Speaker
Description
This study presents a deep learning framework for the inverse reconstruction of open-cell porous metamaterials, targeting specific hydraulic properties such as porosity and intrinsic permeability. Utilizing a combination of synthetic 3D porous microstructures and CT-scan-based representative volume elements (RVEs) of porous open-foam specimens, the framework leverages a property-variational autoencoder (pVAE) to establish robust mappings between 3D microstructures and their corresponding hydraulic properties. The pVAE integrates a variational autoencoder (VAE) with a regression network, providing a compact latent space that enables efficient interpolation and optimization. Moreover, a convolutional neural network (CNN) is trained to predict the hydraulic properties, thereby generating a computationally efficient dataset to train the pVAE, circumventing the high cost of direct numerical simulations.
In particular, the VAE comprises a convolutional encoder, which compresses input structures into a low-dimensional latent space modeled as a Gaussian distribution, and a decoder, which reconstructs 3D geometries. The training process minimizes a composite loss function that includes a reconstruction loss, ensuring fidelity to the input data, and a Kullback-Leibler (KL) divergence term, which regularizes the latent space for smoothness and semantic continuity. Additionally, a regression loss term aligns the latent space with target hydraulic properties, optimizing the model's ability to predict and generate structures with specific attributes.
By leveraging the continuous and interpretable latent space for optimization and sampling, the pVAE generates porous structures tailored to desired hydraulic properties. This approach enables scalable, data-driven design of porous metamaterials for multiscale and multi-functional engineering systems, advancing structure-property mapping methodologies.
REFERENCES
[1] Heider Y, Aldakheel F, Ehlers W (2025) A multiscale cnn-based intrinsic permeability prediction in deformable porous media. Meccanica “under review”
[2] Wang L, Chan YC, Ahmed F, Liu Z, Zhu P, Chen W (2020) Deep generative modeling for mechanistic-based learning and design of metamaterial systems. Comput. Methods Appl. Mech. Eng. 372:113377,
https://doi.org/10.1016/j.cma.2020.113377
[3] Jones RE, Hamel CM, Bolintineanu D, Johnson K, Buarque de Macedo R, Fuhg J, Bouklas N, Kramer S (2024) Multiscale simulation of spatially correlated microstructure via a latent space representation. J. Solid Mech. 301:112966, \newline
DOI https://doi.org/10.1016/j.ijsolstr.2024.112966
[4] Zheng L, Karapiperis K, Kumar S, Kochmann DM (2023) Unifying the design space and optimizing linear and nonlinear truss metamaterials by generative modeling. Nat Commun 14, 7563.
https://doi.org/10.1038/s41467-023-42068-x
