LFP memory effects thread

[evm1024]

Thanks Delfin for your observations.

I've been reading up on OCV hysteresis and other such effects. Lots of info there.

The few papers I've read indicate that (especially) at high charge rates that there is a gradient in ion deposition in the active films. This gradient causes rapid voltage rise on charge but after resting the ions redistribute in the active film somewhat resulting in a further ability to accept charge. This imprecisely stated of course.

At the moment I am going to do a few cycles of low current charge and discharge with at least 24 hours rest between cycles to see if I can recover any charge capacity.

Sorry I cannot express my thoughts on this better at this time. I've just gone through about 6 papers and it takes a while to sort through them.

N.B. There have been a number of folks who have contacted me and who are experiencing the same effect as the few of us who have posted.

[OceanSeaSpray]

Delfin, what is your resting cell voltage 4 hours after termination with no load? This is what would give you a real idea of how high your cells recharge.

It is not because you do "this and that" that they will effectively recharge to full. They don't and this is our issue.

[Delfin]

Originally Posted by OceanSeaSpray

Delfin, what is your resting cell voltage 4 hours after termination with no load? This is what would give you a real idea of how high your cells recharge.

It is not because you do "this and that" that they will effectively recharge to full. They don't and this is our issue.

I thought you said in a previous post that memory effect was "by definition" reversible.

To your question, 3.35 +/-. Hence my lifting end charge voltage to 3.65 with 30 minutes absorption.

According to Lithionics, 3.4 vpc after <8 hours rest equals a fully charged battery. 1 hour is <8 hours, as is 4 hours.

[Cpt Pat]

Delfin: I measure 3.37 volts OCV after resting at a 100% SOC, so our numbers match closely. I'm using GBS cells.

I'm doing coulomb counting (counting amp hours in/out). What I have observed, after 3 years and about 500 cycles, is the cell voltages needed to reach 80% SOC (my chosen termination SOC) gradually rising from 3.40 volts/cell to 3.45 volts/per cell charging voltage (not resting voltage) to reach 80% SOC. But since I'm counting amp hours - not voltage - to reach charge termination, it was at first just a curiosity that I attributed to amp/hour counter drift. This caused me to recalibrate the counter frequently (about every 25 cycles) to 100% SOC.

But after each counter "recalibration", where I manually charged to 100% SOC (using charge current taper to detect 100% SOC), that 80% charging voltage returned to 3.40 volts/cell. It appears I am seeing the memory effect, but since I charge to 100% rather frequently, I am erasing the memory with a "memory-release cycle."

I've asserted above in this thread that terminating charging at a voltage threshold instead of counting amp hours is the root of this problem. (As well as presenting the danger of severe overcharging.) I don't believe I have "confirmation bias" by interpreting what I have seen as verification of that assertion.

It appears to me the mechanism of the memory effect is this: when using terminal voltage alone during charging to determine the threshold to stop charging, that voltage progressively produces a diminishing amount of charge and gradually "walks the battery down" to reduced capacity. Eventually, a "memory" sets in that limits capacity.

There is a good discussion of the memory effect at the bottom of this page: http://nordkyndesign.com/practical-characteristics-of-lithium-iron-phosphate-battery-cells/.

The author also explains why terminating charging based on only a voltage threshold will fail: "LFP cells simply don’t really charge at voltages up to 3.3V and then fully charge already at 3.4V and upwards. The transition is so abrupt that claiming to control the charging process by adjusting the voltage is purely and simply bound to fail."

The technique I am using is and approximation of what is discussed in this paper:https://arxiv.org/ftp/arxiv/papers/1803/1803.10654.pdf, though the paper doesn't appear to be specific to LiFePO4 chemistry batteries.

[Delfin]

Originally Posted by Cpt Pat

Delfin: I measure 3.37 volts OCV after resting at a 100% SOC, so our numbers match closely. I'm using GBS cells.

I'm doing coulomb counting (counting amp hours in/out). What I have observed, after 3 years and about 500 cycles, is the cell voltages needed to reach 80% SOC (my chosen termination SOC) gradually rising from 3.40 volts/cell to 3.45 volts/per cell to reach 80% SOC. But since I'm counting amp hours not voltage to reach charge termination, it was at first just a curiosity that I attributed to amp/hour counter drift. This caused me to recalibrate the counter frequently (about every 25 cycles) to 100% SOC.

But after each counter "recalibration", where I charged to 100% SOC, that 80% OCV (after resting) returned to 3.40 volts/cell. It appears I am seeing the memory effect, but since I charge to 100% rather frequently, I am erasing the memory.

I've asserted here (above) that terminating charging at a voltage threshold instead of counting amp hours is the root of this problem. I don't believe I have "confirmation bias" by interpreting what I have seen as verification of that assertion.

It appears to me the mechanism of the memory effect is this: when using terminal voltage during charging alone to determine the threshold to stop charging, that voltage progressively produces a diminishing amount of charge and gradually "walks the battery down" to reduced capacity.

The technique I am using is discussed in this paper:https://arxiv.org/ftp/arxiv/papers/1803/1803.10654.pdf, though it doesn't appear to be specific to LiFePO4 chemistry batteries.

Thanks for that, even if I don't think my brain is large enough to absorb the data from the paper referenced.

I've also been using coulomb counting to determine charge termination, but I'm not sure that alone gives the best results. If I set bulk voltage at 3.55 vpc, then I will see charge acceptance drop to 2% or so of capacity. However, if I set it at 3.65 vpc, I see that same drop of acceptance rate as well, just a bit later and the resting voltage ends up a bit higher, so I have to think that the latter regime is increasing stored energy and if there is a memory effect, it is this regime that will diminish or erase it. On a daily use basis, I charge to 3.55 cpv and terminate when absorption drops to 3% or so. And as noted and based on the learning from this and other threads, a "conditioning" charge to the higher voltage with an absorption time of 30 minutes or so periodically makes sense. I can't go above 3.65, as the BMS starts to shunt cells above that.

[Cpt Pat]

Originally Posted by Delfin

If I set bulk voltage at 3.55 vpc, then I will see charge acceptance drop to 2% or so of capacity. However, if I set it at 3.65 vpc, I see that same drop of acceptance rate as well (...),

Please correct me if I'm wrong, but that doesn't look like coulomb counting to me. It appears to me that you are using voltage thresholds instead. Any voltage greater than 3.4 volts, if applied long enough, will overcharge the battery. You can "tune" the charge current to a termination voltage to approximate a desired SOC, but if anything changes in the charging current or the battery (like series resistance or memory effect), you will not achieve the desired SOC (resulting in over or under charging). In operation, with coulomb counting, the voltage just ends up where it ends up at my set SOC threshold (around 3.5 volts per cell charging voltage for 100% SOC). I do have a last ditch "something went very wrong" termination escape voltage of 14.4 volts (3.60 volts per cell) which has never been reached.

My capacity testing shows I still have a little more than rated capacity (because my discharge currents are very fractional: zero to 0.10C. The batteries are fused at 0.25C.)

I chose to count amp hours in and out instead because I have variable charge currents from zero to 0.2C produced by solar and water impeller power. Voltage thresholds won't work reliably or repeatedly with variable charging currents. There's a danger of overcharging at very low currents and greater than 3.4 volts per cell. (This is why trickle charging is deadly to lithium batteries: anything over 3.4 volts per cell, even if only a few milliamps, will eventually overcharge the battery.)

I can only calibrate my counter when I use a shorepower charger with a constant current output, and I have no way of reliably detecting the 100% SOC charge current taper on any other charge source.

[OceanSeaSpray]

Originally Posted by Delfin

I thought you said in a previous post that memory effect was "by definition" reversible.

To your question, 3.35 +/-. Hence my lifting end charge voltage to 3.65 with 30 minutes absorption.

Yes, it is by definition reversible, once you manage to do it. What we are currently investigating here is what is required to achieve that once it has deeply set in over years of partial cycle use.

If your cells are sitting at 3.35V after charging, then you have in them in the upper plateau and they could be around 85% SOC. However, the voltage is very flat again in that region and 3.36V would be full. Your voltmeter might have a bigger error than 1mV.
You would know if you had metered the charge like we did. Here, they might be recharging just fine and show no memory effects. You haven't got the supporting data.

Originally Posted by Delfin

According to Lithionics ...

Bad start. Source information from people interested in battery technology, not sales.

Originally Posted by Delfin

... 3.4 vpc after <8 hours rest equals a fully charged battery. 1 hour is <8 hours, as is 4 hours.

Let's see, 3 minutes after you finish charging is less than 8-hour rest too and the voltage will be higher than 3.4V... so is that fully charged?

[Delfin]

Originally Posted by OceanSeaSpray

Let's see, 3 minutes after you finish charging is less than 8-hour rest too and the voltage will be higher than 3.4V... so is that fully charged?

Now, you're simply being argumentative, to what end I am not sure.

And thanks for the tip on not listening to vendors regarding their products after you have bought them and are no longer a sales target, but simply interested in how to get the most out of what you just purchased. I've actually heard this theory before from a poster who shall remain nameless. His theory is that LFP vendors want to sabotage their products so that you will have to buy them again. Brilliant.

[Delfin]

Originally Posted by Cpt Pat

Please correct me if I'm wrong, but that doesn't look like coulomb counting to me. It appears to me that you are using voltage thresholds instead. Any voltage greater than 3.4 volts, if applied long enough, will overcharge the battery. You can "tune" the charge current to a termination voltage to approximate a desired SOC, but if anything changes in the charging current or the battery (like series resistance or memory effect), you will not achieve the desired SOC (resulting in over or under charging). In operation, with coulomb counting, the voltage just ends up where it ends up at my set SOC threshold (around 3.5 volts per cell charging voltage for 100% SOC). I do have a last ditch "something went very wrong" termination escape voltage of 14.4 volts (3.60 volts per cell) which has never been reached.

My capacity testing shows I still have a little more than rated capacity (because my discharge currents are very fractional: zero to 0.10C. The batteries are fused at 0.25C.)

I chose to count amp hours in and out instead because I have variable charge currents from zero to 0.2C produced by solar and water impeller power. Voltage thresholds won't work reliably or repeatedly with variable charging currents. There's a danger of overcharging at very low currents and greater than 3.4 volts per cell. (This is why trickle charging is deadly to lithium batteries: anything over 3.4 volts per cell, even if only a few milliamps, will eventually overcharge the battery.)

I can only calibrate my counter when I use a shorepower charger with a constant current output, and I have no way of reliably detecting the 100% SOC charge current taper on any other charge source.

Well, I could certainly be wrong, but I think it is "coulomb counting" in the sense that I terminate charge voltage based on charge acceptance rate, not voltage. My point is that I can reach that rate at different voltages. So, I presume that if that acceptance rate happens at a higher voltage that likely represents a higher internal storage of energy. If it doesn't, then it means that if bulk voltage is set at 3.45 vpc and a CAR of 2.5% is reached, the battery has the same charge as when that CAR is reached at 3.65 vpc. Since in my experience, resting voltage is lower is the former case than the latter, I have to assume the latter regime results in a higher charged state for the bank.

Does that not make sense?

[OceanSeaSpray]

Coulomb counting (CC) = charged passed in Ah
Charge acceptance rate, i.e. residual current = charge current in A

The two are completely different. CC is based on the recent history of the cell, current is a just a measurement taken now.

The higher the termination voltage for a given residual current, the higher the SOC achieved indeed.

CC is complete garbage unless it has been recently recalibrated, so the notion that it is better or more useful or what not for charge termination makes no sense. You will overcharge and undercharge with it just because it is out of calibration.

However, once the capacity left to be charged to clear the memory has been determined, CC is an obvious candidate to control the charging from this point on and in this particular case.

MaineSail has recently reached 10 years of cycling using standard termination on voltage and residual current, returning to full for a capacity test at least every 50 cycles, and it has been working throughout while CC drifted out readily.
Even at very low charge currents, the cell voltage does eventually rise up to the termination point.

[OceanSeaSpray]

Originally Posted by Delfin

Now, you're simply being argumentative, to what end I am not sure.

No. I was hoping that you would think about it. When something looks silly, sometimes it is because it is. So here we go:

"The cell voltage should still be holding above a certain value after a given time".

OCV curves for LFP suggest 3.36V for a full cell and cell voltages were generally found to be almost completely stabilised after 4 hours or so. I have seen brand new cells holding up above 3.5V just after top balancing.

[Cpt Pat]

Originally Posted by OceanSeaSpray

Coulomb counting (CC) = charged passed in Ah
Charge acceptance rate, i.e. residual current = charge current in A

The two are completely different. CC is based on the recent history of the cell, current is a just a measurement taken now.

CC is complete garbage unless it has been recently recalibrated, so the notion that it is better or more useful or what not for charge termination makes no sense. You will overcharge and undercharge with it just because it is out of calibration.

Thanks for explaining coulomb counting. And you are correct: the counter must be accurately calibrated or you will get "GIGO" -- garbage in, garbage out. This is true of any calculation. Saying that really isn't saying much. It's simply stating the obvious.

For my own GBS cells, I've found that setting the peukert function to 1.05 (nearly none) and the charge efficiency to 99% (nearly perfect) yields an accurate count within +/- 2% (measured by reading the pack resting OCV) over 25 cycles, at ambient temperatures that never go outside 15C to 25C parameters. Since I am limited by the counter design to setting only 100%, I've been charging to 100% about every 25 cycles over the past three years - always immediately after discharging the pack to around 20% SOC. Unbeknownst to me, I hypothesize that I was also clearing the memory effect in that process.

I sail often enough, and calibrate often enough, that the pack self-discharge rate doesn't impact accuracy.

I've adapted a Victron BVM-702 to this task. I do use terminal voltage as a low voltage disconnect (threshold = 12.7 volts).

Terminating charge by detecting residual current (charging current taper) isn't possible for me while underway. My charging sources (solar and impeller generator) are too variable in output current. I don't have the luxury of an engine-driven alternator. I am forced to use a shore powered charger as a constant current source for counter calibration.

Terminating charge based on only terminal voltage (what seems to be the most common method) will also be highly inaccurate where the charge current isn't high and constant. Example: Charging to 3.65 volts per cell at 0.5C (a common factory specification) will yield a full charge (100% SOC), but charging at 0.05C (or less) to the same terminal voltage will yield a seriously overcharged cell. The cell voltage detection method is fine for electric vehicles with factory supplied chargers, but very inappropriate for fractional C applications, such as house batteries.

LFE users have been progressively lowering their termination voltages from 4.2 volts to 3.65 volts, and then even down to 3.50 volts, trying to mitigate their fried batteries. All the while, I believe, they haven't understood that terminating charge using terminal voltage alone will eventually ruin their batteries. It's a technique that is bound to fail. Any charge, however small (and greater than the self-discharge rate) applied with a voltage over 3.4 volts per cell, will eventually overcharge their batteries.

[john61ct]

The two methods for determining / defining 100% SoC that are, truly accurate are

A. Stop-charge profile defined by a precise CV setpoint, hold Absorb to a precise endAmps

B. Ah actually stored in the bank, constant current CC load / discharge tests to a precise 0% SoC resting voltage. Yes just like a SoH test, but

used to verify / test / benchmark different stop-charge profiles, seeing how far removed their resulting SoC% is from that of the max theoretical 100% Full resulting from using the max CV setpoint until current tapers to Zero amps.

C. Voltage at rest can be a bit deceptive, with the "surface charge" phenomenon, no significant increase in B although finish voltage increased.

(A) is not "just using voltage", holding CV / Absorb to a precise endAmps does give precision. Using "charge to" CC only will indeed give variability to the finish SoC as current varies.

For (C), different brand cells slightly different formulation chemistries may also vary, enough to account for the difference between 3.35V resting and 3.40V resting after isolated for X hours.

Waiting 8 hours can be shortcut to remove the surface effect, by just removing say .001C off the top as a standard.

The "gold standard" for method A, and maybe OK as a maintenance protocol not IMO normal cycling:

3.65V taper to 0A

what I call "vendor / theoretical" 100% SoC

If you use method (B) CC load testing down to 2.99V, and compare your usual stop-charge profile to that vendor / theoretical one,

that will show precisely how much actual %SoC "above working capacity" you've been sacrificing to the longevity goal.

examples from MS' tests:

* 3.45V taper to 0A, delivers 99.94% SOC
* 3.45V taper to .06C = 98.69% SOC
* 3.40V taper to 0A & then float there for 24 hours = 85.3% SOC
* 3.45VPC taper to 0A and drop float to 3.4V for 24 hours, carrying a 10A load = 98.5% SOC

Note that using coulomb counting = shunt-based Ah totalizing with even the most accurate battery monitor (not a high bar), will **not** accurately show **retrievable Ah actually stored** in the bank,

because, in the last few percent approaching that "vendor / theoretical" 100% SoC, a higher proportion of the Ah that **appear** to be going into the bank are actually getting **dissipated** as heat and other non-helpful chemical activity.

Which is why we stop earlier than that in normal cycling.

Same with higher than healthy charging C-rates.

But in trying to "break through" a hopefully-temporary capacity-loss artifact of the memory effect,

or occasionally using it **to prevent** the memory effect is IMO justified.

1/25 cycles as Cpt Pat suggests or even more often

doing so may indeed be useful.

[Cpt Pat]

John,


I agree, that is a good protocol. Very nice for charging on a bench or while tied up to the dock using shorepower, with the battery isolated from all loads. But....

How do you implement that protocol while underway, with loads connected, alternately charging and discharging the battery at variable rates, while never isolating and resting the battery (i.e., in the real world) where no current taper can be accurately detected?

If I may, I will borrow terms from medicine and call your method an "in vitro" ("outside a living organism") protocol. Mine is an "in vivo" (in actual living operation) protocol.

[evm1024]

I would like to remind folks that the thread is a working thread not a theory thread. Simply stated (in post #1):

This thread is to report observations and actions. ...

...For those who have LiFePO4 banks and who are experiencing capacity loss and who are trying to see if they can recover that lost capacity.

We are not looking at getting to 100% SOC or even defining what 100% SOC is or how to measure it. There are other better threads for that discussion. We are discussing 30% loss of capacity. Any discussion of the difference between 99.94% SOC and 99.69% SOC are far too subtle for this hands on thread.

And oh by the way the difference (between 99.94 and 99.69) is around 0.25 gram of Li ions transferred. Which is meaningless in the context of this thread.

Let's stay focused.