Machine learning for better metals

When humans learned to extract metals from their ores and mix them into  such as bronze, brass and steel, technology took great leaps forward. Now researchers are turning to artificial intelligence to find the next generation of alloys.

Scientists are already finding new alloys with increased strength and other improved features. A research team based in China have now published such discoveries in the journal .

Explaining the origins of their work, researcher  of the Beijing Advanced Innovation Center for Materials Genome Engineering cites as his inspiration the success of  in mastering the strategy game . He also references the algorithms and models used to create expert .

“This showed us the power of data and data-driven machine learning,” says Su.

Any mixture of two or more elements is an alloy, but the team focused their attention on . These contain close to equivalent amounts of at least five different elements.

There is a vast range of possible compositions for these alloys, depending on which elements are used and the precise proportions in their composition. Rather than using time-consuming trial and error methods to analyse them, the researchers devised software that would allow a computer to sift through more than a million possibilities in search of promising mixtures.

In their current article, they discuss their success in identifying alloys with exceptional hardness, using the elements aluminium, cobalt, chromium, copper, iron and nickel. They ran just seven iterations of the machine learning procedure and found compositional details for new alloys that were more than 10% harder than any used to train the software.

Hardness was only an initial target used to test and assess the strategy. The researchers emphasise that the same strategy should also be able to optimise other desirable properties. These include combining hardness with lightness and making specialised alloys known as , which have high electrical resistance, as well as resistance to corrosion. Metallic glasses are applicable to many technologies, including microelectronics, the manufacture of surgical instruments and magnets, and nuclear waste disposal.

Su points out that the team’s research is just one example of how machine learning techniques are changing the traditional methods used to design new materials. These changes are driven by the need to reduce the time and materials invested when exploring the many complex options.

“Our method should discover useful new materials in less time, at less cost, and using much smaller test samples,” Su concludes. The research team are already moving on to target many more possibilities for their machine learning techniques, exploring a wider range of materials and properties

A flow chart outlining the machine learning procedure.
A flow chart outlining the machine learning procedure.

Article details:

Su, Y. et al.: “Machine learning assisted design of high entropy alloys with desired property” Acta Materialia (2019)

Acta Materialia is part of the family of Acta Materialia Inc journals, which also includes Acta Biomaterialia, Scripta Materialia and the newly launched Materialia