Orthogonal experimental designs for screening: construction and analysis
31 May 2018
University of Antwerp - Campus Middelheim, Building A, Room 143 - Middelheimlaan 1 - 2020 Antwerp (route: UAntwerpen, Campus Middelheim
Alan Vazquez Alcocer
Prof. dr. Peter Goos, prof. dr. Eric Schoen
PhD defence Alan Vazquez Alcocer - Faculty of Applied Economics
Experimentation in business and industry is an essential step in the creation of new products and processes, because it is the only way to establish a cause-effect relationship between the controllable characteristics of a product or process and its performance measures. However, experimentation in business and industry is not only time consuming but also expensive. For instance, building prototypes to test new products is a major undertaking. Therefore, it is of crucial importance to plan experiments well. What experimenters desire is to perform a limited number of tests that yield a lot of information and allow them to identify the key experimental factors that drive the performance of their product and process. The field of Design of Experiments focuses on theory and methods to find experimental plans that tell the experimenters which tests to perform, given their limited budgets.
The best known experimental plans are screening designs, which allow the experimenter to identify the most important factors from long lists of potentially important ones. Generally, screening designs have two primary goals: (1) to correctly identify any factor that has a large influence on its own, and (2) to correctly detect whether the factor effects depend on the settings of other factors. To achieve these goals for modern products and processes, larger screening designs are required than those readily available in the literature. These larger screening designs are hard to construct and form, to a large extent, unexplored research territory.
The present dissertation deals with the construction and analysis of large screening designs. The first two contributions of this dissertation are two construction methods for large designs with factors at two levels. In such a design, each of the factors is either set at a low or at a high level in each test. The third contribution of this dissertation is the construction of screening designs with factors at three levels. These designs are intended to study quantitative factors each of which is either set at a low, at a medium or at a high level. The last contribution of this dissertation is a method to analyze the data collected from a screening design with factors at two, three or more levels.