Battery capacity measurement gives wildly varied results?

[Artlav]

I have accumulated (no pun intended) dozens of AA NiMH cells, and i wanted to know what state they were in, what they were basically capable of, and hopefully sort some good ones into a couple 12V packs by matching capacity.

For that i built a simple discharge logging rig - an atmega88 logs the battery voltages with it's ADC, and sends the data back to a PC.
Each battery is connected in series with a 3 Ohm resistor and a N-FET to cut it off once it reaches 0.8V.
Simple, right?

Well, several days later i don't think so any more.
The discharges give wildly different results from run to run.
Same cells can show up to half or twice the capacity from the previous runs.
Same cell can show a lot on run 1, a little on run 2, and a lot again on run 3, while the other in the same batch shows the opposite.
There is some consistency - like one weak cell always shows up somewhat behind the others (but vary by itself too).

Naturally, i suspected the rig.
It's on a breadboard, so there are many options for loose connections.
However, all the numbers seem to check out - the voltage drops are consistent, total of 0.05V difference between the resistor and the battery terminals.
The voltage drop across the resistors is consistent, meaning a consistent load.
There are no loose or unstable connections - an mV or 5 if i roll the battery inside a socket is the most i could produce, certainly nothing sufficient to account for tens of % of difference.
Finally, i checked the rig on a fresh set of alkalines, and got nearly-identical results on all cells.
The rig appears solid.
Are there any apparent flaws in it?

I suspected the procedure.
The batteries are always put into the same slot (and charger) on both charger and discharger each time.
The batteries are charged, then set aside for 5 to 10 hours to cool down and stabilise, then discharged, then cooled off for 30 minutes to several hours, and put back into a charger.
I tried to keep the times consistent for a couple of cycles i can pull off within one day (one discharge takes 6 hours), with no effect.
Are any of these times important?

The last suspect is the charger.
The chargers, two identical ones, are your basic chineese microprocessor-controlled 8 cells chargers with dV detection.
They neither look nor were especially cheap.

I figured out they can fully charge a cell, but by now the chargers are the only remaining suspect.
Is it at all plausible that this kind of a charger can partially charge a cell just like that?

All in all, i'm no longer sure what else can be wrong.
Anyone have any other ideas?
The only remaining assumption is that a NiMH cell should always have the same capacity under the same load, at least within a few cycles.

[ruffy91]

It's possible that the chargers dV detection is really sensitive and has false positives (especially when the signal/noise ratio is low and the dV only is 2 or 3 lsb of the ADC) and stops charging to soon.

Edit: See here for reference. Especially the zero slope method. http://www.prc68.com/I/BatChg.shtml

[bfritz]

I agree that the charger, terminating charge early, is a possible culprit.  The fact it is looking for a zero slope or negative slope, depending on the algorithm in the charger, could be a factor.

You will see some change is capacity as a function of temperature.  For example at 23C, you may get as much as 5% less than at 30C.  The bigger effect is at cold.  At 0C you may get only 30-40% of the capacity available at 23C.  So if you are doing this outside in a garage, and it gets cold in the garage during the night, that might be influencing your error.

It sounds like you are only monitoring time, and not current.  This could be the issue, coupled with using the proto board.  One possibility, is that the resistance of the voltage drops between the proto board and wiring connections are changing, as the cell discharges.  This would change the total current the cell sees during a time period, and impact results.

[HKJ]

I will recommend you try with another type of charger, some chargers are very unstable with their termination.

[nuno]

The discharges give wildly different results from run to run.
Same cells can show up to half or twice the capacity from the previous runs.
Same cell can show a lot on run 1, a little on run 2, and a lot again on run 3, while the other in the same batch shows the opposite.

Looks like bad cells to me 

I see you have it on a breadboard, maybe you have some unintended voltage drop in there. Will it be good if testing only 1 cell at a time? Other that than I would also suspect the "trigger condition" as already mentioned by previous posters. Should wait to have X cosecutive readings below threshold before deciding it's done.

[ConKbot]

IIRC the MH-C9000 is a goto recommendation for quality battery chargers on candlepowerforums for Ni-MH cells
http://www.amazon.com/PowerEx-MH-C9000-WizardOne-Charger-Analyzer-Batteries/dp/B003DIGKOG

Pricy, but it also does conditioning, discharge and capacity testing too. 

There is also the MH-C800S or MH-C801D which charge 8 batteries at a time. The 801 charges faster than the 800, but both support conditioning (discharging the cell fully before recharging it)

[David Hess]

As mentioned above, NiMH charge termination is finicky and this is especially the case at low change rates with dV/dT=0 termination so if you have a programmable charger, try using a higher charge rate like 4 hours minimum.

[Conrad Hoffman]

Certainly the charging issues above. Also, 3 ohms is a heavy load so the end cap connections can have a big influence if they have any residual resistance. You need very good battery holders to get consistency.

[Artlav]

Ok, i got the Maha MH C808M charger, as recommended on candlepower forum.
It's apparently the same as MH-C800S/MH-C801D, only for 8 cells, and get good reviews.
It charges at about 0.75C (2A rated), compared to the older ones at about 0.2C (took 5 hours to charge).
Also, i'm going to make a proper PCB for the rig, to rule out all sort of breadboard issues.

Some more cycles in, it would appear that one of the older chargers give much more consistent results than the other.
Even more curiously, the worse one have something rattling on the inside.
Hm...

So, i'll cycle the cells a few time on the new charger and the old rig, then on new charger and a new, PCB, rig, then on the old chargers and a new rig.
That should pin-point the issue and determine which, if any, of the old chargers will get an appointment with a pathologist for a dissection. 
Hopefully.

[Conrad Hoffman]

Another thought- when I've done battery tests in the past, I used a constant current load instead of constant resistance. Got reasonably consistent results.

[IanB]

For that i built a simple discharge logging rig - an atmega88 logs the battery voltages with it's ADC, and sends the data back to a PC.
Each battery is connected in series with a 3 Ohm resistor and a N-FET to cut it off once it reaches 0.8V.
Simple, right?

I think you have too many mechanical contacts and points of unaccounted resistance in that setup.

I have done constant resistance capacity testing successfully before using a logging multimeter (see graph and experimental setup attached), but I used a four wire voltage measurement right at  the battery terminals and made sure I had a good stable resistance measurement for the current path through the load resistor.

The measurements I have obtained have been essentially the same as reported by a C9000. Note that the eneloop graphed has a low capacity because it was old and tired.

A really important factor in capacity measurement is the charging. Before measuring capacity the cells should be cycled through a few charge/discharge cycles to remove any residual voltage depression and then they should be charged CC at 0.1C for 16 hours before the discharge test.

[Artlav]

So far so good.
Now i get reasonably consistent results.
There are some variations, but these i can track down to the contacts on the rig - wiggle the wire, 10mV less, 2% off.

Never the less, with a breadboard and a new charger i'm getting +-3% accuracy between the runs on average, 5% worst case.
Much better than +-50% i was getting before.
Time to etch that PCB.

That's quite a discovery - i never realised a simple battery charger can be so incompetent at charging batteries. 

I think you have too many mechanical contacts and points of unaccounted resistance in that setup.

I figured the resistance is accounted for in bulk by measuring voltage drop cell to ADC and difference between voltage across the resistor and across the cell.

but I used a four wire voltage measurement right at  the battery terminals

Huh?
How do 4 wire measurement apply to measuring voltage?

In any case, i naturally have the current loop of each discharge cell separate, with all of them touching a common ground and an ADC wire touching the battery terminal side.

[IanB]

but I used a four wire voltage measurement right at  the battery terminals

Huh?
How do 4 wire measurement apply to measuring voltage?

Good question. In general, I am trying to make an accurate voltage profile of the battery in the presence of significant currents in the circuit. Since I do not want my voltage measurements to be contaminated by I²R terms in the measurement path, I measure the voltage using separate wires arranged so that no current flows in them. This might not seem so important if only measuring battery capacity, but in this case I also I want to ensure that my 0.9 V cut-off point in the discharge is a repeatable and accurate measurement of the cell voltage unaffected by the discharge current in the circuit.

Even the contact resistance between the battery terminal and the measurement point can contaminate the measurement. I made contact with the battery using tinned copper foil and clean polished contact surfaces, held in place with strong magnets.

[qno]

Have a good look at the battery holders.
They are not the best quality.
A bad contact can ruin your measurements