Tech

New alloy deep learning

Shown is a data-driven workflow for mapping elastic properties in a high entropy alloy space. Credit: Chenetal.

When is more than the sum of those parts? Alloys exhibit such a synergistic effect. Steel, for example, has revolutionized the industry by taking iron, adding a small amount of carbon, and making the alloy much stronger than any of its components.


Supercomputer simulations help scientists discover a new type of alloy called High entropy alloy..Researchers Stampede 2 Texas Advanced Computing Center (TACC) supercomputer assigned by the Extreme Science and Engineering Discovery Environment (XSEDE).

Their study was in April 2022 Npj calculation data.. This approach can be applied to find new materials such as batteries and catalysts without the need for expensive metals such as platinum and cobalt.

Wei Chen, an associate professor of materials science and engineering at the Illinois Institute of Technology and a senior author of research, said:

In a nutshell, the term “high entropy” refers to the reduction in energy gained by randomly mixing multiple elements with similar atomic fractions, which can stabilize new ingredients resulting from “cocktails”. I can do it.

For this study, Chen et al. Investigated a large space of 14 elements and combinations that produced high-entropy alloys.They perform high-throughput quantum mechanical calculations, with alloy stability and Elastic propertiesThe ability to recover the size and shape of over 7,000 high-entropy alloys from stress.

“This is, to our knowledge, the largest database of elastic properties for high-entropy alloys,” Chen added.

Then get this big dataset, Deep set Architecture, this is an advanced deep learning architecture to generate Predictive model About the characteristics of the new high entropy alloy.

“We have developed a new machine learning model and predicted the properties of over 370,000 high-entropy alloy compositions,” said Chen.

In the final part of their study, we used what is called association rule mining. This is a rule-based machine learning method used to discover new and interesting relationships between variables, in this case individual or combinations of elements, affecting the properties of high-entropy alloys.

“We have derived some design rules for the development of high-entropy alloys and have proposed some compositions that experimenters can synthesize and make,” Chen added. ..

High entropy alloys are a new frontier for materials scientists. Therefore, there are few experimental results. Therefore, this lack of data has limited the ability of scientists to design new data.

“Therefore, we investigate a large number of high-entropy alloy spaces and perform high-throughput calculations to understand their stability and elastic properties,” Chen said.

New alloy deep learning

TACC Stampede2 supercomputer. Credit: TACC

He mentioned more than 160,000 first-principles calculations in this latest study.

“It’s basically impossible to perform huge numbers of calculations on individual computer clusters or personal computers,” Chen said. “Therefore, we need access to high-performance computing facilities like TACC assigned by XSEDE.”

Chen was awarded time at Stampede 2 TACC supercomputer via XSEDE. This is a virtual collaboration funded by the National Science Foundation (NSF) that facilitates free, customized access to advanced digital resources, consulting, training and mentorship.

Unfortunately, EMTO-CPA The code Chen used to calculate quantum mechanical density functional theory is well suited for high-performance computing parallelism, which typically performs large-scale calculations and divides them into smaller calculations that are performed simultaneously. did not.

“”Stampede 2 TACC via XSEDE provided a very useful code called Launcher. This helps you pack individual small jobs into one or two large jobs and makes the most of them. Stampede 2High Performance Computing Node “.

The Launcher Scripts developed at TACC allowed Chen to combine about 60 small jobs into one and run them simultaneously on high-performance nodes. This has improved computational efficiency and speed.

“Obviously, this is a unique application for supercomputers, but it’s also very common in many material modeling issues,” says Chen.

For this task, Chen et al. Applied a computer network architecture called DeepSet to model the properties of high-entropy alloys.

The deep set architecture can use the elemental properties of individual high-entropy alloys to build predictive models to predict the properties of new alloy systems.

“The framework is so efficient that most of the training was done on the student’s computer,” said Chen. “But I used TACC Stampede 2 To make predictions using the model. “

Chen gave an example of a widely studied Cantor alloy. It is a nearly equal mixture of iron, manganese, cobalt, chromium and nickel. What’s interesting about it is that it doesn’t become brittle even at very low temperatures.

One of the reasons for this is what Chen called the “cocktail effect.” It produces surprising behavior compared to its constituent elements when mixed in approximately equal proportions as a high entropy alloy.

Another reason is that when multiple elements are mixed, almost unlimited design space is opened up to find new construction structures and completely new materials for applications that were previously impossible.

“Hopefully, it will help more researchers use computational tools to narrow down the materials they want to synthesize,” Chen said. alloy It can be made from easily procured elements and, if possible, can replace precious metals and elements such as platinum and cobalt that have supply chain problems. These are actually strategic and sustainable materials for the future. ”


The team eliminates guesswork from the discovery of new high-entropy alloys


For more information:
Jie Zhang et al, Composition design of high entropy alloys with deep set learning, npj calculation material (2022). DOI: 10.1038 / s41524-022-00779-7

Quote: New Alloy Deep Learning (July 21, 2022) was obtained from https://phys.org/news/2022-07-deep-alloys.html on July 21, 2022.

This document is subject to copyright. No part may be reproduced without written permission, except for fair transactions for personal investigation or research purposes. The content is provided for informational purposes only.



New alloy deep learning

Source link New alloy deep learning

Show More

Related Articles

Back to top button