Abstract:The need for computational characterization and reconstruction of volumetric maps of stochastic microstructures for understanding the role of material structure in the processing-structure-property chain has been highlighted in the literature. Recently, a promising characterization and reconstruction approach has been developed where the essential idea is to convert the digitized microstructure image into an appropriate training dataset to learn the stochastic nature of the morphology by fitting a supervised learning model to the dataset. This compact model can subsequently be used to efficiently reconstruct as many statistically equivalent microstructure samples as desired. The goal of this paper is to build upon the developed approach in three major directions by: (1) extending the approach to characterize 3D stochastic microstructures and efficiently reconstruct 3D samples, (2) improving the performance of the approach by incorporating user-defined predictors into the supervised learning model, and (3) addressing potential computational issues by introducing a reduced model which can perform as effectively as the full model. We test the extended approach on three examples and show that the spatial dependencies, as evaluated via various measures, are well preserved in the reconstructed samples.Lay description Computational materials science plays an indispensable role in designing advanced materials by simulating virtual samples instead of physical ones. The goal of this paper is to extend our recently developed approach to efficiently generate a batch of virtual three-dimensional (3D) samples which are statistically equivalent to a given 3D image of a stochastic microstructure. These samples can subsequently be used for investigating material properties such as permeability and percolation. The accuracy of the estimated properties hinges on the statistical equivalency between the samples and the original microstructure and therefore one needs to make sure (1) the inherent stochasticity of the original sample is accurately characterized (or, in machine learning parlance, learned), and (2) the learned stochasticity is well preserved when generating (reconstructing) virtual samples based on the learned characteristics. Motivated by these challenges, a handful of methods have previously been developed for microstructure characterization and reconstruction. Bostanabad etal. () have recently developed a versatile and promising approach for microstructure characterization and reconstruction based on supervised learning that is applicable to a wide range of material systems (e.g. isotropic, anisotropic and porous with high/low volume fraction) and is much faster than other methods in the literature. The fundamental idea is to first convert the original microstructure image into an appropriate training dataset and then to fit a supervised leaning model to this dataset and subsequently use the model to generate statistically equivalent microstructures with arbitrary size. In this paper, we improve and generalize the approach in three important ways: (1) characterization and reconstruction of 3D microstructure images: as most important material properties require 3D simulations, we extend the approach to characterize 3D images and demonstrate its computational superiority in 3D reconstruction over other existing methods. (2) Incorporation of user-defined predictors into the supervised learning model: in Bostanabad etal. (), individual pixels were used to learn the material morphology and user-defined predictors (combinations of pixels) were not included into the model. In this paper, we illustrate how such predictors can be defined and incorporated into the model. (3) Addressing potential computational issues that may arise: if the original image is complex and has a high resolution (large number of pixels/voxels), the aforementioned training dataset will require a large memory. This issue is of particular concern in 3D and in this work we demonstrate how the size of the dataset can be reduced with no adverse effect on the results.

Self-Supervised Learning and Prediction of Microstructure Evolution with Recurrent Neural Networks

Self-supervised Learning and Prediction of Microstructure Evolution with Convolutional Recurrent Neural Networks

Spatiotemporal Evolution of Grain Microstructure: A CNN Perspective

A Transfer Learning Approach for Microstructure Reconstruction and Structure-property Predictions.

Time series forecasting of multiphase microstructure evolution using deep learning

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

Spatiotemporal Prediction and Characterization of Microstructure Evolution During In-Situ Heat Treatment Using a Deep Learning Algorithm

Extreme time extrapolation capabilities and thermodynamic consistency of physics-inspired Neural Networks for the 3D microstructure evolution of materials via Cahn-Hilliard flow

Spatiotemporal prediction of solidified dendrites based on convolutional long-short-term neural network

Prediction and characterization of microstructure evolution based on deep learning method and in-situ scanning electron microscope

Recurrent Neural Networks for Prediction of Electronic Excitation Dynamics

Stochastic Microstructure Characterization and Reconstruction Via Supervised Learning

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

Efficient Surrogate Models for Materials Science Simulations: Machine Learning-based Prediction of Microstructure Properties

Exploring Time-Series Transformers for Spatio-Temporal Prediction of Microstructural Evolution of Polycrystalline Grain

Improving direct physical properties prediction of heterogeneous materials from imaging data via convolutional neural network and a morphology-aware generative model

Stochastic reconstruction and performance prediction of cathode microstructures based on deep learning

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

Characterization And Reconstruction Of 3d Stochastic Microstructures Via Supervised Learning

Automatic Detection of Grains in Partially Recrystallized Microstructures Using Deep Learning

Prediction of microstructural-dependent mechanical properties, progressive damage, and stress distribution from X-ray computed tomography scans using a deep learning workflow