L’explosion récente des volumes de données disponibles a fait de la complexité algorithmique un élément central des méthodes d’apprentissage automatique. Les algorithmes d’optimisation stochastique ainsi que les méthodes distribuées et décentralisées ont été largement développés durant les dix dernières années. Ces méthodes ont permis de faciliter le passage à l’échelle pour optimiser des risques empiriques dont la formulation est séparable en les observations associées. Pourtant, dans de nombreux problèmes d’apprentissage statistique, l’estimation précise du risque s’effectue à l’aide de U-statistiques, des fonctions des données prenant la forme de moyennes sur des d-uplets. Nous nous intéressons tout d’abord au problème de l’échantillonnage pour la minimisation du risque empirique. Nous montrons que le risque peut être remplacé par un estimateur de Monte-Carlo, intitulé U-statistique incomplète, basé sur seulement O(n) termes et permettant de conserver un taux d’apprentissage du même ordre. Nous établissons des bornes sur l’erreur d’approximation du U-processus et les simulations numériques mettent en évidence l’avantage d’une telle technique d’échantillonnage. Nous portons par la suite notre attention sur l’estimation décentralisée, où les observations sont désormais distribuées sur un réseau connexe. Nous élaborons des algorithmes dits gossip, dans des cadres synchrones et asynchrones, qui diffusent les observations tout en maintenant des estimateurs locaux de la U-statistique à estimer. Nous démontrons la convergence de ces algorithmes avec des dépendances explicites en les données et la topologie du réseau. Enfin, nous traitons de l’optimisation décentralisée de fonctions dépendant de paires d’observations. De même que pour l’estimation, nos méthodes sont basées sur la concomitance de la propagation des observations et l’optimisation local du risque. Notre analyse théorique souligne que ces méthodes conservent une vitesse de convergence du même ordre que dans le cas centralisé. Les expériences numériques confirment l’intérêt pratique de notre approche. / With the increasing availability of large amounts of data, computational complexity has become a keystone of many machine learning algorithms. Stochastic optimization algorithms and distributed/decentralized methods have been widely studied over the last decade and provide increased scalability for optimizing an empirical risk that is separable in the data sample. Yet, in a wide range of statistical learning problems, the risk is accurately estimated by U-statistics, i.e., functionals of the training data with low variance that take the form of averages over d-tuples. We first tackle the problem of sampling for the empirical risk minimization problem. We show that empirical risks can be replaced by drastically computationally simpler Monte-Carlo estimates based on O(n) terms only, usually referred to as incomplete U-statistics, without damaging the learning rate. We establish uniform deviation results and numerical examples show that such approach surpasses more naive subsampling techniques. We then focus on the decentralized estimation topic, where the data sample is distributed over a connected network. We introduce new synchronous and asynchronous randomized gossip algorithms which simultaneously propagate data across the network and maintain local estimates of the U-statistic of interest. We establish convergence rate bounds with explicit data and network dependent terms. Finally, we deal with the decentralized optimization of functions that depend on pairs of observations. Similarly to the estimation case, we introduce a method based on concurrent local updates and data propagation. Our theoretical analysis reveals that the proposed algorithms preserve the convergence rate of centralized dual averaging up to an additive bias term. Our simulations illustrate the practical interest of our approach.

abstract: Geochronology and thermochronology are valuable tools for investigating the synergy between the deformational and erosional processes that shape mountainous terrains. Though numerous techniques have been developed to probe the rate and timing of events within these settings, the research presented here explores how scientists can use fewer samples to produce richer data products with broader contextual importance. The beginning of this compilation focuses on establishing laboratory techniques to facilitate this goal. I developed a novel laser ablation ‘double dating’ (LADD) technique that rapidly yields paired U/Pb and (U-Th)/He dates for the accessory minerals zircon, titanite, and apatite. The technique obviates the need for geometric corrections typically applied during (U-Th)/He data reduction, enables the analysis of a broader spectrum of detrital crystals, and provides the opportunity for additional mapping and isotopic analyses that are traditionally challenging to procure and/or fraught with assumptions. Despite the technique’s promise, I also found it essential to weigh several considerations of relevance when attempting to date young (≤ Miocene) accessory minerals with low concentrations of U + Th. Consequently, I discuss the impact that such variables have on the magnitude of analytical imprecision and the data’s flexibility for geologic interpretation. Beyond the lab, I collected a suite of bedrock and detrital samples from small catchments draining the southeastern Sierra Nevada mountains of California. Using the techniques described above as well as conventional methods for (U-Th)/He zircon dating, I compared the utility of both bedrock and detrital approaches for extrapolating local exhumation histories. I additionally tested the ability to employ detrital datasets to extrapolate cooling histories that span from mineral crystallization to rock exhumation through the upper crust. Employing principal mode dates from a combination of zircon and apatite LADD dates and detrital hornblende 40Ar/39Ar dates, I was able to derive thermal models that demonstrate the existence of significant variability in the cooling histories of various intrusive units along the eastern Sierra Nevada. While these results only scratch the surface of what’s possible within the realm of detrital-based research, this contribution demonstrates the utility of expanding the temporal and spatial scope of traditional detrital methodologies. / Dissertation/Thesis / Doctoral Dissertation Geological Sciences 2019

In this project, urea-formaldehyde (UF) resins were produced and investigated. UF resins are commonly used indoors as wood adhesives in, e.g. particle boards, in different furniture applications and flooring. UF resins are produced by alternating methylolation and condensation reactions, thus reacting urea and formaldehyde with each other and creating longer polymeric chains. The number of alternations, i.e. number of condensation reactions can be varied. The focus laid on the effect of the formaldehyde/urea molar ratio during synthesis. This includes the effect of the molar ratio on both the composition and structure and in turn their effect on the properties of the resin. UF resin was synthesized via two different methods. In method one, a reference adhesive was synthesized, this adhesive was produced using three condensation reactions. In method one, another resin was also produced using only two condensation reactions but with the same F/U molar ratios. In method two, only two condensations reactions were performed for each resin. In the start of the reaction, the F/U-molar ratio was varied in three different values. Towards the end, however, different amounts of urea were added to make sure that the final F/U-molar ration was kept constant. The results showed that the F/U molar ratio during synthesis will have an effect on both the composition of the resin and the structures being formed. With less urea added in the beginning as in the case with a high starting molar ratio, more dimethyl ether bridges and methylol groups, but less methylene bridges were formed in the end-product. This was formed together with a more branched UF structure with a higher polydispersity index. The reason for this probably originates from the more highly substituted amino groups formed due to the lower amount of urea in the beginning in the sample with a high F/U molar ratio. The composition and structure of the resin with a higher F/U molar ratio also seem to favor a stronger tensile strength. The increased amount of methylol groups and the increased PDI are thought to have a larger effect on the increased tensile strength. The shelf life of the finished resin also showed dependency with the F/U molar ratio, since resins produced with a low molar ratio gelled within 24h compared the two other variations of F/U ratios, where it took about 90 days to gel at the same temperature.

Bostadsbristen i Sverige påverkar 75% av alla kommuner. För att kunna lösa krisen snabbt och effektivt bör byggbranschen ta nya idéer och byggteknik på allvar. Att bygga bostäder av fraktcontainer innebär att bygga snabbt och miljövänligt.Denna typ av projekt behöver utveckla för att anpassa det till det svenska hårda vädret och uppfylla kraven som ställs av boverket BBR. Isolering av en containeryttervägg anses som ett problem när det gäller denna typ av projekt. För att flera ska bygga bostäder av fraktcontainer behöver problemet att lösa. Genom att samla in information och ta inblick i redan byggda bostäder av fraktcontainer ska två olika alternativ för ytterväggars isolering jämföras. Risk för köldbryggor, U-värde och kostnad ska jämföras. / In Sweden, 75% of municipalities are affected by a shortage of housing. To solve the housing crisis, the construction industry must adopt effective and innovative construction techniques and methods. By building housing out of shipping containers, the result will be fast, environmentally friendly, and costeffective. However, this type of housing needs a building technique developed and adjusted for the harsh Swedish climate and fulfill the National Board of Housing, Building, and Planning requirements. One significant problem hindering the process of this project is isolating the container’s exterior wall. Solving this problem will encourage the construction industry to build this type of housing. Through data collection and considering shipping container housing built already, this paper will make a comparison between two different options for wall insulation will be compared. Further the U- value, costand thermal bridges risk will be compared.

No description available.

[pt] Foram determinados experimentalmente fatores de concentração de tensões teóricos para placas e eixos com entalhes simétricos e circunferenciais em U sujeitos à combinação de esforços de flexão e tração, para as placas e, flexão e torção para os eixos. Os fatores obtidos foram plotados em gráficos para comparação com fatores determinados aplicando-se o princípio da superposição. Foram utilizadas as relações geométricas D / d = 1,5; 1,25; 1,10 e r/d = 0,30; 0,25; 0,20 para as placas. Para os eixos utilizaram-se apenas as relações 0/d = 1,3 e r/d = 0,25. Também foram obtidos fatores de concentração de tensões para os esforços puros e seus valores comparados com resultados já publicados. Os resultados para as placas foram obtidos a partir da fotoelasticidade bidimensional. Foi empregado o método fotoelástico tridimensional de congelamento de tensões para os eixos. Neste caso, os modelos foram cortados em finas fatias para separação das tensões. Para o estudo tridimensional foi utilizado um material desenvolvido no laboratório de Fotomecanica da PUC-RJ. / [en] Theoretical stress concentration factors were determined experimentally for bars and shafts with symmetrical and circunferential U – shaped grooves subjected to combinations of bending – tension loads for the bars and bending – torsion loads for the shafts. The stress concentration factors obtained were plotted to compare with the factors determined by application of the superposition effects. The geometric relations used for the bars were Dd = 1,5; 1,25; 1,10 and rd = 0,30; 0,25; 0,20. The geometric relations used for the shafts were Dd = 1,3 e rd = 0,25. The stress concentration factors were also determined for single loaded specimens and they were compared wuth data already published. The stress concentration factors for the bars were determined by the two – Dimensional photoelastic method. In the case of the shafts they were determined by the three – Dimensional photoelastic Method of stress Freezing. In this method, the models developed in the laboratory of Photomechanics of the Universidade Católica.

Determining how loads are distributed in a structure has long been a way for engineers to ensure that the final product will work as intended. As components become more and more complicated so does this analysis. The advantages of quickly identifying critical features in a design are significant. By gaining this knowledge it is possible to explore the design space more thoroughly. This was previously done by collecting information over a long period of time and gradually build up the knowledge which can take several weeks or sometimes even months. The theory of the U* index as a numerical analysis method was formulated in 1986 by Takahashi [30] but due to the computational capabilities at the time there was no practical applications. It aims to give the same information compared to the current iterative process in a significantly faster way by calculating the relative stiffness. In its current form it can however be computationally heavy and has not yet seen any extended applications in the engineering sector. As the U* index is both an algorithm for calculating the load paths but also a broader theory of energy based identification of the stiffest route in a structure, it is referred to both as a method and a tool. In this thesis we initially set out to identify needs in the product development processfor a engineer team. Then we will investigate the capabilities of the U* index method, improve said capabilities by optimizing computation time and compatibility, establish a foundation for future development, and finally propose an integration of the tool into an advanced aero-engine development process. The main focus of this thesis is on the concept solutions phase as the needfinding indicated this phase to be in most need of improvement. Other applications of U* are mentioned but will not be examined in detail. In the concept solutions phase one would want to evaluate design features from a larger perspective to explore thedesign space. This can be done with U* by identifying which features are carrying the loads and how close the calculated path is to the optimal one. A possible application in optimization in the preliminary solution phase were also identified. In this case one can for example analyze the contour of the U* field and use U* sum to identify the relative load carrying contribution of certain regions. The initial state of the technology was determined to be at a technology readiness level (TRL) of 5 in the ModSim scale, meaning that the key elements have been demonstrated on a realistic problem. As the goal of applying the tool on real large models was achieved, the complete system level capabilities were demonstrated resulting in TRL 6 being reached at the end of the development. The more commonly used NASA version of TRL started out at TRL 3 which was reached with the proof of concept conducted by Ramesh [24]. According to these definitions the technology didn’t progress any further due to the specific definitions being primarily oriented towardshardware and the requirement for TRL 4 involved laboratory validation. As the TRL have been assessed with both versions, any potential future work has the option tochoose which method fit that particular work. The successful implementation of theinspection load method for calculating U∗ was critical in being able to compute thesolutions in reasonable time frame. There is however more room to optimize time byimplementing algorithms for better suited meshing and post processing. The tool is applicable on large components but the solutions still require a lot of time which reduces its usefulness in the early, time critical phase. The post-processing routines similarly take time and require extensive manual labor to determine the proper load path. This is also sub optimal for visualizing the load paths because it relies on assumptions. Future work should therefore aim to maximize the automation part of the post-processing. Further validation needs to be conducted on more complicated structures to ensure correlation between physical components and simulation. Furthermore, the quantifying of margins, uncertainties and sensitivities of the simulation is important to better understand the limitations of the tool.

No description available.

This research introduces parallel U-shaped assembly lines with crossover points. Crossover points are connecting points between two parallel U-shaped lines making the lines interdependent. The assembly lines can be employed to manufacture a variety of products belonging to the same product family. This is achieved by utilizing the concepts of crossover points, multi-line stations, and regular stations. The binary programming formulation presented in this research can be employed for any scenario (e.g. task times, cycle times, and the number of tasks) in the configuration that includes a crossover point. The comparison of numerical problem solutions based on the proposed heuristic approach with the traditional approach highlights the possible reduction in the quantity of workers required. The conclusion from this research is that a wider variety of products can be manufactured at the same capital expense using parallel U-shaped assembly lines with crossover points, leading to a reduction in the total number of workers. / M. S.

An experimental program was conducted at the Ferguson Structural Engineering Laboratory of The University of Texas at Austin, under the auspices of Texas Department of Transportation (TxDOT) Research Project 5831, to investigate the tensile stresses that develop in the end regions of pretensioned concrete U-beams at transfer of prestress. Understanding the effect of these “bursting” and “spalling” stresses is essential in order to design standard details that might lead to reliably-serviceable end regions. Two full-scale beam specimens, designed to be worst-case scenarios for bursting and spalling, were fabricated. Each beam had one square and one highly skewed end. Extensive instrumentation, including strain gages on transverse and lateral reinforcing bars, was employed in the end regions of these U-beams. Experimentally determined bursting and spalling stresses in these bars were compared to results of past projects (from the literature) investigating I-beams and inverted T-beams. Preliminary recommendations are made for changes in the standard reinforcing details for U-beam end regions. Such recommended details will be tested in the upcoming phase of Research Project 5831. / text