Accelerating Materials Science with Big Data and Machine Learning: Session II
Program Organizers: Huan Tran, Georgia Institute Of Technology; Muratahan Aykol, Toyota Research Institute
Tuesday 2:00 PM
October 19, 2021
Room: A123
Location: Greater Columbus Convention Center
Session Chair: Christopher Kuenneth, Georgia Institute of Technology
2:00 PM Invited
Data Science as Bridge – Materials Characterization and Modeling: Maria Chan1; 1Argonne National Laboratory
In materials and chemical science, the combination of high throughput computational modeling and experimentation has given rise to significant challenges and opportunities. Data science techniques such as machine learning, artificial intelligence, and computer vision have made a significant impact in the ease, scope, and speed of understanding of known materials and discovery of new ones. In this talk, we will discuss how we use data science approaches in conjunction with theory-based modeling to interpret experimental characterization data (such as x-ray scattering, spectroscopy, scanning probe microscopy, and electron microscopy) and carry out materials design (such as in the space of classic and hybrid perovskite optoelectronic materials). The role of computer vision and pattern recognition in the analysis of microscopy data will also be discussed. In addition, we will discuss computer vision and natural language processing approaches to gather and harness microscopy data from scientific literature.
2:40 PM
Learning Synthesis: Engineering Metal Nanoclusters for Specific Material Properties: Ryan McCarty1; 1University of California Irvine
Metal nanoclusters can have remarkable properties, including fluorescence, self-assembled 2D crystalline morphologies, anti-microbial activity, and useful magnetic properties. Nucleating and stabilizing these metal nanoclusters with desired properties has been challenging, but polymers, particularly short strands of DNA (6-30 nucleotides long) that serve no biological function, appear promising. These DNA sequences are highly tailorable, but establishing the connection between the sequence used, the resulting nanoclusters, and resulting properties remains particularly challenging, due in part to the complexity of short DNA sequences. This talk will present an approach using machine learning and combinatorial design to identify polymer structures and compositions which can be correlated to families of nanoclusters likely to yield valuable materials properties.
3:00 PM
Characterization of Microscopic Deformation of Materials Using Deep Learning Methods: Kavindu Wijesinghe1; Janith Wanni1; Natasha Banerjee1; Sean Banerjee1; Ajit Achuthan1; 1Clarkson University
Advanced experimental capabilities that enable detailed characterization of microscopic deformation of coupon specimens under simple loading conditions for studying the influence of specific microstructural features on mechanical properties is a growing need in the field of materials science. Building such capabilities require powerful data analysis methods to extract complex characteristics of microscopic deformation hidden in the raw image data. In this presentation, we report the development and demonstration of a data analysis framework using deep learning methods. The framework consists of a trained Mask R-CNN model combined with a regional instance segmentation algorithm for feature detection, an intersection over union based multi-object tracking algorithm to track segmented features as they deform, and kinematics models to extract the material characteristics of the deforming instances. For validation, we characterized the microscopic deformation of an additively manufactured 316L stainless steel coupon specimen under quasi-static tensile testing.
3:20 PM
A Data-driven Simulator for High-throughput Prediction of Electromigration-mediated Damage in Polycrystalline Interconnects: Peichen Wu1; William Farmer1; Kumar Ankit1; 1Arizona State University
Electromigration (EM) induced diffusional transport of metal atoms, which manifest as grain-boundary slits and voids in the metal line, often result in failure of an entire electronic component. Formulating preventive strategies and their efficient implementation involves the analysis of failure mechanisms in 4D microstructures via tedious in situ X-Ray tomography characterization as well as large-scale phase-field simulations, both of which are resource-intensive. Here, we present a data-driven simulation (DDS) technique, which for the first-time couples Artificial Neural Networks with microstructure modeling, to enable a high-throughput and an accurate prediction of defects’ evolution in progressively degrading interconnects. Our approach for validating the DDS-predicted EM failure rates by leveraging existing 4D datasets for a range of surface and grain boundary energies, crystal structure, grain texture, and electrical conductivity, as well as process parameters that include current density and temperature, will be discussed.