Behaviour of cells in parallel.

[Wolf]

I'm starting this thread to start a discussion on how cells behave in parallel in a pack.
Before any of you poo poo this and say that all this has been studied and there are many papers out there on this I will concede and agree.
But of all the studies that I have read they all seem to use only the same cells from the same manufacturer whether Lipo, LiIon, LiFePO4and are not mixing cells of different capacity, manufacturers and IR. Neither are they mixing low drain and high drain cells.
https://www.sciencedirect.com/science/article/pii/S0378775316309995
https://www.sciencedirect.com/science/article/pii/S0378775316300015
https://www.sciencedirect.com/science/article/abs/pii/S0378775315306911

The studies all come to a very similar conclusion that as cells degrade the interaction among them causes the cells to be stressed.
I am here to help us understand this phenomenon with down to earth layman's analysis and hopefully a good discussion.
I am no mathematician and most of the formulas the studies use are a complete mystery to me but I do understand charts and experiments.

I have built a board with an ESP32 Dev module and hooked up 4 INA219 I2C Bi-directional DC Current Power Supply Sensors to it. All the INA219
chips are from the same lot so deviation should be minimal if any at all. Along with the cell holder and the ESP32 with the INA219s attached
there is a Buck CC/CV charger and a ZB206+ discharger on it.




The data is going to an influx database to record the4 parameters that the INA219 produces.
Buss Voltage
Load Voltage
Milliamps
Shunt Voltage
The cells will be charged to 4.2V until mA cutoff saturation and then discharged to whatever voltage we want to 3.0V to 2.8 or whatever some of you want to experiment with. IRand voltagewill be measured before the test and recorded.
We can experiment with a charge Voltage of 4.0V and a discharge to 3.2V etc. etc. etc. The range of parameters is unlimited.
You can participate in this study with suggestions of what cells to test in combination of each other.
All of this data is then graphically represented on a grafana dashboard for analysis.
I have added annotations and will continue to do so at the start of each cycle on the milliamp chart to indicate the cell part number IR and Voltage.
As this data accumulates you are more than welcome to look at it at any time http://wolftech.mynetgear.com:4562/...s?orgId=1&from=1565775270104&to=1565786145474 username guest password guestw
To view the data correctly choose a time frame



Once you learn how to use grafana you can drill down to the time frame you want to see and then can extrapolate your own conclusions. Anything prior to 8/11 is experimental and should be disregarded.

The first set of cells you see in this graph are



Discharge curve at 2 Amps for the pack and charge curve at 4 Amps for the pack



The second set are all UR18650FMs which are undergoing a 1A per pack discharge right now. Combined mAh of the 4 is 9615mAh so it will take at least 9 hrs do discharge at that rate. They have been discharging for almost 4hrs.






Questions, comments, rants, and raves bring them on.

Wolf

 

[jdeadman]

This is a Cool experiment. Always wondered with varying Cell health the interaction between the cells.

 

[completelycharged]

Brilliant. Data !!!!!

Suggestions
1. Same cell make/year but more than one cell where the capacities are different, i.e. a cell with 2000mAh another with 1900mAh and a third 1800mAh. Reason : if the accuraacy of your readings are good it may well show up some other traces to actually measure the remaining cell life..
2. Measure changes in IR with different discharge / charge currents, through the voltage range, due to internal chemical reaction differences/distributions. Again this may end up showing a cell life correlation

Loads more suggestions (like over charge and undercharge voltage profile and IR)...

Brilliant.

---

Separate thought... you need the data in excel/open office calc in order to actually play with the numbers and see the real gems... The charts are great, but so it a 6ft target from 100m away, it's the small subtle differenec and detail which may actually prove to be a real will that everyone has seen but nobody has noticed.

 

[Wolf]

Brilliant. Data !!!!!

---

Separate thought... you need the data in excel/open office calc in order to actually play with the numbers and see the real gems... The charts are great, but so it a 6ft target from 100m away, it's the small subtle differenec and detail which may actually prove to be a real will that everyone has seen but nobody has noticed.

---

First off thanks

Ah yes the close up view.
Well I can view the chart by the minute if need be. And so can you!



Or here is a 5 minute x axis view of a different time section.




Alternatively the data can be exported from influx to a csv file.

Better yet as a guest you can also export the data once you select a time frame to CVS format.




Select "Series ascolumns"



Import into Excel or any other sheet program




And Bobs your uncle. Now you can view, manipulate, graph, and lord knows what else you want to do withthe raw data to your hearts content.

Update:
I have created another dashboard with just the raw data that you can download on each individual cell and a combined chart of all the cells.
http://wolftech.mynetgear.com:4562/...anced&orgId=1&refresh=30s&from=now-24h&to=now
same username guest and password guestw



And the export of the All Cells Raw data is perfect. So have fun.






Wolf

 

[jdeadman]

Very Nice dashboard. I'm going to have to look into this for my setup

 

[completelycharged]

The real interesting calculations will be the trace of difference between the cell voltage/IR for a given charge/discharge Wh for different cell ages.

If you have 4 identical cells (different mAh capacities) I would be really interested playing with the data to see where it goes.....

I think there may be a way to determine cell life and the data you are hopefully able to generate could prove this.
Not sure if you read my earlier thread that had no responses, lol.
https://secondlifestorage.com/t-Charge-and-discharge-cycle-data-help

 

[Wolf]

The real interesting calculations will be the trace of difference between the cell voltage/IR for a given charge/discharge Wh for different cell ages.

If you have 4 identical cells (different mAh capacities) I would be really interested playing with the data to see where it goes.....

I think there may be a way to determine cell life and the data you are hopefully able to generate could prove this.
Not sure if you read my earlier thread that had no responses, lol.
https://secondlifestorage.com/t-Charge-and-discharge-cycle-data-help

Completlycharged,

Just read your post from06-17-2018, 09:03 PM
https://secondlifestorage.com/t-Charge-and-discharge-cycle-data-help
Well that was posted just a little before before my time (I joined 09-25-2018, 07:27 AM)but somehow the concept of this study has struck the same cord in both of our minds.

Ideally I would need 1 second measurement data of voltage and current and probably a couple of PDh students looking for a thesis... so anyone interested in helping get in touch, pm or post.

Well I did put a 2 second delay at the end of the loop to give the database a breather from all the data thrown at it, but I can probably remove that.I do have a 960GB SSD drive I will be installing on this laptop soon and that should improve writability a bit.
Phd well sorry you are stuck with me.

I will break apart some packs and try to get a set of identical cells with reasonable IR that I know will pass ,be marginal, and fail.
As you know I have a pretty good handle on IR and know how to judge a cell by it.
So I will try to get 4 cells with different IR readings out of the same pack hence the cells should be from the same lot.
I will capacity test them first on my SKYRC MC3000 and leave them at a state of discharge. Then charge them on my (Phd )board at
4A (1A each) and discharge at 2A (500mA) each. Of course recording V ,mAh.and IR throughout all the steps.
Lets see what happens. I should be able to get to another pack break open session this weekend.

Let me know if that is what you are looking for and anything else you can think of.

Wolf

 

[Oz18650]

This is very interesting Wolf!
I am looking forward to the results.
I think it may shed some light on how to select or reject cells to use and/or build good, usable packs.

 

[daromer]

You can easily push data into inlfux every 10ms if you want... With normal drive... Influx is a time series database.... Just make sure you set it up correctly in terms of resolution and cache in front so it doesnt write each value but instead for instance 5sec at a time.

Doing this you can like example the raspi project i use... I write 500 values per second into that On SD card.

Im not going to comment the work right yet.... Its a well layed out work for sure and I think what you will find is "suprising"
Im very happy you do it so people get to see it. I was thinking of it but life catched up with me unfortunately 2 years back and never got to show my findings.

Keep it up guys! VERY well done so far.

 

[Wolf]

You can easily push data into inlfux every 10ms if you want... With normal drive... Influx is a time series database.... Just make sure you set it up correctly in terms of resolution and cache in front so it doesnt write each value but instead for instance 5sec at a time.

Thanks daromer.
I'm looking at the influxdb.conf file and there are a bunch of variables that can be set for the database. I am pretty much running it in default mode right now. Let me know what settings you would change. I assume they are in [data] and the TSM engine.If I run my query at 1s I get data at about every 3 to 5 seconds which isn't that bad. I do get a bunch of nulls though.



I will eliminate the 2s delay in the sketch at the end of the loop and see what happens.

I have downloaded the raspi image from your website but haven't played with it yet.Don't have a pi yet been concentrating on ESP8266 and ESP32.

Wolf

 

[daromer]

You get the nulls because you arent sending the data the database expects The series you have there is set up to have data every second and since it doesnt it show nulls in between. Its not a problem persue.
Note that if you push data faster than 1 second the data will be interpolated to 1 second values since thats how its created.

I guess the "GUI" that you use creates the nulls because of that. If you do a normal select they wont really show up unless the "time" have any data in it.

https://docs.influxdata.com/influxd...s/#does-the-precision-of-the-timestamp-matter

So in your example you only have a resolution of +-1 Second at most. If you want it coarser you need to make sure to save with such. I hope you get it. There is like tons of documentation of it.

 

[rev0]

Glad to see I'm not the only amateur battery scientist here. I did a similar experiment a while back which you can find here: https://secondlifestorage.com/t-Different-Size-Parallel-Cell-Experiment I did another more specific test paralleling a Leaf cell (LMO chemistry which has a higher average voltage) with standard Li-Ion (NMC or similar chemistry). Results showed what you might expect, the Leaf cells provided most of the power for the top majority of the SoC, then the lower (average) voltage Li-Ion cells kicked in in the last 30% or so of SoC. This is problematic since they will be taking a majority of the current drawn from the overall pack during the lower 30% or so of pack capacity.

 

[Wolf]

Glad to see I'm not the only amateur battery scientist here. I did a similar experiment a while back which you can find here: https://secondlifestorage.com/t-Different-Size-Parallel-Cell-Experiment ........................

Welcome to the club or should I say I'm glad to be joining the club.

Wow completlycharged put out a commission on06-17-2018 for acells in parallelstudy
and you (rev0) on02-19-2018, starteda very similarexperiment.
That will teach me too dig some deeper into the archives
If I can ever figureout the search function.

That being said here are the latest results on a discharge on 4 batteries.
These batteries where parallelcharged on a CC/CV board to 4.2V and sat for about30 min before being inserted into the testing board.
Here are the parametersof the cells.




Notice in the beginning of the chart there is very little balance interaction of the cells as they are practically identical in voltage.
At 6:08 AM I hit the discharge switch at 2.5A and the reaction of1 of the13Qs in the beginning and the other 13Q at the end is interesting.
The 20Cs just more or less lumber along and at the end 2.8V cut off at ~8:00 AMare now charging the 13Qs as the 4p pack is slowly coming to equilibrium. That basically shows me that the 13Qs (understandably so) ran out of gas first and brought the bus voltage to 2.8V which caused the discharge to stop. The 20Cs still had plenty of gas to go as theywere initially filling up their brethren after the 2.8V cut off with ~400mA and as of noon 4 hours later are still supplying them with ~60mA.
So that test begs for a different test,only 13Qs and only 20Cs with the same 2.5A discharge and we will see how they do on time. Also it will be interesting to see how their interaction is after the discharge cutoff. Theoretically the 13Qs should last ~2 hrs and the 20C ~3.2 hrs.I will run those tests in the next couple of days. Before I do that I will run a charge curve on this pack. One thing I have noticed that even though I am charging these packs at 4A,1A each the do not warm up in the slightest. I will see if I can find a "heater" and mix it in and see what transpires.
Man this has opened up a can of worms so many different variable to consider testing for.
Here is the discharge chart



And here are all the parameters.



BTW I have updated Grafana to 6.2.3 so some of the menu items have changed but not by much if you visit the site and notice a difference.
Wolf

---

You get the nulls because you arent sending the data the database expects The series you have there is set up to have data every second and since it doesnt it show nulls in between. Its not a problem persue.................
....................... I hope you get it. There is like tons of documentation of it.

@daromer
No I get it and it is not a huge problem as you can see from my grafana graphs the data is good and represents itself accordingly.
I will take out the 2 second delay at the end of my loop to speed it up a bit. All in all I am very satisfied with the resolution that I am getting from all of this.
I will read up more on influx and such as I have time but for now I think we are doing quite well.

Thanks

Wolf

 

[Korishan]

That will teach me too dig some deeper into the archives
If I can ever figureout the search function.

Just use this search: https://secondlifestorage.com/search.php Most people miss the Advanced Search function.

 

[daromer]

Grafana reads the values. Your GUI to show the data does not. It reads all positions On your current setup i guess you have "second" resolution in the series.

 

[Wolf]

Grafana reads the values. Your GUI to show the data does not. It reads all positions On your current setup i guess you have "second" resolution in the series.

Not sure what you mean by my GUI the picture is of the import into Excel and yes the series wasset to 1 second.
I have played with that and have changed to 10s, 15s, 30s, 1m, and so on. Just to experiment.

Wolf

 

[completelycharged]

Brilliant,

P - personal
H - help with
D - data



Ah, I see how the data is being written. needs to be time syncronised to make the analysis easier, otherwise it's a bit of fliddling to get the numbers consistent. With the tollerances on the values it should be ok to start off with though.

+1 for the cache setting

Awesome..

---

I missed the rev0 post, interesting.

The analysis needs to be done in relation to Wh (or scale for mWh) because the mA is hard to correlate against other cells without a volt context as it removes or aligns the time variable by using Wh.

It's all about the net energy in / out, i.e. Wh...

Separate thought for another type of test, during the discharge process maybe run the discharge for 5 minutes, rest for 2 minutes and then start the discharge again for 5 minutes and repeat. This should show how well the internal chemistry recoveres from a discharge by way of the cell voltage recovering.. But.... the +ve and -ve bus is comon so this test would need to be done one cell at a time.

---

Quick calc...
This is the Wh difference between the cells for the discharge cycle (first cell is the baseline)





Notice the separation half way through.... cell 2 and cell 4 following the same path.

Cell 1 discharges 4.61Wh with the red cell only 3.76Wh

But, this variation is due to the IR differences or cell contact difference.... i think...

---

This is where you want to be..... this is for cell 1.



DischargeWh deltavs Voltage as a scatter

Explanation...
Work out the mWh per time interval the cell has discharged
Create a running total of the cumulative mWh through the discharge period
The take two points, say 60 seconds appart and calculate the difference in the discharge
Then.... plot the value as a series in the first instance

End of the curve..... collapse in discharge rate.

Each cell has a different fingerprint due to the chemistry distribution, age, etc...

---

Will let the chart do the talking....

 

[Wolf]

completlycharged,

Interesting. I'm going to have to read that post about 10 times to figure out what you are saying.:angel:

Meanwhile at 6 AM EST another round is discharging.
This time it's all 18650-20C cells discharge at 2.5A for the pack.




You can already tell the even discharge between the cells.



Theseinitial tests are to get a baseline and understand how cells interact with each other. The more "fun" ones with different chemistries ,and all kinds of variations of IR, and mAh ratings voltage ranges etc will be in the wings.

I hope those who are following this have bookmarked the site where all the data is being displayed.

Wolf

 

[daromer]

Wolf: Upload issue being fixed Thanks

 

[Wolf]

Wolf: Upload issue being fixed Thanks

Great I will edit my post.
Thanks
Wolf