Materials Processing and Fundamental Understanding Based on Machine Learning and Data Informatics: AI-guided Microstructure Study
Program Organizers: Kathy Lu, University of Alabama Birmingham; Jian Luo, University of California, San Diego; Xian-Ming Bai, Virginia Polytechnic Institute and State University; Yi Je Cho, Sunchon National University
Monday 8:00 AM
October 10, 2022
Room: 311
Location: David L. Lawrence Convention Center
Session Chair: Kathy Lu, University of Alabama Birmingham
8:00 AM Invited
AI/ML-Driven Multi-Scale Modeling and Design of Structural Materials: Pinar Acar1; Sheng Liu1; Mahmudul Hasan1; Arulmurugan Senthilnathan1; Hengduo Zhao1; 1Virginia Tech
This study addresses the role of artificial intelligence/machine learning (AI/ML) in the multi-scale modeling and design of structural materials. Of particular interest is the design of microstructures to improve the mechanical properties at the component level with the help of data-driven and physics-based/informed AI/ML approaches. Example applications will be discussed for the design of cellular mechanical metamaterials (CMMs) and polycrystalline microstructures (PMs). The CMM applications will explore the extreme properties of the representative volume elements (RVEs) and perform inverse design to achieve target RVE properties using Generative Adversarial Networks (GANs). The PM applications will study the large deformation (crystal plasticity) behavior by imposing constraints arising from the microstructural orientation space using physics-based/informed ML and explore the processing-(micro)structure-property relationships of conventionally and additively manufactured aerospace-grade alloys using Gaussian Process Regression.
8:30 AM
AI Driven Microscopic Analysis to Predict the Local Structure in Zirconia Ceramics: Prachi Garg1; Kristofer Reyes1; Baishakhi Mazumder1; 1University at Buffalo
Tetragonal yttria stabilized zirconia (t-YSZ) is widely applied in dental restoration and solid oxide fuel cells due to their excellent properties. To increase the phase structure and phase stability of t-YSZ, a critical vacancy concentration is required. Annihilated vacancies cause the degradation of the oxide material driving the importance of vacancy mapping and detection. However, it is challenging to detect vacancies in oxide materials using conventional microscopy tool. Atom probe microscopy, a three-dimensional nano-analytical tool can provide atomic positions along with structural and chemical information, however is limited to vacancy detection. We applied deep learning model on the microscopy data, which was trained on synthetic data generated by crystal simulations using an empirical ball-and-spring model. The trained AI model will automatically analyze the real APT data to learn and predict the local structure including vacancies to predict the mechanical stability of the material.
8:50 AM
Graph Neural Network Modeling of Deforming Polycrystals: Darren Pagan1; 1Pennsylvania State University
Here the applicability of using graph neural networks (GNNs) to predict grain-scale elastic response of polycrystalline metallic alloys is assessed. Using GNN surrogate models, the stresses within embedded grains in Low Solvus High Refractory Nickel (LSHR) Superalloy and Ti 7wt%Al (Ti-7Al) in uniaxial tension are predicted for both synthetic and measured 3D microstructures. A transfer learning approach is taken in which the GNN surrogate models are trained using crystal elasticity finite element modeling (FEM) simulations and then the trained surrogate models are used to predict the mechanical response of microstructures measured with high-energy X-ray diffraction microscopy (HEDM). The performance of using various microstructural and micromechanical descriptors for input nodal features to predict grain stresses is explored. The effects of elastic anisotropy on GNN model performance and outlooks for extension to plasticity will be discussed.
9:10 AM
High-throughput Machine Learning Experiments with Graph Neural Networks for Predicting Abnormal Grain Growth in Polycrystalline Materials: Ryan Cohn1; Elizabeth Holm1; 1Carnegie Mellon University
Abnormal grain growth (AGG) significantly affects the properties of materials, but is not well understood for many processes. In this study, Monte Carlo was applied to generate a large dataset of AGG simulations. After representing the microstructure as a graph of connected grains, graph neural networks were trained to predict the occurrence of AGG using only the initial state of the system as input. The preliminary results indicated that a simple graph network outperforms a standard computer vision approach used for comparison. Further exploration required high-throughput experiments, leading to the development of a flexible, containerized, and cloud-native approach for running experiments. Extensive parameter sweeps were conducted to interpret the relative feature importance of the inputs, providing physical insight to the mechanism of AGG. The results motivate ongoing efforts to replace the binary classification approach with statistical predictions, and using a more sophisticated set of features generated through deep learning.
9:30 AM
Microstructure Characterization and Reconstruction by Deep Learning Methodology: Satoshi Noguchi1; Junya Inoue1; 1The University of Tokyo
For the establishment of process–structure–property linkage, we propose an image-based general methodology for the characterization and reconstruction of material microstructures using two deep learning networks, a vector quantized variational auto-encoder a vector quantized variational auto-encoder (VQVAE) and a pixel convolutional neural network (PixelCNN). VQVAE is used for the extraction of spatial arrangements of geometrical features corresponding to input micrographs, and PixelCNN is used for the determination of spatial correlation among the extracted geometrical features depending on process parameters and/or material properties. We applied our framework in the generation of low-carbon-steel microstructures from the given material processing. The results show good agreement with the experimental observation qualitatively in terms of the basic topology and quantitatively in terms of the volume fraction and the average grain size, demonstrating the potential of applying the proposed methodology to forward/inverse material design.
9:50 AM
Large Scale Atomistic Simulation of the B1-B4 Phase Transition of GaN with the Machine Learning Potential: Qiang Zhu1; Pedro Santos1; Yansun Yao2; 1University of Nevada, Las Vegas; 2University of Saskatchewan
In this work, we introduce the construction of machine learning potential (MLP) for the purpose of solid-solid phase transition studies. We will explain how to generate the high quality training data from quantum mechanical (QM) simulation to adequately describe the potential energy surface between multiple solid forms. The data is then used to train the accurate MLP based our recently developed NN-SNAP scheme. Using the GaN as a model system, we apply the newly developed MLP to investigate the B1-B4 phase transition under high pressure based on metadynamics simulation. Varying the size of starting models, we can clearly identify both homogeneous and hetrogenous nucleation mechanisms that cannot be accounted from the past studies based on QM simulation.