HEA 2023: High-throughput and Machine Learning II
Program Organizers: Andrew Detor, DARPA/DSO; Amy Clarke, Los Alamos National Laboratory
Wednesday 9:00 AM
November 15, 2023
Room: Riverboat
Location: Omni William Penn
Session Chair: Raymundo Arroyave, Texas A&M University
9:00 AM Break
9:30 AM Introductory Comments
9:35 AM
Design of High Entropy Superalloy FeNiCrAlCu with Stacking Fault Energy Modeling using Thermodynamic Calculations, First-principles, and Machine Learning: Tria Achmad1; Farrel Baskara1; Muhammad Aulia1; 1Bandung Institute of Technology
High entropy superalloys (HESA) are promising materials with unique properties, but their development and optimization remain ongoing. This study focuses on Fe-based HESA FeNiCrAlCu and their stacking fault energy (SFE), a critical parameter influencing deformation mechanisms and creep resistance. Leveraging machine learning and computational thermodynamics, we propose a novel approach for predicting SFE using big data analysis. However, thermodynamic data of some rare elements like Zr is limited. Then, we use first-principles to investigate further the effect of adding Zr on elastic properties, SFE, and electronic structure. Our research establishes an optimal design guide for achieving desired SFE values: Ni (20-25 at%), Cr (15-36 at%), Al (5-20 at%), and Cu (9-20 at%). We achieve an impressive 0.98 accuracy in classifying SFE types by employing a deep learning neural network model. This work advances HESA design, provides valuable insights into their mechanical behavior, and improves creep resistance for demanding applications.
9:55 AM
Designing High-entropy Alloys: Aluminum Refractory B2 Phases: Diego Ibarra1; Jie Qi1; Xuesong Fan2; Debashish Sur1; Rui Feng2; John Scully1; Peter Liaw2; Joseph Poon1; 1University of Virginia; 2The University of Tennessee, Knoxville
Ordered high-entropy body-centered-cubic (BCC) phases with the B2 structures, particularly those formed with refractory elements and aluminum (Al-RHEAs), are of interest due to their high strength, corrosion resistance, and oxidation resistance. The incorporation of Al lowers the density and promotes the long-range atomic ordering, which in turn stabilizes the B2 formation, and strengthens the material but usually deteriorates ductility. B2 Al-RHEAs are screened using machine learning (ML) models to predict B2 formation and toughness. High prediction accuracy is achieved. Several Al-RHEAs have shown large compression plasticity and high strength and exhibit some tensile ductility.
10:15 AM
Ab-initio Tensile Tests Applied to BCC Refractory Alloys: Vishnu Raghuraman1; Michael Widom1; Michael Gao2; 1Carnegie Mellon University; 2National Energy Technology Laboratory
Refractory metals exhibit high strength at high temperature, but often lack ductility. Multi- principle element alloys such as high entropy alloys offer the potential to improve ductility while maintaining strength, but we don’t know a−priori what compositions will be suitable. A number of measures have been proposed to predict the ductility of metals, notably the Pugh ratio, the Rice-Thomson D-parameter, among others. Here we examine direct ab − initio simulation of de- formation under tensile strain, and we apply this to a variety of Nb- and Mo-based binary alloys and to several quaternary alloy systems. Our results exhibit a variety of ductility mechanisms including slip, stacking faults, transformation and twinning. We relate these deformations to otherpredictors of ductility, and we correlate these with each other.
10:35 AM Break
10:55 AM
AI-accelerated Materials Informatics Method for the Discovery of Ductile Alloys: Max Hodapp1; Ivan Novikov2; Olga Kovalyova2; Alexander Shapeev2; 1Materials Center Leoben; 2Skoltech
In computational materials science, a common means for predicting macroscopic (e.g., mechanical) properties of an alloy is to define a model using combinations of descriptors that depend on some material properties (elastic constants, misfit volumes, etc.), representative for the macroscopic behavior. The material properties are usually computed using density functional theory (DFT). However, DFT scales cubically with the number of atoms and is thus impractical for a screening over many alloy compositions.Here, we present a novel methodology which combines modeling approaches and machine-learning interatomic potentials. Machine-learning interatomic potentials are orders of magnitude faster than DFT, while achieving similar accuracy, allowing for a predictive and tractable high-throughput screening over the whole alloy space. The proposed methodology is illustrated by predicting the room temperature ductility of the medium-entropy alloy Mo-Nb-Ta.
11:15 AM
Stability and Growth Kinetics of Deformation Twin Embryos in BCC Complex Concentrated Alloys: Ganlin Chen1; Liang Qi1; 1University of Michigan
We employed computational tools to understand and further tune the effects of diffusionless phase transformations on twinnabilities in BCC complex concentrated alloys. First-principle calculations were firstly performed to study the atomic structures and energy stability of different metastable phases at their local-minimum states in different alloy compositions. Secondly, we applied atomistic simulations with classical interatomic potentials to further analyze the atomic structures of alloys at finite temperatures. Both first-principle calculations and atomistic simulation results are employed to construct the order parameters that can effectively describe the energy landscape and diffusionless phase transformation paths between multiple phases. Thirdly, with the aid of atomistic simulations and crystallographic theories, we investigate the energy landscape of the nucleation and growth dynamics of deformation twinning under external loading conditions at different temperatures and further explore how the diffusionless phase transformations affect the deformation twinning activities in BCC complex concentrated alloys.
11:35 AM
ML-based High-throughput Search to Identify Refractory High Entropy Alloy with Trade-off Mechanical Properties: Debasis Sengupta1; Stephen Giles1; Hugh Shortt2; Peter Liaw2; 1CFD Research Corp; 2University of Tennessee, Knoxville
Identifying Refractory High Entropy Alloy (RHEA) compositions with desired high temperature strength and room temperature ductility/plasticity from the vast composition space is a challenging task. Generally, compositions with higher yield strength tend to have lower room temperature ductility or sometime show brittle behavior. In this work, we first present machine learning (ML) based models for compressive yield strength and room temperature plasticity. These models were extensively validated against experiments. The two completely independent models were able to reproduce the well-established fact that an increase in plasticity comes at a cost of reduction in strength. We then used the two models and applied the state-of-the-art sampling method to generated approximately 100,000 RHEA compositions, and computed their strengths and plasticities. We then designed a “Figure of Merit” to identify the promising compositions. Some selected compositions were synthesized and characterized for their mechanical properties.