Modern machine learning methods open up opportunity to significantly speed up the material design process. Conventional approach of trial and error based on expert knowledge becomes very inefficient after many large material databases, such as the PolyInfo database in Japan, emerged. Instead of manually hypothesizing and testing different prediction models for new material discovery, machine learning is a promising tool to create effective empirical models that can be integrated into a data-driven material design framework. However, sufficient data may not be available for all desired properties during a design process. In this study, we will demonstrate a specific implementation to integrate information from existing physical models, various expert knowledge and the PolyInfo database to efficiently search for high thermal conductivity polymers. Our method aims at providing an end-to-end material design process that incorporates Bayesian inversion, machine learning models, and experimental design concepts.

Integrating prior knowledge and machine learning for efficient design of high thermal conductivity polymers

255th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.