Clustering recovered 18650 lithium-ion cells to improve homogeneity in second-life battery pack construction

Second-life use of lithium-ion batteries depends on assembling packs from recovered cells that behave as consistently as possible; however, recovered cells are often heterogeneous because of mixed models and different aging histories. This study aimed to improve pack-level homogeneity by clustering recovered cylindrical 18650 cells using laboratory-measured electrical variables and validating the approach through the construction and testing of a second-life pack. Cells were extracted from discarded laptop batteries, screened using safety and basic performance criteria, and characterized under a standardized charge-discharge protocol. For each retained cell, discharged capacity and internal resistance were measured, and health indicators were computed from nominal specifications and experimental results. Correlation analysis was used to identify the most informative variables associated with health status and to avoid redundant predictors prior to clustering. Clustering performance was evaluated across different numbers of groups using a cohesion-separation index, and two iterative selection algorithms were applied to select non-overlapping, highly homogeneous subsets of cells for series-parallel pack assembly. Using the same pack topology, the refined algorithm produced a second-life pack with higher usable capacity (8.98 Ah versus 8.72 Ah), lower equivalent internal resistance (76.38 mΩ versus 78.90 mΩ), and higher estimated maximum power (716.97 W versus 694.08 W) compared with the baseline algorithm. Overall, the results show that clustering guided by discharged capacity and internal resistance, combined with iterative refinement, enables the construction of more homogeneous and electrically consistent second-life packs from heterogeneous recovered cells.