Alloys, materials made by combining two or more metals are usually themselves metallic and crystalline. US scientists have now developed a shortcut to making amorphous, glassy, alloys that could lead to new materials that are stronger, but less dense than steel and more resistant to corrosion and wear than crystalline alloys.

Researchers at Northwestern University, SLAC, and the National Institute of Standards and Technology used machine learning to help them quickly make and screen hundreds of sample materials at a time. The process generated three new leads for metallic glassy materials. [Wolverton et al. Sci Adv (2018) 4, 4, eaaq1566; DOI: 10.1126/sciadv.aaq1566]

"It typically takes a decade or two to get a material from discovery to commercial use," explains Northwestern's Chris Wolverton who is pioneering machine learning and artificial intelligence in materials discovery. "This is a big step in trying to squeeze that time down. You could start out with nothing more than a list of properties you want in a material and, using AI, quickly narrow the huge field of potential materials to a few good candidates."

Wolverton's ultimate goal is to be able to screen hundreds of sample materials and get almost instantaneous feedback from the algorithm as to whether or not the material is suitable and to then generate another set of samples ready for testing immediately afterwards. Given that scientists have investigated about 6000 of the possible millions of combinations that might form metallic glasses, this approach allowed the team to make and screen 20000 alloys in a year.

"The unique thing we have done is to rapidly verify our predictions with experimental measurements and then repeatedly cycle the results back into the next round of machine learning and experiments," team member Apurva Mehta explains. With machine learning, no prior knowledge is required. Moreover, it should be possible to drive the whole process of making and testing without human intervention, leaving the scientists free to develop and optimize the processes and apply the science.

On the basis of what the team's algorithms "learned" in the first round of testing a databank of thousands of materials accumulated over 50 years, the scientists could make two sets of sample alloys using two different methods and then obtained detailed structural information using the SSRL X-ray beam. The results, fed back into the algorithms, then generates new machine learning information and seeds a second generation, of improved alloys to make and test. And, so on.

By the time the team had done three rounds, they had a success rate in finding metallic glasses that was as low as one in 300 or 400 samples in the first round to one in every two or three samples tested being a metallic glass. Such a hit rate bodes well for the rapid development of novel three-metal glassy alloys that might be selected for particular characteristics for a wide range of materials science and engineering applications in the future.

David Bradley blogs at Sciencebase Science Blog and tweets @sciencebase, he is author of the popular science book "Deceived Wisdom".