Get access

Prediction of mechanical properties of compatibilized styrene/natural-rubber blend by using reaction conditions: Central composite design vs. artificial neural networks

Authors

  • Natsupa Sresungsuwan,

    1. Center of Excellence for Petroleum, Petrochemicals and Advanced Materials, Department of Chemical Engineering, Faculty of Engineering, Kasetsart University, Bangkok, Thailand
    Search for more papers by this author
  • Nanthiya Hansupalak

    Corresponding author
    1. Center of Excellence for Petroleum, Petrochemicals and Advanced Materials, Department of Chemical Engineering, Faculty of Engineering, Kasetsart University, Bangkok, Thailand
    • Center of Excellence for Petroleum, Petrochemicals and Advanced Materials, Department of Chemical Engineering, Faculty of Engineering, Kasetsart University, Bangkok, Thailand
    Search for more papers by this author

Abstract

A polystyrene (PS)/rubber blend compatibilized with PS-g-rubber copolymer, prepared via emulsion polymerization using redox initiator system, is used to investigate the utilization of central composite design (CCD) and artificial neural network (ANN) approaches in correlating polymerization conditions to mechanical properties (tensile strength and abrasion loss) of unfilled compound vulcanizates. The conditions were manipulated by changing four factors: reaction temperature and time, percentage of deproteinized rubber in the mixture containing natural rubber, and amount of chain transfer agent. The results show that the relationships between the conditions and the mechanical properties for compatibilized PS/rubber blend are too complex to be explained by polynomials, but are well described by the ANN models, developed for each response. In addition, simulation results for the tensile strength response as a function of those factors using the obtained ANN are in agreement with literature, whereas those results for the abrasion loss do not quite agree with literature due to the interference of the large measurement error. This suggests that only experimental data with high precision should be used to train an ANN to achieve a model with not only best performance but also high reliability. © 2012 Wiley Periodicals, Inc. J Appl Polym Sci. 2013

Ancillary