Simulated Alloys

By Kimberly Mann Bruch, SDSC
An image of chunks of zinc alloy in a pile.

Amina Bendjennat, SDSC Communications, contributed to this story.

Chemical patterns in metallic alloys can now be analyzed using a novel machine-learning approach that has been developed on resources funded by the U.S. National Science Foundation (NSF) ACCESS program. Researchers from the Massachusetts Institute of Technology (MIT) have recently used their ACCESS allocations on Expanse at the San Diego Supercomputer Center (SDSC) at UC San Diego to develop a machine-learning method that efficiently decodes short-range order patterns in metallic alloys. 

“Thanks to ACCESS allocations on SDSC resources, we have developed computational models with unprecedented physical fidelity to investigate chemical patterns in alloys. These patterns are very difficult to observe in experiments, so any information that can be extracted directly from calculations is valuable,” MIT Assistant Professor of Materials Science and Engineering Rodrigo Freitas said. “The Expanse resources enabled us to perform highly accurate simulations and develop computational tools that can now predict what experimentalists observe in their measurements.”

Led by Freitas, the MIT team recently published their work in a paper titled Quantifying chemical short-range order in metallic alloys in the Proceedings of the National Academy of Sciences of the United States of America (PNAS).

A scientific visualization of the research in this story.
Histogram showing the size of the short-range order patterns observed in their simulations and how they change with temperature. The inset shows one of the more than 1.5M chemical motifs used by their machine-learning algorithm to identify patterns. Credit: MIT

“Metallurgists are always looking for ways to create metals with different properties – and we have been working on making better alloys possible by understanding their chemical arrangements at the atomic level.” Freitas said.

The MIT team used their ACCESS allocations on Expanse to study chemical patterns in metals known as short-range order (SRO). Freitas explained that all alloys in thermal equilibrium favor low-energy chemical motifs, driving them away from complete randomness and promoting the formation of short-range order (SRO). Yet, deciphering such information has remained an open challenge in the field.

“Quantifying SRO has been a challenge for high-entropy alloys due to the diversity in the chemical patterns they exhibit – our solution was a high-fidelity approach capable of capturing chemical patterns and their correlations – up to nanometers in scale – which matches well with experimental results,” Freitas said. “We found that a complete description of SRO requires capturing the prevalence of all possible chemical motifs in addition to the quantification of their organization in space.”

This three-dimensional cut of the five-dimensional “space of chemical motif dissimilarity” illustrates how the MIT team’s machine learning approach “sees” and encodes the chemical patterns observed in their simulations. Credit: MIT

The team used a damage-tolerant material made of chromium, cobalt and nickel (CrCoNi) as the main subject for their work and developed an approach for predicting and quantifying SRO in agreement with experimental measurements.  They used density functional theory (DFT) – a quantum-mechanical method that calculates the electronic structure of atoms – in combination with machine learning to run large-scale simulations without compromising accuracy. Additionally, the team used physics-informed machine learning to uncover complex chemical patterns in the simulation data and were able to statistically quantify SRO across various temperatures. 

“Excellent agreement with DFT predictions was obtained by developing an entirely new training approach centered around the extensive sampling of the chemical-ordering space,” Freitas said. “This approach also generalized much better than existing methods when evaluated on independent test sets under conditions not covered by the training scenarios.”

Thanks to ACCESS allocations on SDSC resources, we have developed computational models with unprecedented physical fidelity to investigate chemical patterns in alloys.

–Rodrigo Freitas, assistant professor, materials science and engineering, MIT

He said that a remaining challenge is the complete control of the SRO during the manufacturing processes. To address this issue, the team plans on conducting simulations of various manufacturing processes that metallic alloys undergo, such as solidification (e.g., casting, additive manufacturing, three-dimensional printing) and thermo-mechanical processing (e.g., cold rolling). 

“Using our new capabilities made possible by ACCESS allocations on Expanse, we will track the creation of SRO during solidification and its evolution as metallurgists process the alloy through mechanical deformation,” Freitas said. “Our specific next steps include examining the formation of SRO during simulations of typical manufacturing processes that alloys experience.”

For this work, the team plans to use Frontier at Oak Ridge Leadership Computing Facility through the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program.

Project Details

Resource Provider Institution(s): San Diego Supercomputer Center (SDSC)
Affiliations: Massachusetts Institute of Technology (MIT)
Funding Agency: NSF
Grant or Allocation Number(s): MAT210005

The science story featured here was enabled by the U.S. National Science Foundation’s ACCESS program, which is supported by National Science Foundation grants #2138259, #2138286, #2138307, #2137603, and #2138296.

Sign up for ACCESS news and updates.

Receive our monthly newsletter with ACCESS program news in your inbox. Read past issues.