Solving Materials’ Small Data Problem with Dynamic Experimental Databases

Materials processing is challenging because the final structure and properties often depend on the process conditions as well as the composition. Past research reported in the archival literature provides a valuable source of information for designing a process to optimize material properties. Typically, the issue is not having too much data (i.e., big data), but rather having a limited amount of data that is sparse, relative to a large number of design variables. The full utilization of this information via a structured database can be challenging, because of inconsistent and incorrect reporting of information. Here, we present a classification approach specifically tailored to the task of identifying a promising design region from a literature database. This design region includes all high performing points, as well as some points having poor performance, for the purpose of focusing future experiments. The classification method is demonstrated on two case studies in polymeric materials, namely: poly(3-hexylthiophene) for flexible electronic devices and polypropylene–talc composite materials for structural applications.