An Artificial Intelligence Constitutive Model for Amorphous Solids Utilizing Graph Neural Networks

JL Tao and YJ Wang, JOM, 76, 5777-5784 (2024).

DOI: 10.1007/s11837-024-06742-9

Constructing an efficient constitutive model for the deformation of amorphous solids has long been a challenging yet important objective in materials science. The difficulty lies in the structure-less characteristics of amorphous materials, in which it is not an easy task to extract physically meaningful knowledge-based descriptors for constitutive equations. In contrast to traditional constitutive modeling, machine learning (ML)-based models do not rely on intricate thermodynamics and kinetics of materials, emerging as an alternative. Here, we propose a graph-based constitutive model employing the cutting- edge graph neural network (GNN) techniques to investigate the deformation behavior of amorphous solids, with Cu50Zr50 metallic glass (MG) as a prototypical amorphous material to test the idea. By integrating atomic strain information with graph topology, the GNN model successfully reproduces stress-strain responses of MGs across all tested temperatures and strain rates and exhibits good transferability, showcasing the potential of GNNs in establishing a universal constitutive law for amorphous solids from a data-driven perspective.

Return to Publications page