Prediction of the shrinkage of injection molded iPP plaques using artificial neural networks

Shrinkage occurs in all polymers, being extremely dependent on processing conditions. The utilization of shrinkage data allows designers to accurately predict the final part dimensions. A numerical prediction of a part shrinkage can be made using simulation packages available commercially. However, this shrinkage is highly dependent on the non-linear material behavior and, thus, its estimation involves significant simplifications. On the other hand, artificial neural networks, ANN, can model highly non-linear systems; thus, it is expected that they can predict a part shrinkage effectively. In this study, a neural network architecture was developed to predict the shrinkage of an iPP injection molded plaque after changing four processing conditions: melt and mold temperatures, holding pressure and flow rate. The experiments were defined through the use of the design of experiments. A simulation code, Moldflow®, was used to establish the processing window; its shrinkage predictions were compared with experimental, neural network and statistical results. It was observed that the ANN had the best performance in the shrinkage prediction, even using limited experimental data, confirming its great capacity to model non-linear systems.