LiFePO4 capacity test

[CatNewBee]

If you want to have somehow reliable figures you have to synchronize the meters e.g to a 100% charged battery or an empty battery from time to time. When running in PSOC all the time (as proposed often for LFP), there will be a drift the one way or the other.

[Delfin]

Originally Posted by Q Xopa

Yes agreed that an SOH less than 100% is lower its 'orginal capacity'. However a clearly defined 'orginal capacity' SOH 100% is not known, by us. So by extension we can really know how relevant the SOH %number is?
Obviously we can deduce more is better for SOH, but not what number should cause what action or worry.
The same is true of the SOC number. Which 100 and 0% definitions are being displayed is not known so how can we be sure its the ones we want to use?

Im not saying its not good. In fact from what we are hearing so far it is.

However I, like most here, like being able to verify this information as much as we can.

Lets suppose its working fine, then suddenly doesnt. Its going to be harder to see, or determine. Worse still is it slowly missreading.

Having blind faith in a black box is not the way most of us are comfortable.

I don't think there's a problem here, but since the algorithms are black box, I can only deduce what is going on from the control inputs. The three setup inputs on the SG200 that are relevant here are taper current, which is the Ah the battery is accepting that is your 100%, the time to reach that taper current, the Peukert value, and the "design capacity". In Cat's example of a 1000 Ah battery that really has 1100 Ah when C rates are modest and the battery is new, you could set the capacity at 1000 amps. That would, after the device "learns" your bank, generate a SoH of 100%. As you cycled the battery, it would continue to show 100% even as some degradation occurred. So at 1,050 Ah true capacity, 100%, same with 1,025, etc. You'd only get 80% in Cat's example when the true capacity was 800 Ah. Alternately, you could set the capacity at 1,200 Ah if you wanted, then the SoH would be relative to that value.

Peukert, time and taper current, as near as I can tell, are used both for the black box calcs of SoC and SoH, but how exactly I don't think Balmar is going to tell us. With an LA battery, as they age, it takes longer to charge them fully, so I suspect time to reach taper current is a more important factor in determining health than it might be for LFP.

If Chris W. from Balmar is monitoring this discussion, he can correct any misinformation I'm offering up.....

[Delfin]

Originally Posted by CatNewBee

If you want to have somehow reliable figures you have to synchronize the meters e.g to a 100% charged battery or an empty battery from time to time. When running in PSOC all the time (as proposed often for LFP), there will be a drift the one way or the other.

Supposedly, drift of other meters is one of the problems the SG200 isn't subject to.

[CatNewBee]

Originally Posted by Delfin

Supposedly, drift of other meters is one of the problems the SG200 isn't subject to.

That is exactly what I am saying, too good to be true.

[Q Xopa]

Originally Posted by john61ct

So use multiple boxen, and benchmark both measures with manual load tests.

The better SoC units let you calibrate Peukert and CEF , keep at it until they agree with each other, toss out the hopeless cases, or maybe just keep the one that guesstimates most accurately.

And be sure to post your protocols and results back here!

For those who think the above would not be worthwhile (much less fun), no need to say so.

But I think any of the respected meters are still better than flying blind. . .

Good idea.

I havent got one, yet, maybe one day soon. So it not going to be a project Im going to be doing anytime soon.

But I like your suggestion, why dont you do it?

Lets assume you were to implement your suggestion, Im still not quite sure how you can get a comparison with another SOH?

But as you say we would all be interested in the results.

[Q Xopa]

Originally Posted by Delfin

I don't think there's a problem here, but since the algorithms are black box, I can only deduce what is going on from the control inputs. The three setup inputs on the SG200 that are relevant here are taper current, which is the Ah the battery is accepting that is your 100%, the time to reach that taper current, the Peukert value, and the "design capacity". In Cat's example of a 1000 Ah battery that really has 1100 Ah when C rates are modest and the battery is new, you could set the capacity at 1000 amps. That would, after the device "learns" your bank, generate a SoH of 100%. As you cycled the battery, it would continue to show 100% even as some degradation occurred. So at 1,050 Ah true capacity, 100%, same with 1,025, etc. You'd only get 80% in Cat's example when the true capacity was 800 Ah. Alternately, you could set the capacity at 1,200 Ah if you wanted, then the SoH would be relative to that value.

Peukert, time and taper current, as near as I can tell, are used both for the black box calcs of SoC and SoH, but how exactly I don't think Balmar is going to tell us. With an LA battery, as they age, it takes longer to charge them fully, so I suspect time to reach taper current is a more important factor in determining health than it might be for LFP.

If Chris W. from Balmar is monitoring this discussion, he can correct any misinformation I'm offering up.....

Agreed Im not suggesting there is any issue, or drift etc. From what I hear and have reason to doubt it works well.

But like most things on boats it can and would be suprised if it never does. Then we have less clues to recognise this.

In aviation, for example, there are several instruments that add to the weight of evidence that they are all working well. Experienced pilots can quickly see when and which instrument is not correlating well with the others.

Yes not arguing about 100% SOC reset values. However Ive heard a number of different formulae for determining a 100% SOC. Ie charging tailing current as X % of bank capacity etc. Ive seen different X%s, X Voltage, and other versions.
So without knowing which definition is being used we are have less clues to keep an eye on and verify things.

[Q Xopa]

Originally Posted by Delfin

I don't think there's a problem here, but since the algorithms are black box, I can only deduce what is going on from the control inputs. The three setup inputs on the SG200 that are relevant here are taper current, which is the Ah the battery is accepting that is your 100%, the time to reach that taper current, the Peukert value, and the "design capacity". In Cat's example of a 1000 Ah battery that really has 1100 Ah when C rates are modest and the battery is new, you could set the capacity at 1000 amps. That would, after the device "learns" your bank, generate a SoH of 100%. As you cycled the battery, it would continue to show 100% even as some degradation occurred. So at 1,050 Ah true capacity, 100%, same with 1,025, etc. You'd only get 80% in Cat's example when the true capacity was 800 Ah. Alternately, you could set the capacity at 1,200 Ah if you wanted, then the SoH would be relative to that value.

Peukert, time and taper current, as near as I can tell, are used both for the black box calcs of SoC and SoH, but how exactly I don't think Balmar is going to tell us. With an LA battery, as they age, it takes longer to charge them fully, so I suspect time to reach taper current is a more important factor in determining health than it might be for LFP.

If Chris W. from Balmar is monitoring this discussion, he can correct any misinformation I'm offering up.....

Yes not arguing about 100% SOC reset values. However Ive heard a number of different formulae for determining a 100% SOC. Ie charging tailing current as X % of bank capacity etc. Ive seen different X%s, X Voltage, and other versions.
So without knowing which definition is being used we are have less clues to keep an eye on and verify things

Agreed Im not suggesting there is any issue, or drift etc. From what I hear and have reason to doubt it works well.

But like most things on boats it can and I would be suprised if it never does play up. Then we have less clues to recognise this.

In aviation, for example, there are several instruments that add to the weight of evidence that they are all working well. This the concept John was suggesting this concept.

Experienced pilots can quickly see when and which instrument is not correlating well with the others.

There was a famous crash a few years ago where the speed sensors (Pitot static) iced up, which falsely indicated the aircraft was going too fast. So the inexperienced co piltots reduced power and pulled the nose up. The aircraft stalled and gave warnings which confused them. All this pitch up and down woke the experienced captain. He got through the security door realised what was happening, pushed the nose forward to get it flying again. He almost made it.

[Delfin]

Originally Posted by Q Xopa

Yes not arguing about 100% SOC reset values. However Ive heard a number of different formulae for determining a 100% SOC. Ie charging tailing current as X % of bank capacity etc. Ive seen different X%s, X Voltage, and other versions.
So without knowing which definition is being used we are have less clues to keep an eye on and verify things

Agreed Im not suggesting there is any issue, or drift etc. From what I hear and have reason to doubt it works well.

But like most things on boats it can and I would be suprised if it never does play up. Then we have less clues to recognise this.

In aviation, for example, there are several instruments that add to the weight of evidence that they are all working well. This the concept John was suggesting this concept.

Experienced pilots can quickly see when and which instrument is not correlating well with the others.

There was a famous crash a few years ago where the speed sensors (Pitot static) iced up, which falsely indicated the aircraft was going too fast. So the inexperienced co piltots reduced power and pulled the nose up. The aircraft stalled and gave warnings which confused them. All this pitch up and down woke the experienced captain. He got through the security door realised what was happening, pushed the nose forward to get it flying again. He almost made it.

I use a Link20 coulomb counter to cross check SoC on the Balmar for that reason. The Link20 is useless as a SoC by itself since it does drift, but its amp counting is quite accurate. No rocket science there. For that reason, I have good confidence in the SoC of the Balmar once it learns my bank. I am HOPEFUL the SoH is useful, although I think I will limit my interpretation of usefulness to a trend indicator possibly indicating an issue with my management of the bank if I see it falling more than I think it should over time. That is really the only reason I installed it.

Incidentally, because of drift on the Link20, I installed a momentary switch to reset it, which I do frequently.

[Delfin]

Originally Posted by CatNewBee

That is exactly what I am saying, too good to be true.

I'll be leaving for a few months on the hook, and intend to log data on each cycle. It won't be hard to validate whether the unit is, or isn't stable with respect to SoC.

[nebster]

I think the way SG200 works is through a combination of

a) measured voltage
b) measured current
c) a preprogrammed understanding of the characteristic charge curve for LFP (and other chemistries)
d) a preprogrammed understanding of the short-term voltage dynamics vs current for LFP (and other chemistries)
e) possibly, temperature

What I suspect it does is measure the voltage curve (a) vs time and current (b), apply some correction for recent current history based on (d) and possibly (e), and then attempt to fit the resulting shape to (c).

In a broad sense, this is not too different from how you or I would calculate the capacity of a cell that was handed to us with no documentation. We might discharge it to 2.8V or so, and then charge it up to 3.5V or so and wait for the current to taper. Along the way, we can integrate current to infer energy. We don’t need a manufacturer spec sheet, because it’s a physical object with understood properties. We don’t even need to know the approximate capacity before we start (although it’s handy for establishing a rate).

Similarly, you or I can estimate SOC with no information other than voltage. We simply apply a very small discharge current for a few moments, and then isolate the cell completely and let it rest for a while. If we’re not sure or unable to apply a small flow, we let it rest even longer. Then we measure the voltage, and we look up that voltage on our reference charge curve.

The computer can do a better job of SOC estimation if it has an accumulated base of reference data to work from. It can observe the voltage drops and rises as current changes, and it can relate those to the previous estimated SOC at those moments in time. It can use those to grow a base of understanding about how the battery behaves and what voltages it exhibits as it is exposed to various rates.

Over time, I expect that a combination of shunt-based local state estimation and a global learning model would allow a computer to converge pretty well on fully modeling the battery’s performance.

Now, as for SOH, let’s assume that the people who designed SG200 use SOH in the conventional sense: a ratio of [the capacity of this cell today] to [the original design capacity of this cell].

I believe SG200 (or a good implementation of the approach above, if theirs turns out not to be) should be able to calculate the numerator of that value as described above. The denominator could be determined in one of two ways:

First, the user could program it in at the start. If I tell the SG200 that my pack is “300Ah”, and then it empirically determines that it actually only holds about 250Ah after observing the charge curve over time, then I think we get 83% on the display eventually.

In this scenario, with CatNewBee’s overprovisioned Winston cells, if he tells SG200 that his pack is 1000Ah, it might well converge on an SOH > 100%. (Since CatNewBee “lied” to the computer and they are really 1200Ah or so.)

Or, the computer could figure out the capacity “now” and then store that value as a baseline for the future. It could, then, continue to reassess capacity over time and compute SOH with reference to that baseline. In that case, I would expect SG200 to show us its calculated capacity somewhere.

In short, I don’t see anything in the physics or math here that would preclude a learning algorithm from doing very well with this problem. This is a kind of data fitting problem that our current tools in computer science can handle very well.

Whether SG200 actually does any of this, I don’t know. But I suspect it does all of this.

[Q Xopa]

Originally Posted by nebster

I think the way SG200 works is through a combination of

a) measured voltage
b) measured current
c) a preprogrammed understanding of the characteristic charge curve for LFP (and other chemistries)
d) a preprogrammed understanding of the short-term voltage dynamics vs current for LFP (and other chemistries)
e) possibly, temperature

What I suspect it does is measure the voltage curve (a) vs time and current (b), apply some correction for recent current history based on (d) and possibly (e), and then attempt to fit the resulting shape to (c).

In a broad sense, this is not too different from how you or I would calculate the capacity of a cell that was handed to us with no documentation. We might discharge it to 2.8V or so, and then charge it up to 3.5V or so and wait for the current to taper. Along the way, we can integrate current to infer energy. We don’t need a manufacturer spec sheet, because it’s a physical object with understood properties. We don’t even need to know the approximate capacity before we start (although it’s handy for establishing a rate).

Similarly, you or I can estimate SOC with no information other than voltage. We simply apply a very small discharge current for a few moments, and then isolate the cell completely and let it rest for a while. If we’re not sure or unable to apply a small flow, we let it rest even longer. Then we measure the voltage, and we look up that voltage on our reference charge curve.

The computer can do a better job of SOC estimation if it has an accumulated base of reference data to work from. It can observe the voltage drops and rises as current changes, and it can relate those to the previous estimated SOC at those moments in time. It can use those to grow a base of understanding about how the battery behaves and what voltages it exhibits as it is exposed to various rates.

Over time, I expect that a combination of shunt-based local state estimation and a global learning model would allow a computer to converge pretty well on fully modeling the battery’s performance.

Now, as for SOH, let’s assume that the people who designed SG200 use SOH in the conventional sense: a ratio of [the capacity of this cell today] to [the original design capacity of this cell].

I believe SG200 (or a good implementation of the approach above, if theirs turns out not to be) should be able to calculate the numerator of that value as described above. The denominator could be determined in one of two ways:

First, the user could program it in at the start. If I tell the SG200 that my pack is “300Ah”, and then it empirically determines that it actually only holds about 250Ah after observing the charge curve over time, then I think we get 83% on the display eventually.

In this scenario, with CatNewBee’s overprovisioned Winston cells, if he tells SG200 that his pack is 1000Ah, it might well converge on an SOH > 100%. (Since CatNewBee “lied” to the computer and they are really 1200Ah or so.)

Or, the computer could figure out the capacity “now” and then store that value as a baseline for the future. It could, then, continue to reassess capacity over time and compute SOH with reference to that baseline. In that case, I would expect SG200 to show us its calculated capacity somewhere.

In short, I don’t see anything in the physics or math here that would preclude a learning algorithm from doing very well with this problem. This is a kind of data fitting problem that our current tools in computer science can handle very well.

Whether SG200 actually does any of this, I don’t know. But I suspect it does all of this.

Yes sounds like a reasonable guess.
However as you say it is all speculation so brings me back to we dont know.

So we cant really use it as a reliability comparison.

As an apprentice, too many moons ago, we were always told that if we understand how things work we can recognise when and what the fault is.

[Delfin]

Originally Posted by nebster

I think the way SG200 works is through a combination of

a) measured voltage
b) measured current
c) a preprogrammed understanding of the characteristic charge curve for LFP (and other chemistries)
d) a preprogrammed understanding of the short-term voltage dynamics vs current for LFP (and other chemistries)
e) possibly, temperature

What I suspect it does is measure the voltage curve (a) vs time and current (b), apply some correction for recent current history based on (d) and possibly (e), and then attempt to fit the resulting shape to (c).

In a broad sense, this is not too different from how you or I would calculate the capacity of a cell that was handed to us with no documentation. We might discharge it to 2.8V or so, and then charge it up to 3.5V or so and wait for the current to taper. Along the way, we can integrate current to infer energy. We don’t need a manufacturer spec sheet, because it’s a physical object with understood properties. We don’t even need to know the approximate capacity before we start (although it’s handy for establishing a rate).

Similarly, you or I can estimate SOC with no information other than voltage. We simply apply a very small discharge current for a few moments, and then isolate the cell completely and let it rest for a while. If we’re not sure or unable to apply a small flow, we let it rest even longer. Then we measure the voltage, and we look up that voltage on our reference charge curve.

The computer can do a better job of SOC estimation if it has an accumulated base of reference data to work from. It can observe the voltage drops and rises as current changes, and it can relate those to the previous estimated SOC at those moments in time. It can use those to grow a base of understanding about how the battery behaves and what voltages it exhibits as it is exposed to various rates.

Over time, I expect that a combination of shunt-based local state estimation and a global learning model would allow a computer to converge pretty well on fully modeling the battery’s performance.

Now, as for SOH, let’s assume that the people who designed SG200 use SOH in the conventional sense: a ratio of [the capacity of this cell today] to [the original design capacity of this cell].

I believe SG200 (or a good implementation of the approach above, if theirs turns out not to be) should be able to calculate the numerator of that value as described above. The denominator could be determined in one of two ways:

First, the user could program it in at the start. If I tell the SG200 that my pack is “300Ah”, and then it empirically determines that it actually only holds about 250Ah after observing the charge curve over time, then I think we get 83% on the display eventually.

In this scenario, with CatNewBee’s overprovisioned Winston cells, if he tells SG200 that his pack is 1000Ah, it might well converge on an SOH > 100%. (Since CatNewBee “lied” to the computer and they are really 1200Ah or so.)

Or, the computer could figure out the capacity “now” and then store that value as a baseline for the future. It could, then, continue to reassess capacity over time and compute SOH with reference to that baseline. In that case, I would expect SG200 to show us its calculated capacity somewhere.

In short, I don’t see anything in the physics or math here that would preclude a learning algorithm from doing very well with this problem. This is a kind of data fitting problem that our current tools in computer science can handle very well.

Whether SG200 actually does any of this, I don’t know. But I suspect it does all of this.

That's how I see it.

[nebster]

Originally Posted by Q Xopa

Yes sounds like a reasonable guess.
However as you say it is all speculation so brings me back to we dont know.

So we cant really use it as a reliability comparison.

As an apprentice, too many moons ago, we were always told that if we understand how things work we can recognise when and what the fault is.

I’m not quite sure what a “reliability comparison” is, but I agree that we don’t know the details of how SG200 works, and that certainly makes it harder to reason about it.

I wrote my guess about how it could work mostly to suggest that that at least some of the limitations or concerns mentioned in this thread may not apply. A good design can probably work around most of the issues.

[Q Xopa]

Originally Posted by nebster

I’m not quite sure what a “reliability comparison” is, but I agree that we don’t know the details of how SG200 works, and that certainly makes it harder to reason about it.

I wrote my guess about how it could work mostly to suggest that that at least some of the limitations or concerns mentioned in this thread may not apply. A good design can probably work around most of the issues.

Yes 'reliability comparison' maybe not the best turn of phrase. Trying to say that its hard to compare to another Battery meter that we know, or atleast are reasonably confident in because we dont know what the SG200 is actually displaying.

[john61ct]

If you get set up so that doing a precisely controlled repeatable load test is not inconvenient, always use the same floor as your 0% say 2.99Vpc

make sure to restore charge current soon as you hit that, ideally all automated,

then you can get a true objective and accurate measure of SoC and SoH anytime.

In fact that is **the only** way to so, besides taking some black-box meter' word for it, that is the equivalent of a reference voltage chip in calibrating a voltmeter.

So comparing different meters' accuracy wrt a specific bank model becomes just a matter of taking the time to do so, and

of course for SoH you would ideally start from new rather than just using the vendor benchmark, and won't get a result for many years unless measuring the effects of gross abuse.

But a pretty thorough comparative test of SoC accuracy should not take more than 24 hours per sample batt.

That's just rating one meter against another in a limited set of circumstances, not making a generalized judgement about one meter's "objective accuracy" across a wide range of usage patterns, temperature ranges etc, that would take maybe a week, and would still only apply to that one battery model.

IMO any meter generally staying within say 5% of actual SoC would be extraordinary and put it right up there near the top.

And SoH is a pie in the sky goal IMO, any competence at that across a range of chemistries, would be **very** impressive in my book.

And no, I'm not volunteering to do such rigorous testing, but Bruce @OceanPlanet, did express an interest in doing so sometime last year. Long before SG200 was released though. . .