January_AMP_Digital

A D V A N C E D M A T E R I A L S & P R O C E S S E S | J A N U A R Y 2 0 2 0 9 MACHINE LEARNING | AI of other compounds. The novel ap- proach with the machine learning em- pirical potential can be applied to new molecules in milliseconds, enabling re- search into a far greater number of com- pounds over much longer timescales. lanl.gov. MAKING POLYMERS WITH SOFTWARE AND ROBOTS A team of engineers from Rutgers University, New Brunswick, N.J., devel- oped an automated way to make poly- mers, offering an easier path to creating advanced materials aimed at improving human health. The innovation is a crit- ical advancement for researchers who want to explore large libraries of poly- mers, including plastics and fibers, for chemical and biological applications such as drugs and regenerative medi- cine through tissue engineering. While a human researcher may be able to make a few polymers a day, the automat- ed system—featuring custom software and a liquid-handling robot—can create up to 384 different polymers at once, a huge increase over traditional methods. Synthetic polymers are widely used in advanced materials with spe- cial properties and their continued de- velopment is crucial to new technolo- gies involving diagnostics, medical de- vices, electronics, sensors, robots, and lighting. “Typically, researchers synthe- size polymers in highly controlled en- vironments, limiting the development of large libraries of complex materials,” says assistant professor Adam Gorm- ley. “By automating polymer synthesis and using a robotic platform, it is now possible to rapidly create a multitude of unique materials.” Robotics has automated many ways to make materials as well as dis- cover and develop drugs. But synthe- sizing polymers remains challenging because most chemical reactions are extremely sensitive to oxygen and can- not be done without removing it during production. Gormley’s lab features an open-air robotics platform, which car- ries out polymer synthesis reactions that tolerate oxygen. The group devel- oped custom software that allows a liquid-handling robot to interpret poly- mer designs made on a computer and carry out every step of the chemical reaction. Another benefit is that the newly automated system makes it eas- ier for non-experts to create polymers. rutgers.edu. BUILDING BETTER BATTERIES WITH MACHINE LEARNING Even using the most advanced su- percomputers, scientists cannot pre- cisely model the chemical characteris- tics of every molecule that could serve as the basis of a next-generation bat- tery material. Instead, researchers at the DOE’s Argonne National Laborato- ry, Lemont, Ill., have turned to machine learning and artificial intelligence to dramatically accelerate the process of battery discovery. The team first creat- ed a database of roughly 133,000 small organic molecules that could form the basis of battery electrolytes. To do so, they used a computationally intensive model called G4MP2. This collection of molecules, however, represents only a small subset of 166 billion larger mol- ecules that scientists wanted to probe for electrolyte candidates. Because us- ing G4MP2 to resolve each of the 166 bil- lion molecules would have required an impossible amount of computing time and power, the team used a machine learning algorithm to relate the precise- ly known structures from the smaller data set to much more coarsely modeled structures from the larger data set. “When it comes to determining how these molecules work, there are big tradeoffs between accuracy and the time it takes to compute a result,” says Ian Fos- ter, director of Argonne’s data science and learning division. “We believe that ma- chine learning represents a way to get a molecular picture that is nearly as precise at a fraction of the computational cost.” To provide a basis for the machine learningmodel, Foster and his colleagues used a less computationally taxing mod- eling framework based on density func- tional theory, a quantum mechanical framework used to calculate electronic structure in large systems. Density func- tional theory provides a good approxima- tion of molecular properties, but is less accurate than G4MP2. Refining the algo- rithm to better determine information about the broader class of organic mole- cules involved comparing the atomic po- sitions of the molecules computed with the highly accurate G4MP2 versus those analyzed using only density functional theory. By using G4MP2 as a gold stan- dard, the researchers could train the den- sity functional theory model to incorpo- rate a correction factor, improving accu- racy while keeping computational costs down. anl.gov. An automated approach enables rapid exploration of new polymer materials. Courtesy of Matthew Tamasi. Scientists are using machine learning to find potential candidates for electrolytes used in lithium-ion batteries. Courtesy of Shutterstock/Cherezoff.