Good thesis on preserving the life of LiFePo4 lithium batteries

[CatNewBee]

The BMV are digital measurement devices that measure the voltage drop over the analogue shunt. They do many frequent measurement and integrate the value measured, then build a mean value per time frame, adjust the digital ADC reading to a real Amp value by a transformation to a Voltage and calculation of Amps for a given shunt factor and add this up to the internal Ah counter, multiply the value with the mean voltage reading on the Input and add the value to the internal Wh counter. Depending on current direction (charging / discharging) the value is corrected while charging by the charge efficiency and Peukert factor. There is also an additional noise filtering that you can set to suppress small readings to zero.

All this adds up to the Error Drift over time.

Just to explain, assuming the measurement (sampling) frequency is very low like every 5 seconds, the one BMV measures at time point t+2s, the second measures at t+4s, current is 0 at t=0 and is switched on with 5A at t=4 and off at t=10.

The first counter would measure at t=2 0A
The second would measure at t=4 5A
the first counter measures at t=7 5A
the second measures at t=9 5A
the first measures at t=12 0A
the second measures at t=14 0A

The aggregation of counter one is 5A for 5 seconds (1 measurements)
The aggregation of counter two is 5A for 10 seconds (2 measurements)

The reality was, the current was 5A for 6 seconds.

So there is a drift of 100%, the second counter has the double reading and neither has an exact reading (first -20%, second +66%).

Usually the frequency of measuring is much higher and the duration of the current draft is much longer, so the error is very small and also stochastic.
But if you think about a PWM controller what effectively pulses current while charging and the sampling frequency is near or a multiple of the PWM frequency you can have this huge systematic drift due to interference effects.

There are a lot of other influences on the reading like temperature drift of the ADC reference voltage, noise etc. so even with identical settings there will be a drift over time. The only way to mitigate this is to have something that re-sets the counters frequently before the drift becomes too significant.

You can re-set them on lover voltage points, but then they will be not so accurate, if you reset them on a balanced battery near the upper shoulder, you have a much more precise, reproducible set point, because small changes in SOC produce a big difference in Voltage.

[john61ct]

But going high-voltage reduces lifespan, the more regularly / frequently over time the more so.

To me there is no point in setting a 100% level so much higher than I want my bank to ever be.

In fact, removing surface charge with say a quick 1AH draw would make the calibration more useful to me.

Finally, I can't see the value in a more precise calibration, if going higher does even give that.

[CatNewBee]

Of course, it is up to you...

I would suggest to do the initial calibration when living on board and cycling / testing the battery and then maybe PSOC cycling for later use accepting some drifts, and do sync the counters when manualy going to 100% for balancing every few months or so.

You can program the solar charger or inverter combo setting for (manual or timed) equalization at 14.5..14.6V to kick the balancing thresholds of your BMS then and also do the BMV sync.

[john61ct]

Originally Posted by CatNewBee

do sync the counters when manualy going to 100% for balancing every few months or so.

You can program the solar charger or inverter combo setting for (manual or timed) equalization at 14.5..14.6V to kick the balancing thresholds of your BMS then and also do the BMV sync.

My LFP do not go over 3.45Vpc in use, that **is** 100% for me, I do not consider that PSOC cycling, leftover lead thinking that.

Initial balance at 3.6Vpc, never yet seen any need to re-do it.

[Maine Sail]

Originally Posted by nebster

How many "cycles" in PSOC are we talking about? How far out of sync does it get for you? (How do you know how far out of sync it is, when you conclude that it is?)

For the last 8 or so years all we have really had decent access to in the is Coulomb counting monitors without going to an expensive BMS with built in SOC calculation..



They certainly work ok but Ah counting is imprecise over the long term and most all of these designs, and software algorithms, were designed for lead acid. For example the Peukert, on most of these, is based on a 20 hour rate and only works/subtracts for discharge rates above the 20 hour rate but they don't add for Ah capacity when discharging below the 20 hour rate. Some LFP cells are rated at 1C, .5C, .25C, .2C etc. but I have not yet seen any rated at .05C. Mine gets out of sync in 5-10 cycles more than I am willing to accept and I have had both a Victron BMV and a Link-Pro installed on our boat and tested both on my test bench. It is currently a Link-Pro and there's no sense changing it when I'll just replace it with the new SG-200.

Originally Posted by nebster

When you talk about all of the counters being inaccurate, are you adjusting their model inputs over time? For example, on Victron BMV, do you tweak the charge efficiency and peukert exponent? Those are our two levers into the mathematical model inside that specific device, for example.

Absolutely and I have experimented at length with a myriad of different settings for LFP and tested multiple different Ah counters. For more info on my background with Ah counters (article related to LA) this may give you some more insight.

Making Your Battery Monitor More Accurate

Originally Posted by nebster

Do you use the same shunt for all the ones you've tested?

I only use the correct manufacturer supplied shunt even if they are both 500A / 50mV etc..

Originally Posted by nebster

Btw, I don't mean to hammer you specifically with questions, and really we could start a whole separate thread on this topic to get broader input.

Not an issue, and yes this could be a very long thread...

Originally Posted by nebster

But my experience has been that current-based SOC estimation is still pretty good after 10 days of PSOC cycling.

I guys my question would be what is this experience based on and how did you arrive at it and can we quantify "pretty good"..?

For LFP I consider good to be within 3.5% and quite good to be within 2.5% and excellent to be within 1.5%, below that I don't think measurement is accurate enough to say either way unless an SOC monitor were to move in increments below 1% jumps at a time. My issue is not that an Ah counter can't achieve this resolution it is that it can't do this long term without owner intervention.

Originally Posted by nebster

Certainly good enough to accommodate decision-making along the lines of, "hmm, we're at 40% right now; let's charge for an hour and bring the pack up to 65% for the day."

Ah counters can be great for this, if kept in sync with the bank, but I want something that can do better and that can work for long duration's without requiring a trip to 100% SOC in order to re-sync. Of course you have to know your banks actual capacity, or darn close, as a starting point, or it's GIGO garbage in > garbage out.

Originally Posted by nebster

I'd like to understand what's going on that's making your experience so different.

I don't think it's any different, I need to sync every 5-10 cycles if I want the accuracy I am looking for. I don't want to have to do that nor do I want to have to explain how to do all that and how to capacity test etc. to customers.

My testing to arrive at this has not been on our boat. That would not yield a platform I would consider accurate enough for good data unless I moved all my test equipment on-board. I suspect my wife would draw the line on that one...

I've conducted the testing & drifting on my testing bench with various cells, counters and with varying cycling and charge rates. For example I will simulate our boats loading and cycling and let it rip for 15-20+ days, never charging back to 100%, then discharge the bank back to 0%. From that I can easily see how far out the Ah counter is from the actual battery SOC when I stopped testing. It is not uncommon for it to be 5-10% off.

I want a device I can comfortably send a customer off with, who's not an EE, and know the monitor is telling them the truth, or darn close, about SOC and that I know has not drifted.

I know this is an extreme example but this was a customer boat looking at bank 2 data. Bank 2 is a small start battery. Voltage was fine at 12.70V


The Ah counter was showing it down by -264Ah....... It's not even a 60Ah battery. While this is extreme drift, and lead acid which is much worse than LFP to track, it is still a long tedious explanation to the customer.

This is why I desire something that requires less human intervention and that won't drift over time on SOC..


Today I am still working on Alpha testing the Balmar SG-200 for lead acid. The SG-200 is between the computer and the battery and is uses it's own proprietary shunt. On the shelf above the bench is a long term float test of LFP cylindrical cells held at 3.450V.... On the shelf above that is a dead "drop-in" LFP, the kind with internal BMS that "can't be ruined" though folks seem to find a way.. As can be seen I've started to dissect it......

[nebster]

Originally Posted by Maine Sail

For the last 8 or so years all we have really had decent access to in the is Coulomb counting monitors without going to an expensive BMS with built in SOC calculation..

Some fancy BMSes, to the best I can tell with the documentation I've read, seem to use a mixture of current measurement (usually through hall effect sensors, which are way less accurate) and voltage for state estimation. That is what I am assuming your new beta test gadget will do as well, in some way.

They certainly work ok but Ah counting is imprecise over the long term

While I agree that counting could be imprecise over the long term, and that could be one of the many reasons we see inaccuracy, note that your lead-in on the article you linked to states the exact opposite.

and most all of these designs, and software algorithms, were designed for lead acid. For example the Peukert, on most of these, is based on a 20 hour rate and only works/subtracts for discharge rates above the 20 hour rate but they don't add for Ah capacity when discharging below the 20 hour rate.

Irrelevant for LFP. The contribution from any Peukert-like effect in a lithium pack is almost certainly not a substantial contributor to drift. I know this because I turn that compensation off and I get pretty good results. And there are others also getting acceptable results.

Absolutely and I have experimented at length with a myriad of different settings for LFP and tested multiple different Ah counters. For more info on my background with Ah counters (article related to LA) this may give you some more insight.

Making Your Battery Monitor More Accurate

Thanks, I read your site a long time ago and appreciated its depth and many of the issues you bring up in various areas. However, I don't think most of what you discuss in that page has anything to do with LFP, because we don't have non-linear coulombic efficiency (in the linear voltage regime) nor do our chemistries behave like Peukert's characterization.

That said, what settings did you settle on for your Victron BMV?

I only use the correct manufacturer supplied shunt even if they are both 500A / 50mV etc..

At least Victron monitors are compatible with a wide range of shunts. They even switch which shunts they source over time for the same SKU.

I asked about shunts because it is possible that your shunt is contributing to your issue in some way. If your monitor and loads will support it, you might consider a smaller shunt for more precision, too.

For LFP I consider good to be within 3.5% and quite good to be within 2.5% and excellent to be within 1.5%, below that I don't think measurement is accurate enough to say either way unless an SOC monitor were to move in increments below 1% jumps at a time. My issue is not that an Ah counter can't achieve this resolution it is that it can't do this long term without owner intervention.

So you'd be happy with a drift of about 4% over... 10 round trip cycles from 25% to 75% SOC? Or is it 20 cycles? (Or infinite cycles? )

Obviously everyone's tolerance is going to vary, but I'm asking you to be precise. Then, maybe some of us can mimic what you think is a good target and see what we achieve for the sake of having a comparison.

Ah counters can be great for this, if kept in sync with the bank, but I want something that can do better and that can work for long duration's without requiring a trip to 100% SOC in order to re-sync. Of course you have to know your banks actual capacity, or darn close, as a starting point, or it's GIGO garbage in > garbage out.

I think what would be helpful would be a configurable sync button on our monitors. One that, when pushed, sets the SOC/consumed energy to a specific value. Victron, and presumably others, have a way to force it to 100%.

If we had that, then you could achieve a sync at many more points along the charge curve with reasonable accuracy. Moreso if you have a higher voltage pack. I have shared in another thread that, to first order, I can look at the steady-state voltage of my pack to 4 significant figures and tell you its SOC within 5% if you tell me the current and if that current has been steady for the last minute or two.

Taking it one step further, I see no reason a computer shouldn't be able to do the same thing, but better and constantly. It can look at the current, wait for a period where the load is very stable for long enough to be meaningful, observe the voltage curve over that time period, and project the asymptotic voltage based on the shape of the curve at that point. From there it is a direct lookup to SOC, and surely we can reach your 3.5% preference at that point.

My testing to arrive at this has not been on our boat. That would not yield a platform I would consider accurate enough for good data unless I moved all my test equipment on-board. I suspect my wife would draw the line on that one...

Hehe, it gets even worse with a large pack. It's just unrealistic, even on the bench at home, to do many full capacity drawdowns. Unless you're willing to patch the inverter into your house's panel!

I've conducted the testing & drifting on my testing bench with various cells, counters and with varying cycling and charge rates. For example I will simulate our boats loading and cycling and let it rip for 15-20+ days, never charging back to 100%, then discharge the bank back to 0%. From that I can easily see how far out the Ah counter is from the actual battery SOC when I stopped testing. It is not uncommon for it to be 5-10% off.

So, what were your settings on your Victron BMV for an LFP pack? (Let's leave lead acid results and quandaries to the side: they are much more complex to model and can add additional sources of inaccuracy that I think are irrelevant for this thread.)

I want a device I can comfortably send a customer off with, who's not an EE, and know the monitor is telling them the truth, or darn close, about SOC and that I know has not drifted.

So I think there are only two automated possibilities for this dream system.

Option 1. Configure a charge regime that is achieved regularly and that uses the chemistry to enforce a reliable SOC. Tie the traditional battery monitor's 100% sync to that stopping point. For this to be okay, we must decide:

a) that the stopping SOC's affect on the lifetime use of the pack is acceptable and

b) that the charging source(s) can achieve this syncing SOC reasonably often enough to be fruitful.

Whether (a) and (b) are true is going to vary. But I contend at least in some cases it can be done at relatively safe stopping SOCs, well below the true 100% that keeps getting mentioned here.

Option 2. Use a more sophisticated algorithm to derive state from voltage by leveraging the time domain. The regular monitors we have today don't do this. Such a model will still have estimation error, but the sources of it will be very different, and I don't think those sources would generate the kind of systematic drift that drives you and your customers nuts.

I suppose option 3 would be a mix of these two. I know at least a couple of BMSes that do it this way.

[Maine Sail]

Originally Posted by nebster

While I agree that counting could be imprecise over the long term, and that could be one of the many reasons we see inaccuracy, note that your lead-in on the article you linked to states the exact opposite.

The counting is good, it's just a calculator, it is what it is counting it against that can be problematic. With LFP much less so but still needs re-sync..

Originally Posted by nebster

Irrelevant for LFP. The contribution from any Peukert-like effect in a lithium pack is almost certainly not a substantial contributor to drift. I know this because I turn that compensation off and I get pretty good results. And there are others also getting acceptable results.

But what is good and acceptable and how has this been confirmed? Victron recommended 1.05 (their cells are also Winston) and I found that is less accurate than PC OFF/1.00 or 1.01.

Originally Posted by nebster

That said, what settings did you settle on for your Victron BMV?

I think I was happiest with PC 1.00 and CEF 100% but I played with a lot of different settings. Victron strongly recommends at least bi-monthly syncs with LFP and I tend to agree with this.

Originally Posted by nebster

At least Victron monitors are compatible with a wide range of shunts. They even switch which shunts they source over time for the same SKU.

I asked about shunts because it is possible that your shunt is contributing to your issue in some way. If your monitor and loads will support it, you might consider a smaller shunt for more precision, too.

I am likely not seeing more drift than you in 5-10 cycles but I want better. I don't believe this is a shunt issue because I have tested this with multiple BMV's and they all perform similarly.

Originally Posted by nebster

So you'd be happy with a drift of about 4% over... 10 round trip cycles from 25% to 75% SOC? Or is it 20 cycles? (Or infinite cycles? )

Sure, 4% is fine, but I don't want to have to re-sync or get back to 100% SOC that often. I would be happy to never go see 100% SOC in 5 or 6 months.

Originally Posted by nebster

Obviously everyone's tolerance is going to vary, but I'm asking you to be precise. Then, maybe some of us can mimic what you think is a good target and see what we achieve for the sake of having a comparison.

Again, it is not about how accurate it can be for a few cycles between re-sync, I want easy tracking of SOC that customers can't mess up.

Originally Posted by nebster

I think what would be helpful would be a configurable sync button on our monitors. One that, when pushed, sets the SOC/consumed energy to a specific value. Victron, and presumably others, have a way to force it to 100%.

My counters are almost always manually re-set so PV and other situations can trick them into a premature reset. On the BMV series you simply press and hold SETUP then hit the + button until you get to SYNC then hit SELECT then press and hold SETUP until it resets to 100%. With the Link-Pro it is easier and all you to is press the left and right arrows simultaneously...

Originally Posted by nebster

So I think there are only two automated possibilities for this dream system.


Option 2. Use a more sophisticated algorithm to derive state from voltage by leveraging the time domain. The regular monitors we have today don't do this. Such a model will still have estimation error, but the sources of it will be very different, and I don't think those sources would generate the kind of systematic drift that drives you and your customers nuts.

I suppose option 3 would be a mix of these two. I know at least a couple of BMSes that do it this way.

I think you might be be very surprised with the new SG-200 when it finally hits the streets. The original Smartgauge simply does not work for LFP but the new one was specifically designed from the ground up to track LFP SOC and also provide an SOH feature...

[john61ct]

Is Merlin involved or a Balmar inside design?

[nebster]

Originally Posted by Maine Sail

The counting is good, it's just a calculator, it is what it is counting it against that can be problematic. With LFP much less so but still needs re-sync..

I'm not willing to let the counting off the hook. I think the time-integration could certainly be subject to sampling error, especially high frequency aliasing as discussed above.

It would be nice to understand better how much various issues contribute to the accumulated error.

But what is good and acceptable and how has this been confirmed? Victron recommended 1.05 (their cells are also Winston) and I found that is less accurate than PC OFF/1.00 or 1.01.

Yeah, Victron recommends a lot of stuff that is merely inadequate, and even sometimes abjectly incorrect.* It's up to the installer to get it right, like most stuff.

I think I was happiest with PC 1.00 and CEF 100% but I played with a lot of different settings. Victron strongly recommends at least bi-monthly syncs with LFP and I tend to agree with this.

The fact that their suggestion is specific to a time value and not the things that likely introduce substantial error (current, current-related follow-on effects like heating, etc.) points out how rule-of-thumb it must be.

That said, every two months seems pretty reasonable to take a pack up to a high SOC. At least for someone living on board.

I am likely not seeing more drift than you in 5-10 cycles but I want better. I don't believe this is a shunt issue because I have tested this with multiple BMV's and they all perform similarly.

I don't think the shunts can produce much systematic drift... whatever imprecision they introduce is symmetric; the misread inbound current gets offset by the misread outbound current later.

Sure, 4% is fine, but I don't want to have to re-sync or get back to 100% SOC that often. I would be happy to never go see 100% SOC in 5 or 6 months.

So would 90% be okay, do you think? I think 90% is pretty achievable, but you then have to accept that the monitor is not measuring total ("true") SOC.

I think 85% might be an achievable sync point on a 16s pack, but I think 90% is very conservative already and more than safe for cell longevity. (We can agree to disagree there; I plan to be dead in <2000 cycles so my horizon may be shorter than others'.)

Again, it is not about how accurate it can be for a few cycles between re-sync, I want easy tracking of SOC that customers can't mess up.

I think there is the possibility of setting an automated sync point that is low enough that it happens regularly without requiring the chemistry to be juiced beyond recognition. I'm not advocating for it as the final and best solution; I'm saying that I believe it's a reasonable setup for many that should basically allow for hands-off monitoring.

I also think there are some folks who will prefer to stay topped up to a safe 90% or whatever, whenever possible. The person who prefers to stay at 50% SOC one day, and then 80% the next day, and then is okay with 20% for a few days after that, seems pretty rare. There appears to be a huge diminishing return garnered from keeping the pack lower versus 80 or 85 or 90%, and keeping the pack more full makes for less fussing about with it day to day.

You know your customers better than I do, though.

My counters are almost always manually re-set so PV and other situations can trick them into a premature reset. On the BMV series you simply press and hold SETUP then hit the + button until you get to SYNC then hit SELECT then press and hold SETUP until it resets to 100%. With the Link-Pro it is easier and all you to is press the left and right arrows simultaneously...

I'm surprised you have any "tricking" going on. I don't see that at all. Maybe your customers have other gear that is not well-behaved.

I'm aware of the manual sync feature, and I'm suggesting that it should be possible to have a manual sync that doesn't sync to 100%. Then, someone who wants precision but doesn't want to go to 100% ever can make a steady-state estimation at some intermediate voltage and then punch it in, recalibrating the BM.

I think you might be be very surprised with the new SG-200 when it finally hits the streets. The original Smartgauge simply does not work for LFP but the new one was specifically designed from the ground up to track LFP SOC and also provide an SOH feature...

I can't see any reason why a meter couldn't use voltage-versus-time-and-current data to better estimate state. So I won't be surprised, but I will be pretty excited if it works.


Meanwhile, I'm going to start collecting drift data using CatNewbee's suggestion and my second BMV running off the same shunt. I'm not in a position to achieve more than a dozen PSOC cycles, probably ever, but it should be possible to see a trend at least.

[CatNewBee]

Another systematic error isderived from the voltage - cell temperature curve.

If you have a fix voltage to calibrate / reset SOC and this voltage is reached at different ambient temperatures you may reset the counter to different real SOC so your calibration would drift. Look at the charge / discharge diagram for temp and cell voltage at Winston.

If you calibrate at the shoulders, there is little difference on SOC because of the steep curve, the error will be bigger on the shallow areas, where temperature difference moves the curve parallel up an down and it results to larger SOC differences at a given fix voltage.

With solar during the summer and counter sync at lets say 3.55 ... 3.6V (95% real SOC) you are already on the steeper part and because of the sun as source, the temp during the sync is very similar, usualy at the same time of the day and same weather conditions, and the charge current is in the same range - so there is a quite stable sync environment. It is easier tò calibrate in a controlled lab environment if you need the precision.

For normal use where exact calibration is not relevant and drift is frequently compensated, this is not relevant. But if you want to tweak a counter to high precision without sync it will be better to do it in a lab environment.

http://en.winston-battery.com/images...P1000AHC-2.png

[rgleason]

Maine Sail wrote:

"I want a device I can comfortably send a customer off with, who's not an EE, and know the monitor is telling them the truth, or darn close, about SOC and that I know has not drifted. "

That would make it much easier, perhaps with a few alarms too, maybe even a secondary/primary low and high no nc cutoffs, separate fron the bms?

[john61ct]

Originally Posted by nebster

The person who prefers to stay at 50% SOC one day, and then 80% the next day, and then is okay with 20% for a few days after that, seems pretty rare.

Depends on sources.

If you catch as catch can, especially if inputs are only sometimes keeping up with consumption, yes.

But an easy ICE power on demand setup that refills 2-3 days' usage in ann hour or two, might as well only do so as the need for lots of consumption is immediately anticipated and otherwise let SoC stay in the lower healthier ranges.

[nebster]

Originally Posted by CatNewBee

Another systematic error isderived from the voltage - cell temperature curve.

That's a good point. But, that error should not accumulate. It might move a sync by a few percentage points of SOC, but then the offset will be constant until the next re-sync -- which, presumably, will be at another temperature and likely no further away from the truth.

Interestingly, a voltage-based estimation like we are speculating the new Balmar unit uses would be susceptible to being off because of temperature. Maybe it has a battery temp probe it can use to compensate.

[CatNewBee]

The temp variation on sync is only important during calibration. You need a fix, reproduceable point in SOC to set the comparing unit, while let the other unit drift over cycles.

Then after any amount of acceptable charge / discharge PSOC use, you sync the fix unit by charging in reference conditions and then compare the SOC of your fix to the SOC reading of your drifting un-synced probe and adjust then charge efficiency, while leaving peuckert to a minimum - 1.0 or 1.01. to get as close as possible to the reference counter.

[rgleason]

So the BMS should have cell temp? Or are you doing this manually every month? I assume not from your previous posts.