Humans have been making alloys for millennia. Records as early as 3300 BC describe how bronze, an alloy that left its mark on civilisation, was created by mixing tin with copper. This traditional approach of adding a small amount of one element to a large quantity of another base element hasn’t altered significantly and can limit the diversity of alloys that can be formed.
In the 1980s, materials scientists proposed a radically new method of alloy-making—blending five or more metals in near-equal proportions. The result was a novel class of materials known as high entropy alloys (HEAs). What sets HEAs apart from other alloys is that they are exceptionally strong and pliable because more than five elements with almost equal amounts coexist in a single phase.
The significant challenge lies in how to build a phase selection rule to delineate the single-phase region in the vast temperature-compositions space of HEAs. Randomly combining metallic elements in the periodic table into a furnace is unlikely to produce HEAs because the high-fidelity phase selection rules are not established yet for materials scientists. Kewu Bai and Yong-Wei Zhang of A*STAR’s Institute of High Performance Computing (IHPC) set out to break new ground in the field by developing a new ‘recipe’ for lightweight, high-strength, single-phase HEAs.
First, the team picked a panel of eight starting metals. They then defined their HEA manufacturing protocol by combining machine learning approaches with a computational tool called CALPHAD. CALPHAD is a staple in the materials scientists’ toolkit, used to predict the phases in multi-element materials.
“The [CALPHAD predictions] can then be used as input for a machine learning model to derive high fidelity phase selection rules of single-phase HEAs,” Bai explained.
The researchers extracted data on their metallic building blocks from a most reliable thermodynamic database and used CALPHAD to predict their equilibrium phases. In total, they mapped the phase stabilities of 20 families of five-element HEAs, formed by varying the proportions of each component, across an array of temperatures.
Among the parameters with the highest predictive power for generating single-phase HEAs was equilibrium temperature. Interestingly, previous studies overlooked this parameter, which may explain why few single-phase HEAs were reported.
“The equilibrium temperature, which was neglected in the past, must be included in the machine learning descriptors,” advised Bai.
The team also used their phase selection rules to construct the blueprints for 213 potential single-phase HEAs with the lightweight strength required for building planes and vehicles. In the future, the researchers plan to extend the HEA-design capabilities of their approach to develop increasingly complex HEAs to serve more manufacturing industries.
The A*STAR-affiliated researchers contributing to this research are from the Institute of High Performance Computing (IHPC).