OK. They did replace an entire section, as I thought. Section 4 is the bottom one of two sections stacked under the rear seat. Section 5 sits on top of it, and usually runs warmer, per Torque Pro. Section 4 contains module 4, made up of "cells" 31-40, and module 7, made up of "cells" 59-66.
Making a 3 pouch "cell", a module, or a section of higher capacity would be a complete waste. They are all in series. The entire pack can only deliver the capacity of the weakest cell or cells in series. There is no cost effective way to bypass low cells, or transfer energy from higher to lower cells in a pack. The only solution is very consistent cells. From the data scans I have seen online, the thousands of cells in a Tesla pack are more closely matched than the much larger cells in other manufacturers packs. Seems counter-intuitive, but I suspect the much more automated small cylindrical cell manufacturing systems are easier to control for consistent results.
It's not that the 2170 cells in a Model 3 are intrinsically more reliable than the pouch cells other folks use, I'd argue it's the configuration, and the trade-offs that configuration coices bring.
- The Tesla Model 3 has 4,416 cells wire in a 96s46p configuration (split across four modules, two with 25 bricks of 46 cells and two with 23 bricks of 46 cells).
- The Bolt has 288 pouch cells in 96s3p configuration — 96 cell-groups connected in series where each cell-group containing 3 cells wired in parallel, grouped into 10 modules arranged in 5 sections with two modules per section; there are 8 modules that have 10 cells and two that have only 8 cells.
The key thing to note is that both the Bolt and the Model 3 wire together in series 96 groups of cells, in the Bolt they are cell-groups of three cells and in the Model 3 they are in bricks of 46 cells.
There are trade-offs to each arrangement. I'll explore this. (FWIW, the math I'll be using is just basic probability and binomial distributions, so if you know this stuff you can recreate it yourself easily enough, and if you don't, you can just accept that the math works.)
Let's imagine that the our battery cells (both pouch and 2170 cylindrical cells) have a 1 in 1000 chance of being a poor-performing “weak cell”.
If GM sells 10,000 Bolts, the relevant math tells us that, 7497 will have no weak cells at all, 2161 will have one, 310 will have two, 30 will have three, and two will have four. The average number of weak cells per Bolt is 0.288 with a standard deviation of 0.536. In other words, most Bolts have no weak cells, but a significant fraction of Bolts have a few; in the 95th percentile, we see two weak cells, and in the 99th percentile we see one weak cell.
If Tesla sells 10,000 Model 3 vehicles, we find that only 121 will have no weak cells at all, and one vehicle will have 14 weak cells. On average, a typical Model 3 will have 4.416 weak cells with a standard deviation of 2.100; in the 95th percentile, we see 8 weak cells, and in the 99th percentile we see 10 weak cells.
This seems bad for Tesla, but let's imagine that for a cell group
(a.k.a. brick) to be bad, more than one third of its cells need to be weak.
For a Bolt, the chance of having a three-cell group where two or more cells are weak is 1 in 333556 (if the odds of a single cell being weak is 1 in 1000). It's that same chance each of the 96 cell groups. If GM sells 10,000 cars, there will actually be three cars sold where this happens.
For a Tesla, the chance of having a 46-cell group where 16 or more cells are weak is 1 in 1037474903308485595872778813499779305. Tesla will need to sell 10 million billion billion billion cars before it sells one with a weak cell group.
Thus wiring up 46 cells in parallel is going to be better for handling weak cells than wiring up three. (Perhaps that's obvious without doing any math!)
Now let's assume there is a 1 in ten million chance that a manufacturing defect will cause a cell to explode.
If Chevrolet makes 10,000 Bolts, they'll use 2,880,000 batteries and most likely no Bolts will explode, although there is still a 25% chance that we will see catastrophic failure.
If Tesla makes 10,000 Model 3 vehicles, they'll use 44,160,000 batteries, and about four of those Teslas will explode.
In general, this means that to have the same chance of exploding as a Bolt, Tesla has to make its 2170 cells so they are more than fifteen times
less prone to catastrophic failure than pouch cells. I suspect they can't do this.
So, overall in the space of trade-offs, a Bolt owner is more likely to have to bring their car into the repair shop to get a module swapped out due to weak cells, and a Tesla owner is more likely to have their car suddenly catch fire.
It also shows that the reliability of cells needs to be really high. We want a really
low chance of catastrophic failure (even better than 1 in ten million, especially for Teslas), and the less chance of a weak cell, the better (especially for Bolts, less so for Teslas).