Algorithm Development in Materials Science and Engineering: Machine Learning and Atomistic Algorithms
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS: Integrated Computational Materials Engineering Committee, TMS: Phase Transformations Committee, TMS: Computational Materials Science and Engineering Committee
Program Organizers: Mohsen Asle Zaeem, Colorado School of Mines; Garritt Tucker, Colorado School of Mines; Charudatta Phatak, Argonne National Laboratory; Bryan Wong, University of California, Riverside; Mikhail Mendelev, NASA ARC; Bryce Meredig, Travertine Labs LLC; Ebrahim Asadi, University of Memphis; Francesca Tavazza, National Institute of Standards and Technology
Tuesday 2:00 PM
February 25, 2020
Room: 31C
Location: San Diego Convention Ctr
Session Chair: Garritt Tucker, Colorado School of Mines; Bryce Meredig, Travertine Labs LLC
2:00 PM Invited
Bridging the Electronic, Atomistic and Mesoscopic Scales using Machine Learning: Subramanian Sankaranarayanan1; 1Argonne National Laboratory
Molecular dynamics (MD) is a popular technique that has led to breakthrough advances in diverse fields, including tribology, energy storage, catalysis, sensing. The popularity of MD is driven by its applicability at disparate length/time-scales. Nevertheless, a substantial gap persists between AIMD, which is highly accurate but restricted to extremely small sizes, and those based on classical force fields (atomistic and CG) with limited accuracy but access to larger length/time scales. In this talk, I will present some of our recent work on the use of machine learning (ML) to seamlessly bridge the electronic, atomistic and mesoscopic scales for materials modeling. Our ML approach showed marked success in developing force fields for a wide range of materials from metals, oxides, nitrides, hetero-interfaces to two-dimensional (2-D) materials and even water (arguably the most difficult system to capture from a molecular perspective).
2:30 PM
Designing High-strength Carbon-nanotube Polymer Composites using Machine Learning Algorithms Integrated with Molecular Dynamics Simulations: Aowabin Rahman1; Prathamesh Deshpande2; Matthew Radue2; Michael Czabaj1; S Gowtham2; Susanta Ghosh2; Gregory Odegard2; Ashley Spear1; 1University of Utah; 2Michigan Technological University
Carbon-nanotube (CNT)-based composites have great potential in modern aerospace applications requiring high-strength, lightweight structural materials. However, one factor that limits the potential of CNT composites is the inefficiency in load transfer between CNTs through a polymeric resin, attributed to low CNT/polymer interfacial strength. We present a modeling framework that uses a machine-learning (ML) algorithm in conjunction with molecular-dynamics (MD) simulations to make design modifications at the CNT/polymer interface for improving the strength of CNT composites. The proposed framework uses a modular approach consisting of: (i) a methodology to insert reactive groups at the CNT-polymer interface and perform MD simulations of CNT pullout, (ii) an ML model to map effects of functionalization on MD model configuration (process to structure), and (iii) an ML model to predict pullout force from the functionalized CNT-composite model output (structure to property). The framework thus enables fundamental exploration of design space to develop high-strength CNT composites.
2:50 PM
Monte Carlo Study of Paired-spin Kagome Artificial Spin Ice Lattices: David Friedman1; Frank Barrows2; Yue Li2; Charudatta Phatak2; 1University of Illinois, Urbana-Champaign; 2Argonne National Laboratory
Artificial spin ices (ASI) are lattices comprising of magnetic islands that interact through dipolar coupling and approximate the behavior of naturally occurring bulk spin ice materials at room temperature. They exhibit novel excitations such as magnetic monopole excitation as a result of magnetic frustration emerging due to either geometric or configurational frustration. ASIs are also particularly interesting as they may have applications in computer memory and magnetic logic. In this work, we have developed a novel ASI based on a Kagome lattice with each motif comprising of paired magnetic islands. We have studied their energy landscape using standard Markov chain Monte Carlo simulations with parallel tempering. During the simulations, we tracked various thermodynamic parameters such as specific heat, and magnetic susceptibility of the lattices were able to identify three phase transitions. We will discuss these results along with lack of observation of any long-range order in our simulations.
3:10 PM
Functional Uncertainty Propagation with Bayesian Ensembles in Molecular Dynamics: Saaketh Desai1; Sam Reeve2; Alejandro Strachan1; 1Purdue University; 2Lawrence Livermore National Laboratory
Functional uncertainty quantification (FunUQ) is a method to quantify uncertainties in materials models stemming from input functions, as opposed to input parameters. FunUQ has been previously used within multi-fidelity molecular dynamics (MD) simulations, correcting quantities of interest (QoIs) originating from one interatomic model to yield the QoIs for a new functional form without additional simulation. We now use FunUQ to propagate uncertainties originating from interatomic potentials in MD. Starting with an ensemble of potentials obtained via Bayesian calibration, we run a small number of direct MD simulations across the distribution, calculate sensitivity to local changes in the function via FunUQ (the functional derivative), and predict the QoIs for each parameter set in the ensemble. This is the first computationally tractable MD simulation with quantified uncertainties directly from the interatomic model, the main source of uncertainty. All FunUQ codes and examples are available for online simulation in nanoHUB: https://nanohub.org/tools/funuq.This work was performed in part under the auspices of the U.S. DOE by LLNL under contract DE-AC52-07NA27344.
3:30 PM
Nudged Elastic Band Method for Solid-solid Transition Under Finite Deformation: Wei Gao1; 1University of Texas at San Antonio
Solid-state nudged elastic band (SSNEB) methods can be used for finding solid-solid transition paths when solids are subjected to external stress fields. However, previous SSNEB methods are not appropriate for studying transitions accompanied with finite or large deformation under external stress, due to an inaccurate evaluation in enthalpies and barriers. In this talk, a finite deformation nudged elastic band (FD-NEB) method will be presented for finding transition paths of solids under finite deformation.
3:50 PM Break
4:05 PM
Applying Machine Learning to Identifying Packing Defects in Amorphous Materials: Tina Mirzaei1; P.Alex Greaney1; 1University of California, Riverside
We present machine learning algorithms for identifying the “defects” in amorphous materials — structures with local properties that are outliers in the property distribution and so represent sites of weakness. Several approaches are presented for quantifying local structure, and these are used in combination with principal component analysis and cluster analysis to identify defect types.
4:25 PM
Reduced-order Atomistic Method for Simulating Radiation Effects in Metals: Elton Chen1; Chaitanya Deo2; Remi Dingreville1; 1Sandia National Laboratories; 2Georgia Institute of Technology
Atomistic modeling of radiation damage through displacement cascades is deceptively non-trivial. We will discuss a reduced-order atomistic cascade model capable of predicting and replicating radiation events in metals across a wide range of recoil energy. Our methodology approximates cascade and displacement damage production by modeling the cascade as a core-shell atomic structure composed of two damage production estimators, namely an athermal recombination corrected displacements per atom in the shell and an atomic mixing in the core. These estimators are calibrated from explicit PKA simulations and a standard displacement damage model that incorporates cascade defect production efficiency and mixing effects. We illustrate the applicability by providing examples for simulating high energy cascade fragmentation and large dose ion-bombardment.
4:45 PM
An Atomistic Framework to Understand Solute Grain Boundary Segregation in a Polycrystal: Malik Wagih1; Christopher Schuh1; 1Massachusetts Institute of Technology
Solute segregation at grain boundaries (GBs) is a potent approach to thermally stabilize nanocrystalline alloys. A basic requirement for this approach is the ability to correctly predict the magnitude of such segregation in a given alloy system. To date, the main thermodynamic models used to solve for the extent of segregation treat GBs as a “lump” region that only has one site-type available for segregation. This treatment represents a major simplification, and is in fact oversimplified for even the most ordered types of GBs; it takes into account neither the variety of types of GBs present in a polycrystal nor the variation of atomic sites within each individual GB. This talk will outline a thermodynamic and computational framework to determine the distribution of GB segregation energies in a polycrystalline binary alloy and highlight the significance of such distributions on the prediction of solute segregation at GBs.
5:05 PM
Quasiparticle Approach to Study Solute Segregation at Tilt grain Boundaries in Bcc Iron: Helena Zapolsky1; Antoine Vaugeois1; Renaud Patte1; 1Gpm, Umr 6634
Grain boundaries (GBs) are common defects in crystalline materials and especially play a major role in determining physical, mechanical, electrical, and chemical properties in nanomaterials. GB connectivity, which is strongly correlated with segregation of solute atoms at grain boundaries, and grain boundaries structure have a significant influence on these properties. In this paper, the quasiparticle approach [1] has been applied to study the evolution of grain boundaries structure with temperature as well as the segregation of interstitial solute atoms at GBs in bcc Fe. The simulation results are compared with atom probe data and HRTEM images. [1] Lavrskyi, M., et al., npj Computational Materials, 2, 15013