I don't know array size or serviceability of cells, but I would remove all swollen cells first.
"Good" Batteries are balanced for mAh <1% when new. This imbalance obviously increases with aging such that the weakest cell dies 1st due to more rapid aging from rising ESR during charge and discharge and more from undercharge condition and cold temps.
After-charging, ESR values rises very slightly with each charge cycle then rapidly towards End of Life EOL, causing excess heat and bulging. This depends on CC rates used and expected life cycles, that have a wide value depending on Depth of discharge , DoD i.e. 200 to 2000 cycles or more.
The job of the BMS is to extend the string battery life by preventing overcharge or undercharge during use (if capable).
Low power BMS products may not be able to bypass all the potential power during max current in the field with rapid charging i.e. V*I=Pd bypass, during CC mode and may rely on balancing during CV mode when current is declining.
e.g. If the BMS uses passive balance, and limited to 5W load resistor or 1.5A balance load and you are still charging in CC mode, > 1.5A , you are going to have infant failure mode.
State of the art ($$) BMS designs use patented SMPS methods with dual bridges and series choke to bypass cell charging current without much loss or have greater Amp bypass capacity for rapid charge.
I would suggest the way to identify the weak link is a low duty cycle high pulse current and store peak-peak voltage swings of each cell rather than average. This depends on the ability of the BMS to monitor Vbat (min,avg,max) in real time and store these values. Otherwise use a Vdc, Vac method of your choice for each node and determine where the Vac rises the most. and try to keep these balanced by logging and sorting and use a spreadsheet for each cell serial# or label# for quality control and improvement tracking.
I would expect LPF cells to be 3.2V nominal after charge and rated for 2k cycles before failure when used within specs.
All core Emus BMS functions only use the minimum, maximum, or average values rather than individual cell data, therefore only these aggregated values are normally stored during the periodic cell monitoring process. However, the individual cell data may still be useful for diagnostic purposes, therefore it can be acquired by request over USB, RS232, or CAN interfaces.
If you are not already using this software, I would get it. Then use the above method to obtain individual cell voltages using a pulse current test method at appropriate intervals to collect the data for all cells after initialization.