Searcharxiv⌕ Search

arXiv · 2611.14158

Neural network scaling and generalization for 3D strain fields in heterogeneous materials

Abstract

Computational homogenization and coupled multiscale simulations of heterogeneous materials require repeated solution of mechanical fields on Representative Volume Elements, which becomes prohibitively expensive for high-resolution three-dimensional microstructures. Machine-learning surrogates can accelerate these computations, yet most existing approaches predict fields for fixed macroscopic loadings and provide limited guidance on how to scale network architectures for high-resolution 3D tensor-field prediction. This work presents a U-Net surrogate trained to predict the full 36-component strain localization tensor field in voxelized concrete mesostructures at resolutions up to $256^3$. Training localization tensors are generated with an FFT-based linear elastic solver, and an extensive hyperparameter optimization is performed to derive scaling rules and an accuracy-speed trade-off analysis. The results show that only network width and depth need to be scaled with problem complexity, while several conventional U-Net design choices, such as doubling channels after downsampling, are suboptimal for this problem. The final neural network achieves a Mean Absolute Error of 1.118% on both the validation and the held-out test sets, and an inference speedup of approximately 8.7x over the FFT-based solver on an Nvidia H100 GPU. Generalization experiments reveal transfer across aggregate volume fractions and highlight the importance of training on diverse morphologies. To safeguard reliability, a Maximum Mean Discrepancy metric computed on internal feature maps is proposed for detecting out-of-distribution microstructures without costly reference computations.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Lucas Fourel, Paul Devianne, Marine Debouvry, Tristan Destruhaut, Russell Gunn, Ashwini Gupta, Lori Graham-Brady, Jean-François Molinari. 2026-11-10. Neural network scaling and generalization for 3D strain fields in heterogeneous materials. https://searcharxiv.com/abs/2611.14158

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Data-driven reconstruction of band dispersion and quantum geometry via Koopman dynamic mode decomposition

We present a data-driven framework for reconstructing band structures using Koopman operator analysis and dynamic mode decomposition (Koopman-DMD). Instead of deriving spectra from an explicit Hamiltonian, the approach reconstructs band dispersion and modal dynamics directly from spatiotemporal data, including wavefunctions and observables. This framework establishes a correspondence between Hamiltonian Floquet-Bloch decomposition and Koopman-DMD, whereby the extracted DMD modes encode frequencies, decay or growth rates, spatial profiles and projection weights. These quantities allow the reconstruction of spectral functions, local density of states, and delocalized-to-localized measures such as the inverse participation ratio. Also, these extended DMD modes enable inference of quantum-geometric and topological properties, including the quantum metric, Berry curvature and geometric phases. Applications to prototypical one- and two-dimensional tight-binding models, including disordered Su-Schrieffer-Heeger model and its Floquet and non-Hermitian variants, graphene and Haldane models, demonstrate that Koopman-DMD provides a unified route for the data-driven analysis of wave propagation, localization, and topological phases in condensed matter, photonics, and related fields.

physics.comp-ph↗

Predictive beam-lattice reduction for higher-order topological modes in a 2D SSH phononic crystal

We develop a mechanically faithful reduced model for a two-dimensional topological phononic crystal composed of rigid square masses connected by slender elastic ligaments. Exploiting Euler Bernoulli beam theory, we derive a Hermitian 12 degree of freedom dynamical matrix that retains in-plane translations, rotations, and ligament eccentricity. This reduction captures effects that are absent from scalar mass spring SSH models while remaining computationally much more tractable and more easily interpretable than full finite element simulations. Dimerizing the ligament widths produces a mechanical 2D SSH lattice with a full band gap and a quantized bulk polarization. The sign of the dimerization controls the transition from trivial to non trivial phases, while ligament eccentricity provides an additional purely geometric mechanism for changing the topology. Ribbon and finite cell calculations predict in gap edge and corner modes, quantified by localization measures and confirmed by finite element simulations. Measurements on 3D printed samples show an evanescent response in the trivial structure and enhanced boundary/corner response in the non trivial structure within the predicted gap. The results provide a validated route for designing topological elastic metamaterials using a continuum informed discrete model rather than either idealized mass spring networks or brute force numerical optimization.

physics.comp-ph↗

Online reduced-order modeling from primal solver trajectories for PDE-constrained optimization

Repeated high-fidelity state and gradient evaluations make large-scale PDE-constrained optimization computationally expensive. Reduced-order modeling provides a natural way to reduce the cost of these repeated high-dimensional computations. However, existing reduced-order modeling approaches often rely on converged high-fidelity states for model construction or updating, while their reduced spaces may lose accuracy as the design moves beyond the region represented by the available data. To address these limitations, we present Reduced-order Trajectory-based Optimization (RETO), an online framework that constructs local reduced-order models (ROMs) directly from unconverged primal solver trajectories. Each local ROM is built from snapshots collected during a finite number of high-fidelity primal iterations, without requiring full-order convergence, and supports multiple low-cost state evaluations, adjoint-based gradient evaluations, and design updates. New solver trajectories are generated at updated design points to reconstruct the local ROM and continue the optimization. Across the Poisson inversion and logistic resource-allocation problems, RETO achieves solution quality comparable to that of full-order optimization, with speedups of 14.7 times and 8.49 times, respectively. Even for the more challenging airfoil inverse-design problems, RETO retains speedups of 2.52 to 3.13 times under subsonic and transonic conditions. These results demonstrate the value of unconverged primal solver trajectories as an online information source for reduced-order optimization, allowing the reduced basis to evolve with the design rather than requiring a basis that covers the entire design space in advance.

physics.comp-ph↗