Artificial neural network model for prediction of viscoelastic behaviour of polycarbonate composites

Abstract

Polymer composites are the result of incorporation of nanoparticles into the polymers and can lead to improvements even with a very small amount of reinforcement which can be tuned according to the applications. In order to understand the behaviour of these polymer composites we need to perform a number of characterizations and analyses which in turn requires investment of money and time. Thus, to reduce the number of characterizations and analyses for developing polymer composites, computational techniques can prove helpful. By means of a computational technique known as artificial neural network (ANN), prediction of the thermo-mechanical properties was made possible. Here dynamic mechanical analysis (DMA) data set was used for characterization of polycarbonate / calcium carbonate-SiO2 core shell composites (polycarbonate composites). Using the dataset, the selected ANN model consisted of a network of [3-10-1]. The prediction accuracy achieved using ANN method, was around 90%. Applicability and performance of ANN to the existing system was also confirmed by mean squared error (MSE), which is favourably small for this case, in the range of 10-5. The output predicted by ANN had a coefficient of correlation of 0.999. Furthermore, sensitivity analysis confirmed the importance of various input variables in relation with output. An optimization of the variables facilitated to maximize the conditions thus predicting glass transition temperature.