High Entropy Alloys IX: Structures and Modeling
: Modeling and Machine Learning
Sponsored by: TMS Functional Materials Division, TMS Structural Materials Division, TMS: Alloy Phases Committee, TMS: Mechanical Behavior of Materials Committee
Program Organizers: Peter Liaw, University of Tennessee; Michael Gao, National Energy Technology Laboratory; E-Wen Huang, National Chiao Tung University; Srivatsan Tirumalai; Xie Xie, FCA US LLC; Gongyao Wang, Globus Medical
Wednesday 2:00 PM
March 17, 2021
Room: RM 9
Location: TMS2021 Virtual
Session Chair: Yu Zhong, Worcester Polytechnic Institute; Jia Li, Hunan University
2:00 PM Invited
The Application of High-throughput Calculations in High Entropy Alloys
: Yu Zhong1; 1Worcester Polytechnic Institute
There have been extensive investigations in the last two decades on the high entropy alloys (HEAs). However, it is hard to predict the composition window for the next-generation HEAs by its multicomponent nature. In the current talk, we will go through our recent progress on the high throughput (HT)simulations based on the CALPHAD (Computer Coupling of Phase Diagrams and Thermochemistry) approach and combined with machine learning. At first, we will discuss the details of the customized procedure we developed on the HT simulations. Then we will focus on its application in high entropy alloys, including how to find the multicomponent eutectic compositions targeting the eutectic high entropy alloys (EHEAs) and bulk metallic glasses (BMGs), and the composition window for the Transformation induced plasticity (TRIP) HEAs and Twinning induced plasticity (TWIP) HEAs.
2:25 PM Invited
Extrapolation of Machine Learning Models for Designing Multi-principal Element Alloys: James Saal1; Chris Borg1; Clara Nyby1; Bryce Meredig1; 1Citrine Informatics
Machine learning is becoming an increasingly popular method for predicting metal alloy behavior, particularly for systems and properties where experimental data may be lacking. This is especially true for multi-principal element alloys (MPEAs), where the large compositional design space necessitates computational and data-driven methods to guide design. A model’s ability to extrapolate beyond the compositions on which it has been trained and predict properties for novel systems is a critical model performance question that is often overlooked and cannot simply be summarized by an R^2 value. In this work, we use MPEA data to explore metrics for model extrapolability and the importance of adding physical features to model training data.
2:50 PM
Machine Learning Enabled Prediction of Stacking Fault Energies in Concentrated Alloys: Gaurav Arora1; Anus Manzoor1; Dilpuneet Aidhy1; 1University of Wyoming
A unique combination of high strength and high ductility in certain compositions of high entropy alloys (HEAs) has been observed which is attributed to the low stacking fault energy (SFE). While atomistic calculations successfully predict the SFE of pure metals, large variations up to 200 mJ/m2 have been observed in HEAs. The leading cause is the limited number of atoms that can be modeled in atomistic calculations; as a result, various nearest neighbor environments may not be adequately captured resulting in different SFE values. In this work, we use machine learning to overcome this limitation and provide a methodology to significantly reduce the variation and uncertainty in predicting SFEs. We show that the SFE can be accurately predicted across the composition ranges in binary alloys and in multi-elemental alloy. We also elucidate the underlying causes such as charge distribution, nearest neighbor environment, and the composition of an alloy on SFE.
3:10 PM Invited
Optimalizing Properties of High Entropy Alloy by Machine Learning and Multiscale Simulations: Jia Li1; Yang Chen1; Qihong Fang1; 1Hunan University
High-entropy alloys have attracted wide attention due to their excellent properties, which are expected to be used in nuclear energy, aerospace and other important fields and major equipment. Based on the experimental results combined with high-throughput computing and machine learning, the relationship of “composition-structure-property” is established to guide the development and design of new materials, which overcomes the problems of the low efficiency and waste of resources in the traditional "trial and error method". Therefore, a method based on the high-throughput simulations, theoretical model, and machine learning is adopted to obtain high-strength and low-cost medium-entropy alloy. This method can not only obtain a large number of data quickly and accurately, but also help determine the relationship between the compositions and mechanical properties of medium-entropy alloys. The results show that the combination of the high-throughput simulations, theoretical model, and machine learning is of great significance for the development of alloys with expected properties. The present research will give a foundation for the development of high-throughput experimental and theoretical calculation methods to prepare multicomponent alloys with specific properties.
3:35 PM
Accelerated Exploration of Refractory Multi-principal Element Alloys by Machine Learning: Carolina Frey1; Christopher Borg2; James Saal3; Bryce Meredig3; Daniel Miracle4; Tresa Pollock1; 1University of California, Santa Barbara; 2Citrine Informatics ; 3Citrine Informatics; 4Air Force Research Laboratory
Refractory-based multi-principal element alloys (RMPEAs) are of interest for high temperature structural applications due to their high melting points but remain relatively underexplored compared to other MPEA categories due to processing challenges. Search strategies and experimental methods that reduce the number of needed experiments are necessary to more efficiently discover interesting materials. This presentation will discuss the use of random forest machine learning in concert with rapid processing and characterization techniques to guide sequential alloy design. An updated and publicly available database, which doubles the number of recorded compositions and mechanical properties of previously published MPEA databases, was developed for use as training data. A rapid solidification technique, splat quenching, was used to reduce segregation and the grain size of synthesized materials to evaluate a greater range of properties and allow for more rapid experimental exploration of interesting RMPEA compositions. New alloys with high strength will be discussed.
3:55 PM
Ab Initio Modeling on the Elastic Properties of Al-Co-Cr-Fe-Ni High Entropy Alloys: A Case Study with FCC Phase: Songge Yang1; Jize Zhang1; Yu Zhong1; 1Worcester Polytechnic Institute
The Al-Co-Cr-Fe-Ni system has been one of the most thoroughly studied systems in high entropy alloys (HEAs) due to their promising mechanical properties. However, the prediction of mechanical properties in this system with full composition range could be challenging purely based on experiments. In the current study, the high-throughput ab initio modeling is proceeded to predict the elastic properties of the quinary FCC Al-Co-Cr-Fe-Ni HEA single crystals by using special quasi-random structure (SQS) approach. The predictions will start with pure elements of Al-Co-Cr-Fe-Ni system and will be continued with binaries, ternaries, quaternaries, and finally the quinary compositions. More than 100 compositions were simulated. After that, the elastic property database of the FCC phase in this system will be contoured based on the calculated results from the ab initio calculations as well as the machine-learning method.