LFP voltage variation from loads?

[tanglewood]

For those with LFP banks in use, I'm wondering how much voltage variation you see as larger loads come on and off? I'm trying to sort out how much hysteresis to put in my PLC BMS for alarming and other actions. Please provide both the load size and your battery bank voltage and Ah capacity so we can keep responses in context.


My goal of course is to minimize/eliminate nuisance alarms, while at the same time taking appropriate action when called for. One possibility, if it's called for, would be to monitor current and scale all my voltage set points accordingly. But I'm not sure that's necessary.


Thanks in advance

[john61ct]

Has to be based on C rate, not absolute amps.

Bigger the bank less effect for a given current.

Any load big enough to drop bank V by more than a few tenths of a volt (once surface charge / resting V is gone), is big enough to quickly deplete the bank.

The load itself is not effected by any drops, until SoC is close to the bottom.

So to me the fact that the LVD disconnects at a higher SoC with the big loads, and a lower SoC with the normal ones, is actually A Good Thing.

So I would not add the complexity of auto-adjusting the global LVD setpoint based on current, myself.

If that were a concern I would install LVDs at different setpoints for different circuits (lower for essentials vs higher for non) or specific high current loads.

In that way the innermost global-BMS LVD protection setpoint should never be hit, unless the outer user-level LVDs fail.

And of course alarms should notify at points higher than cutoffs.

[Maine Sail]

In my experience about 10 to as much as 30 seconds of hysteresis, for LVD open, should work. Voltage will definitely dip on in-rush but it rebounds pretty fast.



Best is to just test it simulating your largest load and at the lowest temperature you expect to use the cells at.... Take a cell to 80% DOD (or your preferred lowest point you expect to encounter) then slap a Fluke 289 or similar onto a cells pos & neg and capture the peak low and monitor voltage sag duration.



An o-scope would be even better and would give a real glimpse at actual duration. Upon high in-rush LFP cells do sag but they also rebound quickly even under pretty high loads. I would suggest measuring cell voltages for your hysteresis LVD protection programming not just global bank voltage.



Also keep in mind that your cell to cell interconnections will not all be the same resistance, as much as we'd like them to be, so one cells apparent voltage may dip further upon a large in-rush than another. Programming a short hysteresis can also help with this.

[tanglewood]

Originally Posted by Maine Sail

In my experience about 10 to as much as 30 seconds of hysteresis, for LVD open, should work. Voltage will definitely dip on in-rush but it rebounds pretty fast.



Best is to just test it simulating your largest load and at the lowest temperature you expect to use the cells at.... Take a cell to 80% DOD (or your preferred lowest point you expect to encounter) then slap a Fluke 289 or similar onto a cells pos & neg and capture the peak low and monitor voltage sag duration.



An o-scope would be even better and would give a real glimpse at actual duration. Upon high in-rush LFP cells do sag but they also rebound quickly even under pretty high loads. I would suggest measuring cell voltages for your hysteresis LVD protection programming not just global bank voltage.



Also keep in mind that your cell to cell interconnections will not all be the same resistance, as much as we'd like them to be, so one cells apparent voltage may dip further upon a large in-rush than another. Programming a short hysteresis can also help with this.

Thanks. All my monitoring is per cell, or more accurately per nP block of parallel cells. And triggers are based off the lowest or highest, depending on which direction things are going.


As I think about it more, the hysteresis will be for warning when things are getting too high or too low, and then battery disconnect action when things continue to go awry and last-ditch battery protection is required. So another way to look at it is how long can one of these conditions last before it becomes a problem. I'm thinking that 30 seconds would be pretty safe, i.e. if the battery exceeds one of the set points for 30 seconds there won't be much risk of damage. Now this assumes that there isn't a sudden drop to zero or something like that, so I'm counting on any decline in voltage or climb in voltage to happen over more than 30 seconds, and probably no faster than 1 minute or maybe 2. But I'm just thinking out loud and would welcome any suggestions.

[Maine Sail]

Originally Posted by tanglewood

Thanks. All my monitoring is per cell, or more accurately per nP block of parallel cells. And triggers are based off the lowest or highest, depending on which direction things are going.


As I think about it more, the hysteresis will be for warning when things are getting too high or too low, and then battery disconnect action when things continue to go awry and last-ditch battery protection is required. So another way to look at it is how long can one of these conditions last before it becomes a problem. I'm thinking that 30 seconds would be pretty safe, i.e. if the battery exceeds one of the set points for 30 seconds there won't be much risk of damage. Now this assumes that there isn't a sudden drop to zero or something like that, so I'm counting on any decline in voltage or climb in voltage to happen over more than 30 seconds, and probably no faster than 1 minute or maybe 2. But I'm just thinking out loud and would welcome any suggestions.

Bingo.. I too prefer a warning level HVD audible first then physical and LVD audible first then physical and then a finally "oh $hit" global disconnect contactor. The "oh $hit" level is only used if the warning levels protection were to trip up...

[nebster]

Originally Posted by tanglewood

For those with LFP banks in use, I'm wondering how much voltage variation you see as larger loads come on and off? I'm trying to sort out how much hysteresis to put in my PLC BMS for alarming and other actions. Please provide both the load size and your battery bank voltage and Ah capacity so we can keep responses in context.

Pack: 16s6p, 53V nominal, ~630Ah/33kWh.

At a baseline load of 5A, pack voltage at the bus bars measured 53.23, with a true SOC of roughly 92%.

At 40A, V = 53.03.
At 124A, V = 52.55.
At 289A, V = 51.37.

...or, about 7mOhm pack Ri.

15kW is the largest continuous load I can place on the battery pack, although the system is engineered for a 500A peak for 1 second.

Short of putting a scope on the cells, I don't have a way to see a sag transient. Watching casually, though, the voltage stabilizes extremely quickly at these loads. Under two seconds for sure.

I have a headless cell-level monitor (custom firmware Zeva 8-cell monitors, doubled up, times six) set to LVD at 2.60Vpc, and it has never tripped. I will admit I have not tried putting a crazy load on the pack at very low SOC. I also don't know its internal hysteresis model.

My goal of course is to minimize/eliminate nuisance alarms, while at the same time taking appropriate action when called for. One possibility, if it's called for, would be to monitor current and scale all my voltage set points accordingly. But I'm not sure that's necessary.

You know I advocate for defense-in-depth. One thing I do is set up a "soft LVD" on my inverter, where it is readily programmable and also pretty fancy. This is a pack-level voltage, so it cannot see an aberrant cell going wild, but on the other hand I set it quite a bit higher than 41.6V.

The nice thing about Victron now is that the recent firmware has support for "dynamic current control," which is their term for "current-based voltage cutoff." You can program in your own cutoff curve. It does not have any hysteresis window that is configurable, but I assume they do some temporal smoothing internally as well.

Even if you have other discharge sources, for a house bank it is likely that your big inverter (or inverter stack) will present the lion's share of the load. So you can use this type of disconnect feature as a first line of defense.

[nebster]

Originally Posted by john61ct

Has to be based on C rate, not absolute amps.

That's why he asked for load size and bank capacity.

Bigger the bank less effect for a given current.

True.

Any load big enough to drop bank V by more than a few tenths of a volt (once surface charge / resting V is gone), is big enough to quickly deplete the bank.

False.

Load sag scales linearly as function of bank voltage. Bigger pack, bigger drop. A few tenths of a volt is nothing on a 52V pack.

Have you ever run a large house battery bank, John?

The load itself is not effected by any drops, until SoC is close to the bottom.

As stated, objectively false. I don't even know what this means, but I suspect you are thinking about a 12V pack.

So to me the fact that the LVD disconnects at a higher SoC with the big loads, and a lower SoC with the normal ones, is actually A Good Thing.

While it seems like that quirk of the chemistry would help, in most cases it doesn't provide any practical advantage as long as loads can range across something broad like 0C to 0.5C. You either need current-compensated disconnect, or you have a very low disconnect that offers no real protection, or you get nuisance tripping.

And, no matter what, you need cell-level detection for the real problems, and that level can be set low enough that it does not come into play during normal load sag.

[nebster]

Originally Posted by tanglewood

I'm thinking that 30 seconds would be pretty safe, i.e. if the battery exceeds one of the set points for 30 seconds there won't be much risk of damage.

In my pack's cell level monitoring, I have two setpoints each for high and low. The first setpoint for high trips a "stop charge" command to the charger. The first setpoint for low trips a "stop inverting" command to the inverter.

The outer setpoints actually pull the red cord and disconnect the string. At a max charge, my estimate is that the HVD would be reached approximately 3 minutes after the inner point. And on discharge, I have them spaced a little farther apart to achieve about the same timing (since my maximum discharge rate is larger).

I think 30 seconds is fine. Especially if you are staying out of the knees anyway, there is a lot of conservatism already baked in. You have wiggle room.

Where you might run into trouble is if you don't use a rebalancing system and you get a cell or cells that go off the reservation. I don't have experience with that, but I imagine see needing some more aggressive setpoints and then possibly a different hysteresis window there. But, hopefully your pack won't behave like that. Bring some extra cells along just in case.

[nebster]

Originally Posted by Maine Sail

Bingo.. I too prefer a warning level HVD audible first then physical and LVD audible first then physical and then a finally "oh $hit" global disconnect contactor. The "oh $hit" level is only used if the warning levels protection were to trip up...

Indeed. Me, too.

I have warnings in software to send audible alerts/texts/email. Then, load or charge disable. Then, string disconnect.

Also, thermal protection completely out of band with most of the other voltage-based logic. The thermal fuses go straight to the bus-level contactors.

[tanglewood]

Thanks Nebster, for all your comments.


As you and MaineSail have described, I have a two stage alarming system, but I might make it three stage. Let's look and charging, and it can be an exercise for the reader to sort out the corresponding discharge example...


Right now I have two thresholds.


The first is a warning that things are higher than expected. It's per-cell, so the alarm will trip when the highest cell exceeds the threshold. It turns the icon for that cell yellow, and sends me an email, but no other action is taken.


The last threshold is the panic, push the big red button. The goal of course is to never have that happen.


I could use the lower warning threshold to also signal "stop charging". But my objective is to program all the chargers so they stop on their own before that threshold is reached. That's why the threshold is a warning and considered abnormal, and alerts me. As such, I wouldn't want to use it as a normal operation "stop charging" signal to control a charger because I'd be getting warnings every time the batteries reach their charge point, and that's not abnormal and not worthy of notification.


This has prompted me to consider adding a third threshold specifically for "stop charging". It would not be considered abnormal, but rather just a signal to chargers to stop, moving control from the charger to the BMS. I mentioned this movement of control in another thread, and my belief that it's where we will end up as devices mature to accommodate LFP. So this new "stop charging" signal/threshold would be a normal operation control signal to indicate when the batteries are full and that chargers should stop, and the other two higher thresholds would be a high-voltage warning, followed by an over voltage alarm and panic disconnect.


Make sense? Overkill?

[john61ct]

Actually your results confirm what I said.

As a %, the only load with a significant drop was the very high C rate one.

My "few tenths" was meant at normal House 12V, sorry I did not clarify that.

[john61ct]

Originally Posted by nebster

You either need current-compensated disconnect, or you have a very low disconnect that offers no real protection, or you get nuisance tripping.

Not with normal gentle House bank usage, avoiding the voltage shoulders.

Nothing to do with your 50V high C rate context.

In any case irrelevant wrt my approach, as you say in-depth, so the very large non-essential loads are cutoff at much higher SoC than the small more important ones,

and all loads are cut off long before the "BMS functionality",

that last ditch never used unless the above fails, so can be set wherever you think appropriate

[nebster]

Originally Posted by tanglewood

I could use the lower warning threshold to also signal "stop charging". But my objective is to program all the chargers so they stop on their own before that threshold is reached. That's why the threshold is a warning and considered abnormal, and alerts me. As such, I wouldn't want to use it as a normal operation "stop charging" signal to control a charger because I'd be getting warnings every time the batteries reach their charge point, and that's not abnormal and not worthy of notification.

I also let the chargers control their own destiny vis a vis stopping. However, their charging logic is overridden/complemented by a stop-charge message coming from the monitor, too.

Maybe thinking of it in "levels" is a good way to go; e.g.:

Level 0: normal charging
Level 1: charging stopped by voltage/time condition
Level 2: charging stopped by cell monitor
Level 3: string disconnected by cell monitor
Level 4: pack disconnected by thermal monitor

Levels 2-4 are exceptions, and 2 is where the alerting begins.

This has prompted me to consider adding a third threshold specifically for "stop charging". It would not be considered abnormal, but rather just a signal to chargers to stop, moving control from the charger to the BMS. I mentioned this movement of control in another thread, and my belief that it's where we will end up as devices mature to accommodate LFP. So this new "stop charging" signal/threshold would be a normal operation control signal to indicate when the batteries are full and that chargers should stop, and the other two higher thresholds would be a high-voltage warning, followed by an over voltage alarm and panic disconnect.

I think the concept of moving the control to the BMS is orthogonal to how many levels you elect. You could have a Level 1-style determination made in the BMS and still have a Level 2-style stop decision also made in the BMS based on other criteria, for example.

Some of the newer BMSes, as well as at least Victron gear, now support not just sending charge/stop charge but also a max current value over CAN or via another higher-level protocol. They can say "please charge at nA," which I think is pretty neat.

Also interestingly, Victron has started to roll out distributed charge control, wherein you can set it to divvy up the requested charge amongst multiple devices. You can prioritize PV over shore, say, and only supplement with generator if needed, all while conforming to the current maximum being specified by the BMS. These decisions are made at one central computer and then distributed to the chargers, so this is another potential component in the decisionmaking pipeline.