Algorithm Development in Materials Science and Engineering: Poster Session
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 5:30 PM
February 25, 2020
Room: Sails Pavilion
Location: San Diego Convention Ctr
Session Chair: Mohsen Asle Zaeem, Colorado School of Mines
L-1 (Digital): Machine Learning and Computer Vision on Classification of Carbon Nanotube and Nanofiber Structures for TEM Dataset: Qixiang Luo1; Elizabeth Holm1; 1Carnegie Mellon University
We introduce a convolutional neural networks (CNNs) based machine learning and computer vision method to classify CNTs/CNFs and other airborne particles from a TEM image dataset. The model starts with a transfer learning pipeline achieved by hypercolumn extractions on CNN fully-connected layers from both precise edges at shallow and abstract semantics at deep, followed by K-means visual dictionary for creating feature clusters. It then uses VLAD encoder to mitigate the ambiguity between feature clusters and local feature descriptors based on image retrieval, with boosting step for classifiers. The unsupervised t-SNE method was used to visualize classification performance for data cluster analysis. The method achieves 85% accuracy on full dataset and over 90% on pure nano-carbon dataset for 5-fold cross-validation. We also discuss image pre-processing with data augmentation techniques for imbalanced dataset and hierarchical recognition, to utilize a tree-like learning structure to explore attributes from generalized to specific.
L-2 (Invited): Multi-Scale Modelling and Defect Engineering in Boron Carbon-Nitride van der Waals Heterostructures: Siby Thomas1; Mohsen Asle Zaeem1; 1Colorado School of Mines
Two-dimensional (2D) boron-carbon-nitrogen (BCN) van der Waals (vdW) heterostructures are expected to possess interesting properties complementary to those of insulating hexagonal boron nitride and zero bandgap graphene. We integrate electronic structure calculations with molecular dynamics (MD) simulations to investigate the effects of vacancy and Stone-Wales (SW) defects on the electronic and mechanical properties of a newly designed BCN-vdW heterostructure. The results show that B, C and N mono vacancies make the BCN a conductive material whereas its pristine form is a direct bandgap semiconductor. SW defects in BCN retain a high elastic modulus comparable to its pristine form, whereas the mono vacancies show a considerable decrease in the elastic moduli. By applying uniaxial tensile loading in MD simulations, we found that the tensile strength decreases in presence of any type defects, however a higher elastic modulus is observed when SW defects are present.
L-3: An Improved Collocation Method to Treat Traction-free Surfaces in Dislocation Dynamics Simulations: Abu Siddique1; Tariq Khraishi1; 1University of New Mexico
Dislocation dynamics simulations is an inherently multi-scale computational methodology in materials deformation modeling. The authors address an important topic in such modeling which is the treatment of boundary conditions on the computational domain. Specifically, the effect of traction-free surfaces on the plasticity, i.e. the motion of dislocations and ensuing plastic flow, is treated here. To solve this numerical problem, the surface in question is meshed with elements each representing a dislocation loop. The boundary condition is enforced by solving a system of equations at each time step for the Burgers vectors of such loops. This is a collocation method with collocation points on the surface, and therefore the higher the areal density of the points, the better the numerical outcome. Modeling results have been verified and are presented herein.
L-4: Classifying Atomic Environments by the Gromov-Wasserstein Distance: Sakura Kawano1; Jeremy Mason1; 1University of California, Davis
Interpreting molecular dynamics simulations usually involves automated classification of local atomic environments to identify regions of interest. Existing approaches are generally limited to a small number of reference structures and only include limited information about the local chemical composition. This work proposes to use a variant of the Gromov-Wasserstein (GW) distance to quantify the difference between a local atomic environment and a set of arbitrary reference environments in a way that is sensitive to atomic displacements, missing atoms, and differences in chemical composition. This involves describing a local atomic environment as a finite metric measure space, with the additional advantages of not requiring the local environment to be centered on an atom and of not making any assumptions about the material class. Numerical examples illustrate the efficiency and versatility of the algorithm.
L-5: Coupled Light Capture and Lattice Boltzmann Model of TiO2 Micropillars Array for Water Purification: Pegah Mirabedini1; Agnieszka Truskowska2; Duncan Ashby1; Masaru P. Rao1; P. Alex Greaney1; 1University of California, Riverside; 2Rensselaer Polytechnic Institute
TiO2 has been widely studied as a photocatalytic material due to its non-toxicity, chemical inertness, and high photocatalytic activity. Microfluidic reactors based on TiO2 micropillar arrays are being developed to use solar radiation to treat recycled wastewater in long-duration space missions. While the photochemistry of TiO2 is understood, there is still considerable scope for optimizing photocatalytic reactor design efficiency, particularly for use in space missions which presents unique design limitations requiring a low-power, lightweight, and low-volume power matrix. We present a theoretical model to investigate the effect of the micropillar’s geometry on their photocatalytic activity and water purification efficiency. A Monte Carlo ray tracing approach was developed to model light absorption, and the lattice Boltzmann method was used to simulate water flow. The results from both of these are used in a diffusion-advection model of contaminant decomposition.
L-6: Investigation of Fe-O and Fe-N and H-O Bond Formation Process by the Molecular Dynamics Simulations: Jianxin Zhu1; Guannan Guo1; Jian-Ping Wang1; 1University of Minnesota
The nitrogen atoms behaviors in the preparation of the Fe-N permanent magnetic materials have been discussed by the means of molecular dynamics simulations with ReaxFF (surface functionalization) and EAM (metal atom interactions) force field The forming and breaking of the Fe-O bond structures during the oxidation and reduction to prepare the porous Fe precursor and the forming of Fe-N bond structure during the nitriding under pure ammonia are given in the report. The minimum lattice energy of BCC-iron in low temperature nitriding (<150°C) and FCC-iron in high temperature nitriding (>600°C) has been obtained through the EAM force field simulation. Bond energies for Fe-O, Fe-N and H-O bonds are studied through available ReaxFF potentials.
L-7: Machine Learning Driven Functionally Graded Material Designs for Mitigation of Thermally Induced Stress: Zhizhou Zhang1; Zeqing Jin1; Kahraman Demir1; Grace Gu1; 1Department of Mechanical Engineering, University of California, Berkeley
High operating temperatures of engineering components can generate stress concentration at sharp corners and reduce fatigue life. Functionally graded materials can redistribute stresses at critical locations and achieve better mitigation effects compared to commonly used corner fillets. However, simulating all the possible material pattern combinations would be intractable. In this work, to extrapolate optimal patterns around the critical areas of interest, we propose using convolutional neural networks (CNN) where the input is a matrix containing material distribution and property information and the output is the stress profile. Training data are generated through finite element analysis simulations and a genetic algorithm is then applied to search for the top designs. Results show that annulus material patterns at a sharp corner can reduce the peak stress 40% more than just using a notch design. The implication of this work is the potential to extend service life of engineering components in extreme environments.
L-8: Methods to Simulate Grain Boundary Diffusion in Bicrystals and Polycrystals: David Page1; Katie Varela1; Oliver Johnson1; David Fullwood1; Eric Homer1; 1Brigham Young University
Grain boundaries have diffusive properties that can vary significantly from the bulk and from other grain boundaries. We detail methods to calculate unique diffusivity values for both bulk and grain boundary regions in bicrystal molecular dynamics simulations. However, simulating grain boundary diffusivity one boundary at a time is a slow process. As a result, we also present methods that extend the bicrystal techniques to three dimensions where one can calculate both bicrystal and grain boundary diffusivity values in full polycrystalline simulations and obtain unique diffusivity values for each grain boundary. We report methods for validation of the diffusivity measurements in each region and compare the results with a published model for grain boundary dependent diffusion.
L-9: Numerical Simulation for Microstructural Evolution in Solidification Process using CFD-CA (Cellular Automata) Coupled Method: Wonjoo Lee1; Yuhyeong Jeong1; Chanhee Won1; Jae-Wook Lee2; Howon Lee2; Seong-hoon Kang2; Jonghun Yoon1; 1Hanyang University; 2Korea Institute of Materials Science
In the current study, a two-dimensional (2D) CFD-CA model is considered to simulate the microstructural evolution of alloys in fluid flow during directional solidification process. Based on the local equilibrium theory and algorithms of Lattice Boltzmann Method (LBM) and Cellular Automata (CA), the numerical simulation for microstructural evolution is constructed. The LBM is adopted to numerically solve the fluid flow and solute transport by convection and diffusion. Then, the CA determines the dendrites morphology with the composition field previously calculated by LBM. The results demonstrate asymmetrical dendritic growth influenced by convection and dendrite arm spacing related with the mechanical properties of solidified materials. Also, the results were compared with analytical solution and experimental results of microstructure observation.
L-10: PyMob: Software for Automated Assessment of Atomic Mobilities: Katrin Abrahams1; Irina Roslyakova1; Ingo Steinbach1; 1Ruhr University Bochum
Atomic mobilities are essential requisites to perform realistic diffusion simulations. Their composition and temperature dependence can be stored in CALPHAD type databases. Until now assessments of atomic mobilities for new systems or reassessments of already existing systems are done manually and therefore they are time consuming and subject to the “human factor”. In this talk a new software approach to automate atomic mobility assessments will be presented. One key aspect is the automated weighting of input data using outlier detection where the data are weighted against each other without further information about the data, e.g. experimental methods. Furthermore, automatic parameter selection is used for higher order systems and uncertainty propagation to determine the uncertainty of the assessed data.
L-11: Randomness at Scale: Properties of Bulk Nanostructured Materials from Stochastic Representative Volume Elements: Skylar Mays1; Katherine Moody1; Mujan Seif1; John Balk1; Matthew Beck1; 1University of Kentucky
Predicting and/or optimizing properties of bulk nanostructured materials is challenging. Accounting for effects of disorder and randomness is particularly difficult, and of particular current interest in the context of nanoporous materials, fibrous ablatives, and sintered metallic powders. Scaling local properties to aggregate, bulk effective properties is a critical capability for both the design and application of such materials. Leveraging a recently implemented tool for stochastically generating large numbers of simulated RVEs with physically-motivated structural motifs (including network solids, packed powders, and matted or textured fibers), we are able to compute statistically valid distributions of local properties and to probe relationships between local properties and geometrical features. Here we extend previous work and report new results of efforts to homogenize distributions of local properties into bulk effective properties.
L-12: Simulation of Compressive Stress-strain Curve for Additive Manufactured Ti6Al4V Cuboctahedron Cellular Structure: Annadurai Dhinakar1; Chen Jhewn-Kuang1; 1Institute of Material Science and Engineering, National Taipei University of Technology
In this study, compression test done for Ti-6Al-4V cuboctahedron cellular sample. In the experiment, by varying sample porosity number for Selective Laser Melted structure we ended up with brittle and non-linear response. Simulation is performed using Bi-Linear material model with ANSYS workbench. Different simulations are modeled to include non-linear and plastic effects in the results. Results of compression simulation with strain rate of 2.08E-03 /sec and 50% strain are extracted. The usual way of computing stress from force reaction and strain from total deformation does not fit with experiment exactly. To study the equivalent Stress/Strain directly, the numbers of solid element distributed at each Stress/Strain levels is listed and the solid density of the structure is used to transform the Equivalent Stress/Strain for finite element into Stress/Strain for cellular structure. Experimental Stress-Strain curve is compared with all the Stress-Strain curves derived from the various models described in Static Structural module.
L-13: Three-dimensional Modeling of Growth and Motion of Dendrites under Thermosolutal Convection: Seyed Amin Nabavizadeh1; Mohsen Eshraghi2; Sergio Felicelli1; 1University of Akron; 2Department of Mechanical Engineering, California State University
A lattice Boltzmann-phase field model was developed to study columnar dendritic growth under solute-driven convection. The effects of convection on the solidification microstructure and the motion of detached dendrite arms were investigated. The lattice Boltzmann method was used to simulate the translational and rotational motion of the dendrites. The results can be used to predict spurious grain formation during directional solidification of alloys.
L-14: Uncertainty Propagation in CalPhaD Calculations: Nicholas Ury1; Richard Otis2; Vilupanur Ravi1; 1California State Polytechnic University, Pomona; 2Jet Propulsion Laboratory
Calculation of Phase Diagrams (CalPhad) has proven to be a useful computational approach in reducing the time and resources required for the development of alloys. The next step to advancing this methodology is to account for the inconsistency or lack of data when assessing thermodynamic systems. As proof of concept, a generalized framework made in Python was implemented to perform equilibrium calculations and handle model uncertainty. This framework includes the storage and handling of elemental and phase data, and the implementation of Newton’s method to perform Gibbs free energy minimization to create step and phase diagrams and determine the confidence intervals of the results. The Mg-Si system was assessed using a Bayesian approach as a case study for this framework.