Skeletal-based microstructure representation and featurization through descriptors

Modeling process–structure–property relationships using machine learning methods has become a valuable enabler for materials design and discovery. However, the machine learning models rely heavily on the featurization of the materials’ structure. This paper introduces a microstructure featurization framework to compute generic topological and morphological descriptors. The framework relies on our skeletal microstructure representation. The representation allows for the seamless calculation of topological descriptors, which is the main focus of this paper. To demonstrate the efficacy of our featurization framework, we couple it with a feature selection method to establish the structure–property model for organic photovoltaics (OPV). For this goal, we identify a salient set of descriptors and construct the structure–property map with high accuracy.