26650 LiFePO4 8S battery pack with BMS (WHL #27)
This weeks article is another by-product of the ADC testing (WHL #26) from last week. Could’ve packed in there, but I wanted to share some remarks on Chinese battery technology…
So first of all, this isn’t a piece of kit that I bought myself. After testing two 12V/2.2Ah lead batteries from work that were sitting unused on a shelf for five years, I needed another victim. Those quality Panasonic cells were flat as Kansas – the LC-R122R2PG datasheets recommends 8.6V cutoff voltage at 3C = 6.6A or 9.7V at 1C. Fully charged, they delivered 1C at 7V or 1.5C at 2.5V at full charge – slow 10.5V discharging to 0.1C cutoff and recharging didn’t help much. Sadly, those were ready for the bin.
These LiFePO4 cells however were purchased just earlier this year after making contact with the seller at the Embedded World fair in Nürnberg, Germany (I think…or was it at electronica?). We’re trying to make some device go cordless for a short amount of time, and as the power consumption isn’t something you can buffer with supercaps, we needed proper batteries. My partner demanded lithium iron phosphate cells instead of the classic lithium cobalt oxide (LiCoO2) type for safety reasons, as our gear gets abused a lot. I’m amused by the imagination of a violent explosion plus the chemical fire after someone beats the living crap out of our device, but I have to agree that, in terms of loss of reputation and compensation money, a device that just dies quietly is favourable to one that shreds your terrified face – even if you’re an asshole that destroys tools so that you don’t have to work until replacement arrives.
Anyway, these cells were samples for evaluation, and I think they cost around 50 USD per unit (2 units ordered), mostly for commercial shipping with tracking. We specifically asked for a battery pack voltage around 24V (as our main power supply) and cells that could deliver 60W for short amounts of time. So they offered a 8S 26650 pack – LiFePO cells have 3.3V nominal and 3.6V charging voltage, meaning 26.4V nominal and 28.8V fully charged. Adding some regulators could easily make that a 24V battery pack, as required.
We actually didn’t specify battery capacity, but instead asked the company to pick suitable cells for us. Like in LiCoO cells, there’s always the trade-off between maximum capacity (for low currents = higher internal resistance) and slightly lower capacity with the ability to deliver considerable amounts of power whenever needed. As I have no experience with LiFePO cells, we probably were led to buy some above-average batteries, but whatever. These were samples and even in production, optimizing cost for two-digit numbers of battery packs per year doesn’t really pay. Find something that works and go with it until you’re forced to pick a replacement.
What does it look like? Well, shabby. I don’t have any photo of the assembled unit, but instead I will show you the outer shell:
Is that some sort of PCB material?
Thickness is 1.0mm.
So our friends in the China factory slapped together a battery that is encased by PCB production waste and loads of tape. Is the inside any better?
Good Lord…
I have to say, however, I did already modify that thing. Count the number of cells…one was dead at 0.7V. It still happily delivered 100mA (hats off to that!), but the entire pack had a weird charging behaviour and obviously didn’t deliver more current than this weakest of cells. So I removed it and rewired the battery tapping wires for the BMS accordingly. Trust me, it didn’t look any better before that. It really didn’t…
There’s nothing much going on the back side, just a very wide tinned power trace and some tiny traces that didn’t fit the upper side of the main board. I didn’t bother to take that apart, but have a look at the photo – the daughter board just has paralleled 200Ω input resistors for each battery string, a SOT-23 transistor that taps off to the next string, and most likely some op-amp in the SOT-153 case besides that, maybe for detecting under- or overvoltage situations. Not sure what’s happening below, as there are another three lines of paralled-up strings of transistor thingies. But there has to be some cell cutout circuitry, as I’ve seen the steep drop from 18.xV to 16V when the weakest of cells gets removed. And as there are only three pins feeding back to the main board (the upper two are shorted), I don’t think that is happening on the top board. Overvoltage protection that results in a disconnect from the input – yes, likely.
The more beefy DPAKs on the left share the drain body contact, but have different parts of the circuit to drive their gate. The one on the bottom (“Silkron” brand SSF6808D) obviously carries the (-) input from the first cell to the source pin (15 mΩ sense resistor – how does that work when you remove individual cells?) and the upper DPAK connects to the (-) wire that goes outside the box, doubling as reverse polarity protection with the intrinsic body diode. It’s an AUFR3607 part with International Rectifier logo and date code 139A = 39th week of 2011…not sure if that is genuine. Why would you put two different N-Channel MOSFETs on a board and buy one Wan Hung Lo cheapie and choose an automotive grade part (AUIRFR3607) for the other one? That makes no sense. Both have similar specs, IDrain of 79A/80A (silicon limited – that’s way above package limit for these DPAKs), VDSS of 68V/75V and RDSon (typical) of 6.1mΩ/7.3mΩ. Why?
On a side note: One of the SOT-23 transistors on the daughter board is toast (the one with the largest amount of solder flux residue and glue). Complete short-circuit of all pins. Guess that defaults to “always on” and helps a lot in protecting cells. Oh well, fits the picture of the high quality “BMS” nicely.
The least it could do is perform well. Here’s the data, plotted against time:
And in the more common form, plotted against capacity:
Again, as this is a CC discharge, there’s a fixed conversion factor between runtime and capacity. Not doing CC results in arbitrary discharge profiles that are nowhere reproducible, especially not with simple tools such as this one.
Does it perform well? I’d say: Sort of. The trace looks very much like the usual datasheet ones, and this is actually the first graph of mine that does this (so proud my little electronic load works!). However, if you had asked me on the day that I took this data, I would have noted some concerns about calibration of the current reading. After all, recording wrong current (by a fixed factor) will greatly ruin your totals, while having a slightly wrong cutoff voltage will not add (or subtract) that much capacity – have a look at the steep roll-off beyond 2.9V. That’s basically the same that Dave Jones told us about the 0.8V / Batteriser scam – below a certain discharge voltage, the energy left in the battery is miniscule and there’s no point in trying to squeeze out every last bit of juice.
The totals of this run were 2106 mAh capacity and 45105 mWh of energy delivered until the cutoff at 19.707V / 2.815V per cell. And that’s why I’d suspect a calibration issue: These are Wan Hung Lo cells that don’t even have laser-etched or printed manufacturer data below the blue protective/isolating tubing. But they are fresh and somewhat at full capacity – if they claim 3200 mAh, why would they deliver just barely above 2100 mAh? How would this pass the sample testing at the customer and lead to an follow-up order?
Turns out my cal is okay…but another cell is faulty. It’s entirely possible that it was damaged when removing the first battery, as it is an adjacent one – but soldering with my Ersa iron is a breeze and the spot-welded tabs (clearly a manual job!) on the batteries plus their huge thermal mass should have prevented any damage. But I can’t say for sure.
The cell was at roughly 2.5V when I tested the “fully” charged unit three days later. Contrary to the first one, this one totally went down when asking for current – it went down to some dozens of millivolts when connected to a 1kΩ resistor. Flat.
Now after I’ve removed the cell, I’m left with a 6S battery that now has two cells going to full 3.6V when charging, and four that stay at 3.3V to 3.4V. Before that, they all were at 3.4V except one that limited charging by going to 3.6V too quickly. As this BMS doesn’t really allow for a proper balancer to charge cells individually, this is the best that I can do when using comparable equipment to that in our testing unit at work.
And what do you know, I’ve run once again and this time the battery clocked in at 2877 mAh at the same current as before. Sure, total energy was around the same because it is only 6S instead of 7S (yet more amp-hours) but decent capacity was there. If only I hadn’t messed up the timing of the serial logging, which caused the Arduino serial monitor to overflow and lose my results. Well, charge once again and run another one-hour discharge cycle…and another one:
Huh?
Another one dead? C’mon! 
I’ve seen the BMS cut off one cell when it drops below 2.5V, regardless of current. That’s not dandy, even though e.g. 26650 LiFePO cells from the long gone A123 Systems were specified down to 2.0V:
(roughly same scaling for comparison)
So what am I supposed to do? Remove another cell and try again? I’ve been extra careful when doing the last extraction, I even put my thumb on the cell itself when soldering the metal band between the two cells. There’s no way the cell was damaged in the process – and the one that has had multiple neighbours soldered on and off still works fine. This has to be a batch of shitty cells from a Wan Hung Lo factory far far away…or is it the crusty BMS that does some weird things?
I guess I will try again in 5S configuration, and test the removed cells individually. A battery holder for 26650 is on its way and I will add data to the review of that unit. Until then, the pack charges again…that’s what you get when trying to do two similar discharge cycles and prepare a filler post on a lonely evening.
One thing is for sure: The supplier of this sample pack won’t get any future orders from us 
