Variable screening techniques are fast and crude techniques to scan high-dimensional data and conduct dimension reduction before a refined variable selection method is applied. Its marginal analysis feature makes the method computationally feasible for ultra-high dimensional problems. However, most existing screening methods for classification problems are designed only for binary classification problems. There is lack of a comprehensive study on variable screening for multi-class classification problems. This research aims to fill the gap by developing variable screening for multi-class problems, to meet the need of high dimensional classification. The work has useful applications in cancer study, medicine, engineering and biology. In this research, we propose and investigate new and effective screening methods for multi-class classification problems. We consider two types of screening methods. The first one conducts screening for multiple binary classification problems separately and then aggregates the selected variables. The second one conducts screening for multi-class classification problems directly. In particular, for each method we investigate important issues such as choices of classification algorithms, variable ranking, and model size determination. We implement various selection criteria and compare their performance. We conduct extensive simulation studies to evaluate and compare the proposed screening methods with existing ones, which show that the new methods are promising. Furthermore, we apply the proposed methods to four cancer studies. R code has been developed for each method.

A 67" x 104" area of the metal-rich globular cluster M71 was searched for variable stars using 73 CCD frames. Using mean B and V values a colour-magnitude diagram down to V ≈ 22 is constructed. Four variables were discovered, with two more stars classed as possible candidates for variability. Phase diagrams and real time light curves are presented for all variables. One variable blue straggler (or SX Phe star) has been discovered with a period of 0.d05181. Values for the mass depend on the pulsation mode assumed for this star - (0.90 ± 0.13)M. for the first overtone mode, and (1.57 ± 0.22)M. for the fundamental mode. A second variable, of similar period (0.d06053), but with a magnitude l.m75 below the main sequence turn off was also found but it may be a field star. Two candidate eclipsing binary systems were found. The most likely period of one is 0.d37244. This value, and the shape of the light curve, suggest it may be a W UMa type variable. Its position on the CMD suggests that it too may be a field star. No period was obtainable for the other candidate binary due to a lack of phase coverage. Further data is needed to confirm and strengthen these claims. / Science, Faculty of / Physics and Astronomy, Department of / Graduate

Cellular structures are highly evaluated due to their high material efficiency. Both theoretical and experimental studies have done on periodic cellular structures. However, the mechanical performance can be stochastically distributed in the cellular architecture. This thesis presents the design and manufacturing of variable stiffness cellular architecture to achieve optimized topology by changing the unit cell parameters. The author applies image analysis technique to extract and digitize the information from the performance distribution map. Two types of cellular cells are studied for their relationship of stiffness and relative density. The methods of voxelization for both cells are also given in this study. This proposed methodology is then implemented to design a customized mattress and compare with current existing mattress. With the study of the unit cells and voxelization technique, our designed mattress aligns body curve better which provides more recuperation of the body during sleep.

Ces travaux concernent l'étude expérimentale e analytique de la turbulence en phase de développement dans les fluides hétérogènes à densité et à viscosité variable. Ils font appel à des outils de diagnostics expérimentaux (anémométrie à fil chaud, technique de diffusion Rayleigh, Vélocimétrie Doppler Laser), et au formalisme des équations de Navier-Stockes à viscosité variable. L'innovation porte sur l'indépendance de la mesure de la vitesse. Après sa validation, la plate-forme expérimentale est exploitée pour l'étude comparative d'un jet de propane émergeant dans un milieu air-néon, à viscosité et densité variable, avec un jet d'air classique, à même quantité de mouvement injectée initialement. Ce travail se poursuit ensuite par un approfondissement des propriétés dans le champ proche, complétés par une approche analytique à partir des réécritures des équations de Navier-Stokes à viscosité variable. / This work is devoted to the study of the undeveloped turbulence in heterogeneus gaseus mixtures, using experimental tools (Hot-wire Anemometry, Rayleigh Light Scattering, Laser Doppler Velocimetry) and analytical methods (variable-viscosity Navier Stokes equations). A new technique combining HWA and RLS is first adapted to reliabily measure the fluctuating velocity and concentration fields in variable-viscosity flows (herein, a propane-air mixture). A variable-viscosity round jet (propane emerging into an air-neon mixture) is characterized and compared with a turbulent air jet discharging into still air, at the same initial jet momentum. An analytical work is further performed with a particular focus on the jet axis, based on the Navier-Stokes equations including variable viscosity to support the experiments. It is shown that the kinetic energy dissipation rate is enhanced by several additional terms, particularly involving 'viscosity-velocity' correlations.

Companies are no longer judged on financial performance alone, but rather on their “Triple-Bottom Line”, which accounts for social and environmental measures as well. This leads companies to investigate the sustainability of their operations and their products. Between the increasing cost of virgin raw materials and customers demanding post-consumer product content, manufacturers have begun looking at recycled material options. This paper addresses a procurement issue facing a polystyrene packaging manufacturer considering its optimal purchasing strategies between two suppliers – one providing virgin material, the other offering recycled material. A single-period scenario is modeled where each supplier sells product with a known yield distribution at market pricing. The manufacturer must choose whether to sole-source or dual-source, as well as determine how much material to purchase from each supplier to meet deterministic demand. Our results indicate that there is a range of prices from the recycled material supplier where dual-sourcing will lead to higher manufacturer profits compared to sole-sourcing. We show, based on the procurement strategy, the optimal quantities to purchase to maximize manufacturer’s expected profit. We then investigate the area of supplier development and how the manufacturer can improve their expected profit by investing in their supplier’s quality improvement effort. The questions addressed are how much the manufacturer would be willing to invest and how they ensure the proper return on their investment. This paper determines the expected increase in profit for the manufacturer from yield improvement projects at a supplier, which therefore becomes the upper threshold for investment. We also find that a company can err in their project acceptance criteria if they have an approval process that views project acceptance myopically rather than holistically. Lastly, we develop a systematic and comprehensive approach to the supplier selection process. We utilize the fundamental concepts behind W. Edwards Deming’s Plan-Do-Study-Act improvement cycle and apply them to the supplier selection process. We also present analytic and numerical study results that can be used in conjunction with contractual mechanisms to not only overcome issues such as free riding, but to also incentivize suppliers to engage in supplier development projects they may not have otherwise undertaken.

No description available.

This study evaluates the performance of a semiactive magnetorheological primary suspension on a heavy truck application. A set of magnetorheological dampers is designed, fabricated, and characterized. The set of magnetorheological test dampers are implemented on a Volvo VN heavy truck. An embedded controller determines the level of damping to be supplied by the test dampers. The level of damping in each of the controllable magnetorheological dampers is determined according to a skyhook control scheme. Eleven PCB Piezotronics accelerometers are used to measure the acceleration at various points on the truck. The measurement positions include four measurements on the axles of the truck, and four measurements on the frame of the truck. This data is both recorded for post-test analysis and determining the damping level during testing. Q Additionally, three accelerometers measure the roll, heave, and pitch of the truck cab. The performance of the truck equipped with the semiactive magnetorheological suspension is primarily compared to the performance of the truck with the original (stock) passive system. Results from operation with the adjustable dampers fixed in both their on and off states are also given. The performance comparison between the semiactive and the original passive system is performed for two different driving situations. The first comparison between the two suspension types is for a test case where the truck is driven over a speed bump at approximately 6-7 mph. The second comparison is for the test case where the truck is driven at a constant speed along a stretch of straight and level highway at a constant speed of 55 mph. Acceleration data for both of these test cases is analyzed in the time domain (RMS and peak values of acceleration), and in the frequency domain (average peak intensity in different frequency bands). The findings presented here are confined to the specific magnetorheological dampers that were tested on the truck. Little effort was spent on tuning the high and low states of the adjustable dampers. In addition, the controller used was relatively crude, in the sense that it only implemented the on-off skyhook policy. The findings are meant to highlight some of the potential benefits, as well as shortcomings, of the magnetorheological dampers for heavy truck applications. The data for driving the truck over speed bumps indicate that the magnetorheological dampers used in this study with the skyhook control policy have only a small effect on the vehicle body and wheel dynamics, as compared to the passive stock dampers. The highway data shows that magnetorheological dampers and skyhook control policy are effective at reducing the RMS value of the measured acceleration at most measurement points, as compared to the stock dampers. / Master of Science

Dans un anneau de polynômes à $n$ indéterminées $A\n=A[y_1\tr y_n]$ à coefficients dans un anneau commutatif unitaire $A$ on dit qu'un polynôme $p=p(y_1\tr y_n)$ est une variable ou $A$-variable s'il existe un ($A$-)automorphisme $\alpha$ de $A\n$ tel que $p=\alpha(y_1)$. Dans cette thèse, on donne une construction assez générale de variables de $A\n$ par conjugaison d'automorphismes de $A\n$ avec des automorphismes de $(\Quot A)\n$. On définit les variables résiduelles qui désigne des polynômes qui sont des variables modulo $\Max$ pour tout idéal ma\-xi\-mal $\Max$ de $A$, en particulier, lorsque $A=\C\x=\C[x_1\tr x_k]$, on parle de variables $\xb$-résiduelles. Bien entendu les variables sont des variables résiduelles mais la réciproque est-elle vraie? On montre, grâce à un résultat de Daigle et Freudenburg, que les variables $\xb$-résiduelles de $\C\x[y,z]$ sont bien des $\xb$-variables. Les variables interviennent également dans les problèmes d'hyperplans plongés d'Abhyankar-Sathaye; un polynôme $p$ de $A\n$ est un ($A$-)hyperplan si le quotient de $A\n$ par l'idéal principal engendré par $p$, $(p)$ est isomorphe à $A^{[n-1]}$. Les variables sont des hyperplans et on étudie là encore la réciproque. Dans l'article co-écrit avec M.M. Kaliman et Zaidenberg qui fait partie de cette thèse on étudie les hyperplans de $\C[x,y,z,u]$ de la forme $p=f(x,y)u+g(x,y,z)$. À un changement des variables $x$ et $y$ près on montre que ces hyperplans sont aussi des variables $x$-résiduelles et partant de là on montre que ce sont des $x-$plans (i.e. $A$-plans où $A=\C[x]$) de $\Cx[y,z,u]$ et même qu'il existe un automorphisme $\alpha$ de $A[y,z,u,v]$ tel que $\alpha((p,v))=(y,v)$. Dans certains cas, par exemple lorsque $g$ est de degré un en $z$, on parvient à prouver que ce sont des $x$-variables. On donne aussi une généralisation d'un théorème de Wright en montrant qu'un $x$-plan de la forme $f(x,y,z)u^n+g(x,y,z)$ où $n\geq 2$ est une $x$-variable. Cependant le problème reste irrésolu concernant, par exemple, le polynôme $y+x[xz+y(yu+z^2)]$ qui, bien qu'étant un $x$-plan et une variable $x$-résiduelle ne semble pas être une $x$-variable.

Dans cette thèse, nous nous intéressons aux problèmes d'ordonnancement simultané de tâches et de périodes d'indisponibilité. Dans un premier temps, nous réalisons une revue de littérature sur la prise en compte des indisponibilités dans les problèmes d'ordonnancement.Ensuite, nous définissons un modèle général qui englobe des modèles existants de la littérature pour des ateliers à une machine et à machines parallèles. Une approche globale de résolution basée sur les problèmes d'affectation linéaire a été développée. Cette approche permet de résoudre le modèle général comme un simple problème d'affectation. Un grand nombre de critères d'optimisation et de modèles de maintenance peuvent être traités en utilisant cette approche, fournissant ainsi l'accès à tous les modèles qui ont souvent été étudiés séparément dans la littérature. Les résultats élaborés avec cette approche ont permis de résoudre des problèmes d'ordonnancement non traités avant et aussi de généraliser et améliorer des résultats antérieurs.Nous proposons, en dernier lieu, une étude d'un problème flow shop à deux machines en présence d'une période d'indisponibilité sur la deuxième machine. Une étude de complexité est menée sur le problème. Ensuite, nous définissons des propriétés d'optimalité. En se basant sur ces propriétés, trois méthodes de résolution exacte sont proposées; une méthode énumérative, un programme linéaire et une méthode basée sur l'approche de séparation et évaluation B&B. Une analyse expérimentale est présentée afin d'évaluer les performances de ces méthodes. / In classical scheduling problems, machines are assumed to be continuously available. However, in a real manufacturing system, machine may become unavailable during the scheduling period due to preventive maintenance. In this dissertation, we are interested in the problems of jointly scheduling jobs and unavailability periods.We start our study by introducing a general framework for scheduling problems and we present a review of the scheduling problems with unavailability periods.Then, we consider a general model for scheduling jobs on single-machine and unrelated parallel-machines with maintenance interventions. A unified approach is presented to solve this model as an assignment problem. A large number of performance criteria and maintenance models can be treated in this way, thus providing access to models that have often been studied separately in the published literature.Finally, we focus on the problem of a two-machine flow-shop makespan scheduling with the deteriorating maintenance period on the second machine. Then, we establish some conditions of the optimal schedule. In order to solve the problem, we proposed different exact methods: enumerative method, mixed-integer programming (MIP) model and a branch & bound algorithm. Numerical experiments are reported for all the proposed methods.

This doctoral thesis explores HR management through systems of production incentives in the construction sector which has been instrumental to the Spanish economy. The transformation of a time-based payment system to a performance or production-based payment system is difficult if you consider the complexity of the calculation and the environment in which prices are negotiated. In any case, before agreeing to a piece-rate determined price it is imperative that the person responsible for negotiating on behalf of the company knows the relationship of this piece rate compared to expected labour costs as an objective in the opening plans. The main objective of this thesis is to propose an algorithm in flowchart form that facilitates —from the data found in the costs study— knowledge about the price range within which the economic incentive for piece-rate compensation for any work should be established by applying a few simple conversion coefficients proposed by the model itself / Aquesta tesi doctoral es centra en la gestió dels recursos humans a través dels sistemes d'incentius a la producció. La transformació d'un sistema de pagament per temps a un sistema de pagament per rendiment o per producció no és senzill si es té en compte la complexitat del càlcul i l'entorn en el qual es pacten els preus. És imprescindible que abans d'acordar un determinat preu fet, el responsable de la negociació, conegui la relació d'aquest preu respecte els costos de mà d'obra que té previstos com a objectiu en el pressupost de producció. El principal objectiu d'aquesta tesi és proposar un algoritme que faciliti —a partir de les dades contingudes en l'estudi de costos—, conèixer l'interval de preus dins del qual s'hauria d'establir l'incentiu econòmic de remuneració a preu fet d'un treball qualsevol, mitjançant l'aplicació d'uns simples coeficients de conversió que proposa el propi model