Data-driven constitutive models for brittle solids displaying progressive anisotropic damage
File(s)
Published version
Author(s)
Tagarielli, Vito
Ge, weijian
Type
Journal Article
Abstract
We propose and demonstrate a computational framework to obtain data-driven surrogate
constitutive models capturing the mechanical response of anisotropic brittle solids displaying
progressive anisotropic damage. We train the constitutive models on data obtained from the
analysis of a volume element of a material of interest; the data is generated by a constitutive
model for braided composites, displaying a complex anisotropic damage evolution progressively
transitioning from transversely isotropic to orthotropic. Training involves imposing six dimensional random strain histories on the physical model and recording the histories of stress,
strain and homogenised stiffness matrix of the material, obtained by a set of linear perturbation
analyses. Supervised machine learning and dimensionality reduction are applied to the data and
a structure for a surrogate model is proposed. The surrogate predicts the evolution of the stiffness
of the solid consequent to an arbitrary imposed six-dimensional strain increment, thereby
calculating the corresponding increment in stress. The model displays high accuracy and is able
to reproduce the homogenised material’s response via simple neural networks.
constitutive models capturing the mechanical response of anisotropic brittle solids displaying
progressive anisotropic damage. We train the constitutive models on data obtained from the
analysis of a volume element of a material of interest; the data is generated by a constitutive
model for braided composites, displaying a complex anisotropic damage evolution progressively
transitioning from transversely isotropic to orthotropic. Training involves imposing six dimensional random strain histories on the physical model and recording the histories of stress,
strain and homogenised stiffness matrix of the material, obtained by a set of linear perturbation
analyses. Supervised machine learning and dimensionality reduction are applied to the data and
a structure for a surrogate model is proposed. The surrogate predicts the evolution of the stiffness
of the solid consequent to an arbitrary imposed six-dimensional strain increment, thereby
calculating the corresponding increment in stress. The model displays high accuracy and is able
to reproduce the homogenised material’s response via simple neural networks.
Date Issued
2024-10
Date Acceptance
2024-08-02
Citation
Composites Part C: Open Access, 2024, 15
ISSN
2666-6820
Publisher
Elsevier
Journal / Book Title
Composites Part C: Open Access
Volume
15
Copyright Statement
© 2024 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
License URL
Identifier
https://www.sciencedirect.com/science/article/pii/S2666682024000707
Publication Status
Published
Article Number
100501
Date Publish Online
2024-08-08