AI has accelerated the Discovery of Metallic Glass by 200 times.

Artificial Intelligence (AI) has given this world a lot of revolutionary gifts and the list continues to grow as latest reports suggest that metallic glass is much nearer to us than we thought, thanks to AI. Alloys are pretty common these days and the majority of people know that they are formed by mixing two or more metals. The physical appearance and characteristics of these alloys are pretty much similar to metals as the arrangement pattern of atoms is just the same in them. However, sometimes when researchers are in luck, a combination of metals do give birth to a metallic glass. It is an amorphous material with an irregular arrangement of atoms. As a result, they are lighter yet stronger than all the materials known today. Similarly, they are much more resistant to corrosion.

Despite all these benefits, determining the combinations of metals that will produce metallic glass is a frustratingly slow process. As there are more than 100 elements in our Periodic Table, millions of recipes are possible and we have no pre-defined information to work with. The fact that scientists have managed to check only a few thousand blends in the last 50 years is self-explanatory. What’s more disappointing is that there are only a few mixtures that are stable enough to be used. Continuous efforts were being made to speed up this process and it seems as if AI has the answer to this long-lasting query.

A group of researchers working under the leadership of scientists, from the Department of Energy’s SLAC National Accelerator Laboratory, the National Institute of Standards and Technology (NIST), and Northwestern University, has proposed a solution where discovery and improvement of metallic glass are possible at a much swifter rate. The associated cost will also drop dramatically as the entire procedure is to be automated using techniques of Machine Learning and High Throughput Experiments.

They managed to discover 3 new amalgamations of elements that yield metallic glass and it took them 200 times less time than the traditional method. Chris Wolverton, a Northwestern Professor who is a pioneer of using AI to predict new materials and the co-author of the study, said,

“It typically takes a decade or two to get a material from discovery to commercial use. 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.”

20,000 combinations were tested in a year comparing to 6,000 in a span of 50 years.

A system at SLAC’s Stanford Synchrotron Radiation Lightsource (SSRL) aided this team to make this amazing discovery. It has the ability to combine Machine Learning with those experiments that can make and screen hundreds of sample materials at a time. Wolverton also mentioned that the ultimate goal of their research is to develop a system which should scan hundreds of samples and provide immediate feedback by making use of the machine learning algorithms.

The system will also allow the researcher to test another set of samples the coming day or in an ideal case, in the next hour. Another statistical information came from Apurva Mehta, a representative of SSRL, who claimed that they tested 20,000 combinations in a year. In comparison, only 6,000 blends were tested in the last 50 years. He also explained what they did differently to come up with such results. He said,

“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.”

The feature that adds more weight to this technique is its diversity. It can be used for all kinds of experiments including those where researchers lack theoretical evidence to support their claims. As machine learning is used, all the connections are made automatically and research is expected to progress in unexpected directions. A lot of conventions, followed for years, might come to an end following this invention. Jason Hattrick-Simpers, a Material Research Engineer at NIST, said,

“AI is going to shift the landscape of how materials science is done, and this is the first step.”