State of health does not mean 1 thing
State of health is the measurement of how much capacity a used battery has left compared with when it was new. AVILOO, a battery diagnostics company, says the term is measured 2 different ways. Capacitive state of health counts ampere hours, a unit of stored charge, and ignores internal resistance, the friction inside the battery that slows down how fast it can deliver power. Energetic state of health counts kilowatt hours instead, and ties the number directly to the range a driver actually gets. AVILOO gives an example, if a car reads 90% energetic state of health, it keeps 90% of its original range. The company says the capacitive number can overstate the range left in an older car, because it leaves resistance out of the count.
There are different reference values and calculation methods, which can lead to significantly varying results.
Patrick Schabus, Chief Product Officer at AVILOO, a battery diagnostics company. Source 1.
One pack still needs 1 number
A battery pack holds many individual cells, and a grader has to turn all of them into a single score. AVILOO says most operators average every cell together, a method it calls imprecise. The alternative is reading only the weakest cell, because the whole pack cannot perform better than its weakest cell. A peer reviewed study led by Stanford University measured 8 individual cells taken from 2 retired Nissan LEAF battery packs, all built to the same 32.5 ampere hour capacity when new. Once measured, their real starting capacity ranged from 17.42 to 25.04 ampere hours, before any of the cells had aged at all.
All 8 cells tested were built to the same 32.5 ampere hour nominal capacity when new. The 8 measured results ranged from 17.42 to 25.04 ampere hours. This chart shows the lowest, a middle and the highest of those 8, chosen to stay inside the practical bar limit section 5 of the pipeline rules sets.
Show the numbers
| Cell 1.2 | 17.42 |
| Cell 1.1 | 20.27 |
| Cell 2.1 | 25.04 |
A slow test takes months, a fast one takes minutes
Grading a pack the direct way means running a discharge and recharge cycle in a laboratory, a reference performance test. That is the method behind the Stanford led study, which took 15 months of testing on the 8 cells before publishing results. A faster method instead reads signals the battery management system, the computer inside the pack that tracks voltage, temperature and charge, without draining the battery. AVILOO says its own fast test, called FLASH, takes 3 minutes. Recurrent, a company that studies electric vehicle batteries, says a Tesla owner can run a free comparison against expected degradation instantly from inside the car, or a deeper 24 hour diagnostic through a mode built into the car called service mode. A separate tool built by Recurrent needs about 1 week of accumulated driving data to produce a first reading.
Fast tests do not agree with each other either
The Stanford led team did not stop at 1 fast method. It built 4 different statistical models, all trained on the same battery management system signals from the same 8 cells, and compared how close each model came to the real measured capacity. The best model, elastic net regression, missed by an average of 1.38%. The worst, random forest regression, missed by 4.46%, more than 3 times as far off, despite reading identical data from identical cells.
All 4 figures are the average percent error of each model against the real measured capacity of the same 8 cells, from 1 peer reviewed study. The 3.2 times gap between the best and worst figure is arithmetic performed on 2 numbers in that study, not a multiple the study itself states.
Show the numbers
| Elastic net regression | 1.38 |
| Gaussian process regression | 3.03 |
| Support vector regression | 3.26 |
| Random forest regression | 4.46 |
No test in this review carries a price
None of the 5 sources checked for this article states what it costs to run any of these tests on a single pack. AVILOO calls its FLASH test cost effective without giving a figure, and a separate review of testing methods says the direct electrical method needs expensive equipment while a simpler capacity based method needs less expensive equipment, again with no number attached to either. Until a price is published, the honest answer for a buyer asking what grading a pack costs is that nobody says.