Design of Active Brazing Fillers Based on Eutectic High Entropy Compositions by Machine Learning

Brazing, a technique for joining two materials, has long been used across various industries, from small-scale to large-scale operations. A key component of this technique is the filler metal, an alloy which is melted between the two components being joined, and acts to make the bond between them, with the materials themselves largely unaffected by the process. The filler usually has a tailored composition, since its interaction with substrate materials must be suitable for a successful union.

One of the most difficult bonds to form is between metallic and non-metallic materials, and the properties and types of interaction on each side of the joint are very different. To achieve the bond, the brazing material must be active, meaning it must contain metals in its composition that can react with the ceramic.

To widen the capabilities and improve the performance of such joints. New alloys are being sought, and new classes of materials, like eutectic high entropy alloys (EHEAs), attract a lot of interest as candidate fillers. Segregation and large thermal expansions are less likely in these alloys, due to their isothermal transformation and dual-phase structure. Furthermore, because of the character as a eutectic multi-component alloy, high-temperature active metals can be included in the composition, but a low melting point could still be expected.

Empirical trial-and-error methods and thermodynamic modelling used previously have shown low efficiencies in designing new EHEAs, due to the need for extensive physical experiments, or limited in accuracy, because of the unavailability of appropriate assessed binary and ternary diagrams. As a consequence, a new machine learning methodology, based on thermodynamic, electronic and atomic size features, is here developed and used to design novel EHEAs that can be used as active brazing fillers.

With the use of several predicting algorithms, that show predictive accuracy for novel experimental compositions of over 75 % for EHEAs, four new compositions based on active elements have been developed. Furthermore, vacuum brazing samples, of each alloy, have shown successful joints between Kovar and alumina (as an example metal-ceramic joint), by the formation of different oxides.

This research highlights the capabilities of machine learning in alloy design and underscores the potential of eutectic high entropy alloys (HEAs) as active brazing materials for Kovar-alumina.