Artificial Neural Networks for Predicting Mechanical Properties of Crystalline Polyamide12 via Molecular Dynamics Simulations

TL;DR

This paper develops an artificial neural network model trained on molecular dynamics data to accurately predict the anisotropic stress-strain behavior of crystalline Polyamide12, facilitating multiscale simulations of semicrystalline polymers.

Contribution

The study introduces a novel ANN approach trained on MD data to predict mechanical properties of crystalline PA12, enabling multiscale modeling in additive manufacturing.

Findings

ANN accurately predicts stress-strain relations

Model provides 3D molecular-level anisotropic properties

Enables integration into finite element simulations

Abstract

Predicting material properties of 3D printed polymer products is a challenge in additive manufacturing due to the highly localized and complex manufacturing process. The microstructure of such products is fundamentally different from the ones obtained by using conventional manufacturing methods, which makes the task even more difficult. As a first step of a systematic multiscale approach, in this work, we have developed an artificial neural network (ANN) to predict the mechanical properties of the crystalline form of Polyamide12 (PA12) based on data collected from molecular dynamics (MD) simulations. Using the machine learning approach, we are able to predict the stress-strain relations of PA12 once the macroscale deformation gradient is provided as an input to the ANN. We have shown that this is an efficient and accurate approach, which can provide a three-dimensional molecular-level anisotropic stress-strain relation of PA12 for any macroscale mechanics model, such as finite element modeling at arbitrary quadrature points. This work lays the foundation for a multiscale finite element method for simulating semicrystalline polymers, which will be published as a separate study.