Memory effect found in LiFePO4

How do you store and manage your electricity?
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by weber »

One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
coulomb
Site Admin
Posts: 6357
Joined: Thu, 22 Jan 2009, 20:32
Real Name: Mike Van Emmerik
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by coulomb »

I'm pretty sure that they are saying that this is not a capacity issue, but a voltage profile issue. They seem to be saying the only problem is that a voltage-based SOC determination will get it wrong after a shallow discharge, because of the voltage bump that occurs.

[ Edit: to counter the obvious "why not just coulomb count, it's immune to this effect": ] They seem to think that coulomb counting does not work in automotive applications, due to gradual capacity degradation and other problems:
Nature article wrote:For the SOC estimations, the Coulomb-counting method [20] can be used; however, the method is not suitable for LIBs in automobile uses [21]. LIBs always have some side reactions that lead to irreversible capacity, self-discharge and degradation [20],[21]
, especially at elevated temperatures; also, the required accuracy of the SOCs is particularly high for automobile uses [22].
(LIB = Lithium Ion Batteries)

I read the below from a University account (where we have access to many publications as part of some package deal that Universities get), so you may not be able to read the below:

http://www.nature.com/nmat/journal/v12/ ... at3623.pdf
Last edited by coulomb on Tue, 13 Jan 2015, 08:32, edited 1 time in total.
MG ZS EV 2021 April 2021. Nissan Leaf 2012 with new battery May 2019.
5650 W solar, 2xPIP-4048MS inverters, 16 kWh battery.
Patching PIP-4048/5048 inverter-chargers.
If you appreciate my work, you can buy me a coffee.
User avatar
Johny
Senior Member
Posts: 3749
Joined: Mon, 23 Jun 2008, 16:26
Real Name: John Wright
Location: Melbourne
Contact:

Memory effect found in LiFePO4

Post by Johny »

Not really using the correct terms are they?
In referring to Hybrids that use voltage to determine charge/discharge limits.....

"The numerous successive cycles of partial charging and discharging lead to individual small memory effects adding up to a large memory effect, as this new study demonstrates. This leads to an error in the estimate of the current state of charge of the battery, in cases where the state of charge is calculated by software on the basis of the current value of the voltage."

Not memory effect at all. Just plain additive quantum inaccuracy.
TooQik
Groupie
Posts: 179
Joined: Mon, 18 Jun 2012, 00:53
Real Name: Chris
Location: Mount Gambier

Memory effect found in LiFePO4

Post by TooQik »

Johny wrote:Not memory effect at all. Just plain additive quantum inaccuracy.
I was trying to work out where the capacity issue was. Am I correct in saying that there is no capacity loss due to this, but rather a perceived loss due to the variance in voltage?
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by weber »

What I want to know is, how does it relate to this
http://www.cse.anl.gov/us-china-worksho ... %20BMS.pdf
which I was planning to use to determine (by correcting for current times internal resistance) when an individual LiFePO4 cell is at an SoC of 75+-5% and 20+-5%, for maximising battery life in a solar power system.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by weber »

TooQik wrote:I was trying to work out where the capacity issue was. Am I correct in saying that there is no capacity loss due to this, but rather a perceived loss due to the variance in voltage?

I think you are correct in saying that, TooQik. However that's what memory effect is in NiCd too.

I think it really is a memory effect, Johny. But I think they are making a mountain out of a molehill. Of course they want to make their work sound important, rather than admitting it was just to see who won the bet they made over a beer at lunchtime. Image

That's why they have to point out the flaws in coulomb counting. Thanks, Coulomb. Image They are not saying that the memory effect makes coulomb counting impractical. The memory effect doesn't affect coulomb counting at all. The things they say about coulomb counting are just its known limitations, for any type of battery.

They are in fact the reasons I'm trying to use voltage, current and internal resistance to estimate those two particular SoCs (20% and 75%). But I think I have far bigger corrections to worry about than those little memory bumps of a few millivolts per cell.

e.g. I'm presently trying to do it without taking _any_ charge history into account. So I'm not even allowing for the long-known hysteresis (which memory effect is defined as being in addition to) except by assuming that whatever the current is doing now, it has been doing for a long time previously (solar, night and day).
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
TooQik
Groupie
Posts: 179
Joined: Mon, 18 Jun 2012, 00:53
Real Name: Chris
Location: Mount Gambier

Memory effect found in LiFePO4

Post by TooQik »

weber wrote:I think you are correct in saying that, TooQik. However that's what memory effect is in NiCd too.
Cheers for that weber.

Regarding creating an accurate SOC meter, what about creating a multiple axis lookup table in the same way they do fuel injection fuel mapping tables?
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by weber »

I think what it means is if I'm only going to charge to 75% SoC every time, then the rested voltage I have to aim for will go up a few millivolts on each of the first few times around the loop. A simple way to deal with it might be to use a voltage corresponding say a non-memory 80% initially and then after a few times around, that same voltage might correspond to 70%, which would still be OK.

Coulomb tells me that in the full paper they also look at what happens when you go to 100% SoC every time, but have different depths of discharge. The situation is almost symmetrical, but the resulting memory dips on the low side are smaller than the memory bumps on the high side.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
TooQik
Groupie
Posts: 179
Joined: Mon, 18 Jun 2012, 00:53
Real Name: Chris
Location: Mount Gambier

Memory effect found in LiFePO4

Post by TooQik »

What I'm taking away from this discussion is that unless you charge to 100% SoC and discharge to 0% SoC each and every time you will see some form of memory effect. But here's the kick in the pants, in order to know the accurate SoC we need to charge and discharge to a known SoC - bit of a paradox really. Image

Roll on magnetic SoC sensors.

Obviously there is a way to determine where the Li ions are present in a cell, is there a way to incorporate this into a SoC meter, or are we talking stupidly expensive laboratory equipment incorporation that's not feasible?

I guess the other side to this coin is, how accurate is accurate enough for general use?
Last edited by TooQik on Tue, 13 Jan 2015, 10:57, edited 1 time in total.
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by weber »

TooQik wrote: What I'm taking away from this discussion is that unless you charge to 100% SoC and discharge to 0% SoC each and every time you will see some form of memory effect.
Sure. But the magnitude of that effect is nothing like it is in NiCd. We already know it's crazy to try to estimate SoC from voltage with LiFePO4 because of the big flat areas of the curve, and the hysteresis. Memory bumps and dips are just one more annoyance to complexify the model even further.
But here's the kick in the pants, in order to know the accurate SoC we need to charge and discharge to a known SoC - bit of a paradox really. Image
I'm not getting what's paradoxical about that. To know the accurate SoC based on the voltage and the current history, without taking the memory effect into account, we need to charge and discharge to 0% and 100% every time. If we model the memory effect, we don't need to charge and discharge to 0% or 100% ever.

With coulomb counting, the model is dead simple, but you do have to go to 100% often (but not every time) to eliminate drift due to self-discharge etc, and to 0% occasionally to recalibrate the slowly reducing capacity.
Roll on magnetic SoC sensors.
What is a magnetic SoC sensor? How are they supposed to work?
Obviously there is a way to determine where the Li ions are present in a cell, is there a way to incorporate this into a SoC meter, or are we talking stupidly expensive laboratory equipment incorporation that's not feasible?
If you mean, somehow measure it directly, it sounds stupidly expensive. But that's effectively what coulomb counting (amp-hour counting) does, by looking at the movement of the counter-balancing electrons in the external circuit. It just misses out on the electrons that take a short-cut inside the cell.
I guess the other side to this coin is, how accurate is accurate enough for general use?

Well yeah.

So now I'm thinking, do I rely on coulomb counting instead, and program periodic excursions to 0% and 100% to resync and calibrate. Hmm. I think I'll persist with the attempt to recognise approx 20% and 75% states from voltage and current. This is for a solar power system. I think coulomb counting is fine for an EV.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
T1 Terry
Senior Member
Posts: 1952
Joined: Thu, 30 Sep 2010, 20:11
Real Name: Terry Covill
Location: Mannum SA

Memory effect found in LiFePO4

Post by T1 Terry »

Perhaps program in a higher cell end voltage once a mth and use this at the 100% SOC reset, say normal operation is cell end voltage of 3.38v and the once a mth to 3.5v. Still well within the safe voltage area but it would eliminate the memory effect. I have found the memory effect is greater if a high end voltage is regularly used, a trip to even 0.2v higher per cell will require quite a few more Ah if it is only perform once a yr on a daily cycled battery pack, yet minimal if the higher voltage is a once a mth programmed event. It's almost like doing an equalise charge on a lead acid battery, just the voltage is much lower.

T1 Terry
Green but want to learn
TooQik
Groupie
Posts: 179
Joined: Mon, 18 Jun 2012, 00:53
Real Name: Chris
Location: Mount Gambier

Memory effect found in LiFePO4

Post by TooQik »

weber wrote:What is a magnetic SoC sensor? How are they supposed to work?
Here's some links to battery SoC using magnetic field strength:

http://www.cadex.com/_content/New_Advan ... _Final.pdf

http://www.ecnmag.com/articles/2012/01/ ... or-fallacy
User avatar
Johny
Senior Member
Posts: 3749
Joined: Mon, 23 Jun 2008, 16:26
Real Name: John Wright
Location: Melbourne
Contact:

Memory effect found in LiFePO4

Post by Johny »

I haven't taken much part in this discussion because I've stopped getting emails in the last day or so.

Anyway. I have been measuring the Vogue LiFePO4 modules (a module is 16S2P of 10AH cells) on and off for the past 20 months with an eye to doing what weber is suggesting. That is using internal resistance, voltage and current to determine SOC.
I have found that temperature plays such a big parts and my cells are scattered throughout the car - so have pretty mush given up on this approach (I don't have a built in measurement system either so it;'s a big add-on for me).
What has come out of it however is that Open Circuit Voltage after discharge and after 60 minutes or more resting is accurate to about 5%.

Since I don't measure charge coulombs - only discharge - I assume Max charge each time I reset the trip/SOC meter and I use the OCV SOC every few months (for each module then over entire pack) to track pack degradation. So far none after 400 cycles with a usual 30% DOD.

IMO LeFePO4 is so mV sensitive that it's really difficult to "tune out" this "memory effect" with full bi-directional coulomb counting.
It's a lot easy with the 3.7V cell chemistries I think.
TooQik
Groupie
Posts: 179
Joined: Mon, 18 Jun 2012, 00:53
Real Name: Chris
Location: Mount Gambier

Memory effect found in LiFePO4

Post by TooQik »

weber wrote:
TooQik wrote:But here's the kick in the pants, in order to know the accurate SoC we need to charge and discharge to a known SoC - bit of a paradox really.
I'm not getting what's paradoxical about that. To know the accurate SoC based on the voltage and the current history, without taking the memory effect into account, we need to charge and discharge to 0% and 100% every time. If we model the memory effect, we don't need to charge and discharge to 0% or 100% ever.

With coulomb counting, the model is dead simple, but you do have to go to 100% often (but not every time) to eliminate drift due to self-discharge etc, and to 0% occasionally to recalibrate the slowly reducing capacity.
By a paradox, I meant that to charge to 100% or discharge to 0% means knowing exactly where 100% and 0% SoC really is, but these are moving goal posts.

What seems to be the current general consensus is to draw a line in the sand and say "at X.X OCV a cell is at Y % SoC", and then adjust down the track for aging/memory/temperature/etc. and I agree with this approach but can't help think (like most others here) that it would be nice to know the SoC more accurately.
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by weber »

Johny wrote:... using internal resistance, voltage and current to determine SOC.
I have found that temperature plays such a big parts and my cells are scattered throughout the car - so have pretty mush given up on this approach (I don't have a built in measurement system either so it;'s a big add-on for me).
What has come out of it however is that Open Circuit Voltage after discharge and after 60 minutes or more resting is accurate to about 5%.
Glad you're back with us. Thanks for reminding me that I've been ignoring how strongly internal resistance depends on temperature. But I do have temperature measurement for every cell, from the LyteFyba BMS, so I could do it.
Since I don't measure charge coulombs - only discharge - I assume Max charge each time I reset the trip/SOC meter
I assume you do count regen-braking charge coulombs?
and I use the OCV SOC every few months (for each module then over entire pack) to track pack degradation. So far none after 400 cycles with a usual 30% DOD.
How exactly do you use rested OCV to track degradation? Of course it can only help detect one kind of degradation: reduction in capacity and not the other kind: increase in internal resistance (at a given temperature).

Which reminds me that increase in internal resistance with age is another thing I'm not yet taking into account, for the voltage and current based method of SoC determination.

I guess I could try to make every CMU measure its cell's internal resistance opportunistically, whenever a heavy load (kettle, toaster, microwave, water pump) is turned on or off. This would automatically take account of aging and temperature. It could be updated as a weighted average with the previous value so one bad measurement doesn't throw things way out.

It would require that the CMUs have up-to-the-second current information sent to them by the IMU. But this is needed for a CMU to estimate its cell's SoC anyway, even with a fixed internal resistance figure.

By the way, when I write "current" in this thread, you can assume I mean "charge flow rate". I'll use "present" for the other thing. I wish people writing papers about this stuff would do so too.
IMO LeFePO4 is so mV sensitive that it's really difficult to "tune out" this "memory effect" with full bi-directional coulomb counting.
Not sure what you're saying here. The summation-drift of coulomb counting is not a memory effect, even in scare quotes, and it is not what the paper that started this thread is talking about.

And the memory effect described in the paper has no effect on coulomb counting.
It's a lot easy with the 3.7V cell chemistries I think.
Yes.

It should be relatively easy to use voltage measurement to determine the 75+-5% SoC in my "black monolith" solar power system because at that stage the charger is throttled back by the PI controller running in the IMU, to supply little more current than can be bypassed by the CMUs. So internal resistance doesn't enter into it, and therefor I don't have to worry about the effect of temperature or aging.

But when detecting 20+-5% at the low end, I can't throttle back the power to the loads (as I can in the MX-5). I have to deal with highly varying current.

The original plan was to use coulomb counting for this (reset to 75% at the top end). But coulomb counting doesn't tell you what's happening with individual cells (which particularly matters if they are out of balance), and it needs periodic full discharge to recalibrate for the capacity of the least cell (assuming they are in balance).
Last edited by weber on Sun, 28 Jun 2015, 09:12, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by weber »

TooQik wrote:By a paradox, I meant that to charge to 100% or discharge to 0% means knowing exactly where 100% and 0% SoC really is, but these are moving goal posts.
Ah [is that a kind of sigh or am I starting a sentence about amp hours?]. Image IMHO 100% and 0% are the only ones that aren't moving goal posts. They are the only ones that are easy to detect by voltage and current (and a very approximate internal resistance) alone. You don't need to know any history to detect them. This is because the voltage vs SoC curve is so steep near them.

With LiFePO4, it's impossible to use voltage to tell the difference between 80% and 95% or between 45% and 65% because the curve is sooooo flat in those two regions. But outside of those two regions, it may just be possible to get +-5% accuracy, and that is certainly so at 0% and 100%. I just want to bring those charge extremes in a bit, to 20% and 75%, for longevity.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
Johny
Senior Member
Posts: 3749
Joined: Mon, 23 Jun 2008, 16:26
Real Name: John Wright
Location: Melbourne
Contact:

Memory effect found in LiFePO4

Post by Johny »

I assume you do count regen-braking charge coulombs?
Yes - just don't measure charger current as I have 12 chargers - one for each module.

Forinterests sake, here is a snapshot of a graph comparing web info I have gleaned compared with my cells. Mine is incomplete (blue) as there are some SOC values I have never captured. The SOC for my cells is the calculated SOC from AH used since full charge.
Image
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by weber »

Thanks Johny. Can you give links for those 3 web graphs?
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
TooQik
Groupie
Posts: 179
Joined: Mon, 18 Jun 2012, 00:53
Real Name: Chris
Location: Mount Gambier

Memory effect found in LiFePO4

Post by TooQik »

weber wrote:By the way, when I write "current" in this thread, you can assume I mean "charge flow rate". I'll use "present" for the other thing.
Ah [is that a kind of sigh or am I starting a sentence about amp hours?].
I stand guilty as charged your honour. Image
IMHO 100% and 0% are the only ones that aren't moving goal posts. They are the only ones that are easy to detect by voltage and current (and a very approximate internal resistance) alone.
I agree that they the easiest to determine.

You may have seen this before, but I found the data in the tables on this link interesting: http://www.bestgopower.com/technology/d ... -test.html
Johny wrote:Forinterests sake, here is a snapshot of a graph comparing web info I have gleaned compared with my cells.
Very interesting graph there Johny, thanks for sharing that.
User avatar
Johny
Senior Member
Posts: 3749
Joined: Mon, 23 Jun 2008, 16:26
Real Name: John Wright
Location: Melbourne
Contact:

Memory effect found in LiFePO4

Post by Johny »

weber wrote: Thanks Johny. Can you give links for those 3 web graphs?
Sure - I think you have one already. It's difficult to get figures so I often enlarge the graphs and carefully pick off values. I also sometimes curve matching to extrapolate values not present.

Web1=http://www.cse.anl.gov/us-china-worksho ... %20BMS.pdf
[Edit: Broken link above replaced by https://forums.aeva.asn.au/download/file.php?id=1410]

Web2=http://www-personal.umich.edu/~hpeng/pu ... 3_Weng.pdf

Web3=http://www.academia.edu/4672627/Develop ... ery_System
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by weber »

Incidentally, you need a very good multimeter to measure LiFePO4 voltages accurately enough for SoC determinations, or to calibrate other equipment to do so.

After much research, and with many thanks to Dave Jones of EEVBlog, I found that the best value-for-money multimeter on the planet for this purpose, is a 12-to-18-year-old Fluke 87 III. The Fluke 87 III was made (in the USA) from 1997 thru 2003. They sell second-hand on eBay for a little over $200. Ask if they have green or white back-lighting. The white ones are newer. And of course check the reputation of the seller. And make sure you're not buying a plain 87 (sometimes called an 87 I). They are too old and only half as accurate. There is no 87 II or 87 IV. Some sellers clean them up so well you'd think they were new.

Image

Sure you could buy a present-model Fluke 87 V but it will cost you at least twice as much, is no more accurate and has no new features of any interest. [Edit: With the possible exception of the ability to measure capacitance above 5 uF] In fact some of its new "features" are just annoying. Both have a basic DC accuracy of 0.05% +-1 count. They are auto-ranging true-rms 4000-count meters having a 20,000-count mode with a lower update rate.

Coulomb and I both have one now. I'm in love with mine. They are a brilliant general purpose multimeter. I wish I'd never wasted my money on all the other crap multimeters I've bought. How to tell if your multimeter is crap.

The Fluke 87 III just does everything right, including instant beep on continuity, high enough voltage on diode-check to handle LEDs, a wide viewing-angle display with back-lighting, a Cat III 1000 V rating and the magical "Touch Hold" mode so you don't have to try to read the display at the same time you're holding the probes in some difficult or dangerous position. Here's the Fluke 80 Series III User Manual

[Edit: Changed the image to one that shows correct lead polarity -- Thanks Coulomb, for spotting that.]
Last edited by weber on Fri, 12 Feb 2016, 19:17, edited 1 time in total.
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
User avatar
Richo
Senior Member
Posts: 3737
Joined: Mon, 16 Jun 2008, 00:19
Real Name: Richard
Location: Perth, WA

Memory effect found in LiFePO4

Post by Richo »



Yeah good one!

I can tell mine is crap.
But it only crap because I left it in the car after a job and it got to 50DegC+ and killed the LCD.
User avatar
Jeff Owen
Groupie
Posts: 157
Joined: Thu, 13 Nov 2008, 15:53
Real Name: Jeff Owen
Location: Brisbane

Memory effect found in LiFePO4

Post by Jeff Owen »

weber wrote:And make sure you're not buying a plain 87 (sometimes called an 87 I). They are too old

Well thanks Weber. I was very proud of owning a genuine Fluke, and now I will have to hang my head in shame every time I take it out in public.
User avatar
Adverse Effects
Senior Member
Posts: 1172
Joined: Sat, 01 Jan 2011, 03:30
Real Name: Adverse Effects
Location: Brisbane

Memory effect found in LiFePO4

Post by Adverse Effects »

Jeff Owen wrote:Well thanks Weber. I was very proud of owning a genuine Fluke, and now I will have to hang my head in shame every time I take it out in public.


Image
User avatar
weber
Site Admin
Posts: 2936
Joined: Fri, 23 Jan 2009, 17:27
Real Name: Dave Keenan
Location: Brisbane
Contact:

Memory effect found in LiFePO4

Post by weber »

Jeff Owen wrote:
weber wrote:And make sure you're not buying a plain 87 (sometimes called an 87 I). They are too old

Well thanks Weber. I was very proud of owning a genuine Fluke, and now I will have to hang my head in shame every time I take it out in public.

Sorry Jeff. But if someone didn't already own a Fluke 87, and if they needed to determine LiFePO4 cell SoC from rested voltage (which is what this thread, and my post, is about), would you recommend they buy, second-hand on eBay, a 0.1% accurate meter that is 23+-4 years old, or a 0.05% accurate meter that is 15+-3 years old. I figure I'm taking enough of a risk recommending a 15 year old meter. Here is some History of the Fluke 80 series

And you should still be proud. Those old 87 series are just beautifully designed and built pieces of equipment (built in the USA, not China). And you should particularly be proud, since I happen to know it was a gift from your grateful previous employer, CSIRO.

And yes, I expect you were probably being facetious. But I thought I'd fight water with water. Image
One of the fathers of MeXy the electric MX-5, along with Coulomb and Newton (Jeff Owen).
Post Reply