Predicting physical properties of NMCM cathode materials using machine learning guided DFT simulations

Abstract

With the rapid increase in development of electric vehicles and energy storage systems, the demand for long lasting batteries with high energy density is higher than ever before. A crucial aspect of the market-leading lithium battery is the longterm cycling performance – to perform with high capacity even after thousands of chargedischarge cycles with as small degradation as possible. One cause for this degradation is the occurrence of small micro cracks in the cathode material due to small volume changes during charge-discharge cycles. To suppress this effect, state-ofthe- art batteries today use metallic dopants such as aluminum in the cells of the cathode material. This project investigates other suitable dopants in NCM materials by implementing regression and gradient based prediction models on data acquired from supercomputer simulations using density functional theory (DFT). The results, while not fully conclusive, gives indications on what atomic features of dopants are interesting, as well as validates this relatively new machine learning approach in material science.