Data-Driven 21700 Lithium-Ion Cell Sorting and Fair Comparison of UAV Battery Pack Configurations

This paper presents a reproducible data-driven framework for selecting 21700-format lithium-ion cells and interpreting UAV battery-pack configurations. The study uses 289,132 sorting-machine measurement records, of which 288,076 OK records were retained after filtering, and six charge-discharge XLSX files with two different voltage windows. The main scientific question is how large-scale internal-resistance and capacity-screening data can be converted into pack-selection criteria without confusing voltage level with total cell count and stored energy. The results show that internal-resistance screening is useful for cell grouping, but the apparent advantage of 8S and 12S configurations must be separated from the effect of additional cells, higher pack energy, and mission-level mass. A simplified pack-level model and stochastic grouping layer are used to compare same-parallel-count and same-total-cell-count cases. The extended 4.50/1.70 V voltage-window tests are treated as boundary or destructive-validation data and not as a recommended reusable UAV operating regime.