Abstract

SOTO, Diana; ALANA, Jorge  and  OLIVA, Haydee. Regression models for the prediction of high impact polystyrene properties using operational variablesModelos de regresión para la predicción de propiedades de poliestireno de alto impacto usando variables operacionales. Rev. Téc. Ing. Univ. Zulia [online]. 2006, vol.29, n.2, pp.144-158. ISSN 0254-0770.

In this work, we proposed models based on multiple linear regression analysis and artificial neural networks in order to predict some properties used for the quality control of different grades of high impact polystyrenes. These models were based on recipes and operational conditions in an industrial plant. The properties considered were melt flow index, Izod impact resistance, yield stress, break stress, percent elongation. In validation, the sum of the squares errors were 38,4, 7,16 × 10³, 116 and 103 for melt flow index, Izod impact resistance, yield stress and break stress, respectively. According to standardized coefficients of the regression equations, variables with most significant effects on the considered properties were the modifier concentration (tert-dodecylmercaptane), and lubricants (zinc stearate and mineral oil) concentrations. In the studied region, the performances of both models (linear regression and neural networks) were similar.

Keywords : High impact polystyrene; artificial neural networks; multiple analysis regression.