Tuning the Battery Aging Model

[surfingslovak]

So far, I do not recall seeing any such measured data.

I believe that this NREL paper is based on empirical data, and it's very detailed in terms of modeling. I used a simplified version of it to see if it would fit the data we saw up to last summer, and it seemed to work reasonably well: http://1.usa.gov/agingmodel" onclick="window.open(this.href);return false;

Nissan's plot, which they shared with TickTock and is presumably based on empirical data as well, shows a leveling off, as does the Panasonic data chart for the cells that are supposedly used in the Model S. That said, I think we may want to look into using a different approach and see how well it fits the data coming from the field. I'm surprised to see that with the exception of hot locales, Stoaty has detected only very small deviation from the model so far. I would have expected a more significant diversion given the recent reports and discussion.



Click to open

 

[RegGuheert]

So far, I do not recall seeing any such measured data.

I believe that this NREL paper is based on empirical data, and it's very detailed in terms of modeling. I used a simplified version of it to see if it would fit the data we saw up to last summer, and it seemed to work reasonably well: http://1.usa.gov/agingmodel" onclick="window.open(this.href);return false;

Nissan's plot, which they shared with TickTock and is presumably based on empirical data as well, shows a leveling off, as does the Panasonic data chart for the cells that are supposedly used in the Model S.

Again, it's a model based on an extrapolation of losses observed during the beginning of life of the cell. If you ignore linear loss mechanisms, even ones the researchers may think are low by comparison at the beginning of life, you will come to the wrong conclusions when you extrapolate over a long time. Actual testing bears this out. Please have a look at the measured calendar loss plots in the four papers I referenced.

 

[surfingslovak]

Please have a look at the measured calendar loss plots in the four papers I referenced.

I have seen those, and we talked about it at length last year. Accelerated testing versus actual aging could very well be the reason why Nissan and other manufacturers might have a surprise coming. I picked the non-linear t^1/2 model from the NREL paper just to see if it fit the data from the field, and it did. It also appears to cover it pretty well so far, according to the data Stoaty has collected. That said, the LEAF is still near the beginning of its lifecycle, and that may very well be the reason. I'm not opposed to other approaches, and would be curious to see of they would fit the data collected so far. A more linear loss would explain why AZ and TX LEAFs did so much worse than predicted, but the rest of the cars do not seem to fit that curve (yet).

If you ignore linear loss mechanisms, even ones the researchers may think are low by comparison at the beginning of life, you will come to the wrong conclusions when you extrapolate over a long time.

Could the t^1/2 model not account for a more linear decline later in the life of the cell by holding cycling losses steady or increasing this component, while attenuating calendar aging? If the pack was temperature-managed and more heat-resistant, we could indeed see a lot larger cycling loss component, and the capacity decline graph could look much more linear.

 

[TomT]

This may, in fact, be so.

In addition to more cycles, it might also be necessary to consider more frequent charging to 100% as part of the equation... However, there seems to be contradictory conclusions on how much 100% charging adds to battery degradation so a true metric for this may be hard to determine...

What if the cycling loss only made out 20% or 30% of the total capacity loss? Even with 100% charging and more cycles towards the end of the life of the battery, the impact on the overall decline should be fairly modest, and might add "only" a few percentage points. I think the model could be tweaked to include rising number of cycles for cover the same mileage. I think the factor to multiply the line item with would be 1/(remaining capacity). An owner who has lost 15% of battery capacity will consequently need to cycle the battery 17% more to achieve the same number of annual miles.

 

[RegGuheert]

I have seen those, and we talked about it at length last year.

But no one has ever produced a long-term calendar loss measurement which shows slower than a linear degradation while we have several showing linear or super-linear calendar degradation.

Accelerated testing versus actual aging could very well be the reason why Nissan and other manufacturers might have a surprise coming.

For Nissan and their customers, the surprise has already come. Why would we want to cling to a t^1/2 calendar aging rate when we KNOW it does not match the literature OR the LEAF results.

I picked the non-linear t^1/2 model from the NREL paper just to see if it fit the data from the field, and it did. It also appears to cover it pretty well so far, according to the data Stoaty has collected.

No, it does not. It only fits where calendar aging has not gone very far:

Early indications are that the model is quite good in the Pacific Northwest and does a progressively poorer job as the average temperature increases.

That said, the LEAF is still near the beginning of its lifecycle, and that may very well be the reason.

Not in Phoenix. Many LEAF batteries have already completed their life cycle there.

I'm not opposed to other approaches, and would be curious to see of they would fit the data collected so far. A more linear loss would explain why AZ and TX LEAFs did so much worse than predicted, but the rest of the cars do not seem to fit that curve (yet).

I doubt that. If you fit a linear calendar degradation curve (along with cycling) to work in Phoenix, I'm betting the rest will fall in line.

Could the t^1/2 model not account for a more linear decline later in the life of the cell by holding cycling losses steady or increasing this component, while attenuating calendar aging?

No. The curve will continue to shallow forever. A linear curve dominates in the long term.

If the pack was temperature-managed and more heat-resistant, we could indeed see a lot larger cycling loss component, and the capacity decline graph could look much more linear.

I would put that slightly differently: We would see a smaller calendar fade component, so cycling should be more dominant.

Please note that this issue of the shape of calendar losses is most important for high-temperature climates and/or low mileage applications, so Nissan's efforts to address the needs of their customers in high-temperature climates does not resolve this issue for those of us in cooler climates who have low mileage requirements. (Admittedly, the issue is much more acute and costly in the hot areas.)

 

[RegGuheert]

This may, in fact, be so.

In addition to more cycles, it might also be necessary to consider more frequent charging to 100% as part of the equation... However, there seems to be contradictory conclusions on how much 100% charging adds to battery degradation so a true metric for this may be hard to determine...

What if the cycling loss only made out 20% or 30% of the total capacity loss? Even with 100% charging and more cycles towards the end of the life of the battery, the impact on the overall decline should be fairly modest, and might add "only" a few percentage points. I think the model could be tweaked to include rising number of cycles for cover the same mileage. I think the factor to multiply the line item with would be 1/(remaining capacity). An owner who has lost 15% of battery capacity will consequently need to cycle the battery 17% more to achieve the same number of annual miles.

+1

I think we need to use TaylorSFGuy's data to calibrate the cycling loss data and see where calendar losses fall out from there. Because of his cool climate and very high cycling rate, calendar fade should be largely negligible while cycling losses are dominant. In fact, I predict that no other 2011/2012 LEAF will lose the first capacity bar at as high a mileage as he achieved.

 

[Stoaty]

What if the cycling loss only made out 20% or 30% of the total capacity loss? Even with 100% charging and more cycles towards the end of the life of the battery, the impact on the overall decline should be fairly modest, and might add "only" a few percentage points. I think the model could be tweaked to include rising number of cycles for cover the same mileage. I think the factor to multiply the line item with would be 1/(remaining capacity). An owner who has lost 15% of battery capacity will consequently need to cycle the battery 17% more to achieve the same number of annual miles.

The model uses a fixed number of 1.5% loss for every 10,000 miles traveled at 4 miles per kwh. It doesn't take into account increased cycling. Using your suggestion, I added a column to do a rough calculation using the 1/(remaining capacity) where the remaining capacity was from the prior year. The example shown below is for Dallas, TX using the figures from jmh614:

Annual mileage - 19026
Days in sun - 5.75
Miles/kwh - 4.1

As you can see, this doesn't affect the cycling loss calculaton much in the first several years, but later it really gets going.

 

[surfingslovak]

That said, the LEAF is still near the beginning of its lifecycle, and that may very well be the reason.

Not in Phoenix. Many LEAF batteries have already completed their life cycle there.

I'm well aware of that, thanks, even though might think I have not expressed that adequately enough in my post. Also, I believe that you have misunderstood the point I was trying to make about cycling losses dominating over the long term when the calendar aging component is held to smaller portion of the total loss than what we have seen with the LEAF. As I said in response to your long list couple of days ago, which I appreciate, is that I don't think that it's possible to agree on all the finer points and rationalizations. I'm especially not fond of endless arguments over semantics and other things. I've been far too busy to spend this much time MNL, and if you wish to tweak the model or build your own, then more power to you. We had to start somewhere, that's why I picked the NREL model last year.

The only other thing that existed at the time was Mark Larssen's capacity loss tool, which he subsequently proceeded to publicize everywhere he could. What I initiated was meant as a best-effort counterexample to demonstrate that the data could be explained another way. The Arrhenius climatic modeling was done to see if the theory that we will see different aging in different climates will hold, and what the extent of the difference might be. It was well understood that the effect of microclimates in garages will be difficult to spot and to model, and that the results might not be accurate. I also remember have a prolonged debate with you where you claimed that the effective battery temperature will be well above the ambient, and therefore climatic data should not be used or we needed to raise the effective temperature of the battery.

As I said, I don't have all the answers, nor the time to debate this endlessly. I'm sure that the common understanding of this topic and the data will continue to evolve.

 

[Stoaty]

I think we need to use TaylorSFGuy's data to calibrate the cycling loss data and see where calendar losses fall out from there. Because of his cool climate and very high cycling rate, calendar fade should be largely negligible while cycling losses are dominant.

I have already done that as part of the validation (using Gids when he had about 56000 miles on his Leaf). The model predicted his loss very closely:

I am pretty sure I checked about around 70,000 miles plus and the model was still right on track for TaylorSFGuy. Probably there is a post of mine on the update somewhere on the forum, but I'm not going to try to find it.

Edit: I looked at the end of life calculation using my own location, driving pattern, annual mileage, etc. and found that my predicted EOL decreased from 10.4 years to about 9.5 years using the alternate calculation for cycling loss suggested by surfingslovak

 

[RegGuheert]

Capacity fade models typically assume diminishing caledar losses and linear cycling losses,...

While that may be so, please note that this is in direct opposition to fairly recent measurements made by Panasonic on some of their cells. In that paper (that Stoaty linked to), it is quite clear to see that calendar capacity fade is linear while cycling capacity fade slows with cycle number. (Note that this cycling test likely has the same problem as others: it likely cycles between two fixed SOC levels.) These are not the cells in the LEAF, but perhaps they are related to something in the Tesla Model S?

Simply put, typical Li-ion capacity fade models do not match the reality that occurs in the field. Until that is addressed, we will not be accurately predicting the life of our vehicles (except for those who live in the areas where the vehicles are fully aged).

 

[surfingslovak]

These are not the cells in the LEAF, but perhaps they are related to something in the Tesla Model S?

I happen to have the data sheet for the Panasonic cell, which is reportedly used in the Model S. Here is the cycle life they predict at 25 C:

Simply put, typical Li-ion capacity fade models do not match the reality that occurs in the field. Until that is addressed, we will not be accurately predicting the life of our vehicles (except for those who live in the areas where the vehicles are fully aged).

Yes, understood, and that's why early adopters play such an important role. The data and other learnings from the field are very important. And to reiterate, I would view this as a beginning. We have to start somewhere and try something, otherwise less accurate and more questionable approaches and theories might fill the void.

 

[surfingslovak]

As you can see, this doesn't affect the cycling loss calculaton much in the first several years, but later it really gets going.

Thank you for trying that, hopefully it won't affect the calibration you did for the data available last year. I will try to have a look at this later today.

 

[RegGuheert]

I think we need to use TaylorSFGuy's data to calibrate the cycling loss data and see where calendar losses fall out from there. Because of his cool climate and very high cycling rate, calendar fade should be largely negligible while cycling losses are dominant.

I have already done that as part of the validation (using Gids when he had about 56000 miles on his Leaf). The model predicted his loss very closely:

I am pretty sure I checked about around 70,000 miles plus and the model was still right on track for TaylorSFGuy. Probably there is a post of mine on the update somewhere on the forum, but I'm not going to try to find it.

Thanks! That's awesome!

On the other end, we have some LEAFs in Phoenix in low-mileage applications. One that I recall has already lost a bar, but I don't recall if the owner has a meter or not. If metered, those may be the best to cases to use to calibrate calendar losses.

 

[RegGuheert]

Yes, understood, and that's why early adopters play such an important role. The data and other learnings from the field are very important. And to reiterate, I would view this as a beginning. We have to start somewhere and try something, otherwise less accurate and more questionable approaches and theories might fill the void.

Agreed! It certainly has been a bit painful, particularly for some. It has also been a bit manic-depressive.

And I will say again that I appreciate the excellent work that Stoaty and you have done! Many thanks!

 

[TomT]

+1!

And I will say again that I appreciate the excellent work that Stoaty and you have done! Many thanks!

 

[Stoaty]

OK, I have added the data from ColumbiaRiverGorge. Unfortunately, I had to make an educated guess about the Aging Factor, tried to pick a city with a somewhat similar temperature profile. I ended up with Pittsburgh, PA, even though he lives east of Portland, but in a somewhat hotter area. Should be close enough. His actual loss is 3.75% greater than predicted loss, in line with many of the other differences.

 

[TaylorSFGuy]

What can I tell you about my LEAF. Just through 91k miles. One capacity bar gone. I thought I had lost another but misunderstood the message a couple of days ago from our son. From memory GID count was 214 yesterday. Haven't replaced tires yet in spite of plan to get it done over last couple of weeks.

 

[RegGuheert]

Haven't replaced tires yet in spite of plan to get it done over last couple of weeks.

O.K. I won't be riding with TaylorSFGuy this winter! :shock:

 

[FairwoodRed]

Haven't replaced tires yet in spite of plan to get it done over last couple of weeks.

O.K. I won't be riding with TaylorSFGuy this winter! :shock:

Wussy! Steve can take me for a ride anytime he wants! <Disclaimer: Not if it snows!>

 

[Stoaty]

What can I tell you about my LEAF. Just through 91k miles. One capacity bar gone. I thought I had lost another but misunderstood the message a couple of days ago from our son. From memory GID count was 214 yesterday. Haven't replaced tires yet in spite of plan to get it done over last couple of weeks.

Couple of questions:

What is your estimated average miles/kwh?
How many days (if any) do you park in the sun?

Thanks.