2019/09/06
Japan Science and Technology Agency (JST)
Keio University
The University of Tokyo
Under the JST Strategic Basic Research Programs, a research group from Keio University's Faculty of Science and Technology, led by Associate Professor Yuya Oaki and graduate student (at the time) Hiromichi Numazawa, in collaboration with Assistant Professor Yasuhiko Igarashi and others from the Graduate School of Frontier Sciences at The University of Tokyo, has used materials informatics (MI) to establish a new design principle for organic materials to be used as anodes in lithium-ion secondary batteries, and has succeeded in obtaining high-capacity, high-durability materials with an extremely small number of experiments.
To conserve resources in batteries, organic materials that do not use metals are being studied worldwide. The search for anode materials for batteries such as lithium and sodium batteries has traditionally relied on researchers' trial and error, experience, and intuition.
MI is generally a method that applies machine learning to large-scale data (big data) to reduce the involvement of researchers' experience and intuition. A challenge has been how experimental scientists can leverage their own small-scale data and empirical knowledge.
The research group has been studying an "experiment-driven MI" method that fuses small but relatively accurate experimental data with the experience and intuition of experimental scientists, and has previously achieved successes such as improving the yield of nanosheet materials.
In this study, the group first measured the capacity of 16 organic compounds as anodes and extracted the few factors determining capacity using sparse modeling, a data science method. Based on this learning result, they constructed a capacity prediction formula (prediction model) using the extracted factors as variables. Next, incorporating the researchers' experience and intuition, they selected 11 commercially available compounds from a pool of such compounds that were expected to have a certain level of capacity as anodes, and calculated their predicted capacity values before conducting experiments. When they measured the capacity of the three compounds with the highest predicted values, two of them showed high capacity. Furthermore, by polymerizing one of these, a thiophene compound, they were able to obtain a polymer anode material with improved capacity, durability, and high-rate charge/discharge characteristics.
The design principle for organic anode materials established in this study is important for aiming for further performance improvements. Furthermore, the success in discovering high-performance materials by fusing a small experimental dataset, researchers' experience and intuition, and machine learning has demonstrated the effectiveness of integrating experimental science and MI as a method for streamlining material discovery.
The results of this research will be published on September 6, 2019 (Germany time) in the online early view of the international scientific journal "Advanced Theory and Simulations".
For the full press release, please see below.