High-entropy alloys are promising materials because they remain stable in many different compositions. However, their many components make the solid solution phase more energy-efficient and favorable, increasing the number of competing intermetallic compounds. Therefore, it’s essential to computationally identify all intermetallic compounds in systems with many elements to study high-entropy alloys.
Researchers at Skoltech and MIPT have accelerated the search for high-performance metal alloys. They’ve developed a machine learning-based method that quickly identifies promising alloy compositions for lab testing.
This new method speeds up the traditionally slow and complex alloy modeling process. Without it, scientists often guess and may miss valuable alloys. The new approach allows for a thorough search of alloy candidates.
Pure metals often have inferior properties compared to alloys, which are made from several metals and sometimes other elements like carbon or silicon. By changing the composition and ratio of these elements, it is possible to adjust the alloy’s characteristics, such as strength, malleability, melting point, corrosion resistance, and electrical conductivity.
This search for better alloys benefits aerospace technology, mechanical engineering, construction, electronics, medical instruments, and more.
However, new alloys only get engineers ‘ attention after thorough lab testing, which is expensive and time-consuming. Even simulated experiments require so much computing power that researchers can’t explore all possible options.
Professor Alexander Shapeev, who heads the Laboratory of Artificial Intelligence for Materials Design at Skoltech AI, said, “The number of potential candidates is vast because so many variables are involved: what elements make up the alloy, in which proportions, what the crystal structure is, and so on.”
Scientists created 3D maps of the next-generation alloys
“To give you an idea, in the simplest system of two elements, say niobium and tungsten, if we consider a crystal lattice cell with 20 atoms, you’re going to have to model more than a million possible combinations, or 2 to the power of 20, not accounting for symmetry.”
Lead author of the study, Skoltech MSc student Viktoriia Zinkovich from the Data Science program, said, “The current approaches rely on a fundamental physical description of the process in terms of direct quantum mechanical calculations. These are exact but complex and time-consuming calculations.”
“On the other hand, we use machine-learned potentials, which are characterized by rapid computations and make it possible to sort through all possible combinations up to a certain cutoff limit, 20 atoms per supercell, for example. That means we won’t miss the good candidates.”
The researchers tested their new approach on two groups of metals: five high-melting-point metals (vanadium, molybdenum, niobium, tantalum, and tungsten) and five noble metals (gold, platinum, palladium, copper, and silver). Each group looked at three different combinations of these elements.
For instance, the noble metals group considered combinations like copper and platinum, copper, platinum, and palladium, or all five noble metals together. These metals tend to form the same crystal structure, simplifying calculations because the alloy is also assumed to have this structure.
The researchers used their algorithm on six different compositions (three from each group) to find stable alloys by optimizing the energy and enthalpy of formation. Unstable alloys would change into more viable configurations.
The new algorithm helped discover 268 new stable at zero-temperature alloys that weren’t listed in a commonly used industry database. For example, in the niobium-molybdenum-tungsten system, it found 12 alloy candidates that the database didn’t have.
These newly found alloys still need to be tested in more detail to determine their practical applications. Computational modeling has already led to the discovery of many important industrial alloys, like those used in car parts and rocket fuel storage. The researchers plan to expand their algorithm to include other alloy compositions and crystal structures.
Journal Reference:
- Zinkovich, V., Sotskov, V., Shapeev, A. et al. Exhaustive search for novel multicomponent alloys with brute force and machine learning. npj Comput Mater 10, 269 (2024). DOI: 10.1038/s41524-024-01452-x
Source: Tech Explorist
