Machine learning predicts when exotic alloys stretch like glass

Some metals can stretch to many times their original length before breaking, flowing at high temperature almost like molten glass or chewing gum. This remarkable property, known as superplasticity, has fascinated materials scientists for decades because it allows engineers to form complex, intricate shapes in a single step that would otherwise require dozens of machining operations or welded components. Now a team of researchers has turned to artificial intelligence to answer one of the field’s most stubborn questions: given a new alloy, can we predict in advance whether it will be superplastic, without spending months or years in the laboratory?

The study, published in the Journal of Materials Science by Hamed Shahmir and Sina Kooshamanesh of Tarbiat Modares University, Mohammad Sajad Mehranpour of the University of Tehran, and veteran superplasticity researcher Terence G. Langdon, focuses on a class of materials called high-entropy alloys. Unlike conventional metals, which are built around one dominant element with small additions of others, high-entropy alloys mix four, five, or even more principal elements in roughly equal proportions. The result is a chaotic atomic landscape that can produce extraordinary combinations of strength, ductility, and thermal stability, properties that have made these materials one of the hottest topics in metallurgy since their emergence in the early 2000s.

Superplasticity is not a given for any metal. It requires a very fine, stable grain structure, typically with grains smaller than about ten micrometers, and it generally appears only within a specific window of temperature and strain rate. Under those conditions, the dominant deformation mechanism shifts to grain boundary sliding, in which individual crystalline grains slide past one another like bricks in a lubricated wall rather than stretching internally. When this mechanism operates cleanly, elongations exceeding several hundred percent, and in exceptional cases several thousand percent, become possible. The problem is that the conditions are delicate: grains must remain small at high temperature, which demands resistance to grain growth, and the alloy’s microstructure must not degrade catastrophically during deformation.

High-entropy alloys are theoretically ideal candidates for this behavior. Their sluggish atomic diffusion, a consequence of the complex multi-element lattice, helps stabilize fine grains at elevated temperatures, and many of them form dual-phase or multiphase structures that further pin grain boundaries in place. Over the past decade, laboratories around the world have demonstrated superplastic flow in alloys such as the famous equiatomic CoCrFeMnNi, various AlCoCrCuFeNi derivatives, and a growing family of medium-entropy and eutectic compositions, some processed by severe plastic deformation techniques like high-pressure torsion to create the necessary ultrafine grains. But each demonstration has come at the cost of extensive trial-and-error experimentation.

That cost is precisely what the new research set out to eliminate. The authors compiled the available experimental data on superplasticity in nanostructured and ultrafine-grained high-entropy alloys, spanning a wide range of chemical compositions, processing routes, grain sizes, testing temperatures, and strain rates. This curated dataset became the training ground for machine learning models tasked with learning the hidden relationships between a material’s composition and processing history and its superplastic response. Machine learning, in essence, searches for patterns in data that human intuition and classical constitutive equations may miss, and it can do so across a multidimensional space of alloying elements that no experimental campaign could ever explore exhaustively.

The approach reflects a broader transformation sweeping through materials science. In recent years, machine learning has been applied to predict the hardness and grain refinement of severely deformed metals, to screen out high-entropy alloys prone to forming brittle sigma phase, to identify thermomechanical processing parameters that deliver desired combinations of strength and ductility, and to estimate properties such as stacking fault energy in refractory multi-principal element alloys. High-throughput computational screening, coupled with thermodynamic databases and empirical parameters, now allows researchers to explore thousands of candidate compositions on a laptop before a single ingot is melted. The new study extends this toolkit to one of the most mechanically demanding and commercially valuable properties of all: the ability to deform superplastically.

Why does this matter industrially? Superplastic forming is already used commercially in aerospace, particularly for titanium alloys in aircraft structures, where complex double-skin panels and engine components are blow-formed at high temperature. But the process is slow, energy-intensive, and limited to a handful of well-characterized alloys whose superplastic windows were established through decades of empirical work. If machine learning can rapidly identify which high-entropy alloys will exhibit superplasticity, and under what conditions of temperature, strain rate, and grain size, the timeline for qualifying new formable materials could shrink from decades to years. Lighter, stronger, and more corrosion-resistant alloys could enter superplastic forming lines, enabling aircraft and spacecraft structures that are currently impractical to manufacture.

The scientific significance goes beyond engineering convenience. By training models on data from many different alloy families, the researchers can interrogate which compositional and microstructural features matter most for superplastic flow. Grain size emerges as a central variable, as expected from the classical mechanics of grain boundary sliding, where strain rate scales strongly with the inverse of grain size raised to a power of roughly two. But the model can also weigh subtler factors, such as the stabilizing effect of sluggish diffusion on grain growth, the role of second-phase particles in pinning boundaries, and the influence of phase fraction in multiphase alloys. In doing so, machine learning acts not just as a predictor but as a hypothesis generator, pointing experimentalists toward the mechanisms most worth investigating.

There are, of course, important caveats. Machine learning models are only as good as the data they learn from, and superplasticity data for high-entropy alloys remain relatively sparse compared with conventional alloys. The field of small-data machine learning in materials science has developed techniques to handle this scarcity, including careful feature selection, regularization, and physics-informed constraints, but extrapolation beyond the training data remains risky. A model trained on face-centered cubic alloys may not reliably predict the behavior of refractory body-centered cubic systems, and superplasticity depends on processing details, such as the severity of prior deformation and annealing conditions, that must be captured accurately in any dataset. The authors note that their underlying data will be made available on request, a practice that supports reproducibility and future model refinement.

Nevertheless, the trajectory is clear. Twenty years after the CoCrFeMnNi alloy was first reported, high-entropy alloys have matured from laboratory curiosities into serious candidates for structural applications, and superplasticity has emerged as one of their most promising attributes. Combining that experimental maturity with the predictive power of machine learning creates a genuine shortcut: instead of melting, deforming, and testing alloy after alloy, researchers can computationally shortlist the compositions most likely to flow like taffy at high temperature, then devote laboratory resources only to the best candidates. If that workflow becomes standard, the next generation of superformable metals may be designed on a computer before they ever exist in a crucible, and the seventy-five-year-old science of superplasticity may be entering its fastest-moving era yet.

Subject of Research: Machine learning prediction of superplasticity in high-entropy alloys

Article Title: Using machine learning to predict superplasticity in high-entropy alloys

Article References: Shahmir, H., Kooshamanesh, S., Mehranpour, M. S., & Langdon, T. G. (2026). Using machine learning to predict superplasticity in high-entropy alloys. Journal of Materials Science. https://doi.org/10.1007/s10853-026-13754-0

Image Credits: AI Generated

DOI: 10.1007/s10853-026-13754-0

Keywords: high-entropy alloys, superplasticity, machine learning, grain boundary sliding, ultrafine-grained materials, severe plastic deformation, materials science, computational materials science, grain refinement, aerospace forming, CoCrFeMnNi, thermal stability