Characterization: Structural Descriptors, Data-Intensive Techniques, and Uncertainty Quantification: Diffraction, Microscopy & Machine Learning
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS Extraction and Processing Division, TMS: Computational Materials Science and Engineering Committee, TMS: Materials Characterization Committee
Program Organizers: Shawn Coleman, DEVCOM Army Research Laboratory; Tomoko Sano, U.S. Army Research Laboratory; James Hogan, University of Alberta; Srikanth Patala, SASA Institute; Oliver Johnson, Brigham Young University; Francesca Tavazza, National Institute of Standards and Technology

Wednesday 8:30 AM
February 26, 2020
Room: Theater A-3
Location: San Diego Convention Ctr

Session Chair: Shawn Coleman, CCDC Army Research Laboratory; Tomoko Sano, CCDC Army Research Laboratory


8:30 AM Introductory Comments

8:35 AM  
100 Years of Scherrer Modifications: Demystifying Diffractogram Width Analyses for Nanocrystalline Materials: Cody Kunka1; Brad Boyce1; Stephen Foiles1; Remi Dingreville1; 1Sandia National Laboratories
    Since the landmark development of the Scherrer Method 100 years ago, many specialized versions of diffractogram width analysis have originated to nondestructively characterize large volumes of unique nano materials. To facilitate rational applications of width analyses, we present a unified theory and generalized nomenclature using atomistic modeling. We demonstrate the width-method selection, in terms of width type, broadening sources, and functional forms, can affect not only the magnitudes but also the trends of the crystallographic predictions. The present methodology can be applied to various nanocrystalline systems and loading conditions to guide the associated experimental investigations.

8:55 AM  
Machine Learning Approach for On-the-fly Crystal System Classification from Powder X-ray Diffraction Pattern: Yuta Suzuki1; Hideitsu Hino2; Takafumi Hawai1; Kotaro Saito3; Kanta Ono1; 1High Energy Accelerator Research Organization; 2The Institute of Statistical Mathematics; 3Paul Scherrer Institute
    The Crystal system and space group determination in the initial stage of crystal structure analysis is one of the time-consuming processes in materials research. We demonstrate an automated method to predict crystal systems and space groups from X-ray diffraction (XRD) patterns using a machine learning (ML) technique. The XRD dataset was calculated from crystal structures in ICSD (Inorganic Crystal Structure Database) with pymatgen middleware. Our tree-based ML model marks over 90% accuracy for crystal system prediction and 88% for space group prediction with five candidates. We applied this method to an actual XRD experiment for VO2 and confirmed that our method works for actual experimental data. By using an interpretable machine-learning approach, we also succeeded in quantifying empirical knowledge of experts. Our result shows the possibility of the data-driven discovery of unrecognized characteristics embedded in experimental data and will contribute to the realization of on-the-fly data analysis.

9:15 AM  
Indexing of Electron Back-Scatter Diffraction Patterns Using a Convolutional Neural Network: Zihao Ding1; Elena Pascal1; Marc De Graef1; 1Carnegie Mellon University
    We propose a new convolution neural network (EBSD-CNN) with residual block and separable convolution to realize high accuracy and near real-time indexing of EBSD patterns. The integrated output of unit quaternions and a disorientation loss function are implemented to adapt the neural net for crystallographic orientation indexing. In addition to validating on simulated EBSD patterns, data from a series of experiments on Nickel with various exposure time have also been tested to study the network's robustness against pattern noise. The results suggest that a CNN can provide an alternative indexing method to the commercial Hough-transform-based indexing with comparable accuracy and indexing rate. To gain insight into the model, we provide for a visualization of the filters as well as intermediate output in the network. As more features are extracted during the process, the approach also shows potential to measure other material properties that are encoded inside the EBSD patterns.

9:35 AM  Invited
Parametric Models for Crystallographic Texture: Estimation and Uncertainty Quantification: Stephen Niezgoda1; James Matuk1; Oksana Chkrebtii1; 1The Ohio State University
    An Orientation Distribution Function (ODF) describes the orientation of crystals in a polycrystalline material. While non-parametric kernel density estimation provides a quick method for estimating an ODF, there is limited interpretability. In this talk, we instead propose that an ODF takes a parametric form as a mixture of symmetric Bingham distributions. We treat the number of components, the mixture weights, and the scale and location parameters that determine the symmetric Bingham distribution as random variables through the Bayesian paradigm. Posterior distribution inference of the parameters through Bayesian methodology allows for interpretability and structured uncertainty quantification of the parameters of interest. We will additionally discuss the associated reversible jump Markov chain Monte Carlo algorithm which allows one to sample from the target posterior distribution. We will conclude with analyses of various data sets with interpretations of parameters of interest and a comparison with kernel density estimation methods.

10:05 AM Break

10:25 AM  Invited
Large-scale Defect Contrast Simulations for Scanning and Transmission Electron Microscopy: Marc De Graef1; 1Carnegie Mellon University
    Transmission electron microscopy (TEM) has for several decades been the tool of choice for the study of lattice defects such as dislocations and stacking faults. In recent years, electron channeling contrast imaging in the scanning EM (SEM) has provided access to surface penetrating defects. Defect image contrast simulations, on the other hand, have not kept up with the experimental capabilities in the sense that they have only been possible for simple defect geometries, e.g., straight dislocations in an infinite sample. In this presentation, we will review recent advances in the incorporation of defect displacement fields generated by molecular dynamics simulations, phase field simulations, and discrete dislocation dynamics simulations into the electron dynamical scattering formalism. We will review briefly the theory of image formation for TEM and SEM, and provide examples of image simulations for a number of realistic defect configurations generated by these simulation approaches.

10:55 AM  
Machine Learning and Electron Backscatter Diffraction: Alessandro Previero1; Guillaume de Certaines1; Alex Foden1; Thomas Britton1; 1Imperial College London
    Electron backscatter diffraction (EBSD) is a data-rich, analysis technique. Each diffraction pattern contains information about the crystal phase and orientation of the material within the electron beam interaction volume. The advent of high-quality, dynamical-based electron diffraction simulations enables us to explore and test new analysis approaches. We apply (un)supervised machine learning methods, focusing on convolutional neural networks, transfer learning, and principal component analysis to augment the capabilities of the technique. Inspection of the trained networks also provides direction for the next generation of deterministic analysis procedures that we will use to classify phases and crystal orientations, especially for challenging cases such as differentiation of FCC-Cu patterns from austenite (also FCC). We will explore the benefits and disadvantages of these approaches considering current state-of-the-art diffraction pattern analysis methods.

11:15 AM  Invited
A New Crystallographic Defect Quantification Workflow via Advanced-microscopy-based Deep Learning: Yuanyuan Zhu1; Graham Roberts1; Rajat Sainju1; Colin Ophus2; Brian Hutchinson3; Danny Edwards4; Mychailo Toloczko4; 1University of Connecticut; 2Lawrence Berkeley National Laboratory; 3Western Washington University; 4Pacific Northwest National Laboratory
    Crystallographic defects, in particular dislocations, voids and precipitates are critical to the bulk physical and mechanical properties of metals and alloys. Transmission electron microscopy (TEM) is a standard tool for defect characterization at the nanometer scale, however, quantitative analysis of enough TEM images to provide a statistically satisfactory representation of these defects over a range of synthesis and deformation conditions can often be a time-consuming daunting task. In this talk, we propose a new workflow of high-throughput defects analysis demonstrated using a HT-9 martensitic steels. One unique advancement of our approach involves systematic improvement in all three key steps, including the development of a high-clarity defect imaging technique free of bend contour, a novel convolutional neural network (CNN) model-the MetalDefectNet-for defect identification, and a dedicated MATLAB algorithm for defect quantification. We envision that this new workflow can provide a more reliable and statistically meaningful experimental foundation for metallurgy.