Predicting Ionic Conductivity of Solid-State Battery Cathodes Using Machine Learning

Safer and more efficient battery technologies are in soaring demand, with a primary focus on transitioning from flammable lithium-ion batteries to non-flammable solid-state batteries. While solid-state batteries offer enhanced safety features, their power density remains a challenge due to poor ionic conductivity induced by non-optimal cathode microstructures. Laborious experimental processes and time- consuming data analysis algorithms are obstacles to establishing structure-performance correlation and optimizing cathode microstructure. In this paper, we present a machine learning approach to predict the current or resistance of a composite cathode based on scanning electron microscopy (SEM) images, given the inputs as a binary image, a voltage, and a conductivity value. Our results showed that current or resistance can be quickly predicted from input images with high accuracy.