Due to their high theoretical specific energies, lithium-sulfur (Li-S) batteries have drawn significant attention as one of the promising chemistries to replace commercially used Li-ion batteries. However, in Li-S batteries attaining high capacities and long cycle lives are severely hindered by major challenges. Insulating nature of sulfur is one of the main problems that results in low sulfur utilization and requirement of electronically conductive materials in the cathode. In addition, escaping polysulfides from the cathode, in other words the polysulfide shuttle mechanism, result in sulfur loss and lithium corrosion. Hence, efficient material selection and cell design are crucial for attaining high performance Li-S batteries. Even though the battery performance depends on these materials and cell design factors significantly, the critical link between these variables and the performance is not clear. To identify this link, machine learning is a highly effective tool, especially for such complex systems with many variables and large dataset.1 In this study, we use machine learning, association rule mining (ARM) specifically, to determine the critical materials and cell design factors in Li-S batteries providing high performances at prolonged cycle lives. First, a comprehensive literature search is performed and the experimental articles are collected randomly. 353 papers accounting for almost 10 % of the literature, published between 01.01.2020 and 18.07.2018(search day), are used to collect 1660 experimental data. As performance indicators, peak discharge capacity (PDC) and cycle number at which 80% of PDC retained are chosen. 19 important factors, such as anode, encapsulation and conductive materials and their wt.% are determined and extracted from papers. This dataset is analyzed via ARM, which provides single factor associations by presenting three parameters as results: support, confidence and lift. Lift is the ratio of the fraction of a specific factor with high PDC to the fraction of that specific factor in the total data. In our analysis, lift value is chosen as the performance metric to compare the results since it provides both positive and negative correlations of factors with the output (lift values greater and lower than 1 show positive and negative correlations, respectively). The change of lift values with increasing PDCs and cycle numbers are presented as bubble graphs, where bubble size shows the number of datapoints obeying that rule. In addition, other analyses are performed to investigate the rules for high energy density cells by individually restricting the minimum sulfur loading and the maximum electrolyte-to-sulfur (E/S) ratio to 5 mg cm-2 and 5 mL g-1, respectively, for PDCs higher than or equal to 1000 mAh g-1.Figure 1 shows the change of lift values for the encapsulation material type with increasing PDC and cycle number limits. Figure 1a clearly shows the significance of the encapsulation strategy since no encapsulation case has lift values around 0.5. Porous carbons, collection of infrequently used materials such as polyacrylonitrile defined as others, and CNT with additives show promise for PDCs higher than 1400 mAh g-1, since promising factors for high performance should have increasing lift trends with increasing capacity or cycle number limits. Moreover, it is found that the encapsulation materials wt. %’s should be higher than 40% for improved cell performances, which may be due to enhanced sulfur utilization and suppressed polysulfide shuttle mechanism. According to the rest of the ARM results (not shown here), ethylene carbonate:diethyl carbonate (EC:DEC) or tetraethylene glycol dimethyl ether (TEGDME) and LiPF6 salts are proposed to be very efficient as electrolyte materials especially for lean electrolyte conditions. Furthermore, polytetrafluoroethylene (PTFE) and polymer n-lauryl acrylate (LA) are shown to be efficient binders for high PDCs. Similar results are found for the maximization of the cycling performance of the batteries. To sum up, in this study, it was concluded that novel electrode and electrolyte materials are essential for reaching high capacities at prolonged cycle lives.Figure 1. Lift vs. PDC (a) and cycle number graphs (b) for encapsulation type. References A. Kilic, Ç. Odabaşı, R. Yildirim, and D. Eroglu, Chem. Eng. J., 390, 124117 (2020) Figure 1
Read full abstract
Jan 1, 2022
Jiaming Dou + 2
Open Access
