Roadster Owner Based Study of Battery Pack Capacity Over Time

[wycolo]

TOP OFF - No one has mentioned this function lately, so I guess it offers no particular help wrt balancing. Likewise, no particular help wrt making a long-distance trip; if you *really* need that extra 6-8 miles you'd be better off doing a partial RANGE charge for 20 extra miles. So T.O. is mainly for initiating recharging for a Roadster that sits plugged in & idle for weeks at a time, to avoid the need to unplug/plug back in? [ModelS has been hogging the outlet for weeks now; must buy Roadster its own extension!].
--

 

[Doug_G]

Top Off seems to only be useful if you do a Range Mode charge and leave the car sitting for a while. I don't tend to do that.

 

[hcsharp]

This puzzles me. To do so the pack would have to physically disconnect all the high current series connections. If there are 69 strings that's 69 high current connections that have to be broken with large contactors, and then another set of connections in parallel have to be made, though at lower current levels for balancing. When the series connections are open pack voltage would drop to the level of a single string and the car could not be driven. I would think it more likely that Tesla simply bleeds off charge with resistors, or uses charge shuttling from high to low strings.

You misunderstood the design. Breaking the load up into 69 strings is what makes it easy to manage, eliminating the need for contactors. At full power 215kW you have less than 10A per string which is easily switched with electronics. Another advantage of this is that if one cell fails or becomes weak you can disconnect the whole string that it's in with little impact on the operation of the car. This reduces uneven stress on the other cells in the pack, helping prevent what would otherwise be a faster decline of the first few bricks that develop weak or dead cells.

 

[JRP3]

69 strings in series passing only 10 amps?

 

[hcsharp]

69 strings in series passing only 10 amps?

Not even 10 amps. Why would you think it would have more? The strings are not in series with each other. Read my post again and you'll see that the strings are connected in parallel with each other and the cells in each string are in series.

 

[JRP3]

So 99 cells in series x 3.7V = 366.3V at 2.2ah. 2.2 ah x 69 = 151.8 ah x 366.3V = 55.604 kWh. But to get 225kW, 225,000 watts, of power with 366V means pulling 615 amps.
366V x 615 A = 225,090 Watts, 225 kW.

 

[Dave EV]

It's 99 cells in series, 69 strings in parallel.

615A/69 = ~9A per 99 cell string.

 

[tomsax]

The Roadster battery pack consists of 11 sheets. Each sheet has 9 bricks. Each brick has 69 cells in parallel, so each brick voltage is around 4V when charged. The bricks in each sheet are in series, so each sheet produces about 9 * 4 = 36V. The sheets are also in series for a pack voltage around 11 * 36 = 396V nominal.

My understanding is that the cells in each brick are soldered together in parallel and can't be individually addressed. Doing that would require 2x6,831=13,663 transistors capable of 10 amps each. It seems like that would be pretty expensive.

 

[hcsharp]

The Roadster battery pack consists of 11 sheets. Each sheet has 9 bricks. Each brick has 69 cells in parallel, so each brick voltage is around 4V when charged. The bricks in each sheet are in series, so each sheet produces about 9 * 4 = 36V. The sheets are also in series for a pack voltage around 11 * 36 = 396V nominal.

My understanding is that the cells in each brick are soldered together in parallel and can't be individually addressed. Doing that would require 2x6,831=13,663 transistors capable of 10 amps each. It seems like that would be pretty expensive.

That was my understanding for a long time as well. Then recently I was told differently by a Tesla engineer (not a technician). The discussion was the result of questions that I had about what happens when the first cell goes bad in one of the bricks. My concern was that it would cause that brick to wear out significantly faster than the others. In a battery that is only as strong as it's weakest brick, and the weakest brick is getting stressed more than all the others, your whole pack will become useless while most of the cells are still in good condition. That's when I was told that one bad brick cannot be isolated when under power because it's almost impossible to shunt nearly 600 amps. Instead it's designed so one bad cell will not stress any remaining cells more than any others.

I also think this design only requires 6,831 transistors that can handle 500ma or less, in addition to 69 that can handle 10A, which would be inexpensive.

I have not taken a battery apart to verify any of this. But you don't have to think about it very long to realize it makes a lot of sense. If you have evidence that this is wrong, please correct me.

 

[tomsax]

I also think this design only requires 6,831 transistors that can handle 500ma or less, in addition to 69 that can handle 10A, which would be inexpensive.

How is that? The 69 bricks are in series, so each one has to produce about 600A at 36V. The voltages add up to ~400V, but the current through each brick has to be the same.

Because the cells are in parallel in each brick, they have to average 600/69 = 8.7A. You're right that you only need one transistor per cell, but it seems like each one has to be able to handle more than 8.7A.

How do you get 500mA?

 

[hcsharp]

How is that? The 69 bricks are in series, so each one has to produce about 600A at 36V. The voltages add up to ~400V, but the current through each brick has to be the same.

Because the cells are in parallel in each brick, they have to average 600/69 = 8.7A. You're right that you only need one transistor per cell, but it seems like each one has to be able to handle more than 8.7A.

How do you get 500mA?

If you assume that the cells are permanently hard wired in parallel in each brick, then you are correct. On the other hand if a string of 99 cells are permanently wired in series, one cell from each brick, then you only need one small transistor per cell to connect it in parallel to the other cells in the same brick. I guessed at 500ma but that is probably overkill. If the cells in a brick are in a parallel state most of the time then there would never be much current flowing between them. The transistors would mostly be used to disconnect or isolate a string.

 

[richkae]

Here are some charts with all the latest data.
I parsed all the data and some of the log files do not start when the vehicle was new, so the time span is bogus.
It's also clear that any vehicle that doesn't start with 158 or 159 amp hours has a log file that doesn't have data all the way back to new.
Max = the maximum weakest brick amp hour value ( brickahmin ) found in the file
Min = the minimum weakest brick amp hour value ( brickahmin ) found
Last = the last weakest brick amp hour value ( brickahmin ) found
( Thus a vehicle that was very out of balance but then rebalanced will have a last value higher than the min value )

First, amp hour data plotted against the odometer ( miles ).

Second, the amp hour data plotted against the number of days spanned by the log.

Then the same data in percentage form. The percentage is computed assuming 159 amp hours is 100%.

 

[richkae]

Here is a new ( better?) chart.
Min = minimum average brick amp hours ( brickahave )
Max = maximum average brick amp hours ( brickahave )
Last = last average brick amp hours ( brickahave )
Weak = last weak brick amp hours ( brickahmin )

 

[brianman]

@richkae - Are you able to make any "quick" observations and/or guesses about the "min" cases w/r/t usage whether there are charging and/or consumption patterns in common across them? Or is the data too noisy, or do you think it's just variation in quality of the battery packs delivered?

 

[richkae]

@richkae - Are you able to make any "quick" observations and/or guesses about the "min" cases w/r/t usage whether there are charging and/or consumption patterns in common across them? Or is the data too noisy, or do you think it's just variation in quality of the battery packs delivered?

No observations or guesses at this time.
We only have data for 5 really high mileage battery packs.
Unfortunately there isn't a lot of good detailed data, maybe Tom's Plug in America survey can provide some guidance as to what to look for, then we rev the vmsparser and try to gather it from the log files.

 

[markwj]

Btw, I've also provided Tom with a large set of anonymized data from OVMS charge records. I hope it is of some use to narrow down the criteria affecting long-term capacity.

 

[richkae]

Here are a couple of interesting charts.
The first is % battery loss per 10K miles, charted are the 17 vehicles with at least 14K miles.

The line shows that you can expect about 1.75% loss per 10K miles ... if you look on the right at the high mileage cars, the low mileage cars are highly variable.

This one is %battery loss per 10K miles on the horizontal axis.
The stacked charts are % time at different SOC levels. ( The vertical axis is time - totaling 100% not SOC )
T0-15 means time at SOC between 0% and 15% inclusive.
T91-100 means time at SOC between 91% and 100% inclusive.

The vehicles that spend a lot of time >= 86% SOC or below 15% SOC tend to be on the right ( higher %battery loss per 10K miles )
It is consistent with the rule of thumb: don't leave your battery full or empty for long periods of time.
Major caveat: I need to do a bunch of work to verify this data. It might be from a very small sample of the vehicles life and thus not very reliable.

Final note: I defined % battery loss as the % drop of the last "brick ah ave" number in the log file from 159.

 

[brianman]

I really like the second chart. It's hard to tell directly but I'm curious about the middle bands.

More specifically...
1. Exclude T0-15, T16-26, T86-90, and T91-100 (which leaves the two middles: light green and purple).
2. Collect all the green bar heights.
3. Collect all the purple bar heights.
4. Plot green and purple as independent curves, and rescale them so that the total "points" for each curve is matched (i.e. divide each bar height by the total of the bar heights for that color).

I think this would show the percentage within the green and purple bands of each battery loss notch.

I'm curious if the green or purple shows more to the left vs. the right.

I may try eyeballing the numbers on my end to see if I can build the plot myself as well.

 

[hcsharp]

...
Final note: I defined % battery loss as the % drop of the last "brick ah ave" number in the log file from 159.

Thanks again for compiling and posting this data. Interesting indeed. But when I saw your final note, I had to wonder how accurate it is. You assumed all the cars started at 159 Ah. I don't have the data you do but just from informal conversations with other owners I know there is a variance in the Ah when new of about 4 (158 to 162). Maybe you can confirm this with the data you've collected. So if my car started at 158 Ah when new and dropped to 157 at 10K, it would appear on your chart as having lost 1.26% even though it actually only lost half that much - 0.63%. That might also explain why the data points converge on the high mileage cars. How much variance have you seen in the logs for Ah when new?

 

[richkae]

Thanks again for compiling and posting this data. Interesting indeed. But when I saw your final note, I had to wonder how accurate it is. You assumed all the cars started at 159 Ah. I don't have the data you do but just from informal conversations with other owners I know there is a variance in the Ah when new of about 4 (158 to 162). Maybe you can confirm this with the data you've collected. So if my car started at 158 Ah when new and dropped to 157 at 10K, it would appear on your chart as having lost 1.26% even though it actually only lost half that much - 0.63%. That might also explain why the data points converge on the high mileage cars. How much variance have you seen in the logs for Ah when new?

If you look at the x,y plots I posted on 1/24 all of the log file data has a maximum Ah value of 159. Most of the log files show a max value between 156 and 159, none of the ones I have are higher than 159.
That is almost all the data I have ( data from 2 or 3 new vehicles has come in since 1/24 and I haven't made a new plot ) and you can see the full range of values there.
In lots of the log files the first values are not the max values ( they go up 1 or 2 points before declining ) and there is a lot of variation after firmware upgrades.
I assume that any log file I have where the max value is less than 156 is probably incomplete, and is missing some data from the early life of the vehicle.
But you are right, assuming that every vehicle started at 159 is probably not a good assumption.