Abstract:Polycrystal plasticity in metals is characterized by nonlinear behavior and strain hardening, making numerical models computationally intensive. We employ Graph Neural Networks (GNN) to surrogate polycrystal plasticity with complex geometries from Finite Element Method (FEM) simulations. We present a novel message-passing GNN that encodes nodal strain and edge distances between FEM mesh cells, aggregates them to obtain embeddings, and combines the decoded embeddings with the nodal strains to predict stress tensors on graph nodes. We demonstrate training GNN based on subgraphs generated from FEM mesh-graphs, in which the mesh cells are converted to nodes and edges are created between adjacent cells. The GNN is trained on 80\% of the graphs and tested on the rest (90 graphs in total). We apply the trained GNN to periodic polycrystals and learn the stress-strain maps based on crystal plasticity theory. The GNN is accurately trained based on FEM graphs, in which the $R^2$ for both training and testing sets are 0.993. The proposed GNN plasticity constitutive model speeds up more than 150 times compared with the benchmark FEM method on randomly selected test polycrystals. We also apply the trained GNN to 30 unseen FEM simulations and the GNN generalizes well with an overall $R^2$ of 0.992. Analysis of the von Mises stress distributions in polycrystals shows that the GNN model accurately learns the stress distribution with low error. By comparing the error distribution across training, testing, and unseen datasets, one deduces that the proposed model does not overfit and generalizes well beyond the training data. This work outlooks surrogating computationally intensive crystal plasticity simulations using graph data.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to predict stress distribution through Graph Neural Networks (GNN) in Polycrystal Plasticity. Specifically, the paper aims to develop a GNN - based surrogate model to replace the traditional Finite Element Method (FEM) for numerical simulation of polycrystal plasticity. The traditional method is computationally expensive when dealing with polycrystal plasticity because plastic behavior is nonlinear and path - dependent, resulting in high computational complexity. Therefore, the goals of the paper are: 1. **Reduce computational cost**: By using GNN, the paper hopes to significantly reduce the computation time while maintaining the prediction accuracy. According to the results in the paper, the GNN model is more than 150 times faster than the benchmark FEM method. 2. **Handle data with different degrees of freedom**: Different polycrystal geometries will lead to different degrees of freedom (DoF) in the generated finite - element meshes. Traditional regression tools such as Gaussian processes or neural networks usually rely on a fixed number of training points, while GNN can handle graphs of different sizes (i.e., different numbers of nodes and edges), thus solving this problem. 3. **Preserve spatial connectivity**: The spatial connectivity between finite - element mesh elements is crucial for preserving physical properties. Matrix data is difficult to preserve these geometric features, while GNN can effectively handle these connectivities through the message - passing mechanism. 4. **Handle large - scale data sets**: Each polycrystal mesh of 10 grains contains about 10,000 mesh elements, and training such a large - scale data set is an expensive task. The paper proposes to solve this problem by training GNN on sub - graphs of the finite - element mesh. To achieve the above goals, the paper proposes the following methods: - **Graph Neural Network model**: A new message - passing GNN is designed. This model encodes node strains and edge distances, obtains an embedded representation by aggregating this information, and combines the decoded embedded representation and node strains to predict the stress tensor on graph nodes. - **Sub - graph training**: GNN training is carried out by generating sub - graphs from the finite - element mesh graph. This method not only improves the training efficiency but also enables the model to better generalize to unseen data sets. - **Performance evaluation**: The accuracy and generalization ability of the GNN model are verified through the prediction results on the training set and the test set. The paper shows the high performance of the GNN model in predicting stress distribution, especially in the prediction of von Mises stress distribution. In conclusion, by introducing GNN and the sub - graph training method, this paper successfully solves the problems of high computational cost, complex data processing, and poor model generalization ability in polycrystal plasticity simulation.

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

Graph Neural Networks for an Accurate and Interpretable Prediction of the Properties of Polycrystalline Materials

Graph Neural Network for Predicting the Effective Properties of Polycrystalline Materials: A Comprehensive Analysis

A physics-constrained neural network for crystal plasticity modelling of FCC materials

FIP-GNN: Graph neural networks for scalable prediction of grain-level fatigue indicator parameters

Graph neural networks for simulating crack coalescence and propagation in brittle materials

A Microstructure-based Graph Neural Network for Accelerating Multiscale Simulations

Graph Neural Network for Stress Predictions in Stiffened Panels Under Uniform Loading

General Graph Neural Network-Based Model To Accurately Predict Cocrystal Density and Insight from Data Quality and Feature Representation

Connectivity Optimized Nested Graph Networks for Crystal Structures

A data-driven approach to predicting the anisotropic mechanical behaviour of voided single crystals

AnisoGNN: graph neural networks generalizing to anisotropic properties of polycrystals

Materials fatigue prediction using graph neural networks on microstructure representations

From CP-FFT to CP-RNN: Recurrent neural network surrogate model of crystal plasticity

Material-Response-Informed DeepONet and its Application to Polycrystal Stress-strain Prediction in Crystal Plasticity

Generalizability of Graph Neural Network Force Fields for Predicting Solid-State Properties

An Artificial Intelligence Constitutive Model for Amorphous Solids Utilizing Graph Neural Networks

Data-driven modeling of dislocation mobility from atomistics using physics-informed machine learning

A geometric-information-enhanced crystal graph network for predicting properties of materials

Prediction of Effective Elastic Moduli of Rocks using Graph Neural Networks

GrainGNN: A dynamic graph neural network for predicting 3D grain microstructure