Low cost BMS

[rhills]

the LiPO cells I've installed...

Rob, do you really mean LiPo (as in Lithium Polymer), or LFP (Lithium Iron Phosphate, LiFePO₄)? It makes a big difference.

LFP is nominally 3.2, and should not go much beyond 3.6 V.
Nearly all other lithium chemistries are nominally 3.7 or 3.75 V, and should not go beyond 4.2 V.

You're quite correct @coulomb, I was careless in my nomenclature. I meant LiFeP04

[brendon_m]

They most definitely don't start bypassing till the cell reaches 3.8v and stop bypassing at 3.6v. These units shift 5 amps with a cell voltage of 3.8v or higher

The other cell top units start balancing at 3.5V but not full current and work up to the full bypass current (in my case 700mA) which I guess is done using a zener diode set to pass at 3.5V and only the voltage over 3.5 goes across the balance resistor.
Eg the cell is at 3.8V and it's balancing 700mA.
3.8V - 3.5V = 0.3V
0.3V/0.7A = 0.428 Ohms

So at 3.6V its
3.6V - 3.5V = 0.1V
0.1V/0.428 Ohms = 0.234A

And for the 5A balance at 3.8V
0.3V/5A = 0.06 Ohms
Then 3.6v would be
0.1V/0.06 Ohms = 1.667A

Now it could be the units you have don't start to pass until they avalanche at 3.8V and then burn off until they get back down below 3.5v. So they are really all or nothing rather than a small amount of bypass at first and working up to full current only if the cell is a long way out of balance, and because it doesn't do small balancing it will, when it does balance it has alot of energy to dissipate as heat in a short period of time.
Doing a small amount of bypass will add the same amount of heat energy but spread out over hours not minutes so it won't get as hot.

[weber]

@T1 Terry, It's possible that some old models of the EV Power CMUs started bypassing at a higher voltage, but the present-day specification clearly states that it starts bypassing at 3.5 V.

But I agree that the present-day EV Power spec of not turning on the alarm until 4.0 V leaves much to be desired, as does its use of the cell interconnecting links as a heatsink. As I've said, our CMUs alarm at 3.65 V and don't use the cell or interconnects as a heatsink.

Although I was unwilling to take any of my new aluminium-case cells beyond 3.65 V, I have been holding one of my 10.5-year-old plastic-case 40 Ah Sky Energy cells at 4.0 V for the past 5 hours. Admittedly I haven't thermally insulated it, but I am comparing its temperature to an identical but unconnected cell standing beside it. I charged both cells to 3.8 V last night (it felt very wicked ) then I let them rest over night. This morning around 9:00 am I checked that they were both at the same temperature, and connected one of them to a lab supply set to 4.00 V while leaving the other one unconnected.

The current into the connected cell was initially about 0.5 A, but dropped rapidly to near zero within a few minutes.

After 4 hours I measured the current at 1.5 mA ± 0.2 mA and measured the two cell temperatures as identical to within ±0.05 °C (the resolution of my IR thermometer). And at 5 hours the situation was the same.

So you are simply wrong about high voltage alone causing heating of cells. However, that constant 1.5 mA does represent continual slow damage being done to the cell.

[brendon_m]

But I agree that the present-day EV Power spec of not turning on the alarm until 4.0 V leaves much to be desired

The other big downside is not being able to check an individual cell.

[T1 Terry]

@T1 Terry, It's possible that some old models of the EV Power CMUs started bypassing at a higher voltage, but the present-day specification clearly states that it starts bypassing at 3.5 V.

But I agree that the present-day EV Power spec of not turning on the alarm until 4.0 V leaves much to be desired, as does its use of the cell interconnecting links as a heatsink. As I've said, our CMUs alarm at 3.65 V and don't use the cell or interconnects as a heatsink.

Although I was unwilling to take any of my new aluminium-case cells beyond 3.65 V, I have been holding one of my 10.5-year-old plastic-case 40 Ah Sky Energy cells at 4.0 V for the past 5 hours. Admittedly I haven't thermally insulated it, but I am comparing its temperature to an identical but unconnected cell standing beside it. I charged both cells to 3.8 V last night (it felt very wicked ) then I let them rest over night. This morning around 9:00 am I checked that they were both at the same temperature, and connected one of them to a lab supply set to 4.00 V while leaving the other one unconnected.

The current into the connected cell was initially about 0.5 A, but dropped rapidly to near zero within a few minutes.

After 4 hours I measured the current at 1.5 mA ± 0.2 mA and measured the two cell temperatures as identical to within ±0.05 °C (the resolution of my IR thermometer). And at 5 hours the situation was the same.

So you are simply wrong about high voltage alone causing heating of cells. However, that constant 1.5 mA does represent continual slow damage being done to the cell.

Interesting, no gassing or cell swelling? When the charging is disconnected and a few amp load connected across the cell, how long does it remain above 3.6v? I'm just wondering if the internal resistance is so high that @ 4v there is very little current flowing into the cell that would result in the movement of the lithium ions
or
is there no lithium ions left to move out of the LiFe coating into the electrolyte to cause the over saturation? *

I must admit, I've never tried it on an old cell to test the temperature increase, but I know the charging voltage can go very high before the current starts to flow. Sometimes once the current does start to flow the voltage drops back, but that is part of a different contentious issue

*As far as I understand it, the voltage only rises (as does the temperature) when the lithium ions become over saturated in the electrolyte because there is no available spots in the graphite to accept them.

T1 Terry

[weber]

Interesting, no gassing or cell swelling?

Thanks for reminding me. I was interested in this question myself. I deliberately chose a cell that was not already noticeably swollen. I measured its thickness with dial calipers last night before topping it up, as close to the middle of its broad faces as I could get, and wrote it down. 46.8 mm. The nominal cell thickness is 46 mm.

I just went and measured it now. 46.7 mm. No swelling.

I certainly expect that one of the things that 1.5 mA is doing is generating gas, but it is clearly not enough to have a measurable effect in 5 hours.

When the charging is disconnected and a few amp load connected across the cell, how long does it remain above 3.6v?

Only a few [edit: tens of] seconds. But, as expected, after I disconnect the load (a headlight bulb, which drew about 2 A) the voltage slowly rises above 3.6 V again.

I will now try to measure what the final current is after the cell has been held at 3.6 V for many hours.

[T1 Terry]


"When the charging is disconnected and a few amp load connected across the cell, how long does it remain above 3.6v?
Only a few [edit: tens of] seconds. But, as expected, after I disconnect the load (a headlight bulb) the voltage slowly rises above 3.6 V again."

Sounds like a high internal resistance cell, possibly has serious crap broken down from the electrolyte coating the graphite plates. The result is a very high differential between charging voltage where current is accepted into the cell compared to the discharge voltage where current is drawn from the cell.
Back to the subject though, it does put into question my first assumption that it was the high voltage in the electrolyte causing the heating but does lean towards the idea of an excess of lithium ions with no path out of the electrolyte causing the problem ... but just a personal theory really because I don't have a clue as to how to prove or disprove the theory. I do know serious over/under voltage of the cell does cause the partial separation of the electrolyte and coating of the plates, the more often this occurs the greater the coating and therefore the greater to resistance to electron flow in or out of the plates.

T1 Terry

[weber]

Yes. As I tried to explain earlier: Both high voltage and high temperature accelerate the ageing processes, one of which is an increase in the thickness of the otherwise beneficial SEI layer, which causes an increase in internal resistance. High voltage alone doesn't cause heating. Current causes heating. But high voltage accelerates these destructive reactions directly. It doesn't have to do it via heating.

As best I can tell, the final current after holding the nominally 40 Ah cell at 3.6 V for many hours is somewhere between 0 mA and 0.2 mA. That can be compared with the 1.5 mA ±0.2 mA I measured at 4.0 V. [Edit: Corrected "3.8 V" to "4.0 V".]

[T1 Terry]

As a matter of interest, what voltage is required for the current to increase to an amount that would be regarded a charging current, say above 1 amp or more? Once this threshold has been reached, if the voltage is again dialled back to 3.8v, is the current still only 1.5mA?

T1 Terry

[weber]

As a matter of interest, what voltage is required for the current to increase to an amount that would be regarded a charging current, say above 1 amp or more? Once this threshold has been reached, if the voltage is again dialled back to 3.8v, is the current still only 1.5mA?

Sorry Terry, but I've already discharged the cell to about 90% SoC to put it back in storage until it can go into my home energy system. And I need to work on other things. But you reminded me that it was actually 4.0 V at which I measured the 1.5 mA. I have corrected the above post.

At the end of the discharge to approximately 90% SoC, the headlight bulb was still drawing 1.99 A even though the voltage was now 3.269 V (which shows the approximately constant-current nature of incandescent bulbs). On disconnecting the bulb the voltage rose to 3.280 V within two updates of the multimeter (4 updates per second). The third update showed 3.281 V. When rested for a few hours, I expect it will have risen to the magic number of 3.333 V.

That gives an internal resistance of (3.280 - 3.269)/1.99 = 5.5 mΩ. That's about double what it was 7 years ago, as I remember the battery of 218 cells would sag from 720 V to 600 V with a current of 200 A (5C, 120 kW) when we floored the accelerator in the MX-5. (720-600)/200/218 = 2.75 mΩ. Of course internal resistance is highly dependent on temperature. Today the cell was at 26 °C.

[4Springs]

New version of the Low Cost BMS - this one for the LG E63 NMC cells that Francisco has been importing viewtopic.php?f=17&t=5909

Since these cells have the terminals at each end, the old celltop board needed a bit of modification.
This version bolts on to the negative terminal, with a flying lead to the positive. There are two conductors, one for voltage sense, and one to conduct the current to run the bypass resistors.

Screenshot at 2020-03-23 20-57-07.png (16.77 KiB) Viewed 1317 times

The code has had a few changes.
I added a feature to calibrate the cell voltage, as I was annoyed by the differences I saw with all my new cells being exactly the same, but showing up as different! It is almost certainly not a problem, but I decided to fix it anyway. The way it works is that you send a new 'calibrate' command to the celltop board, followed by the 'set' voltage. The celltop sees the command, looks at the set voltage, and calculates an offset to apply to its own voltage value. From then on, when reporting the voltage, the celltop applies that offset to the raw value.
So to calibrate, you measure the voltage on the cell with a good multimeter. Then you tell the BMS Master what the voltage should be, and tell it to set the voltage to suit. This requires re-compiling the Master code - the calibrate function is not something you can choose to do on the fly. Hook this Master up to the celltop board, and the celltop will adjust itself. If you have more than one with the same voltage, daisychain them all together and they'll all be done at once.

Another change is due to the difference between the discharge curve of the old LFP cells and these new NMC cells. With the LFP cells, the BMS would turn on the balancing resistors at a certain set voltage (say 3.6V). Then when all cells reached that voltage, the BMS would turn off the charger. This could only be performed at a high state of charge, because the voltage doesn't change much until those high charge states are reached. The difference between two cells, both at 3.30V, could be 50% state of charge!

With NMC cells, the graph is much more linear. This means that the cells can be balanced at (I assume) any voltage. So I've changed the algorithm, such that everything hinges on the voltage of the cell with the lowest voltage. If a cell is more than 0.05V above the voltage of the lowest cell, the balancing resistors are turned on.
This should mean that the pack can stay balanced, even if it is never charged to 100%. It also means that the initial balancing can happen 24 hours a day, not just in the last few minutes of charging. Much less temptation to help things along with benchtop power supplies or resistive loads!

Re-reading through what I've just written, I've been inspired to check the spec sheet for the E63 cells. I find that the difference in state of charge represented by 0.05 V could be as much as 10%. This is very significant, so the voltage calibration mentioned above is more important than I thought. I might also change the balancing algorithm to get the voltages closer than 0.05 V.

[francisco.shi]

I was working on my BMS yesterday and was working thru the same problems as you, just dealt with them in a different way.
viewtopic.php?p=78179#p78179
I do a two point calibration and the master is the one that does all the calculations. In my BMS the nodes have very simple logic. To avoid any chance of the bypass flattening the cells the bypass only runs for a short time (which is settable by software) so if you stop sending the bypass command it will only bypass for a short time then stop. If there are no coms activity it turns it self off. I have a FET to turn it off completely that means zero power is used when the node is off. Activity on the coms turns it back on.
I am using an Arduino at the moment for master. I am jot sure if I should make a shield or make a whole board.
Maybe we should compare notes.

[francisco.shi]

How much is your bypass current?

[4Springs]

I do a two point calibration and the master is the one that does all the calculations. In my BMS the nodes have very simple logic. To avoid any chance of the bypass flattening the cells the bypass only runs for a short time (which is settable by software) so if you stop sending the bypass command it will only bypass for a short time then stop.

Interesting, and sounds quite similar to the Low Cost BMS.
These celltop boards will not bypass if they lose communications. If a celltop board doesn't hear anything then it sends out a message "I don't hear anything!". This is passed on to the next one, which increments a counter and passes the message on to the next one, which does the same. By the time the message gets back to the Master, it says something along the lines of "The celltop board 23 boards ago said it didn't hear anything". So the Master puts up a fault on the screen saying that board 23 is reporting a problem.
If it was a celltop board that drained the cells, then presumably it must have been a hardware fault (most likely something mis-soldered). I took the board off and tested it on the bench, but could find no fault. I had tested all the boards before assembling. I haven't put that one back on again...

With the 10 Ohm resistors fitted, the bypass resistance is 6.67 Ohms. At 4 V this would be a maximum of 0.6 A, or 2.4 W. When told to bypass, the boards receive a proportion value, and they use PWM to only turn on their bypass the amount that they are told. I tested these boards on these cells, and decided that 40% was as high as I was willing to go before the heat becomes too much.
If a board fault meant that the bypass was on 100%, the board would drain 0.6 Ah per hour. This adds up to 14.4 Ah per day, and 100 Ah per week (but in reality it would drop as the voltage dropped). The 5 cells in parallel would have 315 Ah capacity, but appear to be shipped quite empty. So a faulty board could probably drain the 5 cells in only a few days.

[francisco.shi]

Do the nodes turn off or go into some low power mode to avoid discharging the battery when not being used?
At 40% that means you generally bypass at about 0.24A?
After doing the balancing yesterday I am starting to think that maybe 5A is an overkill. I have also dropped the max current to about 3A. At 3A the transformer gets to about 40°C so I don't think 5A is worth the risk.
My nodes do not send anything unless they get data so if something goes wrong you can not tell which node is faulty.
There is an LED that flashes when data is transmitted so I figured this would be enough to see where the break is.

[francisco.shi]

My calibration compensates for scale and offset.
I take a measurement at one known voltage then at a different known voltage and from that I work out a scale and offset.
I have also made a setup to do it automatically. It is too much trouble to do it by hand.
In my case the master calculates the corrections and then it sends it to each node individually. The nodes then save the values into flash.

[4Springs]

Do the nodes turn off or go into some low power mode to avoid discharging the battery when not being used?
At 40% that means you generally bypass at about 0.24A?
After doing the balancing yesterday I am starting to think that maybe 5A is an overkill. I have also dropped the max current to about 3A. At 3A the transformer gets to about 40°C so I don't think 5A is worth the risk.
My nodes do not send anything unless they get data so if something goes wrong you can not tell which node is faulty.
There is an LED that flashes when data is transmitted so I figured this would be enough to see where the break is.

No, I don't implement any low power mode. There is one on the PIC, but in full flight the board only draws 0.002 A. So about 100,000 hours (11 years) to drain my 5 cells in parallel from full.
Yes, I'd agree that 5 A is overkill. You have plenty of time, so low current over a long time is just as effective, but should be cheaper and not as scary.
Indeed, 0.2 A or so. We'll see if it is enough, I can always increase it. But I reckon it will be heaps. Just this first ever balancing that takes a long time.

Most BMS nodes have indicators of some kind on them, but I've not seen the point. It adds components, cost and time to the build, and in most cases you can't see (or hear) them anyway once assembled. The Master also keeps a log of the faults. So if you have a celltop board with a fault that shows only under acceleration, you don't have to try to see the board while you're driving. Stop the car and look at the fault log.

My calibration compensates for scale and offset.
I take a measurement at one known voltage then at a different known voltage and from that I work out a scale and offset.
I have also made a setup to do it automatically. It is too much trouble to do it by hand.
In my case the master calculates the corrections and then it sends it to each node individually. The nodes then save the values into flash.

One weakness in the calibration algorithm that I've implemented is that it is stored in volatile memory. So if the board is ever disconnected, or if the watchdog timer reboots it, it will lose the calibration.

[4Springs]

One weakness in the calibration algorithm that I've implemented is that it is stored in volatile memory. So if the board is ever disconnected, or if the watchdog timer reboots it, it will lose the calibration.

I've learned a bit more about the PIC microprocessors, and I'm now using non-volatile storage for the calibration offsets. So the calibration can be calculated and entered when the PIC is first programmed.

The NMC balancing is quite different to LFP balancing. It seems to be working well, so time to describe it:
I've set a threshold voltage, over which cells will balance if required. No balancing will ever occur below 3.60 v. This is an arbitrary number chosen by me because I don't like the idea of a half depleted pack depleting itself further. Even if it is only milliamps...
If above 3.60 v, the BMS looks at all the cell voltages and compares them with the lowest one. If more than 0.01 v higher, a cell board will turn on the bypass resistors. Not on fully though, it proportions the current bypassed using PWM. So at 0.02 v difference a cell board will bypass at a very low rate. If the voltage is further out, it will bypass more. I have the maximum set to 0.1 v, so if it is 0.1 v higher than the lowest cell it will bypass at the highest rate possible.
Since NMC cells can be balanced at any voltage, this means that balancing is occurring during charging, discharging and idle times. The algorithm is designed to take advantage of the long time scales available. A very low bypass current is the norm (the lowest current is about 2 mA), and this might be turned on for days. Low heat on the bypass resistors should mean less stress to the components. But if a cell gets a long way out of balance it will bypass more aggressively.

What I've observed is that the cells sit within about 0.02 v of each other, and there are always some cells in bypass. The BMS shows the number of cells bypassing, and that number flicks around all the time from low numbers to high numbers. It might show 3, then change to 40, then back to 12 within a few seconds.

[weber]

That sounds good 4Springs. Although I've never felt the need to use explicit PWM, because when you just turn bypass on or off at the threshold (e.g. your lowest + 10 mV, no hysteresis), they will balance as quickly as possible, and when they get close to balanced they naturally enter into a sort of random PWM at the frequency of the voltage sample and compare loop. And our bypass resistor is a zig-zagging track over almost one entire side of the PCB, so there is no heat concentration.