Machine learning based constitutive modelling on craze yielding in polymeric materials

Abstract

The inelastic behavior of thermoplastic polymers may involve shearing and crazing, and both depend on temperature and strain rate. Traditional constitutive models account for temperature and strain rate through phenomenological or empirical formulas. In this study, we present a physics-guided machine learning (ML) framework to model shear and craze in polymeric materials. The effects of all three principal stresses for the craze initiation are considered other than the maximum tensile principal stress solely in previous works. We implemented a finite element framework through a user-defined material subroutine and applied the constitutive model to the deformation in three polymers (PLA 4060D, PLA 3051D, and HIPS). The result shows that our ML-based model can predict the stress-strain and volume-strain responses at different strain rates with high accuracy. Notably, the ML-based approach needs no assumptions about yield criteria or hardening laws. This work highlights the potential of hybrid physics-ML paradigms to overcome the trade-offs between model complexity and accuracy in polymer mechanics, paving the way for computationally efficient and generalizable constitutive models for thermoplastic materials.