Artificial neural network modelling of PET constitutive behavior in stretch blow molding

Abstract

The stretch blow moulding (SBM) process is the main method for the mass production of PET containers. Understanding and modelling the behaviour of amorphous PET (aPET) is critical to design the optimum product and process. However, due to its nonlinear viscoelastic behaviour, its constitutive behaviour is very complex. This means that the constitutive model will be useful only if it is known to be valid under the actual conditions of the SBM process of interest. The aim of this work was to develop a material model capable of capturing the deformation behaviour of aPET subject to an arbitrary strain history with the help of the Artificial Neural Network (ANN).

In order to achieve this goal, a reliable, rodust and non-touch characterization method based on the data acquisition system and digital image correlation system was utilized to collect the stress-strain relationship of material in deforming preforms during free stretch-blow tests. This characterization is the base of creating a big enough training database to train the ANN models.

By comparing the performance of the pure ANN based constitutive model under both displacement-controlled deformation and load-controlled deformation, the drawbacks of applying the pure ANN model as a constitutive model in load-controlled simulations were pointed out. In order to figure out this problem, some physical equations were embedded with ANN models to enlarge its stabilization under load-controlled scenario and a Hybrid ANN based material model was developed based on a part of the original Buckley and two ANN models.

The results of the single element finite element (FE) analysis using the Hybrid ANN based material model showed a good prediction on the constitutive behaviour of aPET during conventional biaxial stretching. In predicting the deformation behaviour of aPET during the blowing process, the material model demonstrated its capability of capturing the stress response of aPET under arbitrary deformation.

A finite element model which included the validated process parameters and the Hybrid ANN based material model was created to mimic the free stretch-blow process and SBM process with large delay time. The results showed a good prediction of preform shape evolution, no matter whether the deformation tends towards a sequential biaxial stretch or a simultaneous biaxial deformation.

Thesis is embargoed until 31 July 2029.

Date of Award	Jul 2024
Original language	English
Awarding Institution	Queen's University Belfast
Sponsors	Queen's University & China Scholarship Council
Supervisor	Savko Malinov (Supervisor) & Gary Menary (Supervisor)