Tuning the Battery Aging Model

[Stoaty]

OK, for all those who have been eagerly awaiting updates to the Battery Aging Model, here is the OpenOffice version 0.92:

https://dl.dropboxusercontent.com/u/48149991/Leaf%20Battery%20Degradation%20Model%20Version%20092.ods" onclick="window.open(this.href);return false;

Unfortunately, I don't have a Microsoft Excel version, and I don't know how much work it is for JPWhite to do the conversion.

Main Changes:

The Prediction Tab has been significantly re-worked

--Calendar loss is now calculated from date of manufacture (enter 15th of the month, e.g., 4/15/11 for Leaf made in April, 2011)
--Annual mileage is still calculated from delivery date
--As always, the yellow cells are the ones where you need to enter your data
--Cycling loss now has a correction to account for the fact that as you lose capacity, you will have to cycle the battery more for the same number of miles: multiplies by a factor of 1/(remaining capacity) as suggested by surfingslovak
--Predicted years to end of life takes advantage of the cycling loss correction; for Stoaty the EOL decreased from 10.4 to 9.2 years with this correction

Validation Tab has data added using AH capacity from data submitted by MNL members (you have to scroll down a bit to find it, since it is currently a couple of screens down; at the top is some of the older validation data)

Known problems:

1) After making a change which affects the years to EOL and miles driven at EOL, you have to manually re-calculate the whole spreadsheet (Control-Shift-F9 in OpenOffice) to get the correct values for EOL years and miles. I believe this may be a bug in OpenOffice 3.4.1, since I couldn't find any setting that would make this happen automatically like it should. A workaround to this is to enter the change twice. For example, you pick the city from the dropdown, then pick it again; or you can enter a new miles/kwh and then enter the same value a second time. I may check OpenOffice 4.0 to see if this bug has been fixed, but wasn't mentioned in the list of fixed bugs. It may work correctly in Microsoft Excel if JPWhite makes a version for Excel.

2) The Validation data using AH capacity that some have been posting to this thread is still a work in progress. It has the old results (doesn't use the corrected cycling loss). In most cases, the difference is minimal. For TaylorSFGuy it increased the predicted loss by 1.5-2.0% (don't remember the exact number), but as he has been deep cycling the battery his actual capacity loss has exceeded the predicted loss by about 5% (24% loss actual, 19% predicted). I haven't figured out how to do this in the spreadsheet yet for a bunch of different entries.

 

[klapauzius]

Date of Manufacture (month and year): 11/2010

Date of Delivery (month, day and year): 01, 23,2011

Date of P3227 update, if needed (should be at least 3 weeks prior to reading data): July 2013

Geographic location - be specific, must include city and state, otherwise data of no use (also include work location city and state and days Leaf spends at work if climate significantly different than home): Seattle, Wa.

Average miles/kwh for the life of your Leaf: ~ 4 m/kWh

Current odometer reading (i.e. total miles your Leaf has been driven): 22,716 mi

Current AH capacity reading from LeafDD or Leaf Battery App: 60.55

Exact date you took current odometer reading and current AH capacity : 9/13/2013

Days per week parked in the sun: 0

 

[Stoaty]

Date of Manufacture (month and year): No idea where to find this?

As I recall it is on the door frame of the drivers door.

 

[klapauzius]

Found it! See above.
According to Carwings the annual miles/kWh is 5, but that is not in line with the reading on the dash which is more like 4.

Winter 3.3- 3.6
Summer 4.4 - 4.7

 

[TomT]

Not a problem. Excel 2013 imports and handles the OpenOffice version just fine.

Unfortunately, I don't have a Microsoft Excel version, and I don't know how much work it is for JPWhite to do the conversion.

 

[Stoaty]

Found it! See above.
According to Carwings the annual miles/kWh is 5, but that is not in line with the reading on the dash which is more like 4.

You win the prize:

Predicted loss - 10.8%
Actual loss - 8.6%
End of Life (70% capacity) - 14.7 years at current annual mileage

I think there is one other person in the Pacific NW (DaveInOlyWa) who is doing better than the model predicts, although my Leaf is only 0.15% worse than the model.

 

[klapauzius]

Found it! See above.
According to Carwings the annual miles/kWh is 5, but that is not in line with the reading on the dash which is more like 4.

You win the prize:

Predicted loss - 10.8%
Actual loss - 8.6%
End of Life (70% capacity) - 14.7 years at current annual mileage

I think there is one other person in the Pacific NW (DaveInOlyWa) who is doing better than the model predicts, although my Leaf is only 0.15% worse than the model.

Earlier this year (April or so) my capacity was 65 AH.
Since I do not believe I have lost 5 Ah in a mere 5 months, I would take the Ah readings with a big grain of salt. They seem to be quite temperature dependent.

 

[Stoaty]

Earlier this year (April or so) my capacity was 65 AH.
Since I do not believe I have lost 5 Ah in a mere 5 months, I would take the Ah readings with a big grain of salt. They seem to be quite temperature dependent.

I don't believe you have lost the 5 AH either in that time... but I do believe you got the P3227 update a couple of months ago (which is supposed to be much more accurate) so that could be the reason the reading is lower.

 

[DaveEV]

OK, here is the initial data. To my surprise, all predicted percent capacity loss values are within 3.75% with the exception of the readings from Dallas, TX and Ridgecrest, CA. So far, the Battery Aging Model is doing fairly well, except in places that get really hot.

I think you are understating the difference of the model - at least when looking at Ah readings.

While the absolute capacity loss is typically withing 0.5-4% of predicted except for a couple hot locations (Dallas, TX, Ridgecrest, CA), quitea few are in the 2.5-3.5% error difference.

For example, let's take my car where I am currently at 16.18% capacity loss after 2.41 years while the model expected 13.38% - a difference of 2.8%.

Sounds small, but according to the model I'm not supposed to hit 16.18% until somewhere around 3.5 years.

In other words, I've lost a over a year - my battery is aging 40% faster than the model predicts.

The model predicts EOL at 10 years, 90k miles.

If I choose a city which causes the loss to match up with what my Ah readings are (Witchita Falls, TX matches up nearly identically, for example), that pushes EOL up to 7.2 years, 65k miles, or about 30% faster than modeled.

Those are big errors, not small.

FWIW, looking at modeled GID counts, San Diego happens to match up as expected for my 80% GID count (currently at 195, exact same as modeled), but Witchitia Falls, TX happens to match up as expected for my 100% GID count (currently around 235 which is what the model predicts for Witchitia Falls).

 

[opencar]

Tokenride is the first CA owner i've heard that triggered the warranty (altho DaveKern in San Diego looked to be well on his way before turning his car in). Pomona is between riverside and LA so i'd say his case is a pretty big outlier. 4 bars in 2+ yrs is Arizona performance. instead of a degradation plateau there seem to be cases of accelerated degradation rates after a certain threshold.

OK, here is the initial data. To my surprise, all predicted percent capacity loss values are within 3.75% with the exception of the readings from Dallas, TX and Ridgecrest, CA. So far, the Battery Aging Model is doing fairly well, except in places that get really hot.

I think you are understating the difference of the model - at least when looking at Ah readings.

While the absolute capacity loss is typically withing 0.5-4% of predicted except for a couple hot locations (Dallas, TX, Ridgecrest, CA), quitea few are in the 2.5-3.5% error difference.

For example, let's take my car where I am currently at 16.18% capacity loss after 2.41 years while the model expected 13.38% - a difference of 2.8%.

Sounds small, but according to the model I'm not supposed to hit 16.18% until somewhere around 3.5 years.

In other words, I've lost a over a year - my battery is aging 40% faster than the model predicts.

The model predicts EOL at 10 years, 90k miles.

If I choose a city which causes the loss to match up with what my Ah readings are (Witchita Falls, TX matches up nearly identically, for example), that pushes EOL up to 7.2 years, 65k miles, or about 30% faster than modeled.

Those are big errors, not small.

FWIW, looking at modeled GID counts, San Diego happens to match up as expected for my 80% GID count (currently at 195, exact same as modeled), but Witchitia Falls, TX happens to match up as expected for my 100% GID count (currently around 235 which is what the model predicts for Witchitia Falls).

 

[Stoaty]

For example, let's take my car where I am currently at 16.18% capacity loss after 2.41 years while the model expected 13.38% - a difference of 2.8%.

Sounds small, but according to the model I'm not supposed to hit 16.18% until somewhere around 3.5 years.

In other words, I've lost a over a year - my battery is aging 40% faster than the model predicts.

Yes, but I expect the model to "catch up" during the coming cool months. Until we have anniversary readings from date of manufacture, we would expect the model to be "behind" since the Leaf has gone through more summers than Winters. The Gid readings on my Leaf at 80% charge didn't change by even one Gid for a period of 6 months last Fall and Winter. As I said previously, this is a systematic error in my method, since it doesn't work correctly for fractional years. Time will tell, of course.

 

[Stoaty]

Version 0.93 of the Battery Aging Model is available here:

https://dl.dropboxusercontent.com/u/48149991/Leaf%20Battery%20Degradation%20Model%20Version%20093.ods" onclick="window.open(this.href);return false;

Improvements (all on the Prediction Tab):

--Fractional year graph now shows corrected cycling loss (I forgot to update it when I made the refinement to cycling loss)
--Added End of Life Graph showing Relative Contributions of Calendar Loss, corrected Cycling Loss, and Solar Loading Loss
--Added two small tables (one to the right of each graph) that show Relative and Absolute Loss (Calendar, Cycling and Solar) for both Fractional Year and End of Life

Let me know of any bugs found or features you would like to see added (besides increased accuracy, of course).

PS I am wondering what factors (besides heat) are causing some to match the model closely (e.g, DaveinOlyWa and Stoaty) and others to deviate far from it. Thoughts?

 

[klapauzius]

Earlier this year (April or so) my capacity was 65 AH.
Since I do not believe I have lost 5 Ah in a mere 5 months, I would take the Ah readings with a big grain of salt. They seem to be quite temperature dependent.

I don't believe you have lost the 5 AH either in that time... but I do believe you got the P3227 update a couple of months ago (which is supposed to be much more accurate) so that could be the reason the reading is lower.

My experience was (before the update) that with the coming of summer, capacity (i.e. AH) has been dropping like crazy. I think this has been observed for a long time with the leaf. I dont think this has anything to do with the software change, because I saw it happen way before July, when I did the update.

So the "true" capacity is modulated by the LEAFs software. Since short-term temperature changes (and we have significant ones now here , as the days get shorter) do not seem to affect capacity much, this seems to be some sort of long term adjustment.

Naively one would assume that this is highly correlated with the local temperature, in your model you would have to take temperature dependent modulation of the reported readings into account.

Maybe we should submit monthly readings, so the "seasonal" capacity change can be accommodated.
I think in warmer climates (SoCal, Az etc), this might not be as dramatic as here in Seattle. Right from the start our winter range (which roughly equals capacity) has been significantly less (and not because of heater use) than the summer range.

I would bet if my AH reading would be what I posted on June 25(!!!)

"20,000 miles, 2.5 years and according to the ELM327 app @ 101.44 % or 67.21 Ah."

this would be a far outlier on the current age model.

 

[DaveinOlyWA]

Glad to hear I am doing better than the model but have to say nearly all my loss has happened in the last 4 months and that trend seems to be continuing.

as a general rule of thumb; should we resubmit data every 3 months or when we loss "X" amount of capacity or health? Mine seems to be trickling downwards. the health bounces up and down but capacity is showing a general downward trend that is dropping almost daily

 

[DaveinOlyWA]

Earlier this year (April or so) my capacity was 65 AH.
Since I do not believe I have lost 5 Ah in a mere 5 months, I would take the Ah readings with a big grain of salt. They seem to be quite temperature dependent.

I don't believe you have lost the 5 AH either in that time... but I do believe you got the P3227 update a couple of months ago (which is supposed to be much more accurate) so that could be the reason the reading is lower.

My experience was (before the update) that with the coming of summer, capacity (i.e. AH) has been dropping like crazy. I think this has been observed for a long time with the leaf. I dont think this has anything to do with the software change, because I saw it happen way before July, when I did the update.

So the "true" capacity is modulated by the LEAFs software. Since short-term temperature changes (and we have significant ones now here , as the days get shorter) do not seem to affect capacity much, this seems to be some sort of long term adjustment.

Naively one would assume that this is highly correlated with the local temperature, in your model you would have to take temperature dependent modulation of the reported readings into account.

Maybe we should submit monthly readings, so the "seasonal" capacity change can be accommodated.
I think in warmer climates (SoCal, Az etc), this might not be as dramatic as here in Seattle. Right from the start our winter range (which roughly equals capacity) has been significantly less (and not because of heater use) than the summer range.

I would bet if my AH reading would be what I posted on June 25(!!!)

"20,000 miles, 2.5 years and according to the ELM327 app @ 101.44 % or 67.21 Ah."

this would be a far outlier on the current age model.

i think you had already lost most of that 5 ahr before Summer started. I have experienced nearly the same. I did not have the ELM back then but my GID meter saw me averaging in the high 260's in Early May (got warm early this Summer) to dropping to the low 240's by Early July which was a good 3 weeks into the weather change.

I got the SW update and and then started losing capacity again dropping to the low 230's. Since then I am back to upper 230's with an occasional 240-242

i think the update simply modulated the readings caused by adjustments made by the BMS so my 265-270 was probably really only 250ish which would be very understandable considering I had 30 months and over 30,000 miles by then.

the other conclusion would be to say I had very little (2-3%) degradation for 28 months then fell off a cliff

Either way; i plan to resubmit #'s monthly at least until well into Winter weather

 

[klapauzius]

i think you had already lost most of that 5 ahr before Summer started. I have experienced nearly the same. I did not have the ELM back then but my GID meter saw me averaging in the high 260's in Early May (got warm early this Summer) to dropping to the low 240's by Early July which was a good 3 weeks into the weather change.

I got the SW update and and then started losing capacity again dropping to the low 230's. Since then I am back to upper 230's with an occasional 240-242

i think the update simply modulated the readings caused by adjustments made by the BMS so my 265-270 was probably really only 250ish which would be very understandable considering I had 30 months and over 30,000 miles by then.

the other conclusion would be to say I had very little (2-3%) degradation for 28 months then fell off a cliff

Either way; i plan to resubmit #'s monthly at least until well into Winter weather

No, this started way before the update. Also, I do not believe the earlier readings of AH were reflecting the "real" number. I don't think I had true 67 Ah in June 2013 ( nor in January 2013). There was a very slow modulation (over weeks or month) of available AH going on before, inflating capacity at lower temps and deflating at warmer temperatures.
My overall concern about my battery health was unfortunately not large enough to take continuous records.

Maybe the modulation has been adjusted with the update, but the basic algorithm has probably not been changed.


Anyway, I agree, we should submit regular records of AH. Since the temperature modulation is so slow, average temps from the region will do as reference temperature measurement.

 

[Stoaty]

as a general rule of thumb; should we resubmit data every 3 months or when we loss "X" amount of capacity or health? Mine seems to be trickling downwards. the health bounces up and down but capacity is showing a general downward trend that is dropping almost daily

I would suggest to resubmit every 3 months, unless you are losing capacity rapidly, then I would resubmit every month. I am moving the submitted data to a new tab "Calibration" so it is easier to see, plus I figured out a way to get the corrected cycling loss with only a little manual work. I am trying out some changes to the Aging Factor (which is based on climate, but the model scales it to match Nissan tech drawing), but it isn't really helping as some predictions become more accurate and others get worse. If that doesn't help, I may change the underlying assumptions of the model re: rate of calendar loss and cycling loss per 10,000 miles. Fortunately, I can make those changes instantaneously, just have to change a couple of numbers in the spreadsheet and everything gets recalculated. However, I want to go slow here, probably need more data from other owners plus data from different seasons before deciding how to tweak the model towards reality.

 

[DaveEV]

For example, let's take my car where I am currently at 16.18% capacity loss after 2.41 years while the model expected 13.38% - a difference of 2.8%.

Sounds small, but according to the model I'm not supposed to hit 16.18% until somewhere around 3.5 years.

In other words, I've lost a over a year - my battery is aging 40% faster than the model predicts.

Yes, but I expect the model to "catch up" during the coming cool months. Until we have anniversary readings from date of manufacture, we would expect the model to be "behind" since the Leaf has gone through more summers than Winters. The Gid readings on my Leaf at 80% charge didn't change by even one Gid for a period of 6 months last Fall and Winter. As I said previously, this is a systematic error in my method, since it doesn't work correctly for fractional years. Time will tell, of course.

But that was all pre P3227 update data. We don't know how GIDs perform post update - at the very least we know that cool weather readings are less optimistic in that in hot weather it's less pessimistic. And it's not like my battery is extremely hot - it's seeing temps from upper 70s to low-mid 80s right now.

 

[opencar]

i downloaded the file & opened on the desktop w/ Excel. i don't get a version that has functions for calculations; it seems to be a flat data file. anyone have an idea of what i need on my Win7 computer to open the file w/ all the functionality?

Version 0.93 of the Battery Aging Model is available here:

https://dl.dropboxusercontent.com/u/48149991/Leaf%20Battery%20Degradation%20Model%20Version%20093.ods" onclick="window.open(this.href);return false;

Improvements (all on the Prediction Tab):

--Fractional year graph now shows corrected cycling loss (I forgot to update it when I made the refinement to cycling loss)
--Added End of Life Graph showing Relative Contributions of Calendar Loss, corrected Cycling Loss, and Solar Loading Loss
--Added two small tables (one to the right of each graph) that show Relative and Absolute Loss (Calendar, Cycling and Solar) for both Fractional Year and End of Life

Let me know of any bugs found or features you would like to see added (besides increased accuracy, of course).

PS I am wondering what factors (besides heat) are causing some to match the model closely (e.g, DaveinOlyWa and Stoaty) and others to deviate far from it. Thoughts?

 

[tbleakne]

I can open the .83 excel version on the Wiki into Numbers on my Mac, but when I try the .93 link above, I get a zip file ending in Version 093 2.ods When I click on it, I just get another zip file with suffix 2.ods.cpgz. Perhaps I need a link to the excel version.

I believe errors in tracking actual temperature history vs averages is a primary source of discrepancies with any aging model. I have had several months of experience with the battery temperature readings from the ELM327 App. Observations based upon this data:

1. The battery warms relatively slowly when parked in the shade, even if the ambient temperature is 10-15 F hotter than the battery.
2. The battery warms up very fast while driving, even light level driving, when in warm direct sun.
3. Once warm, the battery cools very slowly, even if the ambient temperature is 10-15 F cooler than the battery.
4. Quick Charges warm the battery a lot. This heat stays in the battery for hours if the ambient is warm.
5. A/C inside the car may be keeping one or more of the 4 temperature sensors cooler than the others, but it is hard to tell.
6. Given the placement of the battery, it makes sense that high cabin temperature from leaving the car parked in sun seeps into the battery fairly quickly. Since I try very hard to avoid parking in the sun, I have no data on this effect. My sympathies to those who have to park in direct sun at work, like I used to do.

I believe radiant transfer from a hot road surface to the underside of the car is significant. This process scales as the 4th power of absolute temperature.
The pavement under a parked car quickly cools relative to pavement under moving cars.
Since the battery never gets as hot as the road surface, the reverse process of IR radiation from the battery is insignificant.

This explains the asymmetry of the heating and cooling, and it also helps explain why high mileage in hot climates seems to usually lead to more loss than high mileage in cool climates.

I am not sure how you might incorporate any of this into your model.