12V voltage recovery curve used by battery testers? Probably not.

I was pondering how the little battery testers determine SOH and thought it might have something to do with how the battery voltage recovers once the load is removed. So I measured the voltages and fit them with the tool at mycurvefit.com. This battery is 4 years old and has seen better days. It was driven for about an hour the day before, sat overnight with the hood unlatched and the battery voltage before doing anything was 12.64 V. Before the measurements were started the following operation was performed by a helper (I was measuring at the jump point under the hood).

Push unlock on remote 2 times.
Open passenger side front door and enter, leaving door open.
Put fob in ignition.
Press Power twice with foot off brake
Put all 4 windows down an inch. (voltages as low as 11.8 V observed here)
Power off
Exit car
Lock car

Immediately start taking measurements (minutes,volts):

0 12.56
1 12.59
2 12.60
3 12.61
4 12.62
5 12.62
6 12.62
40 12.64

This data was graphed and curve fitted (I picked the equation)



A few problems are evident. The voltmeter only has 2 digits to the right of the decimal place so it ends up quantizing values in the flat part of the curve - to make an accurate curve this way one would have to measure more accurately, but these devices display a battery voltage which looks like these values, so they probably don't have the extra accuracy needed. As a corollary the curve fit needed to take this accuracy limit into account, and it doesn't, so the curve it fit misses the last point. Finally, It takes a long time for the voltage to fully recover, so finding the value in the exponent (-0.3367459) requires a much longer test than these devices apply.

Anyway, the equation's "a" is just the %SOC as a voltage, "b" is the starting voltage as an offset from "a", and "c" is the interesting part, describing how fast the battery recovers (the value here is in inverse minutes). "b" could be set anywhere on the curve, it doesn't tell us anything about the battery, it tells us about the experiment. Presumably a bad battery recovers slower than a good one and the "c" value would reflect that. I suppose that it's possible the devices make a bunch of fast measurements on the rapidly rising part of the curve, but fitting all 3 values with the limited accuracy that way would be tricky.

 

What have we learned? First that this is almost certainly NOT part of the tests done by the little electronic battery testers. I suspect that if we tested a bunch of batteries from fresh from the factory to on death's door the "c" value would change such that recovery became longer the worse off the battery was. So in theory if such a table existed one could do this measurement and determine where on that spectrum a particular battery fell. I don't know but also suspect that it wouldn't be perfectly correlated with loss of capacity or reduced ability to reach a full charge. Finally, I would guess that if a battery has lost water it would make the recovery time longer, and that adding water back would restore it (at least for the part due to water loss, not for other forms of damage). So it might offer a way to estimate "how dried out" without pulling the battery out of the car and carefully weighing it.

Weighing the battery is next on my agenda. The fairly accurate 50 lb scale just arrived and it was within a few grams of the expected weight on a gallon water bottle.

 

I had to test the 12V on the Accord today and put a current clamp on one of the battery tester leads during the test. This is the HF Viking battery tester, part number 58759. It looked like there was a single load of 0.6A for about 2 seconds. That is C/1000, roughly. Coming off of that there was one other different value seen which was less. It might be that the trailing edge of that load is sloped and the ammeter caught it somewhere on the slope. Before that there was no current at all. The clamp isn't going to be accurate below 10 or 5 mA, especially if the current measured isn't constant, and there had to be a couple of mA because the meter powers itself from the battery. Tested twice, once on the 20A current range, and then again on the 2A range. The thing is, that pulse was very close to the end of the test, which provided almost no time for the battery to relax back to its normal voltage. How they determine from this that the CCA (nominally 630) was only 623 is a mystery to me. Perhaps there was a very short much larger current spike, too fast for the clamp ammeter to catch and display?

Here is a description of one method used (part way down the page, "AC Conductance"):

BU-902a: How to Measure CCA - Battery University