Elsa Olivetti1
Massachusetts Institute of Technology1
Elsa Olivetti1
Massachusetts Institute of Technology1
Given the heuristic nature of experimental materials development, it can be difficult to gain a complete awareness of the challenges that may ultimately jeopardize a given lab-scale materials and device design during the transition to manufacturing. Despite this, even small amounts of additional understanding could meaningfully streamline scale up, particularly relevant for energy storage applications that must scale quite rapidly. This presentation will outline two data-driven approaches to inform discovery through natural language processing (NLP) approaches. NLP offers a powerful way to extract information from the rich and diverse reservoir of published, peer-reviewed literature in precisely targeted ways. First the work will describe use of automated keyphrase extraction, coupled to full-text classification to discover critical processing-outcome relationships. Second, once key manufacturing scale up barriers we use similar methods to identify possible mitigations to in-domain processing challenges may be found in out-of-domain research. The proposed methodology here rests upon representing whole documents with title-, abstract-, and citation-based embeddings. This approach has been applied to solid state battery materials development.