Multiple response optimization for quality improvement. comparative between classic approach with bayesian approach and neural networks

Abstract

The paper aims to review the existing methodologies for multiresponse optimization, integrate them into one and develop a new algorithm that allows to overcome the existing limitations. For this purpose we reviewed statistical optimization methodologies using the traditional response surface methodology with robust design, then reviewed the application of the bayesian approach to that obtained with traditional statistics, and finally reviewed artificial neural network applications to cases of optimization. After performing the analysis and discussion about the three methodologies were integrated into one, having developed a new algorithm to overcome the limitations and shortcomings of the previous methods. Also, we compared the results obtained with other methods with those obtained with the new method, with favorable outcome. Thus we have developed a multi-response optimization methodology that considers linear and nonlinear relationships, which has the qualities of traditional statistical methodologies, bayesian statistics, and artificial neural networks.