Introducing memory decay network for microstructured viscoelastic composites

Funding Sponsor

National Natural Science Foundation of China

Third Author's Department

Mechanical Engineering Department

Find in your Library

https://doi.org/10.1016/j.compstruct.2024.118792

All Authors

Yicheng Lu Zhi Sun K. I. Elkhodary Hanlin Xiao Shan Tang Xu Guo

Document Type

Research Article

Publication Title

Composite Structures

Publication Date

1-1-2025

doi

10.1016/j.compstruct.2024.118792

Abstract

A mechanics-informed data-driven approach is proposed to describe the small-strain microstructured viscoelastic behavior of composites subjected to arbitrary loading paths and loading rates. Our approach couples data science with viscoelastic theory, avoiding the complications of mathematical derivation associated with relaxation functions, while requiring only a small amount of training data. Specifically, viscoelastic behavior is represented by introducing a trained memory decay network (MDN). The proposed MDN is composed of two artificial neural networks (ANNs) to represent the long-term elastic behavior of a material, and two convolutional neural networks featuring trainable kernel lengths (TKL-CNNs) to represent its viscous behavior. Importantly, the MDN is designed to incorporate mechanics-informed constraints to comply with the memory decay effect of viscoelasticity, and the second law of thermodynamics. Finally, the MDN material model is integrated into a finite element method (FEM) framework to solve boundary value problems (BVPs) for viscoelastic composites that contain voids or particles. Illustrative examples demonstrate that our approach can accurately predict the viscoelastic behavior of composites subjected to arbitrary loading paths and loading rates, even when loading time exceeds that of the training data sets. It is also demonstrated that the proposed approach can improve computational efficiency for viscoelastic composites.

Link to Full Text
Find this item online

Share

COinS