Switched capacitor balancer tests

texaspyro

100 kW
Joined
May 12, 2010
Messages
1,408
I did some tests on a simple (and very unoptimized) switched capacitor balancer on a pair of 2300 mAh A123 cells. A switched capacitor balancer wastes no energy as heat. It shuffles charge from the higher voltage cells to the lower ones by switching a capacitor between adjacent cells. It has no precision components and makes no measurements on the cells.

Its main disadvantage is rather small balance currents... this one is around 1 mA per 10 mV of imbalance. But it can run whenever the pack is being charged/discharged/or sitting idle. By keeping it connected while running or charging, it helps keep the pack from becoming unbalanced in the first place.

First I fully charged the cells, then I drained 750 mAh from one and 1300 mAh from the other. I placed them on a 7.3V, 1A cv/cc lab supply to charge. At 1 hour the cell voltages were 3.32 and 3.30 volts. At 1.5 hours the cell voltages were 3.90V and 3.34V. They essentially stopped charging and stayed at these voltages overnight. Next I discharged them to 2.50V. I got 2356 mAH and 1777 mAh. Clearly the cells stayed unbalanced. Bad news if your pack was in this condition.

Next I fully charged and drained the cells to the same condition as in the first test. I then placed them on the lab supply with the capacitor balancer connected. At 1 hour the cells were at 3.38V and 3.36V. At 1.5 hours they were at 3.82 and 3.38V. At 2 hours they were at 3.76 and 3.47V. At 2.5 hours they were at 3.63 and 3.62V... both cells fully charged and balanced. I drained them to 2.5V and got 2256 mAh and 2306 mAh.
 
nice.. would charging over 3.65v be damaging the cells ..
however, for ebike use this might not be so good...
if somehow the balance current could be scaled up 5x-10x
 
The A123 cells are spec'd to a max 4.2V charge voltage. Normally you would shut off the charger when a cell goes over say 3.65V (most shunting chargers go into shunt mode at 3.8 to 3.9V). I was charging with just a lab supply with no HVC monitoring.

The cell voltage differences that I saw are the kind of cell voltage differences that people get when they charge with a dumb bulk charger without a BMS and individual cell monitoring. With a larger series string of cells you have a much greater chance of developing damaging levels of voltage on the high cells.

The cell balancer that I was using was just cobbled together with what was at hand and driven by a lab pulse generator. It was not optimized in any way. One should be able to get 1 mA/mV. I have seen a switched cap equalizer running on a 96V, 32Ah lead acid stack to very good effect.

Also, my actual balance currents may be higher... I measured them with a hand held meter that I know does not particularly like pulsing waveforms on its DC ranges... I need to break out the Tek current probes...
 
Do you suppose these E-City capacitance BMSs are implementing a similar approach? http://www.bmsbattery.com/index.php?main_page=product_info&cPath=4_13&products_id=85&zenid=3be7929847c064fa123af1f42f01d786

I noticed they're not providing anything in the way of specs when it comes "balancing ability".
 
Nope, that is a capacity balancer, not a capacitor balancer. Capacity balancer is just another (rather poor and confusing) name for a cell balancer.
 
texaspyro said:
Nope, that is a capacity balancer, not a capacitor balancer. Capacity balancer is just another (rather poor and confusing) name for a cell balancer.
LOL... marketing fun and games :roll: . Have to admit it peeked my interest a while back when I first saw them :oops: . I thought "well they must be better than the the other types" LOL.
 
texaspyro, nice work.

Can you give some more details - eg, capacitance you used, switching frequency, etc? Is it only a 2 cell system so far?

Also, I think you'll find its not 100% efficient. Its true it doesn't just throw the energy away like a resistive balancer, but there are losses involved in charging and discharging caps. I'd have to do the maths, but I think there's a theoretical limit of 50% efficiency.

Nick
 
Right now I am using a 1KHz switching freq and 5000 uF cap. The FETs are some TO220's I had laying around. It is just a 2-cell perfboard test unit, for now.

A real unit would probably switch at around 20 KHz and use a smaller cap.

Cap charging is HIGHLY efficient, more so than battery charging. There are charge pump chips that are 98%+ efficient. The main losses from the battery are I*R losses in the FETs, cap ESR losses, and shoot-through conduction between the FETs. Shoot-through can be eliminated by using separate clocks for the P and N channel fets with built in dead time. I am using a single clock so I have the clock rate down. If I had 50% losses, stuff would be getting rather warm.

There are also system losses in the cap balancer in the gate drivers and clock generator. These can be fairly minimal since the FETs are small. Standby losses when the thing is not switching is just the cap leakage (and that can be eliminated by keeping both FETs off when idle). When running, the only real wasted power is in the gate drive/gate capacitance.

I measured the balance current on some mostly balanced 3xAA NiMH packs and it was under 50 micro amps. With the batteries not connected, it was not measureable... less than a micro amp (and that was all shoot-through).
 
Ok, I take it back, I'm sorry.

I was thinking of the case where a capacitor is charged from zero to a given voltage. Only half the energy ends up stored in the capacitor. I've just done the maths and if you start with the capacitor partly charged the losses are lower, and the proportion lost tends to zero as the start and finish voltages get closer together.

Its not perfectly lossless, but as the cells get close together the losses become insignificant.

Nick

PS. If you're really interested, when the charge goes in or out of the capacitor, the energy lost as heat compared to the energy transferred is (Vfin-Vstart)/(Vfin+Vstart). So for Vstart=0 as much is wasted as is transferred.
 
No, it's the energy stored on a cap that is: E = 0.5 * C * V * V (i.e. 1/2 C V^2) All the energy that goes in is stored. All that is stored goes out... less what is lost to the electrical friction gods in the form of the capacitor ESR (and to a lesser extent the dielectric dissipation factor). If caps lost that much energy as heat, your 1000W power supplies would be crispy critters.

On my CD welder (see http://endless-sphere.com/forums/viewtopic.php?f=2&t=2633&start=570#p280216) I put 740 watt seconds in (charging 3.7F from 0 volts to 20V), fire it all into a load which drains the cap back to 0 volts, and I get 740 watt seconds out (less a little vig paid to the gods of Rds, ESR, and Interconnection). Been there, measured that... lots and lots of times. I would certainly have noticed 370 watts of heating going somewhere it shoudn't have. Nothing gets even slightly warm except the intended load.
 
texaspyro said:
No, it's the energy stored on a cap that is: E = 0.5 * C * V * V (i.e. 1/2 C V^2) All the energy that goes in is stored.

If you're starting from zero volts, then it takes C V^2 energy from the source and only half of it ends up in the capacitor. So, yes, all the energy that goes into the cap is stored, but the same amount of energy doesn't go into the cap. That's if you connect the cap to a normal voltage source. If you have a switched mode source doing max power transfer tracking then it can be charged efficiently. But as I said, that applies with a large voltage swing on the cap, for small swings the energy loss in each cycle is small.

The CD discharge is a slightly different matter; there you are trying to turn the stored energy into heat and all you have to do is to make sure the loss R is small compared to the load R for the energy to end up in the right place.

Nick
 
Tiberius said:
So, yes, all the energy that goes into the cap is stored, but the same amount of energy doesn't go into the cap.

Huh? :?

I have measured the power going into and out of the cap with VERY precise test equipment (like accurate to well under tenths of a part per million). All the energy that came out of my charger supply got stored in the cap. All the energy that got stored came out of it. Only trivial amounts were lost in the process. There was no unexplained loss due to heat, etc.
 
So is the function of a capacitor shunt similar to that of a buck converter?
 
It's not a shunt... the device works by switching capacitors across adjacent cells. The cap draws charge out of a cell when that cell has a voltage higher than the cap. It dumps charge into a cell when it is across a cell that has a lower voltage. Let it run long enough and the voltage on the two cells will wind up the same. In a long series string of cells, the voltage across the whole pack eventually equalizes.

There are also inductive balancers that use inductors rather than capacitors. They can handle much high balancing currents at the expense of some rather nasty failure modes, more stressful operating parameters, and a more complicated circuit.

Conceptually, the difference between a resistive shunt balancer and a switched capacitor balancer is similar to the difference between a linear and switching power supply.
 
I did another test of the balancer. This time with the cells radically unbalanced. I used a fully charged cell and one drained of 2000 mAh (leaving 300 mAh). Without the balancer, the charge current dropped from 0.80 amps to 1 milliamp in under 10 minutes. Charging effectively stopped with the cells fixed at 3.18 and 4.09V

With the switched capacitor balancer the cell voltages started at 3.25 and 3.93V. Power supply charge current was 0.4A Within two hours it fell to 0.2A with the cells at 3.33 and 3.89V. In 3.5 hours the cells were fully charged and balanced at 3.63V each with the charge current under 1 milliamp.

One thing that is a bid odd is that if you add up the mAh out of the power supply as measured by my hand held meter, you get around 1200 mAh max into the cells, but it took at least 2000 mAh to restore the drained charge from the low cell. I think the meter is having problems with the 1 KHz signal from the switched caps. The meter is known to be very accurate with clean DC. Another (but highly unlikely) possibility was the charge current peaked when I wasn't looking. After the first 30 minutes, I was checking it every hour.

Another (slightly unlikely) possibility is that I have invented cold lithium fusion and the world's energy problems are now solved. :roll: I think I'll go tuxedo shopping tomorrow for that Nobel Prize trip to Stockholm... after all it's sales tax free weekend in Texas.
 
Wow... exciting stuff, 3.5 hours isn't bad at all from my perspective (charging habits/needs). 8) I'll by you steak dinner if you get that prize LOL. You're not related to anyone with the last name of Ponds, are you? :mrgreen:
 
scoot said:
Wow... exciting stuff, 3.5 hours isn't bad at all from my perspective (charging habits/needs).

At the 1 amp charge rate that the power supply was set to, it would normally have taken 2 hours to recharge those cells if the fully charged cell was not blocking the current flow. That is the problem charging series strings of LiFePO cells. Once a cell becomes fully charged, it stops drawing current and blocks the flow of current to the other cells in the string.

Since the current flowing in a series connection is the same through each component in the string, the cells with lower charge stop getting power and eventually stop charging. That leaves them at a lower voltage, The sum of the voltages across each cell must equal the charger supply voltage. The fully charged cells wind up with a higher than normal voltage across them. In my 2 cell string the cells were at 3.3 and 3.9V (still within acceptable limits). In a larger series string with several uncharged cells and a couple of fully charged ones, the fully charged cells can wind up with over 4.2V across them... very bad for the cell. That is why you need some sort of BMS during charging. It needs to monitor individual cells and when any are starting to go over voltage it needs to shut down the charger.

Resistor balancers try to prevent individual cells from going over voltage (and blocking the power to the other cells in the string) by switching a bypass resistor across the fully charged cells. This works to a point, but wastes a lot of power as heat. If you have a BMS that only shunts smallish currents and a higher power charger the shunts may not be able to keep the cell voltage of the fully charged cells from rising. You still need a way for the BMS to shut down the charger. Also, the BMS is usually located near the cells and the resistors can heat up the cells... not a good thing (particularly in enclosed packages).
 
cap_balance_1.jpg


cap_balance_2.jpg


cap_balance_3.jpg


cap_balance_4.jpg


from this reference
 
The standard paper on switched cap balancers is: C. Pascual and P. T. Krein, "Switched capacitor system for automatic series battery equalization." Most links for it are dead, but I got it off the Wayback Machine. Or see US patent 5710504

As far as the suitability of switched cap balancers for large packs, these people used that same circuit in a 40S 5P 50 amp-hour pack for a snowmobile: http://www.mtukrc.org/download/clarkson/clarkson_ze_design_paper_2009.pdf
 
Knuckles, your drawing gets the basic idea, the gate drive scheme won't work (particularly for larger packs). Most switched cap balancers use a P channel high side fet and an N channel low side fet. The gate drives need to be isolated from each other, otherwise you quickly run out of Vgs thresholds on the upper fets. I am going to try capacitor coupled clock line like the input to a charge/pump or bootstrap front end.
 
texaspyro said:
Knuckles, your drawing gets the basic idea, the gate drive scheme won't work (particularly for larger packs). Most switched cap balancers use a P channel high side fet and an N channel low side fet. The gate drives need to be isolated from each other, otherwise you quickly run out of Vgs thresholds on the upper fets. I am going to try capacitor coupled clock line like the input to a charge/pump or bootstrap front end.
Not my drawing ... found it on-line.

Absolutely fascinating topic texaspyro.

I do look forward to some of your conceptual schematics.
 
texaspyro said:
The standard paper on switched cap balancers is: C. Pascual and P. T. Krein, "Switched capacitor system for automatic series battery equalization." Most links for it are dead, but I got it off the Wayback Machine. Or see US patent 5710504

As far as the suitability of switched cap balancers for large packs, these people used that same circuit in a 40S 5P 50 amp-hour pack for a snowmobile: http://www.mtukrc.org/download/clarkson/clarkson_ze_design_paper_2009.pdf

Very cool ... I'm a Clarkson Grad 84' ChE
 
A good overview of balancer circuitry is in http://www.atmel.com/dyn/resources/prod_documents/doc9184.pdf It shows using their ATA6870 BMS chip for doing all three types of balancers.
 
The only thing I would say to some of the comments about dealing with imbalances is that a BMS shouldn't be dealing with a pack with cells that are at very different SOC. When the pack is built it should really be balanced or fairly close to being balanced. If that is the case the SOC between cells shouldn't become so significant through usage that a suitably sized BMS cannot deal with those imbalances. For sure the average BMS cannot deal very well with packs constructed from cells with big differences in SOC, but as long as it isolates the charger as most do, it should prevent any dangerously high voltages on any of the cells and should eventually bring all the cells into balance.

This elithio white paper discusses the pros and cons of active vs passive balancing:
http://liionbms.com/php/wp_passive_active_balancing.php

I'd love to get to grips with some of these BMS chips. I was looking at a BMS I have here that uses an O2 Micro BMS Chip. It actually doesn't look so difficult to implement, with fairly minimal discreet components for each sense and shunt connection.
 
Back
Top