In the realm of materials science, understanding how atoms are arranged in alloys is crucial for developing stronger and more efficient materials. A team of graduate students from the Massachusetts Institute of Technology, Killian Sheriff and Yifan Cao, are at the forefront of this research, utilizing machine learning to delve into short-range order within metallic alloys. Their work is supervised by Assistant Professor Rodrigo Freitas from the Department of Materials Science and Engineering, along with Assistant Professor Tess Smidt from the Department of Electrical Engineering and Computer Science. The findings from their research were recently published in The Proceedings of the National Academy of Sciences.
Short-range order refers to the specific arrangements of atoms over small distances, a concept that has been somewhat neglected in materials science. However, the recent interest in high-entropy alloys, materials that consist of multiple elements in nearly equal proportions, has sparked renewed focus on quantifying SRO. These alloys can contain anywhere from three to twenty different elements, offering a vast design space for creating materials with superior properties. Traditionally, materials scientists would add small amounts of other elements to a base metal to enhance certain characteristics, but high-entropy alloys allow for a more complex approach.
The challenge in studying SRO lies in accurately capturing the intricate arrangements of atoms. Previous methods often relied on small computational models or limited simulations, which provided an incomplete picture of these complex materials. Sheriff explained that high-entropy materials are chemically intricate, requiring more substantial simulations to truly understand their behavior. This complexity is akin to trying to build a large Lego model without knowing the smallest piece, highlighting the need for a deeper understanding of atomic arrangements.
To tackle this issue, the MIT team employed machine learning techniques to quantify SRO atom by atom. By using advanced computational models, they were able to identify chemical motifs, specific arrangements of atoms, within high-entropy alloys. This approach allowed them to analyze the atomic interactions in greater detail than ever before. Cao likened the process to a connect-the-dots puzzle, where understanding the rules for connecting the dots reveals the underlying patterns of atomic arrangements.
The researchers developed a two-pronged machine learning solution that first focused on reproducing the chemical bonds in high-entropy alloys. This model served as a foundational step in accurately quantifying SRO. The second part involved identifying billions of possible chemical motifs from simulation data, which posed a significant challenge due to their varied appearances. The team utilized 3D Euclidean neural networks to effectively categorize these motifs, allowing them to sort and quantify SRO with unprecedented accuracy.
The implications of this research extend far beyond theoretical exploration. The team plans to leverage the U.S. Department of Energy’s INCITE program, which provides access to the world’s fastest supercomputer, Frontier, to explore how SRO changes under various metal processing conditions. This access will enable them to conduct robust simulations that could lead to the development of alloys with predetermined SRO, enhancing their performance in real-world applications.
Sheriff and Cao's work represents a significant advancement in the field of materials science, as it introduces a framework for understanding the complexities of atomic arrangements in high-entropy alloys. By harnessing machine learning, they are not only uncovering the secrets of these materials but also paving the way for the purposeful design of new classes of alloys that could revolutionize industries such as aerospace, biomedicine, and electronics. The research is supported by the MathWorks Ignition Fund and the MathWorks Engineering Fellowship Fund, among others, underscoring the importance of collaboration in advancing technological frontiers.\
