Abstract:Simulating the mechanical response of advanced materials can be done more accurately using concurrent multiscale models than with single-scale simulations. However, the computational costs stand in the way of the practical application of this approach. The costs originate from microscale Finite Element (FE) models that must be solved at every macroscopic integration point. A plethora of surrogate modeling strategies attempt to alleviate this cost by learning to predict macroscopic stresses from macroscopic strains, completely replacing the microscale models. In this work, we introduce an alternative surrogate modeling strategy that allows for keeping the multiscale nature of the problem, allowing it to be used interchangeably with an FE solver for any time step. Our surrogate provides all microscopic quantities, which are then homogenized to obtain macroscopic quantities of interest. We achieve this for an elasto-plastic material by predicting full-field microscopic strains using a graph neural network (GNN) while retaining the microscopic constitutive material model to obtain the stresses. This hybrid data-physics graph-based approach avoids the high dimensionality originating from predicting full-field responses while allowing non-locality to arise. By training the GNN on a variety of meshes, it learns to generalize to unseen meshes, allowing a single model to be used for a range of microstructures. The embedded microscopic constitutive model in the GNN implicitly tracks history-dependent variables and leads to improved accuracy. We demonstrate for several challenging scenarios that the surrogate can predict complex macroscopic stress-strain paths. As the computation time of our method scales favorably with the number of elements in the microstructure compared to the FE method, our method can significantly accelerate FE2 simulations.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to accelerate the simulation of the mechanical response of advanced materials while maintaining multi - scale characteristics in multi - scale simulation. Specifically, although traditional multi - scale simulation methods (such as FE2) can simulate the mechanical response of materials more accurately, the computational cost is very high because it is necessary to solve the micro - finite element (FE) model at each macroscopic integration point. To overcome this challenge, the paper proposes an alternative surrogate model strategy based on graph neural networks (GNN), which can retain multi - scale characteristics and can be used interchangeably with the FE solver at any time step. ### Main problems 1. **High computational cost**: Traditional multi - scale simulation methods require solving the micro - FE model at each macroscopic integration point, resulting in extremely high computational costs and limiting their practical applications. 2. **Full - field micro - quantity prediction**: Existing data - driven surrogate models are usually only able to predict the macro - stress - strain relationship and cannot provide full - field micro - quantities, which is an important limitation in some application scenarios. 3. **Generalization ability**: Existing data - driven models have poor generalization ability when dealing with unseen meshes or complex load paths and require a large amount of training data. ### Solutions The paper proposes a surrogate model based on graph neural networks (GNN), which can: - **Predict full - field micro - strain**: Directly predict the micro - strain field through GNN, rather than only predicting the macro - stress - strain relationship. - **Embed micro - material model**: Embed the constitutive relationship of the micro - material into the model to obtain the micro - stress and obtain the macro - stress through the standard computational homogenization method. - **Multi - time - step prediction**: During the training process, the model not only predicts a single time - step, but continuously predicts multiple time - steps, similar to an autoregressive model, and can maintain the update of historical - dependent variables during the inference process. - **Geometric inductive bias**: By introducing geometric features in the graph, the model can better generalize to unseen mesh structures. ### Key technologies - **Graph neural network (GNN)**: Used to handle arbitrary mesh structures and can transfer and aggregate information between nodes. - **Message passing layer (MPL)**: Through multi - layer message passing, make information propagate in the graph and capture non - local effects. - **Physical embedding**: Embed the constitutive relationship of the material into the model, reduce the training data requirements, and improve the prediction accuracy. - **Loss function design**: A loss function that considers both micro - strain and stress is designed to ensure that the model can achieve high accuracy when predicting strain and stress. ### Experimental verification The paper verifies the effectiveness of this model through a variety of experiments, including the prediction of elastic - plastic behavior under monotonic loading conditions and the prediction of complex macro - stress - strain paths under non - monotonic loading conditions. The experimental results show that this model can not only significantly accelerate FE2 simulation, but also make accurate predictions under different mesh sizes and more time - steps. ### Conclusion The GNN - based surrogate model proposed in this paper significantly reduces the computational cost of multi - scale simulation while maintaining multi - scale characteristics, provides full - field micro - quantity prediction, and has good generalization ability. This method provides a new solution for the practical application of multi - scale simulation.

A Microstructure-based Graph Neural Network for Accelerating Multiscale Simulations

Accelerate Microstructure Evolution Simulation Using Graph Neural Networks with Adaptive Spatiotemporal Resolution

Multiscale graph neural networks with adaptive mesh refinement for accelerating mesh-based simulations

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

Graph-enhanced deep material network: multiscale materials modeling with microstructural informatics

Graph Neural Networks (gnns) Based Accelerated Numerical Simulation

Multiscale Simulation of the Responses of Discrete Nanostructures to Extreme Loading Conditions Based on the Material Point Method

A Machine Learning Method of Accelerating Multiscale Analysis for Spatially Varying Microstructures

On-the-fly construction of surrogate constitutive models for concurrent multiscale mechanical analysis through probabilistic machine learning

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

Microstructure-guided deep material network for rapid nonlinear material modeling and uncertainty quantification

Graph Neural Network Modeling of Grain-scale Anisotropic Elastic Behavior using Simulated and Measured Microscale Data

Accelerated multiscale mechanics modeling in a deep learning framework

Dynamic and adaptive mesh-based graph neural network framework for simulating displacement and crack fields in phase field models

Physically Recurrent Neural Networks for Computational Homogenization of Composite Materials with Microscale Debonding

Neural Network Accelerated Process Design of Polycrystalline Microstructures

Accelerating phase-field-based microstructure evolution predictions via surrogate models trained by machine learning methods

Physically recurrent neural network for rate and path-dependent heterogeneous materials in a finite strain framework

Graph Neural Networks for Mesh Generation and Adaptation in Structural and Fluid Mechanics

Deep Learning for Multiscale Damage Analysis via Physics-Informed Recurrent Neural Network

Multiscale Analysis of Composite Structures with Artificial Neural Network Support for Micromodel Stress Determination