Investigating the effect of important cell design parameters on the performance of zinc-ion battery using machine learning

In this thesis, we use machine learning and data mining on data collected from the literature to improve our understanding of the effect of different cell design parameters on the performance of zinc-ion batteries. The collected dataset includes different cell fabrication parameters, and the performance of the cell is measured in terms of its peak discharge capacity and cyclability, which is expressed as the fraction of capacity remaining after 100 cycles. The dataset is used to train XGBoost and decision tree models for both performance metrics, dividing them into high and low classes. The XGBoost models show average accuracies of 0.693 and 0.632 on the 5-fold cross-validation for the capacity and capacity retention ratio, respectively. The full-dataset-trained models show F1 scores of 0.761 and 0.891, respectively. The feature importance analysis for the cell's capacity shows that the cathode group, the electrolyte salt, and its concentration are the most important, while the cathode material synthesis method and salt and mass loadings are the most important for the capacity retention ratio. The association rules indicate that vanadium oxides, including mixed metal oxides, perform the best for cathode materials. Cathodes utilizing manganese oxides show a slightly lower cell capacity. Zinc triflate, Zn(CF3SO3)2, as a salt and high salt concentration have a high association with high cell capacity and cyclability. The utilization of high conductive carbon fraction, 20% or slightly higher, is shown to be associated with high capacity. The fabrication of composite cathodes with nanostructured carbon materials like graphene and carbon nanotubes improves the cycle life but not the cell peak capacity. Additionally, modified zinc anodes with polymer coating or deposited zinc lead to a significantly higher cycle life.