White Paper - Parallel cells vs. strings in parallel
Cells directly in parallel work better, are more reliable
Given a numebr of cells in a battery pack (such as 100 cells), they can be arranged as sets of cells directly in parallel, which are then connected in series (such as a 50S2P battery), or as strings of cells in series, which are then connected in parallel (such as 2 x 50S1P).
The first approach (cells directly in parallel) is in general far superior, leading to better reliability. It also results in a lower BMS cost (the BMS must monitor 50 voltages in the first approach, 100 in the second approach).
Some times battery designers decide to use multiple strings which are then connected in parallel, because they think that doing so has advantages:
At first sight, all those reasons appear to make sense, but a deeper understanding of how real world cell work, and how users truly behave, will show you that it is much better to connect cells directly in parallel instead.
This page has only an overview of the issues. For an in depth analysis, please see section 18.104.22.168.1 , "Cells in parallel versus batteries in parallel" of the Battery Management Systems for Large Lithium-Ion Battery Packs book.
In the real world, cells have variance (in capacity, resistance), and a few cells may actually be "bad", in the sense that they have significantly lower capacitance and/or higher resistance.
One such cell in a string will obviously limit the performance of that string; but also it will incapacitate the entire pack (including the other, good strings), because the variations in the pack voltage under load will be applied entirely to the bad cell, resulting in either a fire danger (no BMS) or immediate shutdown (with a BMS).
Conversely, when connecting cells directly in parallel, such a bad cell will have a minimal effect; simply put, that is because the other cells in parallel with it will carry the weight of the bad cell.
The idea of carrying just as many batteries as you need on a particular day sound appealing, until you realize that you cannot connect batteries in parallel willy nilly. If the batteries are not at the same State Of Charge, the moment they are connected in parallel, there will be a very large dump of energy between them, and the resulting current could cause significant damage.
You can only "hot swap" batteries if the BMS is able to track the SOC and voltage of each battery, and if each battery has a contactor that the BMS can engage only when the new battery's voltages is the same as the rest.
The idea is that, should a battery fail, it can simply be taken out of service, and the rest of the batteries can continue powering the product. This idea has some merit, but with some limitations.
Redundancy does not improve reliability (as one would expect) because reliability is compromised by having strings in parallel, as explained above.
In our experience, cells do not suddenly go bad; instead, their performance may be gradually reduced over time. From that point of view, redundancy will not improve reliability.
It would appear that in case of failure, the user may be more likely to seek service (during which the bad battery is repaired) instead of disconnecting the bad battery and continuing on. So, the ability to disconnect a battery, does not imply that the user would take advantage of such a feature.