 |
On the modeling of hyperelastic foams using neural networks
1 : Laboratoire de Mécanique des Solides
(LMS)
-
Site web
Ecole Polytechnique, Centre National de la Recherche Scientifique
Route de Saclay, 91128 Palaiseau Cedex -
France
2 : Department of Materials Science and Engineering - Delft University of Technology
* : Auteur correspondant
2628 CD Delft -
Pays-Bas
This study investigates the potential of neural networks (NNs) as constitutive models for foam materials. Synthetic datasets were generated using foam energy functions which may exhibit nonconvexity such as the Shrimali and Blatz-Ko models. Several NN architectures and choices of strain-related inputs were explored and compared. The results show that the widely-used Input-Convex Neural Network (ICNN), while effective for near-incompressible models, cannot properly capture the nonlinearity of foam models when the inputs are deformation-gradient-convex strain invariants.
