Graph neural networks for mechanical property prediction of 2D fiber composites

This work investigates the ability of graph neural networks (GNNs) to homogenize 2D fiber composite microstructures. We use different inhomogeneity and anisotropy indices to motivate and show that the Volume Elements (VEs) used in ML methods should ideally be far from their Representative Volume Ele...

Full description

Saved in:
Bibliographic Details
Main Authors: Erdem Caliskan, Reza Abedi, Massimiliano Lupo Pasini
Format: Article
Language:English
Published: Elsevier 2025-09-01
Series:Materials & Design
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S0264127525009207
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849390584299520000
author Erdem Caliskan
Reza Abedi
Massimiliano Lupo Pasini
author_facet Erdem Caliskan
Reza Abedi
Massimiliano Lupo Pasini
author_sort Erdem Caliskan
collection DOAJ
description This work investigates the ability of graph neural networks (GNNs) to homogenize 2D fiber composite microstructures. We use different inhomogeneity and anisotropy indices to motivate and show that the Volume Elements (VEs) used in ML methods should ideally be far from their Representative Volume Element (RVE) size limit and, consequently, are notably anisotropic. Hence, training only the isotropic limit properties may not be acceptable. Another aspect is the need to normalize elastic stiffness values for ML, especially when high elastic contrast ratios are encountered between composite phases or in the material set. We introduce a normalization technique based on the mean-field method (MFM) to handle such high contrast ratios and train for the entire stiffness tensor. We show that the proposed GNN approaches exhibit high accuracy and efficiency compared to traditional methods and convolutional neural networks, utilizing unstructured graphs constructed from microstructure topology. Our model successfully predicts the stiffness tensor, peak strength under bulk damage, and brittle fracture initiation strength across diverse microstructure configurations while maintaining high accuracy even for extreme material contrasts and volume fractions. We also present a method to improve prediction accuracy for small dataset sizes using Voronoi partitioning.
format Article
id doaj-art-baa0e2ae37be45a8a95f58c245e173ac
institution Kabale University
issn 0264-1275
language English
publishDate 2025-09-01
publisher Elsevier
record_format Article
series Materials & Design
spelling doaj-art-baa0e2ae37be45a8a95f58c245e173ac2025-08-20T03:41:31ZengElsevierMaterials & Design0264-12752025-09-0125711450010.1016/j.matdes.2025.114500Graph neural networks for mechanical property prediction of 2D fiber compositesErdem Caliskan0Reza Abedi1Massimiliano Lupo Pasini2Department of Mechanical, Aerospace and Biomedical Engineering, University of Tennessee Knoxville, 1506 Middle Drive, Knoxville, 37916, TN, USADepartment of Mechanical, Aerospace and Biomedical Engineering, University of Tennessee Knoxville, 1506 Middle Drive, Knoxville, 37916, TN, USAOak Ridge National Laboratory, Computational Sciences and Engineering Division, 1 Bethel Valley Rd, Oak Ridge, 37831, TN, USA; Corresponding author.This work investigates the ability of graph neural networks (GNNs) to homogenize 2D fiber composite microstructures. We use different inhomogeneity and anisotropy indices to motivate and show that the Volume Elements (VEs) used in ML methods should ideally be far from their Representative Volume Element (RVE) size limit and, consequently, are notably anisotropic. Hence, training only the isotropic limit properties may not be acceptable. Another aspect is the need to normalize elastic stiffness values for ML, especially when high elastic contrast ratios are encountered between composite phases or in the material set. We introduce a normalization technique based on the mean-field method (MFM) to handle such high contrast ratios and train for the entire stiffness tensor. We show that the proposed GNN approaches exhibit high accuracy and efficiency compared to traditional methods and convolutional neural networks, utilizing unstructured graphs constructed from microstructure topology. Our model successfully predicts the stiffness tensor, peak strength under bulk damage, and brittle fracture initiation strength across diverse microstructure configurations while maintaining high accuracy even for extreme material contrasts and volume fractions. We also present a method to improve prediction accuracy for small dataset sizes using Voronoi partitioning.http://www.sciencedirect.com/science/article/pii/S0264127525009207Machine learningFiber compositesBrittle strengthHomogenizationGraph neural network
spellingShingle Erdem Caliskan
Reza Abedi
Massimiliano Lupo Pasini
Graph neural networks for mechanical property prediction of 2D fiber composites
Materials & Design
Machine learning
Fiber composites
Brittle strength
Homogenization
Graph neural network
title Graph neural networks for mechanical property prediction of 2D fiber composites
title_full Graph neural networks for mechanical property prediction of 2D fiber composites
title_fullStr Graph neural networks for mechanical property prediction of 2D fiber composites
title_full_unstemmed Graph neural networks for mechanical property prediction of 2D fiber composites
title_short Graph neural networks for mechanical property prediction of 2D fiber composites
title_sort graph neural networks for mechanical property prediction of 2d fiber composites
topic Machine learning
Fiber composites
Brittle strength
Homogenization
Graph neural network
url http://www.sciencedirect.com/science/article/pii/S0264127525009207
work_keys_str_mv AT erdemcaliskan graphneuralnetworksformechanicalpropertypredictionof2dfibercomposites
AT rezaabedi graphneuralnetworksformechanicalpropertypredictionof2dfibercomposites
AT massimilianolupopasini graphneuralnetworksformechanicalpropertypredictionof2dfibercomposites