Accelerated discovery of metallic glasses through iteration of machine learning and high-throughput experiment

Fang Ren, Logan Ward, Travis Williams, Kevin J. Laws, Christopher Wolverton, Jason Hattrick-Simpers, Apurva Mehta; Science Advances. 2018, 4: aaq1566

Ever wonder how Superman gets his strength of steel?  While that question may never be answered, recent leaps in discovery have been made about creating metallic glass with extremely strong and wear resistant properties.  To unlock these discoveries, SLAC researcher Fang Ren and colleagues (2018) have used complex modeling to both identify and test new metallic glasses (MGs).  Using machine learning accelerated the searching process more than 20 times and over a 100-fold increase in MGs synthesis.

Metallic glass is an alloy, or a combination of metals, that has an amorphous configuration rather than an ordered crystal structure.  Limitations in identifying new MGs come from several avenues such as material expense and toxic properties.  However, using machine learning (ML) helps circumvent some of the problems of identifying good MGs.

The process of ML is iterative, meaning the model builds off of previous versions creating a more precise final set of possible MG candidates.  To create their final model, the researchers used three different modeling approaches in the training system.  Furthermore, candidate possible MGs were tested using high-throughput experiments to identify actual glass/not glass properties of the materials.

In particular, the research focused on three metals (cobalt, vanadium, and zirconium) to build a model with viable MG in the Co-V-Zr ternary.  The results combined with the iterative ML process lead to a refined model that discovered two new additional ternaries.  Together the theory and experimentation identified new glass forming regions.

The authors suggest that their method could be far reaching to a wider range of materials.  Ultimately, this work in combination with potential future applications of the method can have great implications to future technological advances that could take advantage of the new substrates.  Perhaps one day we may be able to create an armor that can give us Superman powers.