Chapter 20 - Design of experiment

Abstract

Bioprocess engineers face the issue of not only how to produce a product of certain traits, but how to control the final product quality as well. The objective of product quality control requires that we obtain reliable data, which is the same for carrying out the analysis and reaction/process design. As such, we need to perform well-thought-of experiments, to fit to mathematical models. The design of experiment is intended to ensure that a favorable outcome of the experimentation—reliability, applicability, and validity, with minimum amount of resources. The factorial design places experimental runs uniformly (grid) in the normalized factor space. It is designed for the complete polynomial model, that is, LI parameters, needing LI number of experiments in a factorial design. Usually high order cross terms are not needed. Fractional factorial designs are invented to reduce the number of runs from factorial design. On the extreme, central designs and Taguchi designs assume no factor interaction at all. In screening the importance of variables, linear models are commonly employed. As such Taguchi designs are ideal for parameter screening. The exact selection of Taguchi design lies between the corresponding factorial design and the saturation orthogonal array. The Response Surface Methodology (RSM) is commonly based on quadratic response model analysis. The quadratic response model for I factors has ½(I2 + 3I) +1 parameters, which is useful for the determination of an optimum response. Factorial design, selected Taguchi design (TD), central-composite design (CCD), Box-Behnken design (BBD), Doehlert design (DD), superlative box design (SBD), and quality impartial design (QID) are all good candidates for RSM.