Abstract:A key problem in computational material science deals with understanding the effect of material distribution (i.e., microstructure) on material performance. The challenge is to synthesize microstructures, given a finite number of microstructure images, and/or some physical invariances that the microstructure exhibits. Conventional approaches are based on stochastic optimization and are computationally intensive. We introduce three generative models for the fast synthesis of binary microstructure images. The first model is a WGAN model that uses a finite number of training images to synthesize new microstructures that weakly satisfy the physical invariances respected by the original data. The second model explicitly enforces known physical invariances by replacing the traditional discriminator in a GAN with an invariance checker. Our third model combines the first two models to reconstruct microstructures that respect both explicit physics invariances as well as implicit constraints learned from the image data. We illustrate these models by reconstructing two-phase microstructures that exhibit coarsening behavior. The trained models also exhibit interesting latent variable interpolation behavior, and the results indicate considerable promise for enforcing user-defined physics constraints during microstructure synthesis.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to synthesize new microstructures in computational materials science based on a limited number of microstructure images or certain physical invariants. Specifically, the research objective is to understand the influence of material distribution (i.e., microstructure) on material properties and develop methods for rapidly generating binary microstructure images. Traditional methods are based on random optimization, which are computationally costly and time - consuming. Therefore, this paper proposes three generative models to accelerate the synthesis of microstructures and ensure that the generated microstructures satisfy physical invariants. ### Specific description of the problem 1. **Influence of material microstructure on properties**: The final properties of materials are closely related to their microstructures. In order to design materials with specific properties, it is necessary to understand and control the characteristics of microstructures. 2. **Limitations of existing methods**: Traditional microstructure synthesis methods rely on random optimization techniques. These methods have high computational complexity and require a large amount of computational resources and time. In addition, these methods are inefficient when dealing with complex multiphase materials. 3. **Data limitations**: The number of microstructure images obtained from experiments is limited, and it is difficult to directly modify specific features through experiments. Therefore, it is necessary to develop methods that can learn from limited data and generate new microstructures that conform to physical laws. ### Methods proposed in the paper To solve the above problems, the paper proposes three models based on generative adversarial networks (GAN): 1. **Wasserstein GAN (WGAN)**: - Use the Wasserstein distance as a loss function to train the generator to generate new microstructure images. - The generated images can weakly satisfy the physical invariants in the original data. 2. **Generative Invariance Network (GIN)**: - Replace the discriminator in the traditional GAN with an invariance checker, which explicitly forces the generated images to satisfy the known physical invariants. - The invariance checker guides the learning of the generator by calibrating the physical characteristics in the training data. 3. **Hybrid (GAN + GIN) model**: - Combine the advantages of the previous two models, and simultaneously utilize the implicit data features and explicit physical constraints. - The generated microstructures not only conform to the statistical characteristics in the data, but also satisfy the user - defined physical invariants. ### Summary This paper aims to overcome the computational bottlenecks of traditional microstructure synthesis methods by introducing efficient generative models and ensure that the generated microstructures conform to physical laws. This will help accelerate the design and optimization of new materials, especially in application scenarios where precise control of microstructure characteristics is required.

Physics-aware Deep Generative Models for Creating Synthetic Microstructures

3D-Aware Image Synthesis Via Learning Structural and Textural Representations

Deep learning for synthetic microstructure generation in a materials-by-design framework for heterogeneous energetic materials

Establishing process-structure linkages using Generative Adversarial Networks

FluxGAN: A Physics-Aware Generative Adversarial Network Model for Generating Microstructures That Maintain Target Heat Flux

3DMaterialGAN: Learning 3D Shape Representation from Latent Space for Materials Science Applications

Generating three-dimensional structures from a two-dimensional slice with generative adversarial network-based dimensionality expansion

Using Physics Informed Generative Adversarial Networks to Model 3D porous media

A physics-informed GAN Framework based on Model-free Data-Driven Computational Mechanics

A Generative Machine Learning Model for Material Microstructure 3D Reconstruction and Performance Evaluation

Evaluating the diversity and utility of materials proposed by generative models

Microstructure reconstruction of 2D/3D random materials via diffusion-based deep generative models

Physics guided deep learning for generative design of crystal materials with symmetry constraints

Generative deep learning for the inverse design of materials

3-D Inorganic Crystal Structure Generation and Property Prediction via Representation Learning

A Data-Driven Framework for Designing Microstructure of Multifunctional Composites with Deep-Learned Diffusion-Based Generative Models

Deep Convolutional Generative Modeling for Artificial Microstructure Development of Aluminum-Silicon Alloy

Deep Generative Models for Materials Discovery and Machine Learning-Accelerated Innovation

Generating three-dimensional bioinspired microstructures using transformer-based generative adversarial network