Mapping PV degradation mechanisms and field performance by leveraging large language models

DuraMAT 2.0

Central Data Resource

Category: Field Data Studies

Recipient LBNL(PI: Jain, Anubhav) Subs NREL

Status Awarded

Abstract The "holy grail" of forecasting is to anticipate the future degradation of new, current, and deployed module technologies. One promising method to advance this goal is to use physical and mathematical models to predict performance for specific module technologies in various environments. However, to use such models, one must first comprehensively map the degradation pathway, i.e., how environmental stressors activate various degradation modes in module materials and devices. Unfortunately, the information needed to develop and validate such pathways is typically scattered across thousands of research articles, in which each article may provide information about only a single material, degradation path, or field validation. This proposal will leverage large language models (LLMs) to accelerate predictive degradation pathway modeling. We use scientific literature data extraction to construct databases of (i) degradation pathways and (ii) field degradation. In addition, PVDeg software will be applied to compare forward modeling of degradation with statistical observations from field data.