AMP_04_May_June_2021_Digital_Edition

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 Y / J U N E 2 0 2 1 7 MACHINE LEARNING | AI MACHINE LEARNING BUILDS BATTERY KNOWLEDGE For the first time, an approach known as scientific machine learning has been applied to battery cycling, says Will Chueh, an associate profes- sor at Stanford University and investiga- tor with the DOE’s SLAC National Accel- erator Laboratory who led a new study on the topic. Instead of using machine learning to speed up scientific anal- ysis by looking for patterns in data, as researchers usually do, Chueh’s team combined it with knowledge gained from experiments and equations guid- ed by physics to understand a pro- cess that shortens the lifetimes of fast- charging lithium-ion batteries. Chueh says the results overturn long-held as- sumptions about how lithium-ion bat- teries charge and discharge and give re- searchers new concepts for engineering longer-lasting batteries. The research is the latest result from a collaboration between Stanford, SLAC, Toyota Research Institute (TRI), and MIT. The goal is to bring togeth- er foundational research and industry knowledge to develop a long-life elec- tric vehicle battery that can be charged in 10 minutes. “By understanding the fundamental reactions that occur with- in the battery, we can extend its life, en- able faster charging, and ultimately de- sign better battery materials,” says Pat- rick Herring, TRI senior research scien- tist. www6.slac.stanford.edu . DEEP LEARNING IMPROVES MICROSCOPY A novel tool based on artificial in- telligence (AI), recently developed at the University of Gothenburg, Sweden, is offering improved ways to analyze images taken with microscopes. Now, a new study published in Applied Phys- ics Reviews titled “Quantitative Digital Microscopy with Deep Learning” shows that the tool could fundamentally change microscopy and pave the way for new discoveries in both research and industry. The work focuses on deep learning, in which a mathematical mod- el is used to solve problems that are dif- ficult to tackle using traditional algo- rithmic methods. A neural network is used to retrieve useful information from a microscope image. Courtesy of Aykut Argun. Researchers incorporated scientific insight into machine learning for battery research, an approach that will speed development of batteries for electric vehicles, laptops, the grid, and more. Courtesy of Greg Stewart/SLAC. Researchers at Osaka University, Japan, used machine learning to design new polymers for use in photovoltaic devices. After screening more than 200,000 candidate materials, they synthesized one of the most promising and found its properties to be consistent with their predictions. The team believes this work may revolutionize how functional materials are discovered . www.osaka-u.ac.jp/en . BRIEF This non-fullerene acceptor solar cell device includes a polymer designed by machine learning. Courtesy of Osaka University. In microscopy, the main challenge is to glean as much information as pos- sible from the data-packed images, and this is where deep learning has prov- en to be very effective, the team says. The new tool involves neural networks learning to retrieve exactly the informa- tion that a researcher wants from an im- age by looking through a huge number of images known as training data. The AI tool simplifies the process of produc- ing training data compared with having to do so manually, so that tens of thou- sands of images can be generated in an hour instead of roughly 100 in a month. www.gu.se.