The development of machine learning has accelerated the emergence of new metal glass

If two or three metals are combined, an alloy that normally looks like a metal is obtained, the atoms of which are arranged in rigidity.

But under special conditions, we might get something new: a futuristic alloy called metallic glass.

The way the atoms are arranged is very similar to the atomic arrangement of the glass. But its glassy nature makes it stronger and lighter than the best steels available today, and it is more resistant to corrosion and wear.

Although metallic glass seems to be a good fit as a protective coating and a substitute for steel, in the past 50 years, there have been millions of combinations, but only a few thousand combinations have been evaluated, and only a few have been invested. use.

A team led by scientists from the SLAC National Accelerator Laboratory, National Institute of Standards and Technology (NIST) and Northwestern University now reports the shortcuts to discover and improve metallic glass, and unlike the past, it is now rare. Time and cost.

Science Advances reports: A research team found and improved the shortcuts for metallic glass by using a system from SLAC Stanford Synchrotron Radiation Source (SSRL), which uses machine learning to quickly create and screen hundreds of sample materials.

With this new system, the team quickly discovered three new blends of metallic glass components that were 200 times faster than before.

Northwestern professor Chris Wolfton is an early pioneer in the use of computational and artificial intelligence to predict new materials. He said: "It usually takes ten or two years to discover new materials to commercial applications, but using AI to predict new materials, This is a big step in the effort to shorten the time. You can choose the list of materials that best suits you from the material list and use artificial intelligence to quickly shrink the huge areas of potential materials into several excellent materials."

His ultimate goal is to enable scientists to scan hundreds of sample materials and get near-instant feedback from machine learning models.

In the past half century, scientists have studied about 6,000 components that make up metallic glass. SSRL staff scientists added: "We are able to produce and screen 20,000 combinations in a year."