbstract:
Includes abstract.Includes bibliographical references.
This thesis presents a deep optical galaxy catalogue of the Vela region in the Zone of Avoidance (I b < 10°.245 < I < 280. This region was searched for galaxy candidates by optical inspection of IIIaJ (3950 A to 5400 A) film copies of the ESO/SRC sky survey as part of an effort to reduce the ZoA.
31
Skibbe, Eric. "A comparison of design using strut-and-tie modeling and deep beam method for transfer girders in building structures." Manhattan, Kan. : Kansas State University, 2010. http://hdl.handle.net/2097/3907.
Full text
32
Maluleke, Vutlhari Absalom. "The effects of boat mooring systems on squid egg beds during squid fishing." Thesis, Cape Peninsula University of Technology, 2017. http://hdl.handle.net/20.500.11838/2528.
Full textAbstract:
Thesis (MTech (Mechanical Engineering))--Cape Peninsula University of Technology, 2017.In South Africa, squid fishing vessels need to find and then anchor above benthic squid egg beds to effect viable catches. However, waves acting on the vessel produce a dynamic response on the anchor line. These oscillatory motions produce impact forces of the chain striking the seabed. It is hypothesised that this causes damage to the squid egg bed beneath the vessels. Different mooring systems may cause more or less damage and this is what is investigated in this research. The effect of vessel mooring lines impact on the seabed during squid fishing is investigated using a specialised hydrodynamic tool commercial package ANSYS AQWA models. This study analysed the single-point versus the two-point mooring system’s impact on the seabed. The ANSYS AQWA models were developed for both mooring systems under the influence of the wave and current loads using the 14 and 22 m vessels anchored with various chain sizes. The effect of various wave conditions was investigated as well as the analysis of three mooring line configurations. The mooring chain contact pressure on the seabed is investigated beyond what is output from ANSYS AQWA using ABAQUS finite element analysis. The real-world velocity of the mooring chain underwater was obtained using video analysis. The ABAQUS model was built by varying chain sizes at different impact velocities. The impact pressure and force due to this velocity was related to mooring line impact velocity on the seabed in ANSYS AQWA. Results show the maximum impact pressure of 191 MPa when the 20 mm diameter chain impacts the seabed at the velocity of 8 m/s from video analysis. It was found that the mooring chain impact pressure on the seabed increased with an increase in the velocity of impact and chain size. The ANSYS AQWA impact pressure on the seabed was found to be 170.86 MPa at the impact velocity of 6.4 m/s. The two-point mooring system was found to double the seabed mooring chain contact length compared to the single-point mooring system. Both mooring systems showed that the 14 m vessel mooring line causes the least seabed footprint compared to the 22 m vessel.
33
Shi, Zhiqun. "Automatic interpretation of potential field data applied to the study of overburden thickness and deep crustal structures, South Australia." Title page, contents and abstract only, 1993. http://web4.library.adelaide.edu.au/theses/09PH/09phs5548.pdf.
Full textAbstract:
Bibliography: leaves 189-203. Deals with two interpretation methods, a computer program system AUTOMAG and spectral analysis, used for studying overburden thickness and density structure of the crust. The methods were applied to the Gawler Craton, Eyre Peninsula.
34
Kodi, Ramanah Doogesh. "Bayesian statistical inference and deep learning for primordial cosmology and cosmic acceleration." Thesis, Sorbonne université, 2019. http://www.theses.fr/2019SORUS169.
Full textAbstract:
Cette thèse a pour vocation le développement et l’application de nouvelles techniques d’inférence statistique bayésienne et d’apprentissage profond pour relever les défis statistiques imposés par les gros volumes de données complexes des missions du fond diffus cosmologique (CMB) ou des relevés profonds de galaxies de la prochaine génération, dans le but d'optimiser l’exploitation des données scientifiques afin d’améliorer, à terme, notre compréhension de l’Univers. La première partie de cette thèse concerne l'extraction des modes E et B du signal de polarisation du CMB à partir des données. Nous avons développé une méthode hiérarchique à haute performance, nommée algorithme du dual messenger, pour la reconstruction du champ de spin sur la sphère et nous avons démontré les capacités de cet algorithme à reconstruire des cartes E et B pures, tout en tenant compte des modèles de bruit réalistes. La seconde partie porte sur le développement d’un cadre d'inférence bayésienne pour contraindre les paramètres cosmologiques en s’appuyant sur une nouvelle implémentation du test géométrique d'Alcock-Paczyński et nous avons présenté nos contraintes cosmologiques sur la densité de matière et l'équation d'état de l'énergie sombre. Etant donné que le contrôle des effets systématiques est un facteur crucial, nous avons également présenté une fonction de vraisemblance robuste, qui résiste aux contaminations inconnues liées aux avant-plans. Finalement, dans le but de construire des émulateurs de dynamiques complexes dans notre modèle, nous avons conçu un nouveau réseau de neurones qui apprend à peindre des distributions de halo sur des champs approximatifs de matière noire en 3DThe essence of this doctoral research constitutes the development and application of novel Bayesian statistical inference and deep learning techniques to meet statistical challenges of massive and complex data sets from next-generation cosmic microwave background (CMB) missions or galaxy surveys and optimize their scientific returns to ultimately improve our understanding of the Universe. The first theme deals with the extraction of the E and B modes of the CMB polarization signal from the data. We have developed a high-performance hierarchical method, known as the dual messenger algorithm, for spin field reconstruction on the sphere and demonstrated its capabilities in reconstructing pure E and B maps, while accounting for complex and realistic noise models. The second theme lies in the development of various aspects of Bayesian forward modelling machinery for optimal exploitation of state-of-the-art galaxy redshift surveys. We have developed a large-scale Bayesian inference framework to constrain cosmological parameters via a novel implementation of the Alcock-Paczyński test and showcased our cosmological constraints on the matter density and dark energy equation of state. With the control of systematic effects being a crucial limiting factor for modern galaxy redshift surveys, we also presented an augmented likelihood which is robust to unknown foreground and target contaminations. Finally, with a view to building fast complex dynamics emulators in our above Bayesian hierarchical model, we have designed a novel halo painting network that learns to map approximate 3D dark matter fields to realistic halo distributions
35
Macleod, Adrian K. A. "The role of marine renewable energy structures and biofouling communities in promoting self-sustaining populations of non-native species." Thesis, University of the Highlands and Islands, 2013. https://pure.uhi.ac.uk/portal/en/studentthesis/the-role-of-marine-renewable-energy-structures-and-biofouling-communities-in-promoting-selfsustaining-populations-of-nonnative-species(0c7f0d89-74e8-4468-83c9-4216e4f2b1a8).html.
Full textAbstract:
Novel environments and biological communities created by the large-scale deployment of Marine Renewable Energy Devices (MREDs) have the potential to promote the spread of non-native species (NNS). Knowledge of how community composition resident on MREDs is shaped by geography, local hydrodynamics and the duration of deployment, will clarify how these technologies will interact with natural habitats, including the provision of suitable habitat for NNS. A network of navigation buoys was used to study biofouling communities in areas proposed for MRED deployment. Significant differences in community structure were observed in different geographic areas. A significant reduction in number of taxa present and community wet-weight was observed where buoys were deployed in greater tidal flow rates. However, overall community composition was not significantly different between ‘high’ (>1 ms-1) and ‘low’ (<1 ms-1) flow areas and for buoys deployed for different time durations (1-7 years). These finding have important implications for the longevity of devices and their interaction with natural habitats, including proposed ‘artificial reef’ effects. In total five non-native species were identified on the buoys sampled, supporting the need to monitor MREDs as the industry grows. Hydrodynamic and biotic features of the epibenthic communities were used to predict the presence of the most prevalent NNS, the amphipod Caprella mutica, in addition to other native amphipod species. Caprella mutica presence was found to be significantly affected by increasing flow speed compared with the native amphipod Jassa herdmani. Behavioural flume studies investigating flow-related processes governing the presence of non-native amphipods supported these findings. This study details how the hydrodynamic and biological environments created by MREDs determine their suitability for the establishment of self-sustaining populations, and therefore their dispersal potential for NNS. These findings inform design criteria and management options to minimise the biosecurity risk that these structures will pose as the industry expands.
36
Meneux, Baptiste. "Evolution de l'agrégation des galaxies dans le sondage VIMOS-VLT Deep Survey." Phd thesis, Université de Provence - Aix-Marseille I, 2005. http://tel.archives-ouvertes.fr/tel-00011320.
Full textAbstract:
Les grands sondages récents de l'Univers ont mis en évidence la présence de structures dans la distribution de la matière, sous forme de filaments et de vides. Pour étudier l'évolution de la distribution spatiale des galaxies, il est nécessaire de connaître leur position précise dans l'espace à trois dimensions. Cette thèse s'est déroulée dans le cadre du sondage profond VIMOS-VLT Deep Survey (VVDS). Son but est de mesurer quelques 100000 redshifts pour étudier la formation et l'évolution des galaxies et des structures à grande échelle de l'Univers jusqu'à z~5. Après avoir fait un état des lieux de la connaissance que nous avons de la distribution des galaxies, puis introduis le sondage VVDS, je présente la mesure et l'évolution de la fonction de corrélation spatiale à deux points à partir des données de la première époque du VVDS, le plus large échantillon (10759 spectres) jamais acquis à I_AB=24. J'ai développé un ensemble de programmes mis à la disposition du consortium VVDS pour mesurer facilement la longueur de corrélation spatiale des galaxies dans un intervalle en redshift donné, avec ses erreurs associées, en corrigeant les effets de la stratégie d'observation du VVDS. Cet outil a permis de mesurer l'évolution de la fonction de corrélation spatiale de la population globale des galaxies jusqu'à z=2. J'ai prolongé cette étude en divisant l'échantillon de galaxies par type spectral et par couleur. Enfin, en combinant les données de GALEX avec celles du VVDS, j'ai pu mesurer l'agrégation de galaxies sélectionnées en ultraviolet jusqu'à z~1. C'est la première fois que de telles mesures sont réalisées sur une si longue plage de temps cosmique. Les résultats présentés dans cette thèse font ainsi office de nouvelles références pour les travaux futurs.
37
Straßburger, Lutz. "Linear Logic and Noncommutativity in the Calculus of Structures." Doctoral thesis, Saechsische Landesbibliothek- Staats- und Universitaetsbibliothek Dresden, 2003. http://nbn-resolving.de/urn:nbn:de:swb:14-1063208959250-72937.
Full textAbstract:
In this thesis I study several deductive systems for linear logic, its fragments, and some noncommutative extensions. All systems will be designed within the calculus of structures, which is a proof theoretical formalism for specifying logical systems, in the tradition of Hilbert's formalism, natural deduction, and the sequent calculus. Systems in the calculus of structures are based on two simple principles: deep inference and top-down symmetry. Together they have remarkable consequences for the properties of the logical systems. For example, for linear logic it is possible to design a deductive system, in which all rules are local. In particular, the contraction rule is reduced to an atomic version, and there is no global promotion rule. I will also show an extension of multiplicative exponential linear logic by a noncommutative, self-dual connective which is not representable in the sequent calculus. All systems enjoy the cut elimination property. Moreover, this can be proved independently from the sequent calculus via techniques that are based on the new top-down symmetry. Furthermore, for all systems, I will present several decomposition theorems which constitute a new type of normal form for derivations.
38
Szeder, Thore. "Active tectonics in the NW German Basin evidence from correlations between the modern landscape and deep geological structures (Lower Saxony, river Hunte) /." [S.l. : s.n.], 2003. http://ArchiMeD.uni-mainz.de/pub/2003/0038/diss.pdf.
Full text
39
Lindenmaier, Falk. "Hydrology of a large unstable hillslope at Ebnit, Vorarlberg : identifying dominating processes and structures." Phd thesis, Universität Potsdam, 2007. http://opus.kobv.de/ubp/volltexte/2008/1742/.
Full textAbstract:
The objective of this thesis is to improve the knowledge of control mechanisms of hydrological induced mass movements. To this end, detailed hydrological process studies and physically-based hydrological modelling were applied. The study site is a hillslope in the Dornbirn Ache valley near Bregenz, Austria. This so called Heumös slope features a deep-seated translational shear zone and surface near creep movements of up to 10 cm a year. The Cretaceous marlstones of the Austrian Helveticum have a high susceptibility for weathering and might form clay-rich cohesive sediments. In addition, glacial and post-glacial processes formed an unstable hillslope. High yearly precipitation depths of about 2100 mm and rainstorms with both high intensities and precipitation depths govern surface and subsurface hydrological processes. Pressure propagation induced in hydrological active areas influences laterally the groundwater reactions of the moving mass. A complex three-dimensional subsurface pressure system is the cause for fast groundwater reactions despite low hydraulic conductivities.
To understand hillslope scale variability, hydrotopes representing specific dominating processes were mapped using vegetation association distribution and soil core analysis. Detailed small-scale soil investigations followed to refine the understanding of these hydrotopes. A perceptional model was developed from the hydrotope distribution and was corroborated by these detailed investigations. The moving hillslope is dominated by surface-runoff generation. Infiltration and deep percolation of water is inhibited through clay-rich gleysols; the yearly average soil moisture is close to saturation. Steep slopes adjacent to the moving hillslope are far more active concerning infiltration, preferential flow and groundwater fluctuations. Spring discharge observations at the toe of the steep slopes are in close relation to groundwater table observations on the moving hillslope body. Evidence of pressure propagation from the steep slopes towards the hillslope body is gathered by comparison of dominating structures and processes. The application of the physically-based hydrological model CATFLOW substantiates the idea of pressure propagation as a key process for groundwater reactions and as a possible trigger for movement in the hillslope.Diese Arbeit soll die Zusammenhänge von hydrologischen Rahmenbedingungen und Massenbewegungen besser erforschen, damit in Zukunft verbesserte Vorhersagen des Versagenszeitpunktes möglich werden. Das Untersuchungsgebiet besteht aus einem ca. 2 km langen und 500 m breiten Hang mit einem maximalen Höhenunterschied von ca. 400 m. Das dort vorkommende Festgestein besteht im Wesentlichen aus Mergelstein. Die vergangenen Eiszeiten haben dieses Gestein überarbeitet und Grundmoränenablagerungen auf dem Hang zurückgelassen. Diese wurden in den letzen 10.000 Jahren von Hangschutt, der aus den benachbarten Steilhängen stammt, überlagert. Der Hangschutt ist sehr verwitterungsanfällig, die Kalkkristalle lösen sich und wandeln den Hangschutt in lehmiges Material. Bewegungsmessungen an der Oberfläche zeigen, dass sich der Hang mit ca. 10 cm im Jahr talabwärts bewegt. Diese Bewegungen werden sehr wahrscheinlich durch kleine ruckartige Ereignisse in ca. 8 m Tiefe ausgelöst. Ziel der Untersuchungen war, den Wasserhaushalt des Hanges so gut wie möglich zu erfassen und mit Computermodellen abzubilden. Dabei spielt die Heterogenität der pedologischen Eigenschaften einen wesentliche Rolle, als Eingangsparameter für die Modelle. Grundwasserstandsmessungen in 5,5 m Tiefe auf dem Hang zeigen schnelle Reaktionen des Grundwasserspiegels nach Niederschlagsereignissen. Das Wasser dieser Ereignisse kann aber aufgrund des Lehms, der nur eine geringe Wasserdurchlässigkeit für Wasser besitzt, nicht in den tieferen Untergrund gelangen, sondern fließt fast vollständig an der Oberfläche ab. Dahingegen führt ein schnelles Versickern von Wasser in an den Hang anschließenden Steilhängen zu einem schnellen Grundwasseranstieg, der aufgrund eines gespannten Grundwasserleiters den Druck in die Hangrutschung weitergibt. Dort wird ein Überdruck aufgebaut, der sehr wahrscheinlich die Bewegungen auslöst. Die vorliegende Arbeit ist eine detaillierte Herangehensweise um Erkenntnisse aus der Hyrologie für die Bestimmung des Wasserhaushaltes von Massenbewegungen heranzuziehen.
40
Latrach, Soumaya. "Optimisation et analyse des propriétés de transport électroniques dans les structures à base des matériaux AlInN/GaN." Thesis, Université Côte d'Azur (ComUE), 2018. http://www.theses.fr/2018AZUR4243.
Full textAbstract:
Les matériaux III-N ont apporté un gain considérable au niveau des performances des composants pour les applications en électronique de puissance. Les potentialités majeures du GaN pour ces applications résident dans son grand champ de claquage qui résulte de sa large bande interdite, son champ de polarisation élevé et sa vitesse de saturation importante. Les hétérostructures AlGaN/GaN ont été jusqu’à maintenant le système de choix pour l’électronique de puissance. Les limites sont connues et des alternatives sont étudiées pour les surmonter. Ainsi, les hétérostructures InAlN/GaN en accord de maille ont suscité beaucoup d’intérêts, notamment pour des applications en électronique de puissance à haute fréquence. L’enjeu de ce travail de thèse consiste à élaborer et caractériser des hétérostructures HEMTs (High Electron Mobility Transistors) afin d’établir des corrélations entre défauts structuraux, électriques et procédés de fabrication. Une étude sera donc menée sur la caractérisation de composants AlGaN/GaN afin de cerner les paramètres de croissance susceptibles d’avoir un impact notable sur la qualité structurale et électrique de la structure, notamment sur l’isolation électrique des couches tampons et le transport des porteurs dans le canal. En ce qui concerne les HEMTs InAlN/GaN, l’objectif est d’évaluer la qualité de la couche barrière. Pour cela, une étude de l’influence des épaisseurs ainsi que la composition de la barrière sera menée. La combinaison de ces études permettra d’identifier la structure optimale. Ensuite, l’analyse des contacts Schottky par des mesures de courant et de capacité à différentes températures nous permettra d’identifier les différents modes de conduction à travers la barrière. Enfin, les effets de pièges qui constituent l’une des limites fondamentales inhérentes aux matériaux étudiés seront caractérisés par différentes méthodes de spectroscopie de défautsIII-N materials have made a significant gain in component performance for power electronics applications. The major potential of GaN for these applications lies in its large breakdown field resulting from its wide bandgap, high polarization field and high electronic saturation velocity. AlGaN/GaN heterostructures have been, until recently, the system of choice for power electronics. The limits are known and alternatives are studied to overcome them. Thus, lattice matched InAlN/GaN heterostructures have attracted a great deal of research interest, especially for high frequency power electronic applications. The aim in this work of thesis consists in developing and in characterizing High Electron Mobility Transistors (HEMTs) to establish correlations between structural, electrical defects and technologic processes. A study will therefore be conducted on the characterization of AlGaN/GaN components to enhance the parameters of growth susceptible to have a notable impact on the structural and electrical quality of the structure, in particular on the electrical isolation of the buffer layers and the transport properties. For InAlN/GaN HEMTs, the objective is to evaluate the quality of the barrier layer. For this, a study of the influence of the thickness as well as the composition of the barrier will be conducted. The combination of these studies will allow identifying the optimum structure. Then, the analysis of Schottky contacts by measurements of current and capacity at different temperatures will allow us to identify the several conduction modes through the barrier. Finally, the effects of traps which constitute one of the fundamental limits inherent to the studied materials will be characterized by various defects spectroscopy methods
41
Kalendra, Vidmantas. "Study of the deep levels induced by the high energy proton and neutron irradiation in the structures of high resistivity Si, SiC and GaN." Doctoral thesis, Lithuanian Academic Libraries Network (LABT), 2009. http://vddb.library.lt/obj/LT-eLABa-0001:E.02~2009~D_20091215_091727-69995.
Full textAbstract:
Investigations made on new materials and their structures for production of particle detectors based on semi-insulating SiC and GaN comprise the technological and applied importance of this study. Innovations in defect control technology, especially, in recognition of extended defects and percolative carrier transport in heavily irradiated detector structures are considered and applied for scientific implementations. These investigations have been performed within a framework of CERN rd50 project. Irradiation by 24 GeV protons varying fluence up to 1016 cm-2 deteriorates rectifying properties of the 4H-SiC particle detectors. Different isotopes produced in 4H-SiC during irradiation by protons have been revealed by gamma spectroscopy. In the non-irradiated GaN material the temperature-dependent variations of leakage current have been unveiled to be caused by the carrier mobility temperature changes. Activation energy values have been extracted for proton radiation induced deep centres in the GaN detectors by thermally stimulated current spectroscopy as well as the isotopes and long-living radio-nuclides have been identified by gamma spectroscopy. In the Si detectors, irradiated by reactor neutrons, the photo-activation energy values have been determined for the deep levels located below the mid-gap by photo-ionisation spectroscopy while isochronal anneals enhance the density of the acceptor-type vacancy-related defects.Disertacijoje išanalizuoti gilieji centrai didžiavaržiuose Si, SiC ir GaN dariniuose, sietini su didelės energijos protonų bei neutronų spinduliuote sudarytais defektais, atskleistos radiacinių defektų transformacijos po iškaitinimų, didelių energijų spinduliuotės įtaka krūvio pernašai ir pagavai medžiagose, tinkamose jonizuojančiosios spinduliuotės detektoriams, tiriamiems pagal Europos branduolinių tyrimų centro (CERN) projektus. 4H-SiC dariniuose, apšvitintuose 24 GeV/c protonais, išanalizuota elektrinių charakteristikų kaita. Iš šiluma skatinamųjų srovių spektrų nustatytos šiluminės aktyvacijos energijų vertės. Taip pat 4H-SiC dariniuose, apšvitintuose protonų įtėkiais, siekiančiais 1016 cm-2, įvertintas skirtingų spinduliuote sukurtų izotopų kiekis. Neapšvitintose GaN dariniuose nustatyta, kad medžiagos elektrinio laidumo parametrų kaitą nulemia krūvininkų judrio kitimas. Apšvitintuose neutronais GaN dariniuose šiluma skatinamųjų srovių spektroskopijos būdu buvo nustatyti dominuojančių defektų lygmenys. Aptikta, kad po apšvitos 24 GeV/c protonų įtėkiais, siekiančiais 1016 cm-2, GaN susidarė 7Be, 22Na ir kiti ilgaamžiai radionuklidai, kurių atominis skaičius A<70, bei žymiai pakito spinduliuotės detektorių krūvio pernašos savybės. Didžiavaržio silicio detektoriuose po apšvitos reaktoriaus neutronais susidarė visa eilė radiacinių defektų, kuriems priskirtinų giliųjų centrų parametrai buvo įvertinti fotojonizacijos spektroskopijos ir tamsinės srovės temperatūrinių kitimų... [toliau žr. visą tekstą]
42
Kamgaing, Souop Landry. "Etude et optimisation du perçage orbital robotisé pour l'assemblage des structures aéronautiques." Thesis, Toulouse 3, 2020. http://www.theses.fr/2020TOU30114.
Full textAbstract:
Ces travaux de thèse traitent du perçage orbital de pièces en alliage d'aluminium AA2024-T351. Ce cas d'étude est issu d'une problématique industrielle rencontrée par différents constructeurs aéronautiques qui souhaitent intégrer ce procédé dans leurs moyens de fabrication. Malgré les multiples avantages du perçage orbital par rapport au perçage axial conventionnel, il subsiste un véritable frein technologique très important : l'abaissement de la tenue en fatigue des pièces en alliage d'aluminium percées. Ceci résulte d'une non optimisation des niveaux des contraintes résiduelles au sein des alésages réalisés par ce procédé. Il n'existe à ce jour aucun moyen d'optimisation des paramètres de coupe du perçage orbital. L'objectif global de ces travaux est d'améliorer la durée de vie en fatigue des pièces percées en introduisant au cours de l'opération des contraintes résiduelles compressives et de l'écrouissage superficiel, tous deux bénéfiques pour la tenue en service des composants. Les travaux présentés se focalisent d'abord sur l'optimisation des paramètres de coupe du perçage orbital au sein de l'alliage d'aluminium AA2024-T351, optimisation menée sur la base de la minimisation des efforts et énergies spécifiques de coupe. Un traitement mécanique de surface novateur a également été introduit au sein du projet : le galetage orbital. La caractérisation de ce procédé a été réalisée au moyen de modèles Éléments Finis. Ces modèles ont permis a posteriori d'étudier l'influence des paramètres de ce procédé sur l'intégrité de surface de l'alésage, notamment sur la répartition des contraintes résiduelles. Une mise en parallèle avec les résultats expérimentaux est effectuée, afin de valider les différentes simulations numériques mises en œuvre. Une caractérisation expérimentale de l'intégrité de surface des alésages, réalisés par perçage orbital en conditions de coupe optimisées et par galetage orbital, ont démontré, non seulement, la faisabilité du procédé de galetage orbital, mais aussi le respect des exigences aéronautiques. L'étude de la tenue en fatigue d'éprouvettes en alliage d'aluminium AA2024-T351 percées ont complété cette caractérisationThis thesis addresses the orbital drilling of AA2024-T351 aluminum alloy parts. This case study arose from an industrial problem encountered by various aircraft manufacturers who wish to integrate this process into their manufacturing processes. Despite the multiple advantages of orbital drilling compared to conventional axial drilling, there is still a very important technological barrier: the lower fatigue strength of drilled aluminum alloy parts. This is due to the non-optimization of residual stress levels within the boreholes produced by this process. There is currently no means of optimizing orbital drilling cutting parameters. The overall objective is to improve the fatigue life of drilled parts by introducing during drilling compressive residual stresses and strain-hardening, which are both beneficial to components fatigue life. The work presented in the thesis focuses firstly on the orbital drilling cutting parameters optimization, based on the specific cutting forces and energies minimization. An innovative mechanical surface treatment was introduced within the project: orbital deep rolling. Its characterization had been carried out using Finite Element Method. These models allowed to study further its parameters influence on boreholes surface integrity, particularly on residuals stress. A comparison with experimental results is achieved in order to validate the various numerical simulations implemented. Experimental characterization of the surface integrity of the boreholes performed with orbital drilling under optimized cutting conditions and with orbital deep rolling proved not only the feasibility of the orbital deep rolling process, but also compliance with aeronautical requirements. All this is completed by a fatigue life study of AA2024-T351 aluminum alloy drilled samples
43
Bibault, Jean-Emmanuel. "Prédiction par Deep Learning de la réponse complète après radiochimiothérapie pré-opératoire du cancer du rectum localement avancé Labeling for big data in radiation oncology: the radiation oncology structures ontology Big data and machine learning in radiation oncology: state of the art and future prospects Deep learning and radiomics predict complete response after neo-adjuvant chemoradiation for locally advanced rectal cancer." Thesis, Sorbonne Paris Cité, 2018. https://wo.app.u-paris.fr/cgi-bin/WebObjects/TheseWeb.woa/wa/show?t=2388&f=17288.
Full textAbstract:
L’utilisation de systèmes informatiques pour formaliser, organiser et planifier le traitement des patients a abouti à la création et à l’accumulation de quantité importante de données. Ces informations comprennent des caractéristiques démographiques, socio-économiques, cliniques, biologiques, d’imagerie, et, de plus en plus, génomiques. La médecine et sa pratique, fondées sur la sémiologie et la physiopathologie, vont être profondément transformées par ce phénomène. La complexité et la quantité des informations à intégrer pour prendre une décision médicale pourrait dépasser rapidement les capacités humaines. Les techniques d’intelligence artificielle pourraient assister le médecin et augmenter ses capacités prédictives et décisionnelles. La première partie de ce travail présente les types de données désormais accessibles en routine en oncologie radiothérapie. Elle détaille les données nécessaires à la création d’un modèle prédictif. Nous explorons comment exploiter les données spécifiques à la radiothérapie et présentons le travail d’homogénéisation et de conceptualisation qui a été réalisé sur ces données, notamment via la création d’une ontologie, dans le but de les intégrer à un entrepôt de données. La deuxième partie explore différentes méthodes de machine learning : k-NN, SVM, ANN et sa variante, le Deep Learning. Leurs avantages et inconvénients respectifs sont évalués avant de présenter les études ayant déjà utilisé ces méthodes dans le cadre de la radiothérapie. La troisième partie présente la création d’un modèle prédictif de la réponse complète à la radiochimiothérapie (RTCT) pré-opératoire dans le cancer du rectum localement avancé. Cette preuve de concept utilise des sources de données hétérogènes et un réseau neuronal profond dans le but d’identifier les patients en réponse complète après RTCT qui pourraient ne pas nécessiter de traitement chirurgical radical. Cet exemple, qui pourrait en pratique être intégré aux logiciels de radiothérapie déjà existant, utilise les données collectées en routine et illustre parfaitement le potentiel des approches de prédiction par IA pour la personnalisation des soinsThe use of Electronic Health Records is generating vast amount of data. They include demographic, socio-economic, clinical, biological, imaging and genomic features. Medicine, which relied on semiotics and physiopathology, will be permanently disrupted by this phenomenon. The complexity and volume of data that need to be analyzed to guide treatment decision will soon overcome the human cognitive abilities. Artificial Intelligence methods could be used to assist the physicians and guide decision-making. The first part of this work presents the different types of data routinely generated in oncology, which should be considered for modelling a prediction. We also explore which specific data is created in radiation oncology and explain how it can be integrated in a clinical data warehouse through the use of an ontology we created. The second part reports on several types of machine learning methods: k-NN, SVM, ANN and Deep Learning. Their respective advantages and pitfalls are evaluated. The studies using these methods in the field of radiation oncology are also referenced. The third part details the creation of a model predicting pathologic complete response after neoadjuvant chemoradiation for locally-advanced rectal cancer. This proof-of-concept study uses heterogeneous sources of data and a Deep Neural Network in order to find out which patient could potentially avoid radical surgical treatment, in order to significantly reduce the overall adverse effects of the treatment. This example, which could easily be integrated within the existing treatment planning systems, uses routine health data and illustrates the potential of this kind of approach for treatment personalization
44
Iwamoto, Roberto Kunihiro. "Alguns aspectos dos efeitos da interação solo-estrutura em edifícios de múltiplos andares com fundação profunda." Universidade de São Paulo, 2000. http://www.teses.usp.br/teses/disponiveis/18/18134/tde-08062006-163117/.
Full textAbstract:
O objetivo do trabalho é mostrar através de um modelo numérico a importância da consideração da interação estrutura - solo na análise global da estrutura. No modelo de estrutura tridimensional utilizado é considerada a contribuição da rigidez transversal à flexão das lajes, a existência de excentricidades das vigas em relação aos pilares e a hipótese de diafragma rígido no plano horizontal de cada pavimento. Serão consideradas fundações profundas em estacas verticais submetidos à carga de compressão axial ligadas ao bloco suposto rígido, nas quais o efeito de grupo de estacas imersas no solo é calculado considerando a continuidade do solo. A análise da interação do sistema solo estrutura será feito através dos ajustes das rigidezes de fundações pelo processo iterativo até que ocorra uma certa convergência nos recalques ou nas reações. Com isso, procura-se mostrar que a análise integrada da estrutura e o solo possibilita uma melhor estimativa dos recalques diferenciais e reações nos apoios, assim como a redistribuição dos esforços nos elementos estruturais com o comportamento mais real da interdependência dos esforços entre a estrutura e o soloThe main aim of this work is to use a numerical model for soil structure interaction and the importance of their consideration in a global structural analysis. For the structure the model considers the contribution of transverse bending stiffness of slabs, the exccentricy of beams in relation to the pile, and the hypothesis of rigid diaphragms in the plane of the slabs. Primary attention is placed on vertically loaded pile under rigid pile cap in which the influence of pile groups imerse in the soil is calculated considering the soil continuity. The analysis of soil structure interaction is done in an iterative process by adjusting the stiffness of the foundation until a certain preestablished convergence of calculated settlements or load reactions are obtained. In this manner its shown that the integrated analysis of the structure and soil medium leads to better results of differential settlements and load reactions of the supports. In the same manner, this analysis procedure leads to a better estimate of the internal forces in the structural elements, showing a more realistic behaviour of interdependence betwen the strucutre and the soil medium
45
Xue, Xin. "Modelling and control of twist springback in lightweight automotive structures with complex cross-sectional shape." Doctoral thesis, Universidade de Aveiro, 2016. http://hdl.handle.net/10773/17766.
Full textAbstract:
Doutoramento em Engenharia MecânicaEste trabalho é dedicado à investigação dos mecanismos / fontes de retorno elástico torsional em estruturas automóveis leves e à identificação de formas de controlar este problema. Em primeiro lugar, para garantir uma correta modelação do retorno elástico torsional, foram utlizados os resultados de vários ensaios do material, incluindo diferentes solicitações de carga/descarga, assim como a utilização de modelos constitutivos adequados. O comportamento mecânico dos materiais submetidos a trajetórias simples e complexas de carga é descrito utilizando leis de encruamento e critérios de plasticidade anisotrópicos. Foi desenvolvido um novo dispositivo de ensaios de corte para os aços DP para realização de ensaios de inversão de carga. Foram realizados testes cíclicos de carga-descarga-carga de tração uniaxial e biaxial assim como testes de dobragem em três pontos em material pré -deformado com vista à determinação da degradação do módulo de elasticidade com o aumento de deformação plástica. O efeito da trajetória de deformação na determinação do valor inicial do módulo de elasticidade e a sua degradação foram registados e analisados. Em segundo lugar, foram selecionados como casos de estudo dois processos clássicos de deformação plástica de metais, nomeadamente embutidura de chapas de aço DP e dobragem por matriz rotativa de tubos de alumínio de parede fina e secção assimétrica, devido ao seu evidente efeito de retorno elástico torsional. Foi proposta uma definição melhorada de retorno elástico torsional baseada nos eixos principais de inércia da secção transversal. A relação entre o momento de torção e ângulo de torção foi introduzida para explicar a ocorrência de retorno elástico torsional. Para melhorar a robustez dos modelos numéricos, foram realizadas várias técnicas de modelação, incluindo a identificação de coeficiente de atrito, a restrição de acoplamento da superfície para mandril flexível utilizando um elemento conector articulado, e a correlação de imagens digitais. O mecanismo de retorno elástico torsional foi analisado tendo em conta a evolução de estado plano de tensão e a trajetória de deformação nos componentes após a enformação por deformação plástica. Em terceiro lugar, foi analisada e discutida a sensibilidade dos modelos constitutivos de materiais no que diz respeito à precisão da previsão do retorno elástico torsional. Além disso, foi investigada a influência dos parâmetros do processo de embutidura profunda (direção de material, “blank-piercing” e lubrificação) e dos parâmetros numéricos do processo de dobragem de tubos (restrição dos limites do mandril flexível e atrito nas zonas de contacto) no retorno elástico torsional. Finalmente, foram propostas duas estratégias de controlo para o processo de embutidura profunda, com base no raio da curvatura da matriz variável e na posição dos freios, para reduzir o retorno elástico torsional de duas peças “Cchannel” e “P-channel”, respetivamente. No caso de dobragem de tubos, o controlo do retorno elástico torsional foi alcançado pela otimização da função do mandril e inclusão de um assistente de impulso de carga. Estas estratégias de controlo, baseadas em FEA, apresentam-se como métodos alternativos para a redução do momento torsor e do retorno elástico torsional em termos de aplicações específicas.
This work is devoted to the investigation of the mechanism/source of twist springback in lightweight automotive structures and to the identification of ways to control this problem. Firstly, to ensure accurate twist springback modelling, a reliable test data of material behaviours under various loading /unloading conditions as well as appropriate constitutive models are necessary. The anisotropic yield criteria and hardening models were adopted to characterize the material behaviours under monotonic and complex strain paths. An enhanced simple shear device was developed to obtain the stress-strain behaviour under reversal loading of DP steels. Uniaxial and biaxial loadingunloading- loading cycle tests and the proposed three-point bend test with prestrained sheets, were conducted to determine the elastic modulus degradation with the increase of plastic strain. A significant effect of the loading strategy on the determination of the initial and the degradation of elastic modulus was observed and discussed. Secondly, two typical metal forming processes, namely deep drawing of DP steel sheets and mandrel rotary draw bending of asymmetric thin-walled aluminium alloy tube, were selected as case studies due to their evident twist springback. A more reasonable definition of twist springback with respect to the principal inertia axes of the cross-sections was proposed. The relationship between torsion moment and twist angle was introduced to explain the occurrence of twist springback. Several key modelling techniques including the friction coefficient identification, surface-based coupling constraint for flexible mandrel using HINGE connector element and digital image correlation were performed for improving the robustness of the numerical models. The mechanism of twist springback was analysed from the evolution of in-plane stress and deformation history in the components after forming. Thirdly, the sensitivities of material constitutive models to the accuracy of twist springback prediction were analysed and discussed. The influence of deep drawing process parameters (material direction, blank piercing and lubrication) and numerical parameters of tube bending (boundary constraint for flexible mandrel and interfacial friction) on twist springback are provided. Finally, two control strategies for deep drawing process, based on variable die radius and partial draw bead design, were proposed to reduce the twist springback of the C-channel and the P-channel, respectively. In case of tube bending, the control of twist springback was reached by the optimization of mandrel nose placement and inclusion of push assistant loading. These FEAbased control strategies appear to be alternative methods to reduce the unbalance torsion moment and the twist springback in terms of particular case.
46
Nishiwaki, Takafumi. "Comparison of Damage Zones of the Nojima and the Asano Faults from the Deep Drilling Project: Differences in Meso-to-microscale Deformation Structures related to Fault Activity." Kyoto University, 2020. http://hdl.handle.net/2433/253096.
Full text
47
El, Haffar Ismat. "Physical modeling and study of the behavior of deep foundations of offshore wind turbines in sand." Thesis, Ecole centrale de Nantes, 2018. http://www.theses.fr/2018ECDN0021/document.
Full textAbstract:
La capacité axiale et latérale des pieux foncés dans du sable de Fontainebleau NE34 ont été étudié à l’aide d’essais sur modèles réduits centrifugés. L’effet de la méthode d’installation, de la densité et de la saturation du sable, du diamètre du pieu, de la géométrie de sa pointe (ouvert /fermé) et de sa rugosité sur la capacité axiale a été étudié. Une augmentation significative de la capacité en traction est observée dans les pieux foncés cycliquement, contrairement aux pieux foncés d’une manière monotone à 100 × g. La saturation du sable dense accélère la formation du bouchon lors de l'installation du pieu. L'augmentation de la rugosité du pieu et de la densité du sable accroissent significativement le frottement latéral des pieux testés. Dans tous les cas, les capacités de pieux sont comparées aux codes de dimensionnement des éoliennes offshore. Une étude paramétrique de l'effet de la méthode d'installation, de l'excentricité de la charge et de la saturation du sable sur la réponse latérale des pieux foncés est ensuite réalisée grâce à l'utilisation d'un pieu instrumentée. Le pieu est chargé d’une manière monotone puis un millier de cycles sont appliqués. Une nouvelle méthode a été développée pour la détermination des constantes d'intégration pour déterminer le profil de déplacement latéral du pieu. La méthode d'installation influence directement le comportement global (moment maximum et déplacement latéral) et local (courbes p-y) des pieux. L'effet de l'excentricité de la charge et de la saturation du sable sur le comportement des pieux est également présenté. Dans chaque cas, une comparaison avec les courbes p-y extraites du code DNVGL est réaliséeThe axial and lateral capacity of piles jacked in Fontainebleau sand NE34 are studied using centrifuge modelling at 100×g. The effect of the installation method, sand density and saturation, pile diameter and pile tip geometry (open or closed-ended) and pile roughness on the axial capacity of piles are firstly studied. A significant increase in the tension capacity is observed in cyclically-jacked piles unlike piles monotonically jacked at 100×g. The saturation of dense sand accelerates plug formation during pile installation. The increase in pile roughness and sand density increases significantly the shaft resistance of the piles tested here. For all the cases, pile capacities are compared with the current design codes for offshore wind turbines. A parametric study of the effect of the installation method, load eccentricity and sand saturation on the lateral response of jacked piles is then realized using of an instrumented pile. The pile is loaded monotonically, then a thousand cycles are applied. A new methodology has been developed for determining of the constants needed in the integration procedure to identify the lateral displacement profile of the pile. The installation method influences directly the global (maximum moment and lateral displacement) and local behaviour (p-y curves) of the piles. The effect of the load eccentricity and sand saturation on the behaviour of the piles is also presented. In each case a comparison with the p-y curves extracted from the DNVGL code is realized
48
Hemme, Christina [Verfasser], and Klaus-Jürgen [Akademischer Betreuer] Röhlig. "Storage of gases in deep geological structures$dspatial and temporal hydrogeochemical processes evaluated and predicted by the development and application of numerical modeling / Christina Hemme ; Betreuer: Klaus-Jürgen Röhlig." Clausthal-Zellerfeld : Technische Universität Clausthal, 2019. http://d-nb.info/1231363142/34.
Full text
49
Megner-Allogo, Alain-Cedrique. "Sedimentology and stratigraphy of deep-water reservoirs in the 9A to 14A Sequences of the central Bredasdorp Basin, offshore South Africa." Thesis, Stellenbosch : Stellenbosch University, 2006. http://hdl.handle.net/10019.1/17400.
Full textAbstract:
Thesis (MSc)--University of Stellenbosch, 2006.ENGLISH ABSTRACT: The Barremian to Albian siliciclastic deep-water deposits of the central Bredasdorp Basin were investigated primarily in terms of their stratigraphic evolution, depositional characteristics and facies distribution. Cores from the deep-water deposits reveal that the facies successions are composed of massive, ripple cross- to parallel-laminated sandstones, conglomerate, massive claystone, alternating laminated to interbedded sandstone/siltstone and claystone, laminated and clay-rich siltstone. These facies are grouped into channel-fill, sheet-lobe, overbank and basin plain deposits, by inference. The application of sequence stratigraphy, based on gamma ray and resistivity log patterns, reveals that all 3rd-order depositional sequences comprise 4thorder cycles. The latter are subdivided into three components (lowstand, transgressive and highstand systems tracts), based on vertical facies changes and internal stratigraphic key surfaces. Taking the 13Amfs as the stratigraphic datum for each well, correlation was possible on a regional basis. Lowstand deposits, comprising thick amalgamated massive sandstones, were interpreted to represent channelfills. Their vertical and horizontal stacking forms channel-fill complexes above Type 1 unconformities. Adjacent thin-bedded intervals, comprising parallel- to ripple cross-laminated sandstones, were interpreted as levee/overbank deposits, whereas clay-rich intervals were interpreted to represent basin plain deposits of hemipelagic origin. Facies associations and their distribution have revealed that channel-fills are associated with overflow deposits and sheet sand units. These deposits, as well as downdip sheet sands associated with small channel-fills within the 9A, 11A/12A, 13A Sequences and the 14A Sequence were interpreted to have been deposited in a middle fan to upper fan setting. A similar association occurs in the 10A Sequence, except that thick conglomerate units are present at the base of proximal channel-fills. This led to interpret the 10A Sequence as being deposited in a base-of-slope to upper fan setting. The thickness of each sequence, as revealed by isochore maps, shows sinuous axial flow path which corresponds to channel-fill conduit. The continuous decrease of this sinuosity upward in the succession was interpreted as being related to basin floor control along the main sand fairways. Successive flows result in erosion-fill-spill processes, which locally favour connectivity of reservoirs over large areas. Recognition of higher-order sequences and key stratigraphic surfaces helps to understand internal stratigraphic relationships and reveals a complex and dynamic depositional history for 3rd-order sequences. However, sparse well control and uneven distribution of boreholes, as well as lack of seismic and other data, limited the models derived for this study.
AFRIKAANSE OPSOMMING: Die Barremiaanse tot Albiaanse silisiklastiese diepwater afsettings van die sentrale Bredasdorp Kom is hoofsaaklik in terme van stratigrafiese evolusie, afsettingskarakteristieke en fasies distribusie ondersoek. Kerne van die diepwater afsettings toon dat die fasies opeenvolgings uit massiewe, riffelkruis- tot parallel-gelamineerde sandstene, konglomerate, massiewe kleistene, afwisselende gelamineerde tot intergelaagde sandstene/slikstene en kleistene, sowel as gelamineerde en klei-ryke slikstene bestaan. Hierdie fasies word onderverdeel in kanaalopvulsel, plaatlob, oewerwal en komvlakte afsettings. Die toepassing van opeenvolgingsstratigrafie gebaseer op gammastraal en resistiwiteit log patrone toon dat alle 3de-orde afsettingsopeenvolgings uit 4deorde siklusse bestaan. Laasgenoemde word onderverdeel in drie komponente (lae-stand, transgressie en hoë-stand sisteemgedeeltes), gebaseer op vertikale fasies veranderinge en interne stratigrafiese sleutel vlakke. Korrelasie op ‘n regionale basis is moontlik gemaak deur die 13Amfs as die stratigrafiese verwysing vir elke boorgat te neem. Lae-stand afsettings, wat uit dik saamgevoegde massiewe sandstene bestaan, word geïnterpreteer as kanaalopvulsels. Die vertikale en horisontale stapeling van die sandstene vorm kanaalopvulsel komplekse bo Tipe 1 diskordansies. Naasliggende dungelaagde intervalle, wat uit parallel- tot kruisgelaagde sandstene bestaan, word geïnterpreteer as oewerwal afsettings, terwyl klei-ryke intervalle geïnterpreteer word as verteenwoordigend van komvlakte afsettings van hemipelagiese oorsprong. Fasies assosiasies en hul verspreiding toon dat kanaalvul geassosieër word met oorvloei afsettings en plaatsand eenhede. Hierdie afsettings, sowel as distale plaatsande geassosieër met klein kanaalopvulsels binne die 9A, 11A/12A, 13A en die 14A Opeenvolgings, word geïnterpreteer as afgeset in ‘n middelwaaier tot bo-waaier omgewing. ‘n Soortgelyke assosiasie bestaan in die 10A Opeenvolging, behalwe dat dik konglomeraat eenhede teenwoordig is by die basis van proksimale kanaalopvullings. Dit het gelei tot die interpretasie van die 10A Opeenvolging as afgeset in ‘n basis-van-helling tot bo-waaier omgewing. Die dikte van elke opeenvolging, soos verkry vanaf isochoor kaarte, toon ‘n kronkelende aksiale vloeipad wat ooreenkom met ‘n kanaalopvulling toevoerkanaal. Die aaneenlopende afname van hierdie kronkeling na bo in die opeenvolging word geïnterpreteer as verwant aan komvloer-beheer langs die hoof sand roetes. Opeenvolgende vloeie veroorsaak erosie-opvul-oorspoel prosesse, wat lokaal die konnektiwiteit van reservoirs oor groot areas bevoordeel. Herkenning van hoër-orde opeenvolgings en sleutel stratigrafiese vlakke dra by tot ‘n goeie begrip van die interne stratigrafiese verhoudings en ontbloot ‘n komplekse en dinamiese afsettingsgeskiedenis vir 3de-orde opeenvolgings. Beperkte boorgatbeheer en ‘n tekort aan seismiese en ander data het egter ‘n beperkende rol gespeel in die daarstel van modelle vir hierdie studie.
50
Gidaris, Spyridon. "Effective and annotation efficient deep learning for image understanding." Thesis, Paris Est, 2018. http://www.theses.fr/2018PESC1143/document.
Full textAbstract:
Le développement récent de l'apprentissage profond a permis une importante amélioration des résultats dans le domaine de l'analyse d'image. Cependant, la conception d'architectures d'apprentissage profond à même de résoudre efficacement les tâches d'analyse d'image est loin d'être simple. De plus, le succès des approches d'apprentissage profond dépend fortement de la disponibilité de données en grande quantité étiquetées manuellement (par des humains), ce qui est à la fois coûteux et peu pratique lors du passage à grande échelle. Dans ce contexte, l'objectif de cette thèse est d'explorer des approches basées sur l'apprentissage profond pour certaines tâches de compréhension de l'image qui permettraient d'augmenter l'efficacité avec laquelle celles-ci sont effectuées ainsi que de rendre le processus d'apprentissage moins dépendant à la disponibilité d'une grande quantité de données annotées à la main. Nous nous sommes d'abord concentrés sur l'amélioration de l'état de l'art en matière de détection d'objets. Plus spécifiquement, nous avons tenté d'améliorer la capacité des systèmes de détection d'objets à reconnaître des instances d'objets (même difficiles à distinguer) en proposant une représentation basée sur des réseaux de neurone convolutionnels prenant en compte le aspects multi-région et de segmentation sémantique, et capable de capturer un ensemble diversifié de facteurs d'apparence discriminants. De plus, nous avons visé à améliorer la précision de localisation des systèmes de détection d'objets en proposant des schémas itératifs de détection d'objets et un nouveau modèle de localisation pour estimer la boîte de délimitation d'un objet. En ce qui concerne le problème de l'étiquetage des images à l'échelle du pixel, nous avons exploré une famille d'architectures de réseaux de neurones profonds qui effectuent une prédiction structurée des étiquettes de sortie en apprenant à améliorer (itérativement) une estimation initiale de celles-ci. L'objectif est d'identifier l'architecture optimale pour la mise en œuvre de tels modèles profonds de prévision structurée. Dans ce contexte, nous avons proposé de décomposer la tâche d'amélioration de l'étiquetage en trois étapes : 1) détecter les estimations initialement incorrectes des étiquettes, 2) remplacer les étiquettes incorrectes par de nouvelles étiquettes, et finalement 3) affiner les étiquettes renouvelées en prédisant les corrections résiduelles. Afin de réduire la dépendance à l'effort d'annotation humaine, nous avons proposé une approche d'apprentissage auto-supervisée qui apprend les représentations sémantiques d'images à l'aide d'un réseau de neurones convolutionnel en entraînant ce dernier à reconnaître la rotation 2d qui est appliquée à l'image qu'il reçoit en entrée. Plus précisément, les caractéristiques de l'image tirées de cette tâche de prédiction de rotation donnent de très bons résultats lorsqu'elles sont transférées sur les autres tâches de détection d'objets et de segmentation sémantique, surpassant les approches d'apprentissage antérieures non supervisées et réduisant ainsi l'écart avec le cas supervisé. Enfin, nous avons proposé un nouveau système de reconnaissance d'objets qui, après son entraînement, est capable d'apprendre dynamiquement de nouvelles catégories à partir de quelques exemples seulement (typiquement, seulement un ou cinq), sans oublier les catégories sur lesquelles il a été formé. Afin de mettre en œuvre le système de reconnaissance proposé, nous avons introduit deux nouveautés techniques, un générateur de poids de classification basé sur l'attention et un modèle de reconnaissance basé sur un réseau neuronal convolutionnel dont le classificateur est implémenté comme une fonction de similarité cosinusienne entre les représentations de caractéristiques et les vecteurs de classificationRecent development in deep learning have achieved impressive results on image understanding tasks. However, designing deep learning architectures that will effectively solve the image understanding tasks of interest is far from trivial. Even more, the success of deep learning approaches heavily relies on the availability of large-size manually labeled (by humans) data. In this context, the objective of this dissertation is to explore deep learning based approaches for core image understanding tasks that would allow to increase the effectiveness with which they are performed as well as to make their learning process more annotation efficient, i.e., less dependent on the availability of large amounts of manually labeled training data. We first focus on improving the state-of-the-art on object detection. More specifically, we attempt to boost the ability of object detection systems to recognize (even difficult) object instances by proposing a multi-region and semantic segmentation-aware ConvNet-based representation that is able to capture a diverse set of discriminative appearance factors. Also, we aim to improve the localization accuracy of object detection systems by proposing iterative detection schemes and a novel localization model for estimating the bounding box of the objects. We demonstrate that the proposed technical novelties lead to significant improvements in the object detection performance of PASCAL and MS COCO benchmarks. Regarding the pixel-wise image labeling problem, we explored a family of deep neural network architectures that perform structured prediction by learning to (iteratively) improve some initial estimates of the output labels. The goal is to identify which is the optimal architecture for implementing such deep structured prediction models. In this context, we propose to decompose the label improvement task into three steps: 1) detecting the initial label estimates that are incorrect, 2) replacing the incorrect labels with new ones, and finally 3) refining the renewed labels by predicting residual corrections w.r.t. them. We evaluate the explored architectures on the disparity estimation task and we demonstrate that the proposed architecture achieves state-of-the-art results on the KITTI 2015 benchmark.In order to accomplish our goal for annotation efficient learning, we proposed a self-supervised learning approach that learns ConvNet-based image representations by training the ConvNet to recognize the 2d rotation that is applied to the image that it gets as input. We empirically demonstrate that this apparently simple task actually provides a very powerful supervisory signal for semantic feature learning. Specifically, the image features learned from this task exhibit very good results when transferred on the visual tasks of object detection and semantic segmentation, surpassing prior unsupervised learning approaches and thus narrowing the gap with the supervised case.Finally, also in the direction of annotation efficient learning, we proposed a novel few-shot object recognition system that after training is capable to dynamically learn novel categories from only a few data (e.g., only one or five training examples) while it does not forget the categories on which it was trained on. In order to implement the proposed recognition system we introduced two technical novelties, an attention based few-shot classification weight generator, and implementing the classifier of the ConvNet based recognition model as a cosine similarity function between feature representations and classification vectors. We demonstrate that the proposed approach achieved state-of-the-art results on relevant few-shot benchmarks