ANALYSIS AND DESIGN OF EXPERIMENTS

Author: LEON ADAMS GUILLERMO DE.
Year: 2003.
University: POLITÉCNICA DE CATALUÑA [www.upc.edu].
Place of defense: DEPARTAMENT D'EIO, ETSEIB, UPC.
Place of preparation: EDIFICI A0 - OMEGA Campus NORD.

Author: MIRO MARTINEZ PAU.
Year: 2005.
University: POLITÉCNICA DE VALENCIA [www.upv.es].
Place of defense: UNIVERSIDAD POLITÉCNICA DE VALENCIA DEP. INGENIERIA TEXTIL Y PAPELERA.
Place of preparation: UNIVERSIDAD POLITECNICA DE VALENCIA.

Summary: SUMMARY: The textiles and tools Statistical Quality Control has evolved with the various industrial revolutions. The technological level of many companies in the sector is very high, but instead the use of techniques Statistics is very low, considering often are very difficult, unnecessary and whose costs are high. The thesis is presented that arose as a target break with this way of thinking and demonstrate some of the advantages that can make use of statistics to make decisions and improve the quality of products and processes. To achieve this goal, a brief review to the degree of utilization of science in the sector, particularly in some districts of the Valencian Community of focus after the proposals made. This finding provides the clues they need to focus their investigation into two types of different products on the one hand, the household who are traditionally produced in the area, and on the other hand, the technicians who represent an opportunity for innovation. The first proposal is made with the intention of enhancing the value added to traditional products, by incorporating a system of Quality Control for its sensory characteristics. Using techniques Sensory Analysis want to control the quality of the article in the production process as well as define the sensory characteristics that distinguish it and what can be distinguished from its competitors. This proposal applies to the blanket to bed. The second proposal aims to facilitate innovation in technical materials, providing the tools necessary for conducting experiments in a systematic and orderly manner, to analyze and take decisions on a more objective and efficient. It applies to those technical products that are made by mixing different types of fibers. The case study is to optimize the Napa obtained by termofusión depending on the use to which it will allocate for this first apply experiments Mixtures. Once the testing phase is through modeling PLS (Partial Least Square) and the results are optimized using the function Deseabilidad. The contributions want to be an instrument to bring scientific knowledge to be grown in the University to the reality of business in the vicinity, betting on improvement and innovation in the textile sector.

Author: Martín Martín Raúl.
Year: 2005.
University: CASTILLA-LA MANCHA [www.uclm.es].
Place of defense: E.T.S. DE ING. INDUSTRIALES.
Place of preparation: E.T.S. DE ING. INDUSTRIALES.

Summary: The objective of this dissertation is to determine optimal designs if not all explanatory variables are controlled by the experimenter. The values of these variables can be known or unknown before the experiment is done. To solve this problem are considered different approaches, depending on the different situations that we find in real examples. Each chapter is divided into an introduction, which presents the objectives and major contributions, the development thereof, and a final section on the findings. In Chapter 1 presents the basic ideas and overall optimal design of experiments and introduces the notation to be followed along the memory. It defines the main criteria for optimization, D? A?, G? E?, DA? THE optimization and that will be used along the memory. The second part of this chapter introduces an important tool in the theory of optimal design, directional derivative of a function criterion. It also provides some theorems optimization based on results of previous sections. The chapter concludes with a presentation of some numerical techniques that seek optimal from one or more original design. Chapter 2 deals with the theory of Optimum Design variables which are beyond the control of the experimenter. Start chapter considering designs with two types of variables: a controllable and non-controllable otherwise known a priori designs are marginally restricted (MR). Here are considered designs with a variable controllable and non-controllable another, unknown before performing the experiment, which will be called conditionally restricted (CR). Finally we study designs with the three types of variables, designs marginal and conditionally restricted (CFR). For these three types of designs are provided in the respective equivalence theorems. Based on the above results, Lopez Fidalgo and Garcet Rodriguez (2004), proposed algorithms for the calculation of these designs that we gather in this report. Particularizamos past performance and algorit- few of the designs MR, CR and CFR respect to certain criteria. Besides proposing a new peak efficiency of the designs are obtained based on the outcome of Atwood (1969). Chapter 3 proposes a multiplicative class of algorithms for calculating the optimal designs proposed in Chapter 2. Finally got an equivalence between designs MR (CFR) D? And Ds? Optimal. Chapter 4 is devoted mainly to the study of the designs D? Optimal marginally restricted by a method that allows direct sequential experiments comments as they appear in the variable controllable but not known. It is motivated by the example shown in López-Fidalgo and Garcet- Rodriguez (2004) on the design of an experiment to assess the predictive ability of a test exercise preoperative whose aim was to predict surgical mortality and morbidity in patients with a diagnosis of cancer lung undergoing elective pulmonary resection, from the physical characteristics of the patient, physical exercise conducted and the presence or absence of blood oxygen desaturation during the test. Under this framework, assumed how we might improve our design if patients arrive in a sequential manner. It provides an algorithm for calculating the optimal designs. In the second stage, the construction design with discrete and continuous distributions marginal classic supposedly known. Chapter 5 contains a first approximation to the calculation of optimum designs for models correlated with observations when using covariance parameterized functions, and considering that some variables are not controllable by the experimenter. It proposes several real examples with linear and nonlinear modeling, solved some of them, by adapting an algorithm used to calculate designs with observacione 8 s correl 314 adas. This algorithm is an adaptation of the designs MR proposed by Brimkulov and the other in 1980.

Author: Trandafir Paula Camelia.
Year: 2005.
University: CASTILLA-LA MANCHA [www.uclm.es].
Place of defense: E. T. S . de Ingenieros Industriales.
Place of preparation: Escuela Técnica Superior de Ingenieros Industriales.

Summary: In this thesis provides extensions to the design of experiments based on variations of the model Michaelis-Menten used in the Biocinética, with the goal of providing optimal designs that meet the requirements of the experimental research. New designs have been found suitable for the modification of the model of Michaelis-Menten consisting of a term linear add to the initial model. For example, in the case homocedástico as in the heteroscedástico obtained design D-óptimo. To address the problem of choosing a model before having the data, when theirs would choose the model after making a contrast of assumptions, it uses the concept of T-optimización, bringing designs T-óptimos to discriminate between two models homocedásticos and with normal distribution. Extrapolando, provides a new approach to discriminate between two models not normally defined by the distance of Kullback-Leibler, hence it denominemos KL-criterio. It applies to discriminate between the models Michaelis-Menten and modification of the same seen here, with the first and gamma log-normal distributions in the second case. The KL-criterio encompasses existing in the literature for models heteroscedásticos multi-respuesta and models for binary and widespread.