METODOLOGÍA COMPARATIVA DE MODELOS NO LINEALES PARA LA PREDICCIÓN DE LAS PROPIEDADES REOLÓGICAS DE MEZCLAS DE GOMA IN LINEAS INDUSTRIALES

Abstract

Data mining and statistics are applied to predict certain of the properties of rubber-extruded mixtures. These properties are associated to their cure curves using data from the mixing phase at the beginning of the process. The main goal is to automatically obtain the model that provides operators accurate set points to control mixing process. The operators would be able to anticipate possible failures in the quality of vulcanized rubber mixture. There are several strategies to develop optimum models. This work proposes the following methodology to optimize the information extraction from the available data. First, an initial analysis of database attributes is performed seeking for significant information to future model derivation. Second, a wide comparison of different non-parametric methods is carried out to determine which one is the most appropriate. Instead of directly contrasting prediction errors, an automatic statistical system of comparison is included by using several non-parametric techniques. Third, some alternative strategies are tested taking advantage of the specific attributes of the database.