National Technical University of Athens (NTUA)
Κατασκευές και στατιστική ανάλυση παραγοντικών και εύρωστων παραμετρικών σχεδιασμών
Abstract
dc:descriptionThis dissertation deals with the construction and statistical analysis of factorial and robust parameter designs. The study of experimental designs is extremely important in modern industry, science and engineering. In Chapter 1 we provide some basic concepts in factorial designs, optimal designs, and computer experiments. We also provide a concise presentation of the most important points regarding the objectives of the experiments, the steps for an experiment to be successful as well as the elements that influence the good choice of an experimental design. Chapter 2 is devoted to a brief introduction to computer experiments, which are extremely important in modern industry, science and engineering. In Chapter 3 we study U-type, Column-orthogonal and nearly column-orthogonal designs. The nearly column-orthogonal designs are very useful in cases where column-orthogonal designs are unknown. In this chapter we present the construction of new designs suitable for screening experiments. In some cases, the constructed designs are optimal in respect to the alias structure. The structure of the alias effects of the U-type, Column-orthogonal and nearly column-orthogonal designs are calculated and presented in closed form. This fact makes this new approach highly innovative and makes it capable of constructing designs that are different from the existing designs in the literature. In addition, we present an extended multiplication theorem, as well as new infinite families of column-orthogonal designs using periodic golay pairs. The research results presented in this chapter were published in the scientific paper [II]. The following chapter (Chapter 4) introduces some basic concepts for factorial designs. Particular emphasis is given in a special class of fractional factorial designs, the supersaturated designs for which basic concepts are presented regarding their optimization criteria, their construction as well as their analysis. Chapter 5 discloses a method of constructing multi and mixed-level supersaturated designs. In particular, we provide a new lower bound for the optimality of such designs. This bound is proved to be tight and quite general since it can be applied to both balanced and unbalanced designs. In addition, new families of balanced and nearly balanced supersaturated designs are both constructed and evaluated. The new designs are constructed by applying the k-circulant type methodology for multilevel and mixed-level designs. Moreover, additional conditions, that are needed on the generator vectors, are developed and proved in this paper to pre-ensure the optimality of the corresponding k-circulant supersaturated designs. This methodology leads to a series of new families of improved supersaturated designs and provides tools for directly constructing optimal or nearly-optimal k-circulant designs by just checking the corresponding generator vector. The research results presented in this chapter were published in the scientific paper [V]. In the next two chapters we study two methods for analyzing supersaturated designs. More specifically, in Chapter 6 we present a method for analyzing both two-level and mixed-level designs. The development of new statistical methods inspired by machine learning algorithms is increasing rapidly, especially nowadays. One of such methods is based on support vector machines (SVM), which manageto successfully extract important genes in classification problems, achieving extremely high performance. In this chapter, we propose a new method of selecting variables in regression problems, called SVR-RFE (Support Vector Regression Recursive Feature Elimination), for detecting active effects in both two-level and mixed-level designs. The simulation studies that we present demonstrate that this process is quite effective, especially in terms of statistical power. The research results presented in this chapter were published in the scientific paper [III]. Then, in Chapter 7, we present a method of analyzing two-level supersaturated designs. In particular, we present a two-stage procedure for analyzing two-level supersaturated designs assuming a main-effect only model without including interaction terms. The proposed method combines sure independence screening (SIS) with different penalty functions; such as Smoothly Clipped Absolute Deviation (SCAD), Lasso and MC penalty achieving both the down-selection and the estimation of the significant effects, simultaneously. Insights on using the proposed methodology are provided through various simulation scenarios and several comparisons with existing approaches, such as stepwise in combination with SCAD and Dantzig Selector (DS) are presented as well. Results of the numerical study and real data analysis reveal that the proposed procedure can be considered as an advantageous tool due to its extremely good performance for identifying active factors. The research results presented in this chapter were published in the scientific paper [VI]. Chapter 8 addresses some basic concepts for robust parameter designs. Particular emphasis is given on both types of designs that make up the robust parameter design methodology and the modelling techniques, including location and dispersion modelling as well as response modelling. In Chapter 9, we propose a construction of an effective SSD along with an analysis method, in order to deal with the significant problem of the robust parameter design methodology (RPDM). Combining iterative SIS variable selection and a penalized method, namely SCAD, we perform the analysis of the SSDs developed in the present work. The proposed methodology is applied in different models so as to show its effectiveness in many different scenarios, assuming both first and second-order models in a sense of a response surface design. Two illustrative examples as well as numerous numerical experiments are conducted for plenty cases. The results imply that the proposed method is highly effective for identifying the active effects of main factors, two-factor interactions, three-factor interactions as well as the pure quadratic ones, under the assumption of effect sparsity. The research results presented in this chapter were published in the scientific paper [V]. The presentation of the research project for this dissertation is completed with Chapter 10, where we propose a construction for supersaturated split-plot Designs (SSSPD) suitable for RPD experiments. We applied the methodology of compound orthogonal arrays using orthogonal arrays of strength greater than 1, for the split-plot units and any relevant balanced SSD for the whole plot units. The proposed methodology suggests that experiments conducted in a split-plot point of view can be of tremendous value in RPD since they enable the estimation of the contrasts of interest in considerably easier way than the designs proposed by Taguchi. The research results presented in this chapter were published in the scientific paper [IV]. The presentation of this dissertation is completed with the four Appendixes. In Appendix A we summarize the results obtained from Chapter 5. In Appendix B we present the results of the simulations that emerged from Chapter 6. In Appendix G we present the results of the simulations resulting from Chapter 7 and finally in Appendix D we give the proof of Lemma 1 as well as the results of the simulations resulting from Chapter 9.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2019
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Drosou, Krystallenia
- Δρόσου, Κρυσταλλένια
Subjects
dc:subject × 18- Παραγοντικοί σχεδιασμοί
- Υπερκορεσμένοι σχεδιασμοί
- Πειράματα υπολογιστών
- Ορθογώνιοι σχεδιασμοί
- Μηχανική μάθηση
- Εύρωστοι παραμετρικοί σχεδιασμοί
- Στατιστικός έλεγχος ποιότητας
- Factorial designs
- Supersaturated designs
- Orthogonal designs
- Machine learning
- Robust parameter designs
- Statistical quality control
- Computer experiments
- Φυσικές Επιστήμες
- Μαθηματικά
- Natural Sciences
- Mathematics
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/45644