Materials Informatics for Images and Multi-dimensional Datasets: On-Demand Oral Presentations
Sponsored by: ACerS Basic Science Division, ACerS Electronics Division
Program Organizers: Amanda Krause, Carnegie Mellon University; Alp Sehirlioglu, Case Western Reserve University; Daniel Ruscitto, General Electric
Friday 8:00 AM
October 22, 2021
Room: On-Demand Room 2
Location: MS&T On Demand
Invited
Training Deep-learning Models with 3D Microstructure Images to Predict Location-dependent Mechanical Properties in Additive Manufacturing: Ashley Spear1; Carl Herriott1; 1University of Utah
Three-dimensional images of additively manufactured (AM) microstructures were used to train deep-learning models to predict effective mechanical properties and their spatial variability throughout AM builds. Images were acquired from high-fidelity, multi-physics simulations of SS316L produced by directed energy deposition under different build conditions. Microstructural subvolumes and corresponding homogenized yield-strength values (~7700 data points) were then used to train convolutional neural network (CNN) models. For comparison, two types of machine-learning (ML) models (Ridge regression and XGBoost) were trained using the same dataset. The ML models required substantial pre-processing to extract volume-averaged microstructural descriptors; whereas, 3D image data comprising basic microstructural information were input to the CNN models. Among all models tested, CNN models that use crystal orientation as input provided the best predictions, required little pre-processing, and predicted spatial-property maps in a matter of seconds. Results demonstrate that suitably trained data-driven models can complement physics-driven modeling by massively expediting structure-property predictions.
Invited
Understanding Degradation and Failure Mechanisms by Multiscale and Multiresolution Electron Microscopy: Josh Kacher1; 1Georgia Institute of Technology
Mechanical deformation and failure processes such as fatigue crack formation and ductile fracture are inherently multiscale processes, ranging from nanoscale crack nucleation mechanisms to collective dislocation interactions ranging across hundreds of microns. Understanding these processes requires multiscale characterization approaches that reflect the nature of the processes. Advances in electron detector technology, including the advent of direct-electron detectors, and increases in computational processing capacity have transformed electron microscopy-based characterization into a big-data analytics tool capable of multimodal image acquisition and high-resolution property mapping. This includes the ability map out the three-dimensional elastic strain tensor, crystal rotations, and dislocation density at length scales ranging from nanometers to hundreds of microns. In this talk, I will discuss the work my group is doing in applying advanced multiscale and multiresolution electron-microscopy based characterization techniques to understand mechanical deformation and corrosion mechanisms in metals and alloys.
Invited
Graph Neural Networks for an Accurate and Interpretable Prediction of the Properties of Polycrystalline Materials: Minyi Dai1; Mehmet Demirel1; Yingyu Liang1; Jiamian Hu1; 1University of Wisconsin-Madison
Various machine learning models have been used to predict the properties of polycrystalline materials, but none of them directly consider the physical interactions among neighboring grains despite such microscopic interactions critically determining macroscopic material properties. Here, we develop a graph neural network (GNN) model for obtaining an embedding of polycrystalline microstructure which incorporates not only the physical features of individual grains but also their interactions. The embedding is then linked to the target property using a feed-forward neural network. Using the magnetostriction of polycrystalline Tb0.3Dy0.7Fe2 alloys as an example, we show that a single GNN model with fixed network architecture and hyperparameters allows for a low prediction error of ~10% over a group of remarkably different microstructures as well as quantifying the contribution of each grain in a microstructure to its magnetostriction. Such microstructure-graph-based GNN model therefore enables an accurate and interpretable prediction of the properties of polycrystalline materials.
Invited
Open-source Hyper-dimensional Materials Analytics Using Hyperspy: Joshua Taillon1; 1National Institute of Standards and Technology
With modern advances in computer technology, materials characterization techniques such as electron microscopy (EM) are generating vastly increasing amounts of digital experimental data, requiring novel processing strategies and providing challenges for data analysis. Prominent among these challenges is being able to easily and reproducibly develop these new strategies, due to the limitations of existing proprietary software solutions available in the EM community. The open source HyperSpy project address this issue by providing researchers with easy access to data in proprietary formats,reproducible analysis through scripting and "notebook computing", and access to an ever-growing collection of high-quality scientific data processing libraries in the scientific Python ecosystem, including state of the art machine learning strategies. This talk will introduce the HyperSpy project, demonstrate the capabilities of the software, and provide a number of published examples of how HyperSpy has been used for the processing of large multi-dimensional EM imaging and spectroscopy datasets.
Invited
Machine Learning Ferroelectrics: Bayesianity, Parsimony, and Causality: Sergei Kalinin1; 1Oak Ridge National Laboratory
Scanning Transmission Electron Microscopy and Piezoresponse Force Microscopy has opened a window into atomic and mesoscale functionalities of ferroelectric materials. However, this wealth of data necessitates development of pathways to extract the generative physics, either in the form of parameters of mesoscopic Ginzburg-Landau model, or corresponding atomistic descriptors. One such approach is based on the Bayesian methods that allow to take into consideration the prior knowledge the system and evaluate the changes in understanding of the behaviors given new data. The second pathway explores the parsimony of physical laws and aims to extract these from the set of real-world observations. Ultimately, we seek to answer the questions such as whether frozen atomic disorder drives the emergence of the local structural distortions or polarization field instability drives cation and oxygen vacancy segregation, and what is the driving forces controlling the emergence of unique functionalities of morphotropic materials and ferroelectric relaxors.
Multivariate Statistical Analysis (MVSA) for Hyperspectral Images: Chuong Nguyen1; Alp Manavbasi1; 1Novelis
Modern materials characterization techniques generate huge amount of data, often in the form of hyperspectral images. Traditional analyses break down these data sets into static spectra or images, and manually correlate them for information. With the advances of computing power, chemometrics, specifically MVSA for hyperspectral images, is increasingly used to automatically extract information using mathematical and statistical algorithms.This paper presents MVSA of data obtained by Auger electron spectroscopy (AES) and secondary ion mass spectrometry (SIMS) concerning surface treatment of aluminum. The data can be analyzed autonomously, or with human inputs. Ultimately, the analyses revealed important features of high performing surfaces for future applications.