Roadster Owner Based Study of Battery Pack Capacity Over Time

[brianman]

Rich was kind enough to share his data, so I was fiddling with Excel graphing to see if I could notice anything interesting.

Discarding the data from [0%,50%] and [90%,100%] SOC allows me to zoom in on -- what I think is -- the interesting part of the graph. Here is a surface plot with the left axis being vehicles (A through T) and the right axis being time spent at version charge levels (from 51% to 89% SOC). Vehicle A has the least loss, while vehicle T has the most loss (ordered by the LP10K column, Rich, as you had it in the data).

I'm having trouble reading much definitive into the data. Nonetheless, what I think I see in the visualization is:
1. Spending much time at high charge (86-87%) decays range most.
2. Spending little time at medium high charge (85%) decays range second most.
3. Thus, the sweet spot is to spend a "reasonable" amount of time at medium high charge (85%).

#2 (and thus #3) might be explained in that doing so allows the pack the opportunity to rebalance

Note that the plot for (I think) M shoots a hole in #2 and #3 assertions, but the rest of the data seems to align with what I've described. Edit: Actually maybe it doesn't, that spike is around 80-81% SOC -- perhaps below the sweet spot where rebalancing has a chance to do its magic.


Here's the chart:

 

[user497]

Hi all, I just sent Rich my latest logs. I have 15k miles on my roadster but I haven't been able to drive it that much recently. Since I only drive about 20 miles per day, I have mostly been charging with the 110V but I did drain the car down pretty low and then recharged it in range mode a couple times in mid-Nov. Currently when I charge in standard mode I only get 155 miles (note that it has been low for over a year now so I don't think this is just due to cold whether). My brickahmin/brickahave is 145/145. In fact my min/ave has always either been the same or off by 1 (which I think is a good thing because if you had a bad brick I would assume that the min would be much lower than the average - or that means all of the bricks are bad). However, 145 is on the low end of Rich's graph from his posting on 1/24. I am asking Tesla again to figure out what is going on but in the past they have told me that the battery is fine and that the range is just showing low because I drive sure short commutes and then recharge with a 110V and that the battery would really let me drive further. I guess one way to confirm that would be to drive until the battery runs out (maybe I can circle around a parking lot in Milford until I run out?) but I don't want to void my warranty

 

[strider]

You are correct in that if the average and min bricks are the same or close you don't have an unbalanced pack. Do you have the time to drive it to a low state of charge, do a range charge, and then drive it for a bit? That should recalibrate the gauge.

Also, running your car out of juice to shutdown will not void your warranty unless you fail to plug it in soon after doing so. A few members here have done so in the past.

 

[smorgasbord]

The Plug-in American survey thread indicates that CAC is 160 when new.
I find it interesting that Rickae tried correlating capacity with time spent at certain charge levels, and then brianman tried correlating capacity with time spent at different charging levels.

I'd be interested in seeing capacity versus time spent at different temperatures, with charge level being a second factor.

We know that Range mode, which charges the battery the most, goes to great pains to keep the cells cool. When a Range charge is done and the car stays plugged in, you'll find the car's AC system remains engaged to keep the cells cool. I'm speculating that these two factors, temperature and SOC, have the strongest correlation to degradation.

I just sent my data to Rickae.

 

[JRP3]

Sure, both high voltage and high temperatures affect electrolyte breakdown.

 

[hcsharp]

... I'm speculating that these two factors, temperature and SOC, have the strongest correlation to degradation.

You mean after mileage? Because mileage appears to be the biggest factor so far.

 

[richkae]

I had hopes to do lots of analysis of temperature vs battery longevity.
There are 2 data streams in the log files, a "permanent" section with daily data, and a "short term" section with high resolution data. The high resolution data is only a couple months worth.
The high resolution data is where the temperature information is.
If you pull a log file from your Roadster every 2 months or so, then you would have all the high resolution data without loss.

The code I wrote in VMSParser to spit out temperature data needs to correct somehow for the partial sampling to be useful. You'd need to track the time window and how that would extrapolate out to the whole year. You cant compare a log file with summer data to one with winter data.
The VMSParser will process all the log files you have and put the data together, so you might have multiple data segments with voids inbetween, which makes the extrapolation a little more complex. ( it doesnt reprocess the same events - or shouldnt ).

 

[driver_EV]

I submitted the logs I have for my Roadster.
Played some with the Parser. I imagine an issue with weakness in the ESS might be identified with one or more bricks.
When I poked around with the log I pulled shortly after my car was delivered It seemed I could see more information with voltages and/or SOC than I see now, unless I am doing it wrong now. I just performed a full range charge from ~10% charge and the car seemed to perform balancing after that finished (for 1/2 hour or so). I was hoping to be able to see brick voltage samples before and during the charge, and see what changed at balancing stage.

-shortly after charge completed. (stayed like this for a while)

Suddenly dropped to 94% about 1/2 hour after charge was finished.

-wow big images, should have scaled them down.

 

[tomsax]

I submitted the logs I have for my Roadster.
Played some with the Parser. I imagine an issue with weakness in the ESS might be identified with one or more bricks.
When I poked around with the log I pulled shortly after my car was delivered It seemed I could see more information with voltages and/or SOC than I see now, unless I am doing it wrong now. I just performed a full range charge from ~10% charge and the car seemed to perform balancing after that finished (for 1/2 hour or so). I was hoping to be able to see brick voltage samples before and during the charge, and see what changed at balancing stage.

Do you have the latest version of VMSParser, 0.9.18, released on March 4th? It shows minimum and maximum brick voltages, and the corresponding brick numbers, from the 1-minute charge records.

Try this command: VMSParser -c <log_file_name>, or VMSParser -c -t <log_file_name> if you want a tab-delimited file to import into a spreadsheet.

Suddenly dropped to 94% about 1/2 hour after charge was finished.

That's pretty typical. You can't take a reading right after a charge or drive seriously. 225 ideal miles in Range mode is consistent with your CAC value. (It would be great if you'd do an update to your survey data with your range mode number.)

 

[smorgasbord]

There are 2 data streams in the log files, a "permanent" section with daily data, and a "short term" section with high resolution data. The high resolution data is only a couple months worth.

OK, but while I do have holes, I also have a lot of data. And the parser spits out how many seconds the battery has spent at each temperature, as well as how many seconds the battery has spent at each SOC. I don't know the math, but it does seem like one could use that info to build a model that correlates those to CAC. You'd just ignore the holes and hope that the data retrieved is indicative of the data missing (it would be in my case since I have more data than holes).

- - - Updated - - -

You mean after mileage? Because mileage appears to be the biggest factor so far.

Looking at the PI America Survey here's the CAC vs Mileage chart:

While there is a definite top-left to bottom-right overall trend, mileage doesn't tell anywhere near the whole story. There are low-mileage cars with low CACs compared to cars with three times the mileage, as well as high mileage cars that have very well preserved CAC values.

It'd be interesting to come up with a single value to represent "time spent at temperature" and plot CAC against that. Higher temps are progressively worse, so coming up with that single time-at-temp value would probably be logarithmic or similar.

 

[brianman]

Using Smorg's data (Thanks!), here's another flavor of the "CAC by Mileage" chart. I couldn't get it to make the curves I was looking for, but the result was interesting (to me at least) nonetheless.

• CAC - Original data.
• CAC Min - Adjust all "higher than previous" data points to match prevous data point.
• CAC Max - Adjust all "lower than next" data points to match next data point.
• CAC Avg - Midpoint between CAC Min and CAC Max at each data point.

 

[tomsax]

Using Smorg's data (Thanks!), here's another flavor of the "CAC by Mileage" chart.

What is "Smorg's data"? Is there a data set with CAC values other than the Plug In America Survey data?

 

[brianman]

Looks like the same data, Tom. Didn't mean to confuse the attribution. Smorg sent me a copy of the data for his chart, and that's what I meant.

 

[smorgasbord]

What is "Smorg's data"? Is there a data set with CAC values other than the Plug In America Survey data?

As I said in my earlier post, the data was taken directly from the Plug-In America Survey.

- - - Updated - - -

here's another flavor of the "CAC by Mileage" chart.

It would be interesting to see if there's a correlation between battery temps ("seconds at temp" as output by VMSParser) and where the car is relative to the red and blue lines.

 

[tomsax]

Looks like the same data, Tom. Didn't mean to confuse the attribution. Smorg sent me a copy of the data for his chart, and that's what I meant.

That's cool, I just wanted to make sure I hadn't missed something else smorgasbord had done. I wondered if there was another data set.

At some point soon, I'll make it easy for anyone to download the whole PIA survey data set.

 

[Pantera Dude]

That's cool, I just wanted to make sure I hadn't missed something else smorgasbord had done. I wondered if there was another data set.

At some point soon, I'll make it easy for anyone to download the whole PIA survey data set.

Thanks a million for doing this stuff Tom!!!
I really enjoy looking at the info!
Art

 

[smorgasbord]

OK, so I thought about this some more, and talked it over with a smart friend at lunch. As a first stab, calculate the Median Temperature at which the car's battery has been kept. That's the temperature at which half of the life of the battery has been spent below and half spent above. It's different than mean, or average. Now, plot CAC against that median temperature for all cars.

Unfortunately, I don't have temps vs Seconds for all cars, just my own. My car's median is just below 21 degrees C. Here's what my data looks like:

This maybe can be used on cars of similar age and mileage to see why one battery has more CAC than the other, and I think it would be useful to plot this for all Roadsters for which we have CAC numbers. Unfortunately, Richkae's survey doesn't have CAC and Plug-in America's survey doesn't have the Seconds at Temp data. Somehow, we'd want to correlate the two.

This is going to have serious limitations. It doesn't help comparing cars of same age but different mileage or same mileage but different age. I think what we want to do is use median temperature as a factor by which we increase/decrease the effective mileage on a car's battery. My thinking is time at high temps explains much of the CAC differences we see in different cars at the same mileage.

At any rate, I'm stymied by the lack of correlation between this survey and Plug-in America's. Tom and Rickkae, any chance of you two getting together and use reported VINs to correlate the two for me?

Anyone have a better idea of what the temperature factor might be?

 

[djp]

OK, so I thought about this some more, and talked it over with a smart friend at lunch. As a first stab, calculate the Median Temperature at which the car's battery has been kept. That's the temperature at which half of the life of the battery has been spent below and half spent above. It's different than mean, or average. Now, plot CAC against that median temperature for all cars.

And to take it one step further, we could do the same with time spent at SOC. The US Army did a longevity study on LiCoO2 batteries (same chemistry as the Roadster). They measured capacity against three factors: charge cycles (mileage in our case), temperature and SOC. They found the capacity loss caused by high temperatures could be completely offset if the battery is kept at <50% SOC. It would be interesting to reproduce their results based on actual Roadster data.

http://www.cerdec.army.mil/directorates/docs/cpi/Cycled_Aged_LiIon_Cells.pdf

During analysis of the data a series of trends stood out. The first being, whenever the cell was held at an elevated temperature both the capacity and power capability of the cell decreased significantly (Figures 3-5). The one notable exception is when the cell was maintained at 0% to 50% state-of-charge (Figure 6).

The second is when the cell was maintained at a high state of charge (Figures 3 and 4). The capacity of these cells faded at an increased rate. This is especially true for the cells maintained at a state-of-charge of 50% to 100%.

Here's the time spent at SOC for my pack:

 

[smorgasbord]

I'm thinking all of the Roadsters' Time at SOC will look pretty similar, and that Tesla knows what it's doing in that regard. I believe that even at full Standard charge the batteries are just fine, and even at Range charge, thanks to the active cooling, that batteries are also fine. I don't think we'll find a good correlation between Time at SOC and CAC. I do believe Time at Temps, though is a factor.

 

[hcsharp]

I'm thinking all of the Roadsters' Time at SOC will look pretty similar, and that Tesla knows what it's doing in that regard. I believe that even at full Standard charge the batteries are just fine, and even at Range charge, thanks to the active cooling, that batteries are also fine. I don't think we'll find a good correlation between Time at SOC and CAC. I do believe Time at Temps, though is a factor.

I agree that time at SOC probably won't have much correlation with CAC. I used to think, as you do, that time at temp would be a bigger factor especially since the reactions that cause the battery to lose capacity happen exponentially faster as temps rise. But I'm not so sure anymore. This is not a Leaf. The people I know in warm climates don't seem to be losing range any faster than I am in VT where we have about 2 weeks of summer every year. OK, maybe 6. Looking forward to the analysis if we can do it.

Can you tell me the command line you used to get the data for the graph in post 217 (hours at temp)?