Foundation model for composite microstructures: Reconstruction, stiffness, and nonlinear behavior prediction
Source abstract
Abstract
We present the Material Masked Autoencoder (MMAE), a self-supervised Vision Transformer pretrained on a large corpus of short-fiber composite images via masked image reconstruction. The pretrained MMAE learns latent representations that capture essential microstructural features and are broadly transferable across tasks. We demonstrate two key applications: (i) predicting homogenized stiffness components through fine-tuning on limited data, and (ii) inferring physically interpretable parameters by coupling MMAE with an interaction-based material network (IMN), thereby enabling extrapolation of nonlinear stress–strain responses. These results highlight the promise of microstructure foundation models and lay the groundwork for future extensions to more complex systems, such as 3D composites and experimental datasets. • Developed a foundation model for composites by pre-training on short-fiber data to learn transferable features. • Reconstructed composite microstructures to validate the representational capacity of the learned latent space. • Achieved R 2 of 0.959 in stiffness prediction via transfer learning with limited data. • Enabled nonlinear behavior extrapolation by transferring learned features to an interaction-based material network.
Source: openalex_inverted_index · Quality: baseline
Research context
OpenAlex topics: Composite Material Mechanics · Topology Optimization in Engineering · Mechanical Behavior of Composites
SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.
References (27)
Show source-provided references
1. Prediction and optimization of mechanical properties of composites using convolutional neural networks10.1016/j.compstruct.2019.111264 ↗
2. Teaching solid mechanics to artificial intelligence—a fast solver for heterogeneous materials10.1038/s41524-021-00571-z ↗
3. Graph-enhanced deep material network: multiscale materials modeling with microstructural informatics10.1007/s00466-024-02493-1 ↗
4. Pre-trained transformer model as a surrogate in multiscale computational homogenization framework for elastoplastic composite materials subjected to generic loading paths10.1016/j.cma.2024.116745 ↗
5. Structure-based out-of-distribution (ood) materials property prediction: a benchmark study10.1038/s41524-024-01316-4 ↗
6. BERT: pre-training of deep bidirectional transformers for language understanding
7. Language models are unsupervised multitask learners
8. Bommasani
9. A comprehensive survey on pretrained foundation models: a history from bert to chatgpt
10. An image is worth 16x16 words: transformers for image recognition at scale
11. Interaction-based material network: a general framework for (porous) microstructured materials10.1016/j.cma.2021.114300 ↗
12. Micromechanics-based material networks revisited from the interaction viewpoint; robust and efficient implementation for multi-phase composites10.1016/j.euromechsol.2021.104384 ↗
13. Orientation-aware interaction-based deep material network in polycrystalline materials modeling10.1016/j.cma.2025.117977 ↗
14. Decoding material networks: exploring performance of deep material network and interaction-based material networks10.1093/jom/ufae053 ↗
15. Masked autoencoders are scalable vision learners
16. Attention is all you need
17. Liu
18. Microstructure-guided deep material network for rapid nonlinear material modeling and uncertainty quantification10.1016/j.cma.2022.115197 ↗
19. Ls-dyna machine learning–based multiscale method for nonlinear modeling of short fiber–reinforced composites10.1061/JENMDT.EMENG-6945 ↗
20. Random sequential adsorption10.1016/0022-5193(80)90358-6 ↗
21. A comparison of three methods for selecting values of input variables in the analysis of output from a computer code10.1080/00401706.2000.10485979 ↗
22. ANSYS
23. Lee
24. Masked autoencoders for unsupervised anomaly detection in medical images10.1016/j.procs.2023.10.084 ↗
25. Establishing size effects in discontinuous fibre composites using 2d finite element analysis10.1016/j.commatsci.2012.05.067 ↗
26. 2d versus 3d probabilistic homogenization of the metallic fiber-reinforced composites by the perturbation-based stochastic finite element method10.1016/j.compstruct.2013.10.035 ↗
27. Determination of rve size for random composites with local volume fraction variation10.1016/j.cma.2016.03.002 ↗