8th World Congress on Integrated Computational Materials Engineering (ICME 2025): Poster Session
Program Organizers: Victoria Miller, University of Florida; Stephen DeWitt, Oak Ridge National Laboratory
Tuesday 4:40 PM
June 17, 2025
Room: Platinum Ballroom 6
Location: Anaheim Marriott
Advancing Materials Discovery: A Novel Generative Model for Inverse Design of Vanadium Oxide Crystal Structures: Danial Ebrahimzadeh1; Sarah Sharif1; Yaser Banad1; 1University of Oklahoma
The discovery of high-performance materials is vital for addressing technical challenges in modern industries. Inverse design methodologies have revolutionized the search for tailored materials. We present a novel generative model combining an enhanced Generative Adversarial Network (GAN) and improved Variational Autoencoder (VAE) for invertible crystal structure encoding. Unlike traditional methods, our model excels in generating and optimizing crystal structures, validated in V-O compositions. Comparative analysis shows superior efficiency, achieving lower loss and better convergence than previously used GAN model. Our model discovered 151 stable and 107 metastable structures of 1394 DFT-evaluated structures within the stability and metastability range, derived from the Materials Project (MP) dataset. Additionally, our findings suggest updates to the stability range in the convex hull of MP. This work marks a significant advancement in inverse design, paving the way for transformative developments in materials science
Development of Cu-S Interatomic Potentials for Time- and Length-Scale Bridging: Thomas Hardin1; 1Sandia National Laboratories
Sulfidation of copper in the presence of sulfuric gases causes embrittlement and degradation of electrical properties. We report the development and validation of a "ladder" of copper-sulfur interatomic potentials representing a variety of speed/accuracy tradeoffs; we find that by additively decomposing the potential energy surface into effects captured by the Embedded Atom Method (EAM) and a machine learned correction term, risk of overfitting is reduced and overall accuracy is improved. Finally, we report on progress towards bridging time- and length-scales by extracting kinetic process rates from atomistic simulations based on these potentials. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525 (SAND2024-14710A).
Enhanced Thermal Conductivity for Modified hBN and Epoxy Composite Using Electric Field: Sidra Ajmal1; Abul Arif1; 1King Fahd University of Petroleum & Minerals
Hexagonal boron nitride (hBN) exhibits exceptional thermal and dielectric characteristics making it a suitable filler for enhancing thermal conductivity in high-temperature polymer matrix. In this study, hBN/epoxy composites were synthesised and their thermal conductivity was thoroughly examined. Furthermore, hBN particles were functionalizedusing plasma modification to improve the compatibility with the epoxy matrix. The composite fabrication involved compression molding, followed by drying in a vacuum oven at 70°C for 6 hours. An electric field was applied during the process to achieve directional alignment of the hBN particles, further enhancing thermal conductivity. The present work employs a representative volume element (RVE) approach and periodic boundary conditions (PBC) to accurately predict thermal conductivity (K), elastic modulus (E) and coefficient of thermal expansion (α) using 3D finite element (FE) simulations. The resulting composites exhibited significantly improved thermal properties, making them promising candidates for efficient heat dissipation in high-performance electronic devices.
Leveraging Advances in Additive Manufacturing Thermal Models to Predict Behavior During Laser Sheet Metal Forming: Benjamin Begley1; Zoe Lipton1; Daniel Bolden1; Tianchen Wei1; Nathan Fripp1; Victoria Miller1; 1University of Florida
Laser sheet metal forming—along with additive manufacturing—is nearly unique in that it has been developed using an ICME-style research paradigm since its inception. Modeling both the thermal input and the mechanics of bending, then linking those processes together, is critical to realizing the full manufacturing process at an industrial scale. This work describes the development, validation, and application of an analytical, forward-stepping thermal model for laser forming in MATLAB, derived from computationally efficient models for additive manufacturing and welding. Small-scale, high-throughput laser forming experiments are used to link the thermal model outputs across laser forming parameter space with predictions of bending mechanism and extent. Finally, gaps in the current ICME approach are discussed, such as the changing absorptivity during multipass laser forming experiments.
Molecular Dynamics Simulation of Polyurea for Impact-Resistant Applications: Tyler Collins1; Sara Adibi1; 1San Diego State University
Polyurea, a versatile block copolymer, is widely applied in impact-resistant applications like protective gear and structural reinforcements. Its unique molecular architecture, featuring hard segments for rigidity and soft segments for flexibility, provides remarkable elasticity and toughness. Additionally, polyurea dissipates stress from high-velocity impacts through the breaking of hydrogen bonds. This study employs molecular dynamics simulations in LAMMPS to investigate polyurea's nanoscale response to high-velocity impacts. We analyze temperature fluctuations, stress propagation, and failure mechanisms to better understand its performance under extreme conditions. These insights are valuable for designing enhanced materials for impact-resistant applications, such as body armor. By simulating scenarios difficult to replicate experimentally, this work highlights the utility of computational methods in advancing material design for high-performance and safety-critical applications.
Rapid Structure Prediction of Single-Phase High Entropy Alloys Using Graph Neural Network Based Surrogate Modelling: Nicolas Beaver1; Aniruddha Dive1; Marina Wong1; Keita Shimanuki1; Ananya Patil1; Anthony Ferrell1; Mohsen Kivy1; 1California Polytechnic State University
In this study, we developed a rapid, reliable and cost-effective methodology via employing a Graph Neural Network based machine learning approach to effectively predict the crystal structures of single-phase high entropy alloys. Our novel approach searches through the enormous potential energy surface (PES) landscape with an aim to identify stable lowest energy crystal structure. Our approach was tested on 132 high entropy alloys, and the predictions were verified with the experimental data in literature. Overall, we achieved ~ 83 % accuracy in correctly predicting the stable crystal structures. Our prediction accuracy easily betters the prediction accuracy based on valence electron concentration (VEC) approach that is widely utilized for crystal structure prediction. Our proposed method was ultimately applied to predict the structure of a novel cobalt-free high-entropy alloy. Our predicted crystal structure of the alloy matched the one characterized using X-ray diffraction (XRD), scanning electron microscopy (SEM), and X-ray fluorescence (XRF).
"Why Can't we Just use Excel?:" Lessons on Incorporating ICME in the Undergraduate Classroom: Victoria Miller1; 1University of Florida
Developing a data literate workforce is crucial at every level, from K12 to PhD. However, undergraduate students are often resistant to learning basic data and coding skills. In other students, willingness to learn is eclipsed by a lack of basic computer literacy. In this presentation, lessons learned during the modernization of several individual classes and an undergraduate curriculum are discussed. Common pitfalls and strategies for overcoming them are presented.