Reconstructing 3D granular microstructures within volumes of arbitrary geometries from limited 2D image data is crucial for predicting the material properties, as well as performances of structural components accounting for material microstructural effects. We present a novel generative learning framework that enables exascale reconstruction of granular microstructures within complex 3D geometric volumes. Building upon existing transfer learning techniques using pre-trained convolutional neural networks (CNN), we introduce several key innovations to overcome the difficulties inherent in arbitrary geometries. Our framework incorporates periodic boundary conditions using circular padding techniques, ensuring continuity and representativeness of the reconstructed microstructures. We also introduce a novel seamless transition reconstruction (STR) method that creates statistically equivalent transition zones to integrate multiple pre-existing 3D microstructure volumes. Based on STR, we propose a cost-effective strategy for reconstructing microstructures within complex geometric volumes, minimizing computational waste. Validation through numerical experiments using kinetic Monte Carlo simulations demonstrates accurate reproduction of grain statistics, including grain size distributions and morphology. A case study involving the reconstruction of a 4-blade propeller microstructure illustrates the method's capability to efficiently handle complex geometries. The proposed framework significantly reduces computational demands while maintaining high reconstruction quality, paving the way for scalable microstructure reconstruction in materials design and analysis.
Simulation-based approaches to microstructure generation can suffer from a variety of limitations, such as high memory usage, long computational times, and difficulties in generating complex geometries. Generative machine learning models present a way around these issues, but they have previously been limited by the fixed size of their generation area. We present a new microstructure generation methodology leveraging advances in inpainting using denoising diffusion models to overcome this generation area limitation. We show that microstructures generated with the presented methodology are statistically similar to grain structures generated with a kinetic Monte Carlo simulator, SPPARKS.
Crystal plasticity finite element method (CPFEM) has been an integrated computational materials engineering (ICME) workhorse to study materials behaviors and structure-property relationships for the last few decades. These relations are mappings from the microstructure space to the materials properties space. Due to the stochastic and random nature of microstructures, there is always some uncertainty associated with materials properties, for example, in homogenized stress-strain curves. For critical applications with strong reliability needs, it is often desirable to quantify the microstructure-induced uncertainty in the context of structure-property relationships. However, this uncertainty quantification (UQ) problem often incurs a large computational cost because many statistically equivalent representative volume elements (SERVEs) are needed. In this article, we apply a multi-level Monte Carlo (MLMC) method to CPFEM to study the uncertainty in stress-strain curves, given an ensemble of SERVEs at multiple mesh resolutions. By using the information at coarse meshes, we show that it is possible to approximate the response at fine meshes with a much reduced computational cost. We focus on problems where the model output is multi-dimensional, which requires us to track multiple quantities of interest (QoIs) at the same time. Our numerical results show that MLMC can accelerate UQ tasks around 2.23×, compared to the classical Monte Carlo (MC) method, which is widely known as ensemble average in the CPFEM literature.
Thermal spray deposition is an inherently stochastic manufacturing process used for generating thick coatings of metals, ceramics and composites. The generated coatings exhibit hierarchically complex internal structures that affect the overall properties of the coating. The deposition process can be adequately simulated using rules-based process simulations. Nevertheless, in order for the simulation to accurately model particle spreading upon deposition, a set of predefined rules and parameters need to be calibrated to the specific material and processing conditions of interest. The calibration process is not trivial given the fact that many parameters do not correspond directly to experimentally measurable quantities. This work presents a protocol that automatically calibrates the parameters and rules of a given simulation in order to generate the synthetic microstructures with the closest statistics to an experimentally generated coating. Specifically, this work developed a protocol for tantalum coatings prepared using air plasma spray. The protocol starts by quantifying the internal structure using 2-point statistics and then representing it in a low-dimensional space using Principal Component Analysis. Subsequently, our protocol leverages Bayesian optimization to determine the parameters that yield the minimum distance between synthetic microstructure and the experimental coating in the low-dimensional space.
Heterogenous materials under shock compression can be expected to reach different shock states throughout the material according to local differences in microstructure and the history of wave propagation. Here, a compact, multiple-beam focusing optic assembly is used with high-speed velocimetry to interrogate the shock response of porous tantalum films prepared through thermal-spray deposition. The distribution of particle velocities across a shocked interface is compared to results obtained using a set of defocused interferometric beams that sampled the shock response over larger areas. The two methods produced velocity distributions along the shock plateau with the same mean, while a larger variance was measured with narrower beams. The finding was replicated using three-dimensional, mesoscopically resolved hydrodynamics simulations of solid tantalum with a pore structure mimicking statistical attributes of the material and accounting for radial divergence of the beams, with agreement across several impact velocities. Accounting for pore morphology in the simulations was found to be necessary for replicating the rise time of the shock plateau. The validated simulations were then used to show that while the average velocity along the shock plateau could be determined accurately with only a few interferometric beams, accurately determining the width of the velocity distribution, which here was approximately Gaussian, required a beam dimension much smaller than the spatial correlation lengthscale of the velocity field, here by a factor of ∼30×, with implications for the study of other porous materials.
SPPARKS is an open-source parallel simulation code for developing and running various kinds of on-lattice Monte Carlo models at the atomic or meso scales. It can be used to study the properties of solid-state materials as well as model their dynamic evolution during processing. The modular nature of the code allows new models and diagnostic computations to be added without modification to its core functionality, including its parallel algorithms. A variety of models for microstructural evolution (grain growth), solid-state diffusion, thin film deposition, and additive manufacturing (AM) processes are included in the code. SPPARKS can also be used to implement grid-based algorithms such as phase field or cellular automata models, to run either in tandem with a Monte Carlo method or independently. For very large systems such as AM applications, the Stitch I/O library is included, which enables only a small portion of a huge system to be resident in memory. In this paper we describe SPPARKS and its parallel algorithms and performance, explain how new Monte Carlo models can be added, and highlight a variety of applications which have been developed within the code.
Physics-constrained machine learning is emerging as an important topic in the field of machine learning for physics. One of the most significant advantages of incorporating physics constraints into machine learning methods is that the resulting model requires significantly less data to train. By incorporating physical rules into the machine learning formulation itself, the predictions are expected to be physically plausible. Gaussian process (GP) is perhaps one of the most common methods in machine learning for small datasets. In this paper, we investigate the possibility of constraining a GP formulation with monotonicity on three different material datasets, where one experimental and two computational datasets are used. The monotonic GP is compared against the regular GP, where a significant reduction in the posterior variance is observed. The monotonic GP is strictly monotonic in the interpolation regime, but in the extrapolation regime, the monotonic effect starts fading away as one goes beyond the training dataset. Imposing monotonicity on the GP comes at a small accuracy cost, compared to the regular GP. The monotonic GP is perhaps most useful in applications where data are scarce and noisy, and monotonicity is supported by strong physical evidence.
Physics-constrained machine learning is emerging as an important topic in the field of machine learning for physics. One of the most significant advantages of incorporating physics constraints into machine learning methods is that the resulting machine learning model requires significantly fewer data to train. By incorporating physical rules into the machine learning formulation itself, the predictions are expected to be physically plausible. Gaussian process (GP) is perhaps one of the most common methods in machine learning for small datasets. In this paper, we investigate the possibility of constraining a GP formulation with monotonicity on two different material datasets, where one experimental and one computational dataset is used. The monotonic GP is compared against the regular GP, where a significant reduction in the posterior variance is observed. The monotonic GP is strictly monotonic in the interpolation regime, but in the extrapolation regime, the monotonic effect starts fading away as one goes beyond the training dataset. Imposing monotonicity on the GP comes at a small accuracy cost, compared to the regular GP. The monotonic GP is perhaps most useful in applications where data is scarce and noisy or when the dimensionality is high, and monotonicity is where supported by strong physical reasoning.
Thermal spray processes involve the repeated impact of millions of discrete particles, whose melting, deformation, and coating-formation dynamics occur at microsecond timescales. The accumulated coating that evolves over minutes is comprised of complex, multiphase microstructures, and the timescale difference between the individual particle solidification and the overall coating formation represents a significant challenge for analysts attempting to simulate microstructure evolution. In order to overcome the computational burden, researchers have created rule-based models (similar to cellular automata methods) that do not directly simulate the physics of the process. Instead, the simulation is governed by a set of predefined rules, which do not capture the fine-details of the evolution, but do provide a useful approximation for the simulation of coating microstructures. Here, we introduce a new rules-based process model for microstructure formation during thermal spray processes. The model is 3D, allows for an arbitrary number of material types, and includes multiple porosity-generation mechanisms. Example results of the model for tantalum coatings are presented along with sensitivity analyses of model parameters and validation against 3D experimental data. The model's computational efficiency allows for investigations into the stochastic variation of coating microstructures, in addition to the typical process-to-structure relationships.
Grain-scale microstructure evolution during additive manufacturing is a complex physical process. As with traditional solidification methods of material processing (e.g. casting and welding), microstructural properties are highly dependent on the solidification conditions involved. Additive manufacturing processes however, incorporate additional complexity such as remelting, and solid-state evolution caused by subsequent heat source passes and by holding the entire build at moderately high temperatures during a build. We present a three-dimensional model that simulates both solidification and solid-state evolution phenomena using stochastic Monte Carlo and Potts Monte Carlo methods. The model also incorporates a finite-difference based thermal conduction solver to create a fully integrated microstructural prediction tool. The three modeling methods and their coupling are described and demonstrated for a model study of laser powder-bed fusion of 300-series stainless steel. The investigation demonstrates a novel correlation between the mean number of remelting cycles experienced during a build, and the resulting columnar grain sizes.
Conditional Point Sampling (CoPS) is a recently developed stochastic media transport algorithm that has demonstrated a high degree of accuracy in 1-D and 3-D calculations for binary mixtures with Markovian mixing statistics. In theory, CoPS has the capacity to be accurate for material structures beyond just those with Markovian statistics. However, realizing this capability will require development of conditional probability functions (CPFs) that are based, not on explicit Markovian properties, but rather on latent properties extracted from material structures. Here, we describe a first step towards extracting these properties by developing CPFs using deep neural networks (DNNs). Our new approach lays the groundwork for enabling accurate transport on many classes of stochastic media. We train DNNs on ternary stochastic media with Markovian mixing statistics and compare their CPF predictions to those made by existing CoPS CPFs, which are derived based on Markovian mixing properties. We find that the DNN CPF predictions usually outperform the existing approximate CPF predictions, but with wider variance. In addition, even when trained on only one material volume realization, the DNN CPFs are shown to make accurate predictions on other realizations that have the same internal mixing behavior. We show that it is possible to form a useful CoPS CPF by using a DNN to extract correlation properties from realizations of stochastically mixed media, thus establishing a foundation for creating CPFs for mixtures other than those with Markovian mixing, where it may not be possible to derive an accurate analytical CPF.
Machine learning (ML), including deep learning (DL), has become increasingly popular in the last few years due to its continually outstanding performance. In this context, we apply machine learning techniques to "learn" the microstructure using both supervised and unsupervised DL techniques. In particular, we focus (1) on the localization problem bridging (micro)structure (localized) property using supervised DL and (2) on the microstructure reconstruction problem in latent space using unsupervised DL. The goal of supervised and semi-supervised DL is to replace crystal plasticity finite element model (CPFEM) that maps from (micro)structure (localized) property, and implicitly the (micro)structure (homogenized) property relationships, while the goal of unsupervised DL is (1) to represent high-dimensional microstructure images in a non-linear low-dimensional manifold, and (2) to discover a way to interpolate microstructures via latent space associating with latent microstructure variables. At the heart of this report is the applications of several common DL architectures, including convolutional neural networks (CNN), autoencoder (AE), and generative adversarial network (GAN), to multiple microstructure datasets, and the quest of neural architecture search for optimal DL architectures.
A parallel, adaptive overlay grid procedure is proposed for use in generating all-hex meshes for stochastic (SVE) and representative (RVE) volume elements in computational materials modeling. The mesh generation process is outlined including several new advancements such as data filtering to improve mesh quality from voxelated and 3D image sources, improvements to the primal contouring method for constructing material interfaces and pillowing to improve mesh quality at boundaries. We show specific examples in crystal plasticity and syntactic foam modeling that have benefitted from the proposed mesh generation procedure and illustrate results of the procedure with several practical mesh examples.
