Complas 2025

COMPLAS 2025

Investigating Deep Material Networks as Representative Microstructures for the Mechanical Behavior of Composites.

Lenau, Ashley (Sandia National Laboratories)
Robertson, Andreas (Sandia National Laboratories)
Shin, Dongil (Pohang University of Science and Technology)
Lebensohn, Ricardo (Los Alamos National Laboratory)
Dingreville, Rémi (Sandia National Laboratories)

In session: 801A - Machine Learning for Computational Mechanics: Innovations and Applications I

Please login to view abstract download link

Deep material networks (DMN) are tree-like machine learning networks that predict homogenized material properties for a given microstructure. With only linear homogenization training datasets, the DMN can extrapolate non-linear material responses, providing a computationally efficient model used to aid in composite design. The DMN learns an abstract representation of the microstructure during training, but the statistical equivalence of DMNs has not been sufficiently investigated for various microstructure types. In this work, the DMN is exercised with different microstructures to test its ability to produce a microstructure equivalent network that distinguishes features with different length scales or statistical arrangements that impact a composite’s behavior. The DMN’s ability to homogenize material properties that are different from the training homogenization task (as a truly equivalent microstructure would be able to do) is additionally investigated with elastic stiffness and thermal conductivity. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525.