Study On Non-Normal Response Robust Design Based On GLM

Robust design is one of the main approaches for quality improvement and optimization. Traditional Robust design theory is based on a fundamental assumption that the process data are normal. However, in many conditions the process data are non-normal. In this dissertation, the methodologies for Robust Parameter Design (RPD) of the products or processes with Non-normal characteristics are developed. The objective is to develop methods for non-normal response RPD, optimization when noise factors are considered based on generalized linear model (GLM) and uncertainty assessment of model parameters. The major contents and research progress are as follows:Firstly, it relates to the research of Taguchi robust parameter design such as noise factors, interactions and combined array, etc. Moreover, study the issue of the establishment of the mean and dispersion models based on GLM. For practical points of view, joint modeling of mean and variance is put forward based on GLM. And then gives the simplified form of the mean and variance using a second order Taylor series approximation. The implementation and effectiveness of the proposed method are illustrated by an example from the literature.Secondly, it addresses robust optimization issue of the products or processes with non-normal characteristics. It put forward the fuzzy membership functions of mean and variance based on fuzzy theory, gives the optimization equation and then put forward the solving method of the optimization equation based on Generalized Reduced Gradient (GRG). An example is given where some methodologies are applied to the data; the results show that the solutions obtained by proposed methods are more efficient.Thirdly, it studies the uncertainty assessment of model parameters of non-normal response in GLM. It compares four solutions to the above problem, and presents a solution using Markov Monte Carlo (MCMC) method in the uncertainty assessment of model parameters. By analyzing the structure of GLM, the estimate procedure of model parameters under Normal distribution, Binomial distribution and Poisson distribution are presented, and the effectiveness of the proposed method is illustrated by an example from the literature.