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.

Scalability of Graph Neural Network in Accurate Prediction of Force Chain Network in Suspensions

A Graph Neural Network Approach to Dispersed Systems

A Scalable Graph Neural Network Method For Developing An Accurate Force Field Of Large Flexible Organic Molecules

A Microstructure-based Graph Neural Network for Accelerating Multiscale Simulations

Fast Simulation of Particulate Suspensions Enabled by Graph Neural Network

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

Multi-scale Graph Neural Network for Physics-Informed Fluid Simulation

Prediction of force chains for dense granular flows using machine learning approach

Accurate and scalable graph neural network force field and molecular dynamics with direct force architecture

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

Accurate and scalable multi-element graph neural network force field and molecular dynamics with direct force architecture

X-MeshGraphNet: Scalable Multi-Scale Graph Neural Networks for Physics Simulation

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

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

Transferable Coarse Graining Via Contrastive Learning of Graph Neural Networks

NN-EVP: A physics informed neural network-based elasto-viscoplastic framework for predictions of grain size-aware flow response under large deformations

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

Scalable and Consistent Graph Neural Networks for Distributed Mesh-based Data-driven Modeling

Scientific Computing Algorithms to Learn Enhanced Scalable Surrogates for Mesh Physics

Predicting Material Properties Using a 3D Graph Neural Network with Invariant Local Descriptors

Scaling Graph Neural Networks to Large Proteins