Materials Informatics for Polymer Materials: Experiments in The Loop

Materials informatics (MI) is to develop or study materials with an aid of informatics or machine learning. A typical approach is to train a black box model that relates basic descriptors (structure, composition, etc) and FoM (target properties) and predict or design a material with the largest FoM. At Thermal Energy Engineering Lab at University of Tokyo, together with the collaborators, we have been working on MI for heat transfer since 2015. One of the initial works was to design binary multilayered nanostructure to minimize or maximize thermal conductance by coupling thermal transport calculation and Bayesian optimization, which showed excellent efficiency. Later, the search space has been greatly expanded by utilizing quantum annealing. We have applied the methodology to computationally design and experimentally realize aperiodic superlattice that optimally impedes coherent thermal transport and multilayer metamaterial with wavelength-selective thermal radiation [1-8].<br/>More recently, we have extended the machine-learning approach to that for polymers, aiming to functionalize them in terms of the thermal and dielectric properties. When molding or compounding polymers, the final properties are quite sensitive to the process parameters, therefore, the above approach of serially connecting optimal design and experimental realization of materials is not sufficient. To this end, we have been developing a semi-automated MI system, where experimental fabrication and measurement are included in the optimization loop. This allows us to efficiently enhance the polymer properties. The MI optimization with experiments in the loop gives rise to various conceptual and practical questions; for instance, how to design and develop a robotic system that is adoptable to the existing material-development instruments and protocols, how to couple the property simulations and database with automated experiments, or how to appropriately define the search space for process parameters, etc. These aspects will be discussed in the talk.<br/><br/>[1] S. Ju, T. Shiga, L. Feng, Z. Hou, K. Tsuda, J. Shiomi, Phys. Rev. X 7, 021024 (2017).<br/>[2] M. Yamawaki, M. Ohnishi, S. Ju, J. Shiomi, Sci. Adv., 4, eaar4192 (2018).<br/>[3] A. Sakurai, K. Yada, T. Simomura, S. Ju, M. Kashiwagi, H. Okada, T. Nagao, K. Tsuda, J. Shiomi, ACS Cent. Sci. 5, 319-326 (2019).<br/>[4] R. Hu, S. Iwamoto, L. Feng, S. Ju, S. Hu, M. Ohnishi, N. Nagai, K. Hirakawa, J. Shiomi, Phys. Rev. X 10, 021050 (2020).<br/>[5] K. Kitai, J. Guo, S. Ju, S. Tanaka, K. Tsuda, J. Shiomi, R. Tamura, Phys. Rev. Res., 2, 013319 (2020).<br/>[6] S. Ju, R. Yoshida, C. Liu, S. Wu, K. Hongo, T. Tadano, J. Shiomi, Phys Rev. Mater. 5, 053801 (2021)<br/>[7] J. Guo, S. Ju, Y. Lee, A. Gunay, J. Shiomi, Int. J. Heat and Mass Transf., 195, 123193 (2022).<br/>[8] C. Lortaraprasert, J. Shiomi, npj Comput. Mater. 8, 219 (2022).