Resumo

BARRIOS-CORDOVA, Abel et al. Comparative study of multi-response optimization techniques in experimental designs. Ing. invest. y tecnol. [online]. 2020, vol.21, n.2, e1713.  Epub 13-Nov-2020. ISSN 2594-0732.  https://doi.org/10.22201/fi.25940732e.2020.21n2.016.

In this work a proposal is developed to compare different multi-response optimization methodologies applied to response surfaces (RSM) in experimental designs, as solution tools of problems mainly in the industrial area. The following methodologies are studied: desirability function (DES), MOORA (MOO), TOPSIS (TOP), MULTIMOORA (MMO), MOORA AD (MAD), TOPSIS AD (TAD) and multilayer neural networks (with Neuralnet (NEU) and Nnet (NET) packages). Each of these techniques is applied in three cases of commercial or industrial interest with different experimental designs (Taguchi, Box-Behnken and Central Composite Design), in a Monte Carlo simulation study, the techniques to be compared, the type of experimental design and different correlation scenarios are used as factors. The techniques are compared by a metric that evaluates the distance of each estimated response with respect to its ideal or desired value, in order to analyze the advantages and disadvantages of each method. The results obtained were consistent in each of the analyzed cases, it is concluded that Neuralnet Neural Networks (NEU) are the best method, secondly, the Desirability Function (DES) and Nnet Neural Networks (NET). In addition, the proposed MOORA AD (MAD) method was found to have excellent performance in one particular study case. It is recommended in future comparative studies, to use more types of experimental designs and to apply more multi-response optimization techniques available, in order to obtain more information on which scenarios and conditions the methods show a better performance and to make more specific implementation suggestions. All the programs were done using R (R Core Team, 2019) to promote the use of free software for research or commercial development purposes.

Palavras-chave : Optimization; multiobjective optimization; multi-response optimization; desirability; MCDM optimization; artificial neural networks; experimental designs.