Synthesizing controlled microstructures of porous media using generative adversarial networks and reinforcement learning

Abstract

For material modeling and discovery, synthetic microstructures play a critical role as digital twins. They provide stochastic samples upon which direct numerical simulations can be conducted to populate material databases. A large ensemble of simulation data on synthetic microstructures may provide supplemental data to inform and refine macroscopic material models, which might not be feasible from physical experiments alone. However, synthesizing realistic microstructures with realistic microstructural attributes is highly challenging. Thus, it is often oversimplified via rough approximations that may yield an inaccurate representation of the physical world. Here, we propose a novel deep learning method that can synthesize realistic three-dimensional microstructures with controlled structural properties using the combination of generative adversarial networks (GAN) and actor-critic (AC) reinforcement learning. The GAN-AC combination enables the generation of microstructures that not only resemble the appearances of real specimens but also yield user-defined physical quantities of interest (QoI). Our validation experiments confirm that the properties of synthetic microstructures generated by the GAN-AC framework are within a 5% error margin with respect to the target values. The scientific contribution of this paper resides in the novel design of the GAN-AC microstructure generator and the mathematical and algorithmic foundations therein. The proposed method will have a broad and substantive impact on the materials community by providing lenses for analyzing structure-property-performance linkages and for implementing the notion of ‘materials-by-design’.

Quantifying microstructures of earth materials using higher-order spatial correlations and deep generative adversarial networks

“…Chun et al 29 implemented a patch-based DCGAN for 2D reconstruction of heterogeneous energetic materials, and showed that it is possible to better control the micromorphology of reconstructions by introducing two input vectors by which one can manipulate the local stochasticity and global morphology of microstructure. This work was then expanded by 34 by adding an actor-critic (AC) reinforcement learning to 3D DCGAN for generating microstructure with user-defined properties.…”

“…In recent years, the advent of deep learning (DL) has opened up unprecedented opportunities and insight into image reconstruction. Generative adversarial networks (GANs) are among the advanced DL-based generative models which have been successfully applied for 2D 29 and 3D 13,[30][31][32][33][34] microstructures reconstruction. Notably, a growing body of literature has recently investigated 2D to 3D image reconstructions intending to infer 3D morphological and structural properties using features extracted from 2D images in specific orientations [35][36][37][38] .…”

Artificial Intelligence Approaches for Energetic Materials by Design: State of the Art, Challenges, and Future Directions

“…Data‐driven methods can be classified broadly based on how they use the data: supervised learning , which requires a pairing of the input data set with a labeled output, and unsupervised learning , which only requires the input data without a labeled output; semi‐supervised learning sits in between [25]. The supervised method can be further broken down into classification (e. g., assigning the material to a general type or class of “similar” materials given an input sample [93]) or regression tasks (e. g., prediction of a property, such as the permeability of a porous material [57]). A variety of AI/ML methods that fall under these broad categories have been used for various materials applications.…”

“…However, recently, there has been efforts to use RL algorithms as an optimization solver [149], in which RL is seen as a learning‐based heuristic search. The main hypothesis is that once trained on a set of optimization problems, RL can learn a policy to efficiently generate solutions for similar but unseen problems [57]. In fact, traditional optimization approaches need to apply to a general class of optimization problems, such that no problem‐specific patterns and trends need to be taken into account.…”

A Machine Learning Challenge: Detection of Cardiac Amyloidosis Based on Bi-Atrial and Right Ventricular Strain and Cardiac Function

“…The complexity of this diagnostic approach, encompassing a wide spectrum of parameters, demands the computational gold standard termed machine learning. Machine learning is a function of artificial intelligence, used as a promising alternative to overcoming the limitations of traditional prediction approaches [ 17 ]. Recent supervised and unsupervised machine learning approaches have proven highly valuable for the classification, regression and prediction of complex datasets in all industrial sectors, including the medical field of CA diagnostics [ 3 , 18 , 19 ].…”