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 | M A R C H 2 0 2 3 1 0 MACHINE LEARNING | AI Engineers at the University of California, San Diego developed an AI algorithm to predict the structure and dynamic properties of any material almost instantly. Called M3GNet, the algorithm was used to develop matterverse.ai, a database of more than 31 million notyet-synthesized materials with properties predicted by machine learning algorithms. ucsd.edu. BRIEFS MATERIALS INFORMATICS TACKLES POLYMERS Scientists at the Research Organ- ization of Information and Systems (ROIS), Tokyo, recently created a comprehensive database of polymer properties. “Materials informatics (MI), a new branch of materials research that combines materials data with data science, is gaining traction,” says assistant professor Yoshihiro Hayashi. “MI applies machine learning to predict new materials with innovative properties and their fabrication methods from a vast design space. As such, data is the most important resource in MI.” Despite the need, Hayashi says efforts to create a comprehensive database of polymer properties to enable data-driven research have fallen short. “To construct a database of polymer properties by molecular simulations, we developed RadonPy,” Hayashi explains. “It’s the first open-source software that successfully automates polymer physical property calculations using simulations of classical molecular dynamics based on atomistic models, which account for the behaviors and characteristics of individual constituents.” The program takes an assigned polymer and runs calculations to equilibrate it in prescribed system parameters. Once it does, it can then calculate the polymer’s density, radius of gyration, refractive index, thermal conductivity, specific heat capacities at constant pressure and at constant volume, among other information. RadonPy produces and stores the data, which can then be accessed later. The researchers also implemented a machine learning technique called transfer learning to correct biases and variations between the simulated property values and experimental data. Scanning electron microscopy images depict novel nanostructures discovered by AI. Researchers describe the patterns as skew (left), alternating lines (center), and ladder (right). Scale bars are 200 nm. Plots visualizing the distribution of 21 classes of polymer backbones according to specific definitions. Courtesy of npj Computational Materials. The team also identified eight amorphous polymers with high conductivity. Now, the group is using RadonPy to create the world’s largest open database of polymer physics, with more than 100,000 different polymer species. In addition to ROIS, three universities and 19 companies are partnering to jointly develop other databases with RadonPy for a variety of applications in academia and industry. www.rois.ac.jp/en. NEW NANOSTRUCTURES COURTESY OF AI Scientists at the DOE’s Brookhaven National Laboratory, Upton, N.Y., successfully demonstrated an AI-driven technique that led to the discovery of three new nanostructures, including a first-ever nanoscale “ladder.” The structures were formed by self-assembly. Staff scientists now aim to build a library of self-assembled nanopattern types to broaden their applications. “The fact that we can now create a ladder structure, which no one has ever dreamed of before, is amazing,” says group leader Kevin Yager. The team is now applying their autonomous research method to other classes of materials. bnl.gov.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==