Modeling the molecular weight and number average molecular masses during the photo-thermal oxidation of polypropylene using neural networks

Hadjira Maouz, Latifa Khaouane Salah Hanini, Yamina Ammi

Abstract


This work investigates the potential of artificial neural network (ANN) model to predict the molecular weight (MW) and number average molecular masses (Mn) during the photo-thermal oxidation of polypropylene (pp). A set of 116 data points were used to test the neural network, 80%, 10%, and 10% of the database were used, or the training, the validation, and the test of the model. The optimal topology of ANN model obtained with the architecture of (3 inputs, 23 hidden and 2 output neurons). Statistical analyses of neural network model show good agreement with experimental data (a coefficient of correlation equal to 0.9864 and 0.9688, and a root mean square error equal to 11.1250 kg/mol and 3.5284 kg/mol for the predicted molecular weight and number average molecular masses respectively), considering, a three layer feed-forward backpropagation neural network with BFGS quasi-Newton (trainbfg) training algorithm, a hyperbolic tangent sigmoid and logarithmic sigmoid transfer function at the hidden and the output layer respectively. The comparison between the experimental and calculated results show that the ANN model is able of predicted the molecular weight and number average molecular masses during the photo-thermal oxidation of polypropylene.


Keywords


Keywords: molecular weight; number average molecular masses; photo-thermal oxidation; polypropylene; neural networks.

Full Text:

PDF