Multi-fidelity physics-guided graph neural networks for predicting structure–property relationships in polycrystalline metals

Materials & Design 2026 Open access

Read full text
Authors: Minh Tien Tran, Hoang Cuong Phan, Jaimyun Jung, Se‐Jong Kim, Seong-Hoon Kang, Ho Won Lee
DOI: 10.1016/j.matdes.2026.116310
Cited by: 1
References: 59

Abstract

• Multi-fidelity PGNN is developed for prediction of structure–property linkage. • Introducing physics-guided HP module improves accuracy and sample efficiency. • PGNN rapidly and accurately predicts anisotropic properties and yield loci. • PGNN shows high efficiency and well generalizes across diverse microstructures.

Fields of Study & Subjects

Machine Learning in Materials Science Model Reduction and Neural Networks Advanced Graph Neural Networks

Publication details

Journal / source
Materials & Design
Publisher
Elsevier BV
Volume
267
Pages
116310–116310
Publication date
28 May 2026
Language
EN