Multi-scale modelling of composite materials: a hybrid approach integrating asymptotic homogenization and physics-informed neural networks

A combined approach based on the two-scale asymptotic homogenization method (AHM) and physics-informed neural networks (PINNs) is presented to compute the effective elastic properties of two-phase laminated composites with isotropic constituents. Six local problems arising from AHM are formulated as a multi-objective optimization problem and solved via PINNs. The architecture consists of pairs of fully connected neural networks, one for each phase of the material. The total loss function penalizes violations of the governing differential equation, interface continuity and the zero-mean uniqueness conditions, without requiring prior solution data. An adaptive weighting scheme based on exponential moving averages balances the three loss terms during training. A cyclic hybrid optimization strategy is proposed, alternating between Adam, L-BFGS, and NysNewtonCG phases. The methodology is validated on a benchmark configuration with two isotropic phases of contrasting stiffness. Running 10 independent trials using different random seeds, all effective elastic coefficients are recovered with low relative errors compared to analytical values. The homogenized stiffness tensor exhibits transverse isotropy. Parametric sweeps over volume-fraction and material-contrast ratios confirm that the methodology remains stable. PINN predictions are cross-validated against an independent finite-element solution. The proposed PINN-in-the-loop approach provides a validated, physics-driven, data-free alternative for solving AHM local problems.

Authors

Institutions

Publication Details

Journal
Proceedings of the Royal Society A Mathematical Physical and Engineering Sciences
Published
2026-11-18
DOI
https://doi.org/10.1098/rspa.2026.0542
Primary Topic
Composite Material Mechanics
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Multi-scale modelling of composite materials: a hybrid approach integrating asymptotic homogenization and physics-informed neural networks

José Miguel Ormaetxe Merodio, Reinaldo Rodrı́guez-Ramos, O.L. Cruz-González, Boris Mederos et al.
3 links
Proceedings of the Royal Society A Mathematical Physical and Engineering Sciences
Composite Material Mechanics
article

Multi-scale modelling of composite materials: a hybrid approach integrating asymptotic homogenization and physics-informed neural networks

José Miguel Ormaetxe Merodio, Reinaldo Rodrı́guez-Ramos, O.L. Cruz-González, Boris Mederos, Yoanh Espinosa-Almeyda, Carlos Quesada
article en

Abstract

A combined approach based on the two-scale asymptotic homogenization method (AHM) and physics-informed neural networks (PINNs) is presented to compute the effective elastic properties of two-phase laminated composites with isotropic constituents. Six local problems arising from AHM are formulated as a multi-objective optimization problem and solved via PINNs. The architecture consists of pairs of fully connected neural networks, one for each phase of the material. The total loss function penalizes violations of the governing differential equation, interface continuity and the zero-mean uniqueness conditions, without requiring prior solution data. An adaptive weighting scheme based on exponential moving averages balances the three loss terms during training. A cyclic hybrid optimization strategy is proposed, alternating between Adam, L-BFGS, and NysNewtonCG phases. The methodology is validated on a benchmark configuration with two isotropic phases of contrasting stiffness. Running 10 independent trials using different random seeds, all effective elastic coefficients are recovered with low relative errors compared to analytical values. The homogenized stiffness tensor exhibits transverse isotropy. Parametric sweeps over volume-fraction and material-contrast ratios confirm that the methodology remains stable. PINN predictions are cross-validated against an independent finite-element solution. The proposed PINN-in-the-loop approach provides a validated, physics-driven, data-free alternative for solving AHM local problems.

Proceedings of the Royal Society A Mathematical Physical and Engineering SciencesVol. 482(2350)
Universidad Autónoma de Ciudad Juárez (MX), Centre National de la Recherche Scientifique (FR), Universidade Federal Fluminense (BR), Aix-Marseille Université (FR), University of Havana (CU), Institut de Recherche sur les Phénomènes Hors Équilibre (FR), Centrale Méditerranée (FR), Universidad Politécnica de Madrid (ES)
Conselho Nacional de Desenvolvimento Científico e Tecnológico
Composite Material Mechanics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon