Abstract:Driven by the need to accelerate numerical simulations, the use of machine learning techniques is rapidly growing in the field of computational solid mechanics. Their application is especially advantageous in concurrent multiscale finite element analysis (FE$^2$) due to the exceedingly high computational costs often associated with it and the high number of similar micromechanical analyses involved. To tackle the issue, using surrogate models to approximate the microscopic behavior and accelerate the simulations is a promising and increasingly popular strategy. However, several challenges related to their data-driven nature compromise the reliability of surrogate models in material modeling. The alternative explored in this work is to reintroduce some of the physics-based knowledge of classical constitutive modeling into a neural network by employing the actual material models used in the full-order micromodel to introduce non-linearity. Thus, path-dependency arises naturally since every material model in the layer keeps track of its own internal variables. For the numerical examples, a composite Representative Volume Element with elastic fibers and elasto-plastic matrix material is used as the microscopic model. The network is tested in a series of challenging scenarios and its performance is compared to that of a state-of-the-art Recurrent Neural Network (RNN). A remarkable outcome of the novel framework is the ability to naturally predict unloading/reloading behavior without ever seeing it during training, a stark contrast with popular but data-hungry models such as RNNs. Finally, the proposed network is applied to FE$^2$ examples to assess its robustness for application in nonlinear finite element analysis.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to accelerate numerical simulation, especially the high computational cost problem encountered in concurrent multi - scale finite element analysis (FE2). Specifically, the paper focuses on how to use machine - learning techniques to approximate microscopic behavior and accelerate simulation while maintaining the physical characteristics of the material model. ### Main Problems 1. **High Computational Cost**: The traditional FE2 method has extremely high computational costs because it needs to embed a complete microscopic model at each macroscopic numerical integration point for solution, which limits its wide use in practical engineering - scale applications. 2. **Limitations of Data - Driven Surrogate Models**: - **Poor Extrapolation Ability**: Data - driven surrogate models (such as conventional ANN and RNN) often perform poorly outside the training range, especially when dealing with path - dependent materials. - **Lack of Interpretability**: Black - box models are difficult to explain whether their prediction results conform to physical laws, especially when facing different loading paths. ### Solutions To address the above challenges, the paper proposes a new neural network design - **Physically Recurrent Neural Networks (PRNN)**, which aims to combine the advantages of traditional physical modeling and the powerful expressive ability of deep learning. The main features include: - **Introduction of Physical Knowledge**: By embedding the actually used material models (such as elastic, elastoplastic, etc.) in the neural network, the network can naturally capture path - dependence without the need for additional history variables. - **Automatic Memory of Historical Information**: The internal variables in each material model layer will track their own historical states, thereby achieving accurate prediction of unloading/reloading behaviors, even if these behaviors have never appeared in the training data. - **Efficient Data Utilization**: Compared with the traditional RNN, PRNN can achieve better generalization performance with less training data, especially when dealing with complex path - dependent materials. ### Application Verification The paper verifies the effectiveness of the proposed network through a series of numerical experiments and compares it with the state - of - the - art RNN. The results show that PRNN is not only superior to RNN in prediction accuracy, but also shows stronger robustness and generalization ability when dealing with unseen loading paths. In addition, PRNN is also applied to the FE2 framework to evaluate its robustness and accuracy in nonlinear finite element analysis. ### Summary By introducing the physically recurrent neural network, this paper successfully solves the deficiencies of traditional data - driven surrogate models in extrapolation ability and physical consistency, and provides a new solution for accelerating concurrent multi - scale finite element analysis.

Physically recurrent neural networks for path-dependent heterogeneous materials: embedding constitutive models in a data-driven surrogate

A Thermodynamics-Informed Neural Network for Elastoplastic Constitutive Modeling of Granular Materials

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

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

Learning solutions of thermodynamics-based nonlinear constitutive material models using physics-informed neural networks

An indirect training approach for implicit constitutive modelling using recurrent neural networks and the virtual fields method

Self-consistency Reinforced minimal Gated Recurrent Unit for surrogate modeling of history-dependent non-linear problems: Application to history-dependent homogenized response of heterogeneous materials

Thermodynamically Consistent Machine-Learned Internal State Variable Approach for Data-Driven Modeling of Path-Dependent Materials

A deep material network for multiscale topology learning and accelerated nonlinear modeling of heterogeneous materials

Neural Material: Learning Elastic Constitutive Material and Damping Models from Sparse Data

A micromechanics-based recurrent neural networks model for path-dependent cyclic deformation of short fiber composites

Learning solution of nonlinear constitutive material models using physics-informed neural networks: COMM-PINN

Embedding an ANN-Based Crystal Plasticity Model into the Finite Element Framework using an ABAQUS User-Material Subroutine

A physics-informed deep neural network for surrogate modeling in classical elasto-plasticity

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

Hybrid data-driven and physics-informed regularized learning of cyclic plasticity with Neural Networks

Recurrent neural networks and transfer learning for elasto-plasticity in woven composites

Learning the physics-consistent material behavior from experimentally measurable data via PDE-constrained optimization

Data-driven computational mechanics: comparison of model-free and model-based methods in constitutive modeling

A Microstructure-based Graph Neural Network for Accelerating Multiscale Simulations

Learning Physics-Consistent Material Behavior Without Prior Knowledge