Using Hybrid Artificial Neural Network-Particle Swarm Optimization for Prediction of HIPS Mechanical Properties

Document Type: Research Paper


1 Urmia University of Technology

2 Chemical Engineering Department, Islamic Azad University, Ilkhchi Branch, Ilkhchi, Iran



Artificial neural networks (ANN) have emerged as one of the useful artificial intelligence concepts used in various engineering applications. ANN only recently has been used in modeling the mechanical behavior of polymers. The aim of study is to show the applicability of ANNs, to predict and determine the mechanical properties of High Impact Polystyrene (HIPS). Izod, Vicat, and MFI of pure HIPS was considered, and the effect of main useful parameters (25 ones) on them were investigated. By using ANN, a black box model is considered as a calculator of mechanical properties. It is not easy to predict the accurate value of these properties without experimental works, because of the nonlinear behavior of polymerization process, and also the inaccuracy of experimental results of these properties measurement. To overcome these problems, an alternative prediction model was proposed for calculating properties using a hybrid ANN-PSO model. All parameters in various cases have been gathered from industrial DCS system and laboratory for training of the ANN. Firstly, by using the ANN model, the sensitivity analysis of parameters has been performed. It is then filtered using the effective ANN-PSO hybrid parameters. Out of 25 proposed parameters, 14, 10, 11 of them were selected in MFI, Izod, Vicat, respectively, and used to predict new parameters in the modified ANN. Results showed that hybrid ANN-PSO model was powerful in predicting of properties with average relative error 6%, 5%, and 1% for prediction of considered properties between industrial and computed ANN-PSO hybrid model data.