high-throughput screening method for Li-ion batteries cathode material using Machine Learning

Typical metallic cathodes in Lithium-ion batteries contain Cobalt which results in high price, toxicity and high consumption of Cobalt contained materials whereas Cobalt is rare element considering its reserves. For this purpose, it is a great opportunity to explore and discover new materials to be experimented for lithium – ion batteries cathode. Density functional theory (DFT) which has been used in previous studies for material discovery, can contribute as a capable tool in multi-level screening process to select final candidates to be tested experimentally. Due to high computational cost of DFT, Machine learning (ML) can narrow down the space for DFT and experimental studies and play an important role in accelerating screening processes of battery materials. In this study, we want to represent an optimized screening process by developing ML approach. For this study, we used a sample containing more than 26000 datapoint to calculate reduction potential. This property is one of the most critical criteria for cathode selection, will be found for all organic materials based on their functional groups. Several experiments using different statistical and machine learning methodologies such as clustering, principal component analysis and stack learning have been used to optimize model performance. We were able to achieve R2Train =0.78, R2Test =0.74 and mean absolute error (MAE) of MAETest=0.388 V MAETest=0.422 V which is suitable for the first stages of the screening. Trained algorithm proves that it is decent to be used to calculate reduction potential for all organic substances.