Best way to manage this LFP house battery

[jflorey2]

If I end up having to wire up 112 cells of monitoring, that's preferable to having a system fully down while trying to live in it on the road. However, my premise is that I don't need much monitoring. So far I see no indication that I need that, but it will take time and imbalance observation to know for sure.

I would suggest that if you want to observe and monitor cells so you can replace them "on the fly" - you need to monitor them all.

[nebster]

Your coulomb counter will loose its accuracy over time unless you reset it on a regular basis as the battery does not have a coulomb efficiency of 100% (I have logged the coulomb efficiency of my battery at ~99.6%). How did you plan to reset the counter?

Yeah, max also brought this up. Coulomb-counting is notorious for wandering off the path, but fortunately:

a) I'm aware of that inclination.
b) My counter can reset on any number of criteria, so I'll just have it detect the end of a generator- or shore-based CC charge cycle.
c) I'm really only using the SOC as a convenience value. I believe I will tie all the real decisioning to voltage (and temperature, in an emergency).

[nebster]

With reference to the following graph I think the best way to charge to say 90% if you using a generator or some other charger that has a CC output is to set the CV point to the voltage where the voltage reaches 90%SOC at whatever current you are charging at and terminate the charge immediately it reaches the set charge voltage with no absorb/CV charging. With these cells and a charge rate of ~0.2C this would equate to a CV point of ~3.45V/cell. This method will not work with any consistently if you are charging from solar or the charger doesn't charge at a constant current or for low charge rates. One good reason for not having any CV charging when charging from a generator is the extra generator runtime necessary to charge during the absorb phase.

Thanks, I haven't considered going with a lower CC rate, but that's mostly because I anticipate using the genset to charge underway, and I'd like to run that at as high a load for as short a time as possible. That is in conflict with my desire to extend absorb times past 1 hour, though. I don't yet have a mechanism to stop CV beyond a crude timer with a minimum of 1 hour. I've definitely got to work that out!

I am not a fan of running an LFP battery without an individual cell BMS mainly because of unforeseen stuffups and the very real chance of a fault within a cell reducing its capacity which could easily lead to it being overcharged and outside its safe operating zone.

What do you think about only monitoring a handful of the "troublemaker" cells? So far it seems completely consistent: the same couple cells are the ones with presumably the least capacity, so they spike up much earlier than the rest. I'd really like to avoid rigging up 100+ monitoring circuits if I can help it.

How often do you intend to redo the bottom balance?

I was planning to observe the system and deduce when a rebalance was necessary. It would be unacceptable to have to do it every few months, but once or twice a year would be okay. I expect we will cycle the pack roughly twice a week in real life.

I know there are (unmonitored) bottom-balanced LFP packs staying very well balanced after several years in the field, so I have been assuming (hoping) that mine would behave similarly.

[nebster]

Well, you can do that, but then you need 7 BMSes. Seems cheaper/easier to avoid that.

If I end up having to wire up 112 cells of monitoring, that's preferable to having a system fully down while trying to live in it on the road. However, my premise is that I don't need much monitoring. So far I see no indication that I need that, but it will take time and imbalance observation to know for sure.

[jflorey2]

I really want to avoid a single 16s pack so that I can isolate individual strings or cells without shutting down the system. With 100 cells, the odds of a cell (or bus bar, or lug, or whatever) needing work are pretty high.

Well, you can do that, but then you need 7 BMSes. Seems cheaper/easier to avoid that.

[jflorey2]

Each group you parallel, should be matched capacity, so you do not have a lo cap cell bundled with a high cap cell, which would set you up for a certain fire in a couple weeks.

I think I might be misunderstanding you here. You can parallel LiFePO4 cells of slightly different capacities - but you cannot easily series them.

Let's take an example. You have four cells you want to build a 2S2P pack out of - a 100Ah, a 110AH, a 120AH, and a 130AH.

You have two choices. You can try to match the parallel cells as closely as possible first. Then you end up with two paralleled units, one 210AH, one 250AH. You then series these. You will have serious problems operating this pack without a good BMS. With a good BMS you will be able to get 210ah out of it.

Second choice is to choose groupings to match series capacity. Then you end up with two paralleled units, one 230AH (100+130) and the other 230AH (110+120). You then series these. You will have no problems operating this pack without a BMS provided they stay in balance (which they will for a while) - and you will get 230AH out of it.

[nebster]

your picture is not showing for me, not sure about others

it doesn't

I think it must be you?

1600W load at 48V creates about 33A current so your 0.78V 'sag' means 23 mOhm internal resistance over 16 cells + wires between them. I was under impression those cells have x10 less internal resistance but then your wires / connectors might contributed to that.

Yes, that seems high to me as well. The cells themselves average about 0.3mOhm each, so there is considerable additional resistance contributed by the rest of it. My guess is that a big chunk of it is at the terminal fuse. I have a FLIR camera and can't see any asymmetric heating on any of the pack wiring itself, so I suspect it is okay.

I have another 16s string almost ready to go, so I'll soon be able to compare two setups and see if maybe I have a connection making trouble.

All I'm saying when putting them in CV the current will be defined by CC output CV voltage and battery voltage difference divided by those 23 mOhm. Or your charge controller max output current making it actually CC . If controller is powerful enough it is very easy to screw up unless the cells are already on their upper knee where they can increase their voltage to compensate.

I don't quite follow this paragraph. Can you explain for me?

[nebster]

and you should be- try to have some independent dumb 'turn it all off' circuit next to sophisticated BMS just to make sure something can shut it down in case of bugs.

Exactly. There will be numerous safety nets, including a totally "dumb" thermal fuse line wired to the contactor. If all else failed, those fuses will deenergize the coil and, hopefully, the contactors will break the circuit.

Amount of energy you're dealing with here is serious so I'd be careful even working around them in manual mode. Due to high energy density and low thermal capacity the temperature will rise to hundreds degrees in seconds.

Yes, that's right, care is necessary. However, a counterpoint: there are hundreds (maybe thousands) of boats and RVs with similar setups running in the wild now, most with owners who are not paying careful attention day to day. To my knowledge, there are zero known fires or any failure modes amounting to more consequence than cell venting/bloat in these installations. This body of accumulated safety is the main reason I am comfortable with a similar pack.

you'd need to come up with auto- reset procedure for your SOC counter preferably in each cycle. As all integrating things it will drift off with time from any error. You always get 'some' number trouble is to relate it to anything in reality.

My coulomb counter has a configurable reset. Currently it observes the end of one of these CC/CV cycles and assumes the pack is at 100% at the end of it.

I believe I am going to trigger charge cycles on voltage, however. The SOC is for convenience.

Relatively to Li-ion- yes, relatively to lead acid- not at all. As I said due to low thermal inertia and very low internal resistance they can do things. Check recent thread here where Mike posted pics of LFP pack cases completely burned out. I mean cells were gone completely, only metal enclosures remained.

Can you send me a link? I searched for posts by Mike with the word "fire" in them and got no hits.

A completely atomized LFP pack would be a sight to see, indeed. Some kind of major safety failure would have had to exist -- like an unconstrained overcharge, I imagine? -- for something like this to happen.

you got mighty good price on them, all I saw before was x3 times higher, was it due to volume discount?

$300-400 per kWh is pretty standard for all LFP prismatics these days?

than I thought I'd still put fire proof separators between cells: aluminum burns very well in open air if you heat it high enough and those cells pack more than enough starting ignition. The idea is to prevent one bad cell to start chain reaction by igniting nearby cells. No, air channels between cells won't do it for me.

The air channels are not to prevent fires, they're to keep all the cells at the same temperature. Uneven heating might encourage cell level imbalance, but more importantly, the cells degrade much faster if they're operated at slightly higher temperatures.

I'm putting my whole bank into a steel enclosure, but I don't hazard any illusions about it: if the total energy in the thing could dump fast enough, it would be a disaster for sure. We do hold property insurance!

[max2k]

your picture is not showing for me, not sure about others

I'm not convinced LFP chemistry follows a single Peukert curve anyway.)

it doesn't

CC was capped at 50A and stayed in this regime for 103 minutes at about 54.00V average, and then the charger switched to CV at 54.8V after a total of about 4.5kWh charger output. In CV, the highest cell reached 4.600V after 39 minutes, and at that point I stopped the charge manually. The rough energy area under this portion of the curve is about 800Wh, so ~5.3kWh total for a 3.2V*100Ah*16, 5.120kWh pack. Throwing in 7% for charging losses and the 6% overcapacity, I think this looks about right.

My new theory is that I can reduce the absorption voltage threshold even further and move into CV earlier, pushing absorption out well past one hour, and then the timer approach will at least "work" until I decide what other criteria to use to stop charging. I really wish I could just turn off absorption and float, but so far I haven't found a way to do that.

1600W load at 48V creates about 33A current so your 0.78V 'sag' means 23 mOhm internal resistance over 16 cells + wires between them. I was under impression those cells have x10 less internal resistance but then your wires / connectors might contributed to that. All I'm saying when putting them in CV the current will be defined by CC output CV voltage and battery voltage difference divided by those 23 mOhm. Or your charge controller max output current making it actually CC . If controller is powerful enough it is very easy to screw up unless the cells are already on their upper knee where they can increase their voltage to compensate.

[max2k]

Thanks for your thoughts, max2k.
My DVMs are accurate, but the reality is that the automated charge control will happen with... simple, inexpensive voltmeters embedded on various PCBs. I'm far less worried about my test setup, where I'm the human BMS, than any automated framework.

and you should be- try to have some independent dumb 'turn it all off' circuit next to sophisticated BMS just to make sure something can shut it down in case of bugs. Amount of energy you're dealing with here is serious so I'd be careful even working around them in manual mode. Due to high energy density and low thermal capacity the temperature will rise to hundreds degrees in seconds. BTW inexpensive 16 bit ADC has accuracy better than 0.1%, enough for this application.

Interestingly, with my cells at least, I find it really easy to observe the SOC versus voltage. It must be because it's 16s, so I'm getting a 6x increase in resolution for "free." The discharge curve is super linear in the middle of the SOC band. But it's mostly irrelevant, because my coulomb counter is also plenty accurate enough for basic SOC estimation given the safety band I'm willing to apply (from above).

you'd need to come up with auto- reset procedure for your SOC counter preferably in each cycle. As all integrating things it will drift off with time from any error. You always get 'some' number trouble is to relate it to anything in reality.

It's pretty hard to get a true thermal runaway with a fire on an LFP cell, which is why I've chosen them instead of just harvesting Tesla packs or something.

Relatively to Li-ion- yes, relatively to lead acid- not at all. As I said due to low thermal inertia and very low internal resistance they can do things. Check recent thread here where Mike posted pics of LFP pack cases completely burned out. I mean cells were gone completely, only metal enclosures remained.

I went through alibaba and paid about $0.35/Wh including air shipping. I'm pretty happy, because these cells fit the form factor I have available best and they are the newer aluminum-shell design with a much higher volumetric and gravitometric energy density. Their layout also establishes air channels between all the metal cases, and I can see this improved cooling on the FLIR camera when I watch the pack heat under charging.

you got mighty good price on them, all I saw before was x3 times higher, was it due to volume discount?

Please be safe and respect your cells especially in charged state. With the improvements you listed they may be got closer to the 'ideal' battery but that 'ideal' battery besides perfect electrical parameters is also dangerous chunk of energy stored in small volume. If I ever build my own bank and with such prices that time might be closer than I thought I'd still put fire proof separators between cells: aluminum burns very well in open air if you heat it high enough and those cells pack more than enough starting ignition. The idea is to prevent one bad cell to start chain reaction by igniting nearby cells. No, air channels between cells won't do it for me.

Try not to follow path of the early experiments on uranium critical mass where on one occasion a researcher was holding neutron reflector hemisphere off the plutonium core with screw driver he wedged in there with expected outcome:

https://en.wikipedia.org/wiki/Demon_core

[karrak]

Interestingly, with my cells at least, I find it really easy to observe the SOC versus voltage. It must be because it's 16s, so I'm getting a 6x increase in resolution for "free." The discharge curve is super linear in the middle of the SOC band. But it's mostly irrelevant, because my coulomb counter is also plenty accurate enough for basic SOC estimation given the safety band I'm willing to apply (from above).

Your coulomb counter will loose its accuracy over time unless you reset it on a regular basis as the battery does not have a coulomb efficiency of 100% (I have logged the coulomb efficiency of my battery at ~99.6%). How did you plan to reset the counter?

Simon

Off grid 24V system, 6x190W Solar Panels, 32x90ah Winston LiFeYPO4 batteries installed April 2013
BMS - Homemade Battery logger github.com/simat/BatteryMonitor/wiki
Latronics 4kW Inverter, homemade MPPT controller

[karrak]

With reference to the following graph I think the best way to charge to say 90% if you using a generator or some other charger that has a CC output is to set the CV point to the voltage where the voltage reaches 90%SOC at whatever current you are charging at and terminate the charge immediately it reaches the set charge voltage with no absorb/CV charging. With these cells and a charge rate of ~0.2C this would equate to a CV point of ~3.45V/cell. This method will not work with any consistently if you are charging from solar or the charger doesn't charge at a constant current or for low charge rates. One good reason for not having any CV charging when charging from a generator is the extra generator runtime necessary to charge during the absorb phase.

ChargeDischargeCurves.jpg
I am not a fan of running an LFP battery without an individual cell BMS mainly because of unforeseen stuffups and the very real chance of a fault within a cell reducing its capacity which could easily lead to it being overcharged and outside its safe operating zone.

How often do you intend to redo the bottom balance?

Simon

Off grid 24V system, 6x190W Solar Panels, 32x90ah Winston LiFeYPO4 batteries installed April 2013
BMS - Homemade Battery logger github.com/simat/BatteryMonitor/wiki
Latronics 4kW Inverter, homemade MPPT controller

[nebster]

Here's a charge curve from tonight's experimental run. This is a single 16s string that was bottom-balanced last week. Charger was set to 54.8v CV/absorb for max 1 hour, which as I mentioned above seems to be the lowest setting I can achieve right now without getting some kind of software tweak. Pack was discharged to 51.98V with a 100W continuous load (and briefly shown to sag to 51.20V with a 1600W load, visible in the graph). Cells were within 3mV of each other at this point. Coulomb-counting suggested DoD should be about 90% at this point, but secretly we know that all cells came from the factory at about 106Ah, so the SOC estimating is a bit conservative. (I set Peukert coefficient to 1.0 for now, and frankly I'm not convinced LFP chemistry follows a single Peukert curve anyway.)

CC was capped at 50A and stayed in this regime for 103 minutes at about 54.00V average, and then the charger switched to CV at 54.8V after a total of about 4.5kWh charger output. In CV, the highest cell reached 4.600V after 39 minutes, and at that point I stopped the charge manually. The rough energy area under this portion of the curve is about 800Wh, so ~5.3kWh total for a 3.2V*100Ah*16, 5.120kWh pack. Throwing in 7% for charging losses and the 6% overcapacity, I think this looks about right.

My new theory is that I can reduce the absorption voltage threshold even further and move into CV earlier, pushing absorption out well past one hour, and then the timer approach will at least "work" until I decide what other criteria to use to stop charging. I really wish I could just turn off absorption and float, but so far I haven't found a way to do that.

[nebster]

Hey Mike,

First, thanks for moderating the forum.

These are good questions you are asking, but I think you need a lot more study before you are ready to buy and build, one of these time bombs.

Thanks for the sort of sideways compliment, but... I've read most of the forum, I've already "bought" and "built," for better or worse, and I'm about as aware as any amateur can be but am interested in input in case there are great ideas I'm missing.

Each group you parallel, should be matched capacity, so you do not have a lo cap cell bundled with a high cap cell, which would set you up for a certain fire in a couple weeks.
You CANNOT trust the mfg or vendor to do this for you.

I don't really understand these suggestions, so let me challenge you with my "dumb retorts" and see if you can help me understand better:

1) I'm not paralleling any cells at this time. I do intend to parallel all of the 16s strings, and I do want the capacities of each string to be similar, yes. That happens mostly automatically, first since these are brand new cells all manufactured together and second since I'm connecting 16 of them. The large number in series smooths the small amount of capacity raggedness cell to cell (measured 1.7% max to min at the factory, btw). Could you clarify what you mean about matching capacities for paralleling? Further, could you quantify what you mean by "matched," for my pack topology?

2) LFP is really, really hard to set on fire. But I am going to try pretty hard, in the lab, before these get installed in my RV! I do wonder why you think a fire is "certain" in "a couple weeks." What happens in a couple weeks?

3) I'm not sure where I led you to believe that the battery mfg was doing any pack design for me. On the other hand, I would LOVE it if a vendor would have designed this for me, except none of them have anything that fits in my form factor. Most of them are either using more aggressive chemistries (NMC, e.g.) or the older plastic cells with half the energy density, and then they are charging me the standard 3x markup for the convenience. (!)

So, yeah, it's just little old me.

[nebster]

Thanks for your thoughts, max2k.

It sounds like you can manage your initially manually balanced bank just using total battery voltage, especially if you're willing to sacrifice a little of capacity.

Yes, I will probably stay between about [10%, 80-85%] SOC most of the time, with an occasional manual push to 95% or something if something strange is anticipated.

I would never use timer to control charging process. Conservatively the bottom knee in voltage to watch for during discharge is around 3.0V /cell while the upper to watch for during charge is 3.6V / cell. You need to make sure your DMM has enough accuracy (better than 0.5%) to perform these measurements.

My DVMs are accurate, but the reality is that the automated charge control will happen with... simple, inexpensive voltmeters embedded on various PCBs. I'm far less worried about my test setup, where I'm the human BMS, than any automated framework.

Between those voltages LFP has very flat charge/discharge curve making it difficult to guess its SOC using voltage.

Interestingly, with my cells at least, I find it really easy to observe the SOC versus voltage. It must be because it's 16s, so I'm getting a 6x increase in resolution for "free." The discharge curve is super linear in the middle of the SOC band. But it's mostly irrelevant, because my coulomb counter is also plenty accurate enough for basic SOC estimation given the safety band I'm willing to apply (from above).

I think your arrangement of 16s7p would be preferable from overall safety point of view- you can fuse each string individually and in case of a cell developing short in one of them string fuse would blow preventing other cells killing entire string and starting fire.

I do like having many fuses and string independence, yes.

It's pretty hard to get a true thermal runaway with a fire on an LFP cell, which is why I've chosen them instead of just harvesting Tesla packs or something.

Can you share where did you source your cells from and @ what price?

I went through alibaba and paid about $0.35/Wh including air shipping. I'm pretty happy, because these cells fit the form factor I have available best and they are the newer aluminum-shell design with a much higher volumetric and gravitometric energy density. Their layout also establishes air channels between all the metal cases, and I can see this improved cooling on the FLIR camera when I watch the pack heat under charging.