Poynting-aware Constitutive Artificial Neural Networks: Hyperelastic model discovery under simple shear

Simple shear is a fundamental deformation mode for characterizing soft materials; however, invariant-based hyperelastic models exhibit an inherent degeneracy under incompressible simple shear, for which I 1 = I 2 , limiting the independent identification of the corresponding constitutive contributions from shear-stress data alone. In addition, representing changes in the sign of the Poynting response remains challenging, particularly for isotropic materials, where no anisotropic fiber contribution is available to modify the normal-stress response. To address these limitations, this work proposes a Poynting-aware Constitutive Artificial Neural Network (PA-CANN) for interpretable hyperelastic model discovery under simple shear. The isotropic response is represented through a weight-shared Valanis-Landel principal-stretch branch combined with a sign-flexible I 2 correction. The latter is reparameterized to preserve a positive reference-state isotropic shear stiffness while allowing either positive or negative Poynting behavior. For fiber-reinforced materials, the anisotropic branch combines orientation-modulated principal-stretch features with the pseudo-invariants I 4 and I 5 . Sublinear square-root activations are additionally introduced to increase constitutive flexibility and represent stiffness-softening responses. Shear and normal stresses are fitted simultaneously through an energy-based constitutive loss, while L 1 regularization and post-training pruning are employed to promote compact and interpretable strain energy density functions. The framework is evaluated using experimental data for an unreinforced elastomer and unidirectional fiber reinforced composites with fiber orientations of 0°, 45°, and 90°. The PA-CANN provides accurate representations of the shear and normal stress responses, including markedly different Poynting behaviors across fiber orientations, and a single parameter set successfully describes all three orientations when jointly calibrated. An independent isotropic validation further demonstrates the ability of the proposed formulation to reproduce a negative Poynting effect while maintaining positive reference-state shear stiffness. The resulting analytical strain energy density functions remain compact and physically interpretable, supporting the use of PA-CANNs as physics-guided constitutive model discovery tools.

Authors

Institutions

Publication Details

Journal
Computer Methods in Applied Mechanics and Engineering
Published
2026-11-19
DOI
https://doi.org/10.1016/j.cma.2026.119579
Primary Topic
Elasticity and Material Modeling
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Poynting-aware Constitutive Artificial Neural Networks: Hyperelastic model discovery under simple shear

C. S. Moreira, Bernardo Feijó Junqueira, Sarah S. Mendes, L.C.S. Nunes et al.
5 links
Computer Methods in Applied Mechanics and Engineering
Elasticity and Material Modeling
article

Poynting-aware Constitutive Artificial Neural Networks: Hyperelastic model discovery under simple shear

C. S. Moreira, Bernardo Feijó Junqueira, Sarah S. Mendes, L.C.S. Nunes, Germano Guilherme Rodrigues Caldeira
article en

Abstract

Simple shear is a fundamental deformation mode for characterizing soft materials; however, invariant-based hyperelastic models exhibit an inherent degeneracy under incompressible simple shear, for which I 1 = I 2 , limiting the independent identification of the corresponding constitutive contributions from shear-stress data alone. In addition, representing changes in the sign of the Poynting response remains challenging, particularly for isotropic materials, where no anisotropic fiber contribution is available to modify the normal-stress response. To address these limitations, this work proposes a Poynting-aware Constitutive Artificial Neural Network (PA-CANN) for interpretable hyperelastic model discovery under simple shear. The isotropic response is represented through a weight-shared Valanis-Landel principal-stretch branch combined with a sign-flexible I 2 correction. The latter is reparameterized to preserve a positive reference-state isotropic shear stiffness while allowing either positive or negative Poynting behavior. For fiber-reinforced materials, the anisotropic branch combines orientation-modulated principal-stretch features with the pseudo-invariants I 4 and I 5 . Sublinear square-root activations are additionally introduced to increase constitutive flexibility and represent stiffness-softening responses. Shear and normal stresses are fitted simultaneously through an energy-based constitutive loss, while L 1 regularization and post-training pruning are employed to promote compact and interpretable strain energy density functions. The framework is evaluated using experimental data for an unreinforced elastomer and unidirectional fiber reinforced composites with fiber orientations of 0°, 45°, and 90°. The PA-CANN provides accurate representations of the shear and normal stress responses, including markedly different Poynting behaviors across fiber orientations, and a single parameter set successfully describes all three orientations when jointly calibrated. An independent isotropic validation further demonstrates the ability of the proposed formulation to reproduce a negative Poynting effect while maintaining positive reference-state shear stiffness. The resulting analytical strain energy density functions remain compact and physically interpretable, supporting the use of PA-CANNs as physics-guided constitutive model discovery tools.

Computer Methods in Applied Mechanics and EngineeringVol. 464
Universidade Federal Fluminense (BR), Universidade do Estado do Rio de Janeiro (BR)
Elasticity and Material Modeling
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