Computational Discovery and Design of Materials : Session III
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS: Computational Materials Science and Engineering Committee, TMS: Integrated Computational Materials Engineering Committee
Program Organizers: Houlong Zhuang, Arizona State University; Duyu Chen, University of California, Santa Barbara; Ismaila Dabo, Pennsylvania State University; Yang Jiao, Arizona State University; Sara Kadkhodaei, University Of Illinois Chicago; Mahesh Neupane, Army Research Laboratory; Xiaofeng Qian, Texas A&M University; Arunima Singh, Arizona State University; Natasha Vermaak, Lehigh University
Tuesday 8:00 AM
March 21, 2023
Room: Cobalt 502A
Location: Hilton
Session Chair: Yang Jiao, Arizona State University; Duyu Chen, University of California, Santa Barbara; Houlong Zhuang, Arizona State University
8:00 AM Invited
Adaptive Discovery and Mixed-variable Bayesian Optimization of Next Generation Synthesizable Microelectronic Materials: Wei Chen1; Hengrui Zhang1; 1Northwestern University
Materials design is challenged by high-dimensionality of the composition–structure space and mixed qualitative and quantitative design variables. We propose a machine learning assisted adaptive design framework and demonstrate it in the autonomous search for metal-insulation transition (MIT) materials. Starting with natural language processing methods, we extract from the literature the potential materials families exhibiting MITs. Next, we use active learning and first-principles calculations to virtually screen these materials, building a classier and a database for MIT materials. Within two prominent MIT families, we conduct Bayesian Optimization-based adaptive discovery driven by a novel latent variable Gaussian process (LVGP) model, which can predict material properties from the composition, that contains categorical design variables (e.g., elements). Using this framework, we have discovered several potential MIT materials not previously reported, for which experimental syntheses are in process. Our framework can be extended to accelerate materials design beyond MITs.
8:30 AM Invited
Computer Vision Problems in Transmission Electron Microscopy: Huolin Xin1; 1University of California - Irvine
Deep learning schemes have already impacted areas such as cognitive game theory (e.g., computer chess and the game of Go), pattern (e.g., facial or fingerprint) recognition, event forecasting, and bioinformatics. They are beginning to make major inroads within physics, chemistry and materials sciences and hold considerable promise for accelerating the discovery of new theories and materials. In this talk, I will introduce deep convolutional neural networks and how they can be applied to the computer vision problems in transmission electron microscopy and tomographic imaging.
9:00 AM
Developing an Ab Initio-Kinetic Passivation Model for High-throughput Screening of Material Stability: Rachel Gorelik1; Arunima Singh1; 1Arizona State University
With corrosion remaining a significant economic issue, the ability to a priori predict the kinetics of material corrosion remains an important consideration in the field of materials discovery, which often lacks the ability to predict time-dependent corrosion behavior during high-throughput screening. To address this challenge, we have developed an ab initio-kinetic framework for predicting material stability by combining density functional theory and molecular dynamics simulations with a kinetic passivation model called the Point Defect Model (PDM). This non-empirical framework can predict the growth rate of passivation films in any elemental material without prior experimental knowledge. After developing and automating this workflow, we have evaluated its performance through two common metal case studies (Cu and W) and compared with available experimental literature. Finally, we have evaluated the viability of extending this framework to the more than 700 elemental materials which are currently available in the Materials Project database.
9:20 AM Break
9:40 AM Invited
Data- and Physics-driven Approaches to Discovering the Governing Equations for Complex Phenomena in Heterogeneous Materials: Muhammad Sahimi1; 1University of Southern California
Discovering the governing equations for various phenomena involving heterogeneity or stochasticity, has dominated the physical sciences and engineering. The classical approach has been based on the fundamental conservation laws, which are averaged over an ensemble of possible realizations. This is valid only if there is a well-defined representative elementary volume, which might not exist. At the same time, rapid advances in various technologies for data-acquisition software/hardware have also opened new fields of exploration for which the governing equations are difficult to derive. The most obvious examples are biological, and nano- and neurosciences where first-principle derivations are very difficult to carry out, while data are becoming abundant. How do we discover the governing equations that not only honor and better explain the data, but also provide predictions for either the future, or over larger length scales? This presentation describes advances in this new field and discusses two examples in detail.
10:10 AM
An Inverse Materials Design Route Based on Structure-property Linkages Leveraging 3D Convolutional Neural Network and Bayesian Optimization: Xiao Shang1; Yu Zou1; 1University of Toronto
Material properties are macroscale expressions of material microstructures, and the key to materials design lies in identifying the underlying structure-property (SP) linkages. Compared with conventional heuristic- and simulation-based methods, advanced Machine Learning (ML)- based techniques are advantageous for its high accuracy and extremely low computation time, enabling fast high-throughput materials design. In this work, we propose a general route for inverse materials design, where realistic microstructures can be identified with target mechanical properties such as yield strengths. 3D Convolutional Neural Networks (CNNs) are used to mine SP linkages from synthesised microstructures datasets, after which Bayesian Optimization (BO) is used for inversely identifying the optimal microstructures expressing desired mechanical properties. Titanium alloy (Ti6-Al4-V) is used to demonstrate the design route, which is generalizable to other materials systems. Our design route provides a reliable and computational efficient way to achieve “Materials-by-Design” for guiding the design and manufacturing of next generation high-performing materials.