Material-to-system analysis of lithium – sulfur and lithium – oxygen batteries

Developing beyond lithium-ion batteries (LIBs) is critical for fulfilling the high energy demand of future vehicles. In this respect, lithium-sulfur (Li-S) and lithium-oxygen (Li-O2) batteries were studied here to understand the effect of cell design and materials on the battery performance. The studies of Li-S batteries involve three main methods: experimental characterization, system-level performance modeling, and machine learning (ML). Instead, only ML models were developed for Li-O2 batteries. First, association rule mining (ARM) was utilized to understand the cell and materials design effects on Li-S battery performance using the data collected from the literature. For Li-S batteries, the encapsulation material type was the most important feature, and system-level performances were highly improved with ionic liquid (IL) electrolytes. Next, polysulfide (PS) solubility in ILs and IL properties were determined for 36260 ILs using the COnductor-like Screening Model for Realistic Solvents (COSMO-RS) calculations. 6 ILs were experimentally tested to determine the solubility ranges of ILs for high-performance Li-S cells. Afterward, extreme gradient boosting (XGBoost) and ARM were utilized to predict and identify promising ILs and their features. Imidazolium cations with either borates or bis_imide anion group were the most promising IL pairs. Encapsulation cathodes were also studied. UiO66/Graphene nanoplatelet (GNP) composites increased the rate performances of Li-S cells where 50 wt.% UiO66 loading was the optimum for balancing electron transfer and PS chemisorption. Vanadium and Cobalt-doped ketjen black cathodes increased the system-level performances of Li-S batteries. ARM was also used for Li-O2 battery literature data; metallic cathode ingredients LaFe oxides, Ni oxides, and electrolyte solvents significantly impact battery performance. Last, gas solubilities of nearly 30,000 ILs were calculated using COSMO-RS, and random forest (RF) was used to build predictive models. Anion descriptors were more determinative of the gas solubilities.