Divergence-free neural operators for stress field modeling in polycrystalline materials

Mohammad S. Khorrami,Pawan Goyal,Jaber R. Mianroodi,Bob Svendsen,Peter Benner,Dierk Raabe

2024-08-28

Abstract:The purpose of the current work is the development and comparison of Fourier neural operators (FNOs) for surrogate modeling of the quasi-static mechanical response of polycrystalline materials. Three types of such FNOs are considered here: a physics-guided FNO (PgFNO), a physics-informed FNO (PiFNO), and a physics-encoded FNO (PeFNO). These are trained and compared with the help of stress field data from a reference model for heterogeneous elastic materials with a periodic grain microstructure. Whereas PgFNO training is based solely on these data, that of the PiFNO and PeFNO is in addition constrained by the requirement that stress fields satisfy mechanical equilibrium, i.e., be divergence-free. The difference between the PiFNO and PeFNO lies in how this constraint is taken into account; in the PiFNO, it is included in the loss function, whereas in the PeFNO, it is "encoded" in the operator architecture. In the current work, this encoding is based on a stress potential and Fourier transforms. As a result, only the training of the PiFNO is constrained by mechanical equilibrium; in contrast, mechanical equilibrium constrains both the training and output of the PeFNO. Due in particular to this, stress fields calculated by the trained PeFNO are significantly more accurate than those calculated by the trained PiFNO in the example cases considered.

Computational Engineering, Finance, and Science,Materials Science,Machine Learning,Analysis of PDEs

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to develop and compare Fourier Neural Operators (FNOs) for the quasi - static mechanical response modeling of polycrystalline materials. Specifically, the paper explores three different types of FNOs: Physics - guided FNO (PgFNO), Physics - informed FNO (PiFNO), and Physics - encoded FNO (PeFNO). The goal of these models is to accurately calculate the stress field \(P\) based on given material properties (such as Young's modulus \(E\) and Poisson's ratio \(\nu\)) and the deformation state \(F\). #### Main problems: 1. **Improve the accuracy of stress - field modeling**: By introducing physical constraints to ensure that the stress field satisfies the mechanical equilibrium condition (i.e., the divergence - free condition), the prediction accuracy of the model is improved. 2. **Performance comparison of different types of FNOs**: Compare the performance of PgFNO, PiFNO, and PeFNO in training and prediction, and evaluate their effectiveness in dealing with the stress fields of complex polycrystalline materials. 3. **Explore new network architectures**: Especially for PeFNO, by directly encoding the divergence - free condition into the network architecture to achieve more robust and accurate stress - field prediction. #### Specific methods: - **PgFNO**: It is trained only based on the data of the reference model, without introducing additional physical constraints. - **PiFNO**: Add the divergence - free condition as a constraint in the loss function to ensure that the stress field satisfies the mechanical equilibrium. - **PeFNO**: By introducing the stress potential \(A\) and using its curl to construct the stress field \(P\), the divergence - free condition is directly encoded into the network architecture. Through these methods, the paper attempts to solve the challenges faced by traditional neural networks in modeling complex physical phenomena and provide directions for improvement in future research.

Divergence-free neural operators for stress field modeling in polycrystalline materials

Revealing the Predictive Power of Neural Operators for Strain Evolution in Digital Composites

A Spectral-based Physics-informed Finite Operator Learning for Prediction of Mechanical Behavior of Microstructures

Comparison of two artificial neural networks trained for the surrogate modeling of stress in materially heterogeneous elastoplastic solids

Neural Stress Fields for Reduced-order Elastoplasticity and Fracture

3D elastic wave propagation with a Factorized Fourier Neural Operator (F-FNO)

Learning Deep Implicit Fourier Neural Operators (IFNOs) with Applications to Heterogeneous Material Modeling

Fourier Neural Operator with Learned Deformations for PDEs on General Geometries

Domain Agnostic Fourier Neural Operators

An artificial neural network for surrogate modeling of stress fields in viscoplastic polycrystalline materials

Imposing equilibrium on measured 3-D stress fields using Hodge decomposition and FFT-based optimization

Heterogeneous Peridynamic Neural Operators: Discover Biotissue Constitutive Law and Microstructure From Digital Image Correlation Measurements

Stress Predictions in Polycrystal Plasticity using Graph Neural Networks with Subgraph Training

A Physics Preserving Neural Network Based Approach for Constitutive Modeling of Isotropic Fibrous Materials

Enhancing subsurface multiphase flow simulation with Fourier neural operator

StressD: 2D Stress estimation using denoising diffusion model

Neural network-based multiscale modeling of finite strain magneto-elasticity with relaxed convexity criteria

Fourier neural operator based fluid–structure interaction for predicting the vesicle dynamics

A finite operator learning technique for mapping the elastic properties of microstructures to their mechanical deformations

A parallel and performance portable implementation of a full-field crystal plasticity model

Deep neural operator for learning transient response of interpenetrating phase composites subject to dynamic loading