Experiments with mixture for optimization of processes: an application with non-normal answers

In this article a methodology for the design of mixture experiments is presented, illustrated with a real case study of a chemical compound, that constitutes a subsystem of a delay mechanism of a rocket motor. The aim of the study is to find the proportion of each mixture component in order to attend the specified burn time. For this, a linear regression model with normal responses was adjusted, and next considered inadequate, because there was an indication that the response variance is not constant. Therefore, the models were adjusted using the Quasi-Likelihood method. With the developed model, it was possible to find the proportion of each component that optimizes the process. For the process optimization, beside a specified burn time, it was possible to consider the minimization of the prediction variance of this time.