Mixture Experiment Design Using Artificial Neural Networks and Electromagnetism-like Mechanism Algorithm

Abstract

A mixture experiment treats a product that is formed of several ingredients together (e.g. gasoline, detergents, and cookies). A mixture experiment is a special type of parameter in which the factors are the ingredients or components of a mixture. Although there have many researchers proposed various methods to improve the design of the mixture experiments, those cannot provide effective analysis. This study aims to use artificial neural networks (ANNs) and electromagnetism-like mechanism (EM) algorithm to optimizing the mixture design. First, we employ an ANN to build the response function model (RFM) of the experiment for estimating the response at specific mixed ingredient proportions. An EM algorithm is then used to obtain the fitness value of the response function and the optimal ingredients proportion within the constraints of ingredients. An example adopted from the literature is re-analyzed to verify the effectiveness of the proposed method.