Accelerating Materials Science with Big Data and Machine Learning: Session I
Program Organizers: Huan Tran, Georgia Institute Of Technology; Muratahan Aykol, Toyota Research Institute
Monday 8:00 AM
October 18, 2021
Room: A123
Location: Greater Columbus Convention Center
Session Chair: Huan Tran, Georgia Institute Of Technology
8:00 AM Invited
Considerations for Interpretability, Reliability, And Data-efficiency in Machine Learning Properties of Solid-state Materials: Christopher Sutton1; 1University of South Carolina
Advances in machine learning (ML) are making a large impact in many disciplines, including materials and computational chemistry. A particularly exciting application of ML is the prediction of quantum mechanical (QM) properties (e.g., formation energy, bandgap, etc.) using only the structure as input. Assuming sufficient accuracies in the ML models, these methods enable screening of a considerably large chemical space at orders of magnitude lower computational cost than available QM methods. Despite the promise of ML in chemistry, several key challenges remain in both applying and interpreting the results of ML algorithms. Here, we will discuss our efforts in addressing these issues, including our recent work on opening the black box of ML methods by identifying the domain of applicability, i.e., where a given model is reliable.
8:40 AM Invited
Searching for New Ferroelectric Materials Browsing a High-throughput Phonon Database: Maksim Markov1; Louis Alaerts2; Henrique Miranda1; Guido Petretto1; Wei Chen1; Janine George1; Eric Bousquet3; Philippe Ghosez3; Gian-Marco Rignanese1; Geoffroy Hautier4; 1UCLouvain; 2Dartmouth College; 3Université de Liège; 4Dartmouth
Ferroelectric materials are of great fundamental and applied interests with wide applications in many technologies such as electric capacitors, piezoelectric sensors, non-volatile memory devices, or energy converters. In this work, we search for new ferroelectrics by using high-throughput computing and a recently built database of more than 2,000 phonons. Browsing the phonon database, we identify materials exhibiting dynamically unstable polar phonon modes, a signature of a potential ferroelectric. We discuss the structure and chemistries emerging from high-throughput approaches and highlight the challenges in finding ferroelectric materials. We then focus on one new family of ferroelectric materials discovered through high-throughput screening: the anti-Ruddlesden-Popper phases of formula A4X2O with A: a +2 alkali-earth or rare-earth element and X: a −3 anion Bi, Sb, As and P. We show that significant ferroelectricity is present in Ba4Sb2O and demonstrate that Eu4Sb2O is a rare example of a material combining coupled ferroelectricity and ferromagnetism.
9:20 AM
Slip Band Characterization with Microtensile Testing Using Digital Image Processing: Anthony Lombardi1; Elim Schenck1; Subhasish Malik1; Ajit Achuthan1; Sean Banerjee1; Natasha Banerjee1; 1Clarkson University
The characteristics of slip band formation and its evolution on the specimen surface of metals and alloys during a microtensile test provides insight on the influence of various microstructural features on mechanical properties. In this presentation, we discuss two digital image processing approaches to derive slip band characteristics from raw video data of a microtensile test. Our first approach estimates slip line density by detecting lines using Hough, dividing the image region into cells, using Kanade-Lucas-Tomasi feature tracking to track each cell across the video, and counting lines in each cell. We evaluate Hough line detection in grayscale versus black-and-white images, and density detection using binning of lines into pixels versus measuring lengths of cell segments obtained using Cohen-Sutherland clipping. Our second approach matches Gabor filters representing oriented Gaussian-blurred sine functions of varying frequencies to the image for modeling probability distributions of finding various line densities in each video image.
9:40 AM
Materials Graph Ontology for Improving the Standardization and Utilization of Materials Data: Sven Voigt1; Surya Kalidindi1; 1Georgia Institute of Technology
To maximize the utility of the materials data being generated in large volumes, it is necessary to store the data such that it is findable, accessible, interoperable, and reusable (FAIR). Although current materials data repositories, databases, and ontologies partly address FAIR principles, they do not adequately capture the critical metadata on contextual information (e.g., relationships between materials data and terms used by materials scientists such as process, structure, and property). This work develops the materials graph and materials graph ontology to standardize, collect, and organize the relational metadata that allows advanced queries that implicitly improve FAIR characteristics and utilization of the data. Case studies show the proposed materials graph ontology flexibly describes a broad variety of materials data, improves the findability and utility of the different graph-connected material concepts and data, and formalizes a materials data ingest framework that is amenable for extracting process-structure–property relationships.