Ceramics and Glasses Modeling by Simulations and Machine Learning: On-Demand Oral Presentations
Sponsored by: ACerS Glass & Optical Materials Division
Program Organizers: Mathieu Bauchy, University of California, Los Angeles; Peter Kroll, University of Texas at Arlington; N. M. Anoop Krishnan, Indian Institute of Technology Delhi
Friday 8:00 AM
October 22, 2021
Room: On-Demand Room 4
Location: MS&T On Demand
Invited
Looking for Order in Disorder: Topological Data Analysis of Glass Structure: Morten Smedskjaer1; Søren Sørensen1; Christophe Biscio1; Lisbeth Fajstrup1; Mathieu Bauchy2; 1Aalborg University; 2University of California, Los Angeles
An important question to unravel within materials science is the interplay between structure and properties in glass materials. To understand this link, there has been a great interest in pinpointing structural features that correlate strongly with the properties. However, identifying such structural descriptors especially at the medium-range length scale remains a challenging task. In this talk, we present our work on using topological data analysis to reveal hidden medium-range order (MRO) in oxide and hybrid glasses. Specifically, we apply persistent homology, a type of topological data analysis, to categorize and understand MRO structure in these systems, for which the atomic configurations have been generated by molecular dynamics simulations. By using persistent homology to study the size of certain algebraic topological features, we observe similarities to the length scales associated with the well-known first sharp diffraction peak in the studied glasses.
Invited
Graph ODE for Learning Dynamic Systems: Yizhou Sun1; Wei Wang1; Zijie Huang1; 1UCLA
Many real-world systems, such as moving planets, can be considered as multi agent dynamic systems, where objects interact with each other and co-evolve along with the time. Such dynamics is usually difficult to capture, and understanding and predicting the dynamics based on observed trajectories of objects become a critical research problem in many domains. Most existing algorithms assume the observations are regularly sampled and all the objects can be fully observed at each sampling time, which is impractical for many applications. To tackle the above challenge, we present LG-ODE, a latent ordinary differential equation generative model for modeling multi-agent dynamic system with known graph structure, which is the first Graph ODE model in this direction. It can simultaneously learn the embedding of high dimensional trajectories and infer continuous latent system dynamics. Experiments on motion capture, spring system, and charged particle datasets demonstrate the effectiveness of our approach.
Machine Learning as a Tool to Accelerate the Design of Nuclear Waste Glasses with Enhanced Sulfur Loadings: Taihao Han1; Xinyi Xu2; Jie Huang1; Albert A. Kruger3; Aditya Kumar1; Ashutosh Goel2; 1Missouri University of Science and Technology; 2Rutgers, The State University of New Jersey; 3U.S. Department of Energy, Office of River Protection
The nuclear waste with a high concentration of alkali/alkaline-earth sulfates is vitrificated with the direct feed approach. It is difficult for the existing empirical models to predict sulfate solubility in these glasses or design glass formulations with enhanced sulfate loadings, especially for HLW glasses whose composition falls outside of the range encompassed by the database used to develop/calibrate the models. This study harnesses the power of artificial intelligence with a goal to address the limitations of the existing models. Random Forests model is trained using a large database; comprising >1000 waste glasses and encompassing a wide range of glass compositions and processing variables. Next, the RF model is used to quantitatively assess the influence of glasses’ compositional/processing variables on the SO3 solubility loading. Finally, on the premise of such understanding of influential variables, two closed-form analytical models –one highly-parametrized and one with fewer input variables – are developed.
Decomposing the Strength of Hydrated Cement Compositions by Machine Learning: Yu Song1; Gaurav Sant1; Mathieu Bauchy1; 1University of California, Los Angeles
Reducing the carbon footprint in cement production is a pressing challenge faced by the construction industry. To curb the massive environmental impact, it is pertinent to improve material performance and reduce the carbon embodiment of cement. This requires an in-depth understanding of how the strength of hydrated cement is controlled by the chemical composition of cement. Although this problem has been investigated for more than one hundred years, our current knowledge is still deficient for a clear decomposition of this complex composition-strength relationship. Here, we approach this problem using Gaussian process regression (GPR). Among all machine learning methods applied to the same dataset, our GPR model achieves the highest accuracy of predicting the strength of hydrated cement composite based on the chemical compounds. Based on the optimized GPR model, we are able to decompose the influence of each oxide on strength to an unprecedented level.
Toward Revealing Full Atomic Picture of Nanoindentation Deformation Mechanisms in Li2O-2SiO2 Glass-ceramics: Binghui Deng1; 1Corning Inc
Obtaining full understanding of deformation mechanisms in Li2O-2SiO2 glass-ceramics subjected to sharp contact loading is a pressing need for many industrial applications. We herein conduct systematic molecular dynamics simulations to reveal atomic details that are otherwise extremely challenging to probe experimentally. Our study shows that glass-ceramics exhibit a dramatically different plastic deformation map compared to glass, where the nanocrystalline phase dictates the plastic zone shape. Shear flow preferentially nucleates and travels along the interfaces before moving to the surrounding glass and nanocrystals. Dislocations, amorphization zones around the indenter, and shear flow at glass and crystal interfaces help to dissipate contact energy and deconcentrate the deformation, thereby increasing fracture toughness and discouraging the development of catastrophic cracks.
Modeling Polaron Hopping in Ternary Spinel Oxides: Maytal Caspary Toroker1; 1Technion - Israel Institute of Technology
The small-polaron hopping model has been used for several decades for modeling electronic charge transport in oxides. Despite its significance, the model was developed for binary oxides, and its accuracy has not been rigorously tested for higher-order oxides. To investigate this issue, we chose the MnxFe3-xO4 spinel system, which has exciting electrochemical and catalytic properties, and mixed cation oxidation states that enable us to examine the mechanisms of small-polaron transport. Using a combination of experimental results and DFT+U calculations, we find that the charge transport occurs only between like-cations (Fe/Fe or Mn/Mn). Reference: A. Bhargava, R. Eppstein, J. Sun, M. A. Smeaton, H. Paik, L. F. Kourkoutis, D. G. Scholm, M. Caspary Toroker*, R. D. Robinson*, Adv. Mat., 2004490 (2020).
Impact of Irradiation on the Properties of Gel Layer Formed After Aqueous Corrosion of Borosilicate Glasses: Amreen Jan1; N.M Anoop Krishnan1; 1Indian Institute of Technology Delhi
Confinement of radionuclides in borosilicate glass matrix is subject to the durability of vitrified nuclear glass in the aqueous medium on the geological time scales under repository conditions. It is now well accepted that under aqueous condition there is a formation of a gel layer. However, the properties of this gel layer are not well understood yet. In this work using atomistic modelling, a series of borosilicate glasses- pristine and irradiated- are prepared and further, gels are prepared by replacing boron and sodium by hydrogen. These gels are then aged at different temperatures - 500K, 1000K, 1500K and 2000K. It is seen that, indeed, there is a difference in the properties gel (connectivity, short and medium range order) formed from pristine and irradiated glass. For example, the mean square displacement of hydrogen is orders of magnitude higher in the gel formed from irradiated glass as compared to pristine glass.
Elucidating Compositional Governance of Optical Properties of Oxide Glasses Using Interpretable Machine Learning: Mohd Zaki1; Vineeth Venugopal1; Ravinder Bhattoo1; Suresh Bishnoi1; Sourabh Kumar Singh1; Amarnath R. Allu2; Jayadeva1; N. M. Anoop Krishnan1; 1Indian Institute of Technology Delhi; 2Glass Division, CSIR-Central Glass and Ceramic Research Institute, Kolkata
Glasses are wide used ranging from display devices to scientific instruments like microscope and telescopes owing to their excellent optical properties. The optical properties of Abbe number (Vd) and refractive index at 587.6 nm (nd) control how the light behaves when it passes through the glass. To eliminate guesswork in experimentally preparing glasses and accelerate the discovery of glass compositions with desirable Vd and nd, we present machine learning (ML) approach to predict optical properties of inorganic glasses with chemical oxides as input. Due to black-box nature of ML models, it is highly impossible to decipher the composition-properties relationship. Here, we have used Shapely Additive exPlanations (SHAP) to elucidate the ML model predictions and decode the influence of chemical oxides on the predicted values of Vd and nd. Hence, this work presents a roadmap for researchers to decode the compositional control of different properties of inorganic glasses.