Abstract:Concurrent multiscale finite element analysis (FE2) is a powerful approach for high-fidelity modeling of materials for which a suitable macroscopic constitutive model is not available. However, the extreme computational effort associated with computing a nested micromodel at every macroscopic integration point makes FE2 prohibitive for most practical applications. Constructing surrogate models able to efficiently compute the microscopic constitutive response is therefore a promising approach in enabling concurrent multiscale modeling. This work presents a reduction framework for adaptively constructing surrogate models based on statistical learning. The nested micromodels are replaced by a machine learning surrogate model based on Gaussian Processes (GP). The need for offline data collection is bypassed by training the GP models online based on data coming from a small set of fully-solved anchor micromodels that undergo the same strain history as their associated macro integration points. The Bayesian formalism inherent to GP models provides a natural tool for uncertainty estimation through which new observations or inclusion of new anchors are triggered. The surrogate constitutive manifold is constructed with as few micromechanical evaluations as possible by enhancing the GP models with gradient information and the solution scheme is made robust through a greedy data selection approach embedded within the conventional finite element solution loop for nonlinear analysis. The sensitivity to model parameters is studied with a tapered bar example with plasticity, while the applicability of the model to more complex cases is demonstrated with the elastoplastic analysis of a plate with multiple cutouts and a crack growth example for mixed-mode bending. Significant efficiency gains are obtained without resorting to offline training.

What problem does this paper attempt to address?

This paper aims to solve the problem of **excessively high computational cost in parallel multi - scale finite element analysis (FE²)**. Specifically, the FE² method can perform high - fidelity modeling of materials in the absence of a suitable macroscopic constitutive model. However, since it is necessary to calculate the nested microscopic models at each macroscopic integration point, this leads to extremely high computational costs, making FE² infeasible in most practical applications. To solve this problem, the author proposes a surrogate model construction framework based on probabilistic machine learning. The core of this method is to replace the original high - fidelity microscopic model with a machine - learning surrogate model, thereby significantly reducing the computational cost. Specifically, this framework has the following characteristics: 1. **Online training**: The Gaussian process (GP) surrogate model is trained online through a small number of fully - solved anchored microscopic models (these models have experienced the same strain history as their corresponding macroscopic integration points). This method avoids the need for offline data collection. 2. **Uncertainty estimation**: By using the Bayesian form in the GP model, online uncertainty estimation can be carried out naturally. When the uncertainty exceeds a certain threshold, new observations will be triggered or new anchored microscopic models will be added to improve the accuracy of the model. 3. **Gradient - information enhancement**: By introducing gradient information, the GP model can construct a surrogate constitutive manifold with as few micromechanical evaluations as possible. This makes the solution more efficient and robust. 4. **Greedy data selection**: The greedy data selection method is embedded in the traditional finite - element solution cycle to ensure the numerical stability of the nonlinear analysis and minimize the number of new data samplings, thereby maximizing efficiency. 5. **Adaptive framework**: This framework can adaptively construct a surrogate model without relying on offline training and dynamically adjust the model complexity according to the current computational requirements. Through the above methods, the author has successfully reduced the computational cost of FE² while maintaining high - fidelity modeling accuracy. The paper verifies the effectiveness and universality of this framework through multiple numerical experiments, including problems such as strain localization and crack propagation. ### Summary of key formulas - **Gaussian process regression prediction**: \[ E[y^*|x^*] = k^T_*(K_o + \sigma^2_n I)^{-1} t_o \] \[ V[y^*|x^*] = k(x^*, x^*) - k^T_*(K_o + \sigma^2_n I)^{-1} k_* \] - **Covariance matrix including derivative observations**: \[ p\left(\begin{bmatrix} t_o \\ t'_o \end{bmatrix}\right) = \mathcal{N}\left(\begin{bmatrix} t_o \\ t'_o \end{bmatrix} \middle| 0, K_o\right) \] where, \[ K_o = \begin{bmatrix} K + \sigma^2_n I & K_{td} \\ K^T_{td} & K_{dd} \end{bmatrix} \] - **Hyperparameter optimization**: \[ \nabla \ln p(t_o) = -\frac{1}{2} \text{tr}(K^{-1}_o \nabla K_o) + \frac{1}{2} t^T_o K^{-1}_o \nabla K_o t_o \] Through these methods, this paper provides an effective solution, enabling the FE² method to be more widely used in practical engineering design.

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

Bridging The Micro And Macro For Granular Media: A Computational Multi-Scale Paradigm

A Coupled SPFEM/DEM Approach for Multiscale Modeling of Large‐deformation Geomechanical Problems

Concurrent multiscale simulations of nonlinear random materials using probabilistic learning

A Microstructure-based Graph Neural Network for Accelerating Multiscale Simulations

On-the-fly adaptivity for nonlinear twoscale simulations using artificial neural networks and reduced order modeling

Integration of Active Learning and MCMC Sampling for Efficient Bayesian Calibration of Mechanical Properties

A reduced order model for geometrically parameterized two-scale simulations of elasto-plastic microstructures under large deformations

On-the-fly multiscale analysis of composite materials with a Generalized Finite Element Method

Model-data-driven constitutive responses: Application to a multiscale computational framework

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

Functional order-reduced Gaussian Processes based machine-learning emulators for probabilistic constitutive modelling

Autonomous Development of a Machine-Learning Model for the Plastic Response of Two-Phase Composites from Micromechanical Finite Element Models

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

Pre-trained transformer model as a surrogate in multiscale computational homogenization framework for elastoplastic composite materials subjected to generic loading paths

Two-Stage Surrogate Modeling for Data-Driven Design Optimization with Application to Composite Microstructure Generation

Learning nonlinear constitutive models in finite strain electromechanics with Gaussian process predictors

Surrogate PHFGMC micromechanical models for multiscale analysis: AI-enhanced low-velocity impact analysis of hat-stiffened composite panels

Accelerating high-strain continuum-scale brittle fracture simulations with machine learning

Adaptive spatiotemporal dimension reduction in concurrent multiscale damage analysis

Efficient Analysis of Composites Manufacturing Using Multi-fidelity Simulation and Probabilistic Machine Learning