A finite operator learning technique for mapping the elastic properties of microstructures to their mechanical deformations
Publication date
2024-12-09
Document type
Forschungsartikel
Author
Rezaei, Shahed
Asl, Reza Najian
Faroughi, Shirko
Asgharzadeh, Mahdi
Harandi, Ali
Laschet, Gottfried
Reese, Stefanie
Apel, Markus
Organisational unit
Publisher
Wiley
Series or journal
International Journal for Numerical Methods in Engineering
ISSN
Periodical volume
126
Periodical issue
1
Article ID
e7637
Peer-reviewed
✅
Part of the university bibliography
✅
Language
English
Keyword
dtec.bw
Abstract
To obtain fast solutions for governing physical equations in solid mechanics, we introduce a method that integrates the core ideas of the finite element method with physics‐informed neural networks and concept of neural operators. We propose directly utilizing the available discretized weak form in finite element packages to construct the loss functions algebraically, thereby demonstrating the ability to find solutions even in the presence of sharp discontinuities. Our focus is on micromechanics as an example, where knowledge of deformation and stress fields for a given heterogeneous microstructure is crucial for further design applications. The primary parameter under investigation is the Young's modulus distribution within the heterogeneous solid system. Our investigations reveal that physics‐based training yields higher accuracy compared with purely data‐driven approaches for unseen microstructures. Additionally, we offer two methods to directly improve the process of obtaining high‐resolution solutions, avoiding the need to use basic interpolation techniques. The first one is based on an autoencoder approach to enhance the efficiency for calculation on high resolution grid points. Next, Fourier‐based parametrization is utilized to address complex 2D and 3D problems in micromechanics. The latter idea aims to represent complex microstructures efficiently using Fourier coefficients. The proposed approach draws from finite element and deep energy methods but generalizes and enhances them by learning parametric solutions without relying on external data. Compared with other operator learning frameworks, it leverages finite element domain decomposition in several ways: (1) it uses shape functions to construct derivatives instead of automatic differentiation; (2) it automatically includes node and element connectivity, making the solver flexible for approximating sharp jumps in the solution fields; and (3) it can handle arbitrary complex shapes and directly enforce boundary conditions. We provided some initial comparisons with other well‐known operator learning algorithms, further emphasize the advantages of the newly proposed method.
Description
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License (http://creativecommons.org/licenses/by-nc/4.0/).
Version
Published version
Access right on openHSU
Metadata only access