Machine learning enabled high-throughput screening of inorganic solid electrolytes for regulating dendritic growth in lithium metal anodes

Abstract

The Li–S secondary battery system has gained popularity owing to its advantage of a higher specific energy compared to Li-ion batteries. However, it suffers majorly due to Li dendrite formation, and the composition of the electrolyte plays an essential role in the formation of dendrites. The mechanical properties of the solid electrolyte and associated interphases can affect the microstructural evolution of the battery electrode and, consequently, its performance. The application of improved electrolytes having a reduced tendency of dendrite growth could enhance the application of Li–S battery systems as next-generation energy storage devices. Herein we applied machine learning (ML) tools for the generation of parameters and predict the structure–property correlation of electrolyte molecules. We have employed the decision tree regressor, random forest regressor, bagging regressor, and gradient boosting regression. It is, however, observed that gradient boosting regression performed the best in deciding the molecular properties in relation to the electrolyte performance. A total of 9353 materials with known properties were used to train the model before it was applied to 49 154 unknown molecules to predict their desirable properties using 134 descriptors. Finally, the best performing model was applied to the unknown molecules, and 30 compounds were shortlisted with suitable shear and bulk moduli, which are expected to have a lower dendrite forming tendency. The predicted materials are mostly metal borates of d block and f block elements. Considering the cost issue of lanthanides, some of the transition metal borates are finally assumed to be more suitable candidates as electrolytes in Li–S battery systems. The model is essentially successful in generating the structure–property correlation from a number of descriptors and a large pool of data.