Finite Element and Neural Network Modeling of Viscoelastic Annular Extrusion

Abstract

Plastics blow molding has grown rapidly for the past couple of decades. Annular parison extrusion is a critical stage in extrusion blow molding. In this work, numerical simulations on the parison extrusion were performed using finite element (FE) method and the Kaye-Bernstein-Kearsley-Zapas type constitutive equation. A total of 100 simulations was carried out by changing the extrusion die inclination angle, die gap, and parison length. Then a backpropagation artificial neural network (ANN) was proposed as a tool for modeling the parison extrusion using the numerical simulation results. The network architecture determination and the training process of the ANN model were discussed. The predictive ability of the ANN model was examined through several sets of FE simulation results different from those utilized in the training stage. The effects of the die inclination angle, die gap, and parison length on the parison swells can be predicted using the ANN model. The results showed that the die gap has a smaller effect on the diameter swell but a greater effect on the thickness swell. Both diameter and thickness swells increase as the die inclination angle increases. The hybrid method combining the FE and ANN can shorten the time for the predictions drastically and help search out the processing conditions and/or die geometric parameters to obtain optimal parison thickness distributions.