Battery Degradation Behavior

[Ray B]

I have been keeping records for all of my charging, including SoC at the beginning and end of each charge and the kWh used. So I plotted the kWh / change in SoC as a function of the odometer miles. I am not sure if there is a direct relationship between the [kWh/ delta SoC] and the battery capacity, particularly if the buffers undergo any change over the life of the battery. But if there is a direct correlation, then a 1/3 drop in that value will happen around 73K miles according to this linear relation.



But I think the degradation is not linear, and the rate of loss tapers off after the first year or so according to plots published by Honda for other EVs. I expect the battery will probably show less than 1/3 loss in capacity over 10 years or 100k miles.

 

[ClarityBill]

This is incredible information. At 30% degradation, the number would be about 0.10. This seems like something that can be reasonably checked. My chargers do not normally give me KWh, but I occasionally use a public charger that gives KWh.

Are you using SOC from Hondalink?

 

[Ray B]

This is incredible information. At 30% degradation, the number would be about 0.10. This seems like something that can be reasonably checked. My chargers do not normally give me KWh, but I occasionally use a public charger that gives KWh.

Are you using SOC from Hondalink?

Yes, the SoC is taken from HondaLink. I do have an OBD-II unit that reports exactly the same SoC as HondaLink.

 

[ClarityBill]

Yes, the SoC is taken from HondaLink. I do have an OBD-II unit that reports exactly the same SoC as HondaLink.

My OBD-II unit shows up to 11% SOC when HondaLink shows 0% and 0 EV range. I have never seen 0% SOC on my OBD-II

 

[Ray B]

Mine are always in perfect agreement. I use this cheap one: https://www.amazon.com/gp/product/B005NLQAHS/ along with the TorquePro app. But anyway, my recorded SoC data before and after charging is always taken from HondaLink.

By the way, 0% on HondaLink is something I have never come close to seeing. I think 6% is as low as I have witnessed, though I think it must have dipped below 5% during use once as I saw the EV gauge drop to 1 bar, though it recovered back to 2 bars with some high engine RPMs for a few minutes.

 

[Ray B]

Another approach to check the degradation would be to occasionally keep an eye on the time it takes to hit the 15% increments on the HondaLink during recharges at Level 2 chargers as long as it gives a consistent power. Routinely, my 15% interval took 24 minutes at 6 kW early in the car's life, but I began to notice that it was going a little faster and was averaging about 23.5 minutes and now it is more typically at 23 minutes. Obviously you must ignore the 85 - 100% range since the power typically drops above ~95 or 96%, extending the time.

So if I start a Level 2 charge (at a constant 6.0 kW) at 1 PM starting at 10% SoC, I would observe the following when the car was new:
1:00 PM: 10%
1:24 PM: 25%
1:48 PM: 40%
2:12 PM: 55%
2:36 PM: 70%
3:00 PM: 85%

Now after 18 months it is more like:
1:00 PM: 10%
1:23 PM: 25%
1:46 PM: 40%
2:09 PM: 55%
2:32 PM: 70%
2:55 PM: 85%

No I don't ever watch it this closely, but I do estimate the time the car would be ready since I need to retrieve the car when it is done since I charge at a busy public charger, and I've had to adjust my time estimates due to the gradual loss in capacity.

If the loss in capacity is reflected in this way, then the 15% interval would be down to ~16 minutes when the capacity has dropped by 1/3.

 

[Clarity_Newbie]

Yes, the SoC is taken from HondaLink. I do have an OBD-II unit that reports exactly the same SoC as HondaLink.

Data point: The 3 OBDII adapters I've used for data collection have been in agreement with Hondalink 100% of the time.

My OBD-II unit shows up to 11% SOC when HondaLink shows 0% and 0 EV range. I have never seen 0% SOC on my OBD-II

Data point: The 3 OBDII adapters I've used for data collection concurs with ClarityBill. ~11% SOC = 0 EV range.

This is why I refer to SOC 10%/EV 0 when I post a thread about charging etc. Several previous detailed threads cover this issue as well. (ie) last 2 bars - SOC 10% whilst EV miles = 0.

 

[Lowell_Greenberg]

Data point: The 3 OBDII adapters I've used for data collection have been in agreement with Hondalink 100% of the time.



Data point: The 3 OBDII adapters I've used for data collection concurs with ClarityBill. ~11% SOC = 0 EV range.

This is why I refer to SOC 10%/EV 0 when I post a thread about charging etc. Several previous detailed threads cover this issue as well. (ie) last 2 bars - SOC 10% whilst EV miles = 0.

On a separate, but perhaps related question- what determines the estimated mileage the Clarity shows at 100% SOC. It is clear it is not merely temperature dependent- but may be impacted by ones daily route.

Sent from my SM-G975U using Tapatalk

 

[Clarity_Newbie]

On a separate, but perhaps related question- what determines the estimated mileage the Clarity shows at 100% SOC. It is clear it is not merely temperature dependent- but may be impacted by ones daily route.

Sent from my SM-G975U using Tapatalk

Lowell_Greenberg

Common answer to these questions are:
As you stated...ambient temperature whilst charging battery. Warmer temps = greater capacity and vice versa.
Previous drive history is the second most firmly held belief.

Caveat: In the maintenance minder post, I reference the possibility Honda has the ability to access ICE miles through software. If true, then there are things going on in the background and variables Honda chooses to keep from the consumer. Go figure.

 

[jpkik96]

All - I found this post extremely useful! I leased my Clarity in June 2018 and am approaching 31M. My family has enjoyed the car and I am now trying to decide if I should buy it off the lease early and purchase an extended HondaCare warranty. My typical use case is to plug in the Clarity every night into my ChargePoint Level II charger as it is my understanding that Honda has incorporated a protection buffer so you can not fully charge or discharge the battery.

Today I ran the EV range to 0 and racked up 44.7 miles, which includes several days of minimum driving between my wife and myself (she always uses the HVAC and heated seats)! Temperature in NJ the past few days has been approx 60 degrees. Overall not bad range considering multiple drivers and the fact that I replaced the Michelin Energy's with Michelin Cross-Climates back in February.

Using my cheap OBDII Bluetooth adapter along with the TorquePro app, the battery showed 11% available with EV at 0. According to my ChargePoint App, the Clarity required 13.01 KWh to be fully charged and now shows 41 miles of EV available. Hope this helps!!

Stay safe!

 

[Ray B]

Thanks for the feedback!

I put together a simple Excel file that makes way too many assumptions, but I went out on a limb to correlate the charging numbers to the capacity and put in your numbers along with my data from today (I have roughly half of your mileage), and then used a simple linear degradation assumption to estimate when the battery would be down to 33.3% loss (and thus potentially eligible for warranty replacement based upon your specific warranty terms). Again, I am taking a few liberties with my math, and there's no evidence that loss becomes linear, so the estimate will probably be way off.



Anyway, I will post the calculator for fun here. You just need to feed in the starting and ending SoC from a charge cycle, and the amount of power consumed. I recommend using data from a Level 2 charger (6 - 7 kW) for better accuracy. Feel free to use and edit edit it as you like. I make no claims to its reliability or accuracy (no guarantees).

 

[leop]

This is an interesting way to estimate the battery degradation/capacity. One caveat would be that during warmer weather, the battery cooling system pump and fan often go on during charging. This would mean more energy is used for the charging cycle and this would increase the calculated battery capacity. The same caveat applies to those who pre-condition the cabin temperature while charging.

I have been recording the energy used and starting/ending SOC since December of 2018. I wish I had thought to do this from the very first charging cycle so that I could have normalized myself to the known battery capacity (PDI) when purchased in February of 2018.

LP

 

[Cash Traylor]

@Ray B

I really like where you are going with this! MrFixit has a lot of data that could be used to refine this estimation sheet and I am sure there are math gurus out there that can tweak your really nice and user friendly spreadsheet.

I am going to go back and watch the losses more carefully on my HV Charge mode tests regarding what power is actually going into the HV battery. Then I think I can use some trickery to figure out a guestimate of the accessory losses during charge cooling that would impact the charging on level 2. Level 1 charging generates less heat, but you get parasitic losses due to the low total amount of power (the losses represent a greater percentage of charging power over the longer required charging period - hence lower efficiency). On a cool day, with a "cold" car - Level 1 would be the most efficient in that "most" of the input power gets to the battery. Conversely, when the car is hot/warm environment Level 2 works the best, as the amount of time cooling is minimized and represents less of the total percentage of available charging current plus warm batteries absorb a charge more readily.

Just as a WAG, it appears that the ancillary cooling systems draw approximately 120 watts passive (pumps and electronics), and when active (radiator fan on) it jumps to 250 watts. I do not think it is worth computing the losses in the BMS and computers they are likely less than 100 watts. I have to figure out a better rule of thumb. However if you plug in and all the cooling comes on, you can likely subtract 250 watts per hour of charging and be conservative (which isn't really a good thing when you want accuracy). I really just need to go get the meter and shunt a few fused busses to get a better idea under cooling and non-cooling events.

Thanks for putting this together. Obviously like most things, some currently happening, predictions are only as good as the final outcomes. I want to run my numbers, print it out, put it in the glove box and see what happens. As I am one of those people watching for the warranty event...

Cheers,

Cash

 

[Cash Traylor]

Also, maybe I am just paranoid here (ok, ready to accept the "yes you ares") but unlike fuel pumps that are tested and certified for volume accuracy (at a given temperature) by government authorities in most locales - are commercial charging stations? Tracking wattage is technically easy, power companies have a plethora of certifications for power meters required (the old mechanical ones HAD to under represent consumption as they aged by design/code). However, if a charge location is not accurate, say it "overestimates" the power delivered to your vehicle then the costs are "padded." This is similar to a pump saying you took 1 gallon but only delivery .9 gallons. As the costs are likely very small and it really may be a non-issue anyway, I only mention this if you are using a commercial station to tell you that your car took 13.3 or 13.9 kWh of energy - do you believe them? Since that is part of the battery degradation test as presented here.

Maybe I have been locked inside too long....

Cash

 

[Paddy]

I have a OBD II adapter and Torque Pro. But I couldn’t find the SOC. Do I need to add its custom PID? I googled and couldn’t find the PID value.

 

[MrFixit]

MrFixit has a lot of data that could be used to refine this estimation sheet

I like what @Ray B is up to here too.

I do have a lot of data, but I am missing a key parameter that is integral to this approach. That is, the SOC.
I have not been tracking SOC in any way.

As a generalization. about all I can contribute is the fact that my 'full' charge when the car was new was about 14.5 kWh. Now, at around 18 months later, it is more like 13.5. If I make the assumption that the SOC is 11% with an EV range of zero, then I can use Ray's spreadsheet to get the following answer...



I don't think this one point adds any insight though.

Can anyone elaborate on a good way to easily get the SOC? I have a lot of trouble with HondaLink, but does the battery charge there accurately reflect the real SOC? HondaLink only seems to provide whole percentages. I can get it with the AP200, but it doesn't have fractional percentage either. I would rather not always have an OBD2 device connected.

If SOC were trivial to get, I would be more inclined to start logging it.

 

[MrFixit]

The same caveat applies to those who pre-condition the cabin temperature while charging

My instinct is that the cooling pump / fan would have an insignificant impact. Pre-conditioning, on the other hand, would not be insignificant so you would obviously just not use those occurrences when evaluating your battery.

 

[Cash Traylor]

OK, there is a lot to unpack here and I need to do a bit more work and research. There are entire industries trying to solve this problem, and we are not going to do it with an OBDII and MS Excel... However, this is not completely unreasonable as a fun "dartboard" prediction tool and I applaud Ray for this effort. The best part is you can track it's accuracy, if we report current and predicted HV Capacity in amp-hours. You then compare that to readings taken at the dealer. However the data here is not linear and we are missing several factors. I completely agree with Ray and MrF's assumptions, and the premise that this must be based on data the "pro-consumer" owner would have regular access to.

The one variable I noticed listed was "time to charge" and how long it took to reach that SOC. This is not a good metric for Lithium batteries unlike almost all other types. Li-poly, especially those other than LiFePO4 cells, actually take longer to charge the older they get (cycles or age). You can monitor power input and correct for parasitic charge losses (efficiency degradation) but the shorter time may be very misleading due to loss of charge transfer capability. This basically means the battery takes longer to absorb the same amount of energy, and in fact longer to absorb even "less" energy as capacity declines. There are a lot of things going on here effecting those observations, that make them still valid but for different reasons (whether that matters in the end for this remains to be seen).

A good article on that is here from Cadex: https://batteryuniversity.com/index...y_do_old_li_ion_batteries_take_long_to_charge

The current model is too aggressive and missing some parametrics that would shallow (refuse to say "flatten" right now) the curve.

All the data to predict this is being streamed over telemetrics to Honda anyway... our cars are research units for them. They are using our cars to train Tensorflow ML models for EV battery life predictions and consumer use modeling I am sure of it (putting on foil hat now...)

Seriously though - all the data is in the car's CAN buss - geez I wish I could figure more of it out, but I am getting a slow handle on it likely much more time needed than I have right now while home schooling. Steep learning curve for me (not a coder) and I really need to sniff the BUSS while the battery capacity is being requested by the i-HDS to be sure. Then this would be a sub $100 problem for everyone and "check at will." I'm actually pretty convinced that once I know the PID and scaling that the $20 OBDII/Bluetooth/Phone mentioned below would work.

So, a basic ELM327 OBDII probe and mobile phone software in the $20 range (I am working on the reason, has to do with voltage drop under a "known load"). Either an EVSE that reports charge data (likely level 2) or a Kilowatt meter ($35 Amazon) and use their OEM Level 1 charger for all data points (you can't mix and match charger data, it all has to come from a single source) and the basic 5 step guide to use it. This spreadsheet will need an entry table for the data and time range for the pretty output page Ray already created. People can use any charger regularly - work, home, destination, etc - however for any entry of data in this to be accurate for a specific vehicle, they have to use only data from a specific charger, mixing will break the range.

You need to know some absolutes. PDI type stuff to extrapolate and interpolate from. Some beginning point... Date of manufacture or date placed in service/sold, original or first known HV Pack capacity (i-HDS reported) and full charge input power as close to the time the pack capacity was obtained. Don't put EV "range" anywhere on there, that's a rabbit hole beaten to death.

1. Input charge, age, mileage, and known pack capacity are primary
2. Known pack capacity with a full charge cycle data capture within a short time for correlation
3. Actual EV "mileage" (not predicted) from full charge to automatic HV mode is good data relative to an individual driver

These don't have to happen all the time, once a month data points would be predictive enough for this form.

Hmmm, I am not good enough at excel formulas for this....

Oh, and yes I had a glass of Scotch in hand while reading the attached document...

Cheers and Happy Easter,

Cash

 

[MrFixit]

@Cash Traylor -

We have two 'simple' things that [almost anyone] can do with very little effort.

1. Measure the input energy (or estimate it using charge time) for a 'full' charge and scale it to estimate capacity.
2. Measure the input energy for an arbitrary [partial] charge and scale it by using the reported SOC change to estimate capacity.

Either of these cold be extrapolated to the point of battery 'failure', but I think we all agree that this is probably not linear.
Approach 1 is just a subset of approach 2 (doing a 'full' charge just assumes a start of 11% SOC and charging to 100% SOC).

OK, then... Yes, this is clearly an estimate. You have enumerated multiple reasons that this estimate will lack accuracy. I would argue however, that the entire purpose of this was to have a gross idea as to what is happening with your battery over time, whether reduced capacity could possibly explain a reduction in EV range, and whether you should consider a trip to the dealer for a "real" measurement of capacity.

Maybe you have identified that using 'time' is undesirable because bad batteries may actually take longer (as shown qualitatively in the article). My sole purpose of suggesting time was that anyone can do it (no need to have an EVSE that logs kWh).

Beyond this gross check that almost anyone can do, then we want the actual number from the vehicle. The holy grail is the sub-$100 widget that the "Pro-Consumer" could buy and deploy for this purpose (be it an ELM, Panda, or AP200).

To that end, the idea of 'sniffing' the i-HDS during a query, or 'sniffing' the $350 Autel device during a query makes the most sense to me. Then learning enough to be able to make the same query with an ELM, Panda, or AP200 (which many would be happy to invest in) seems like the preferred direction to me. I know this approach is not easy and could utterly fail at some point in the process. I think I will keep my eyes open for a good sale on the Autel MX808. It would give me an immediate solution and a toy to use in pursuit of the greater objective going forward.

 

[Ray B]

Great feedback guys. A lot to digest.

Admittedly the first estimator Excel file based on linear extrapolation was going to be a bad first kick at the can. I meant for it to be applied when there is some extended use and then two data spaced apart by a wide period of use (say charging behavior data at 1 and 1.5 or 2 years). Using data from 0 and 1 year will give a poor estimate of capacity loss at 10 years.

Looking at capacity loss curves from LIB research and Honda itself shows non-linear degradation as Cash described. I found a free paper (https://www.researchgate.net/public..._Battery_Degradation_for_Cell_Life_Assessment) that provides a lot of insight into how best to model the degradation. But in looking at example LIB capacity loss curves:



...and this one from Honda:


It appears to approximate a :

y = -a*x^b

function where b<1. I experimented with a variety of values for b and it seems like when b = 0.5 the curve shape looks right. Meaning when the x axis is battery usage (odometer, # of deep charges, time, or whatever) then the degradation likely approximates a negative square root function.

Here again I am making invalid assumptions and the end result is imperfect, but I think the behavior and degradation rate will get much closer than my previous simple linear model. So if we assume the new battery has 0% degradation, and have an estimate of the battery degradation at mile X, then we can find the constant "a" (knowing y (estimate of capacity loss), x (odometer), and assuming b=0.5). Then a better estimate of the odometer at 33% loss can be fashioned by using the same equation and now we have y (the target capacity loss of 33.33%), the constant "a", and the assumed b = 0.5.

So attached is version 2 of the battery capacity loss estimator. I could someday do a better model by using the functions provided in the research paper I cited above, but that is way too complex at the moment. The 33.3% loss estimate in the new file only calculates based on the 'Result #2' data. Note that for this model the data seems to show that it will be difficult to stress the battery so much that the 33% is achieved before 100,000 miles. Which I think was ultimately Honda's goal.

Obviously, each usage pattern and charging pattern will be different for each car and driver, but the more data that can be shared on this forum, the better our knowledge base can get in terms of what to expect for capacity decline, and offering advice to new members on what to expect and perhaps even how to reduce capacity loss. I will continue to track my own charging cycles and see if the curve does mimic the V2 model or if it will need a new function to better extrapolate into the future.

Thanks for all the work and feedback so far. It is taking time for me to read and comprehend everything. Keep the data and thoughts coming.