Charging an enormous battery presents feedback loop problems?

[Flyback]

Hello,
We are designing a 7.4kW offline LLC battery charger for EV. The (lithium) battery is 300-410VDC.
As you know, We are closing the feedback loop on the output current (into the battery).
Do you believe that this presents a tricky dynamic situation for us?
I am used to closing the loop in previous SMPS’s on output voltage, with a resistance type load. In other words, there is a load pole of 1/(2.pi.R.C)…
Where
R is the load
C is the SMPS output capacitance.

As you know, in output voltage regulated SMPS’s the addition of an enormous load capacitance often brings about instability in an SMPS…due to the big decrease in phase margin that it causes.
Our huge lithium battery is obviously an enormous output capacitance, and we are somewhat concerned about this situation of decreased phase margin.
We are wondering, that instead of using an analog feedback loop, comprising an error amplifier, with feedback compensation R’s and C’s, perhaps we would instead be more advised to implement a very slow “try it and see”, incremental type of feedback loop? That is, we could set up for a very low initial charge current, and then just bit-by-bit, reduce the frequency of the LLC converter, until the output current is the right value, and then just continually monitor the current from there on. This would be handled by a microcontroller. Obviously we would have an overcurrent shutdown comparator ‘overseeing’ the whole process.
Which type of feedback loop would you advise for this?

To give an example, the attached Ltspice simulation shows that slightly reducing the LLC switching frequency from 60.2khz to 58.8hz results in an increase in charging current of 10 Amps

 

[Tony Stewart]

Since large LiPo batteries have very low ESR, this presents an unstable operating condition for wide bandwidth SMPS designed for low step load response.

The design could use an external series resistance which reduces efficiency inside the Vsense loop or a single pole integrator that is lower than any other poles and thus reduces the gain BW product, much like an OP AMP compensation has a pole at 1Hz with high gain (1e6) to yield a fixed gain bandwidth based on voltage feedback ratio and current feedback ratio that is not cycle to cycle regulated.

The other way as you suggested is to have a ramp soft start , rather than a step input with expected overshoot. There is no need to input a step response for charge control and it ought to detect battery SoC 1st by pulse current tests to measure ESR then apply a suitable profile to charge the battery. 1) ramp to fixed CC 2) when it reaches the CV level, switch to CV mode 3) switch to shutoff mode when current drops below 10% of previous peak current or after a certain time limit, indicating a faulty shorted cell.

A bode plot is necessary with any expected load and suitable compensation loop to prevent oscillation. The difference in ESR between Short circuit detection and a battery might be negligible < xx milliohms but at a different voltage.

 

[Flyback]

To give an example , the attached ltspice simualton shows that decreasing the LLC switching frequency from 60.2khz to 58.8khz results in a change in charge current of 10 Amps.

 

[Tony Stewart]

is CC programmable?

 

[Flyback]

constant current level is to be programmable , yes.

 

[crutschow]

In a constant current (current output) supply with proper loop feedback the load capacitance should have only a small effect on loop stability since the capacitance is not directly seen by the loop as compared to a constant voltage supply. (The feedback loop is first-order rather than second-order)
If you write the loop equations, you will see this.

Also you might consider a hysteretic (bang-bang) feedback loop which requires no loop compensation.

 

[Flyback]

I believe the battery has an extremely low dynamic resistance, which means the only way we can do this is with a slow microcontroller based, "set and check" type feedback loop, which is overseen by a comparator due to the slow nature of our repetitive "set and check" feedback

 

[crutschow]

I believe the battery has an extremely low dynamic resistance, which means the only way we can do this is with a slow microcontroller based, "set and check" type feedback loop, which is overseen by a comparator due to the slow ature of our repetitive "set and check" feedback

It's true a battery has a very low dynamic impedance which means the loop could be unstable for a voltage feedback loop, but that's not necessarily true for a current feedback loop, which generally can operate with a zero load impedance. Thinking the two types of loops have the same problem with a low impedance load doesn't make it so.
You need to write the loop equation for a constant-current feedback to see this.
But you mind seems to be made up on the subject so why are you posting here?

 

[Flyback]

As discussed, we are designing a 7.4kW battery charger using an LLC converter (SMPS) charging stage.
On connection of a battery, we will set the charging current by first of all setting the LLC converter’s operating frequency to a very high frequency, and then slowly, slowly, bit-by-bit, we will creep the switching frequency (of the LLC converter) down, until the charging current builds up to the value that we want (20A max).
We will gradually decrement the switching frequency by gradually reducing the current drawn from the “FREQ” pin of the ICE2SH01G LLC control chip.
We will do this gradual frequency changing using a microcontroller. The microcontroller will effect the drawing of current out of the FREQ pin of the ICE2SH01G controller by adjusting the voltage at the output of buffer opamps which feed the FREQ pin of the ICE2SH01G controller, via resistors. This is as shown in the attached schematic.
Can you see any problems with this method?
ICE2HS01G LLC controller datasheet:
https://www.infineon.com/dgdl/Infin...n.pdf?fileId=db3a30432a40a650012a458289712b4c

 

[crutschow]

The data sheet for the LLC controller shows a feedback connection to the LOAD input as well as the FREQ input.
Don't you need to do the same?

I see no particular problem with your scheme otherwise. You are simply slowing down the loop response so that the reactive output response is damped.

 

[Flyback]

thanks, we will have to put a voltage of between 0.2 and 1.8v into the load pin to stop the converter shutting down.
Also, we will have to look into how to stop the sync. rect. FETs from working in light load, as the LOAD pin does that by "looking" at the collector of the feedback opto, though we of course, wont be using a feedback opto.
We will have to look into what happens if the sync. rect. fets are just allowed to keep working during light load, as that may be an option.

 

[Flyback]

This is a new situation but relates to the above...

Do you know what is meant by “pulse charging an electric vehicle battery for the purpose of detecting individual damaged cells”? We understand “cell balancing” and how to detect imbalance, but don’t know why doing it needs the battery charge current to be pulsed on and off repeatedly. This all means that the battery charger design that we have done for our customer to date will have to be scrapped..
We have been designing an offline 3kw battery charger for an Electric vehicle battery which has a voltage of 300-410VDC.
We have designed it with an interleaved Boost PFC followed by an LLC converter. We designed it with an extremely simple, slow, non-dynamic feedback loop because we assumed that since it takes 4 hours to charge the battery, it wouldn’t matter if the microcontroller takes some 30 seconds to get the current up to the maximum charging current level. (obviously this is all “Overseen” by an overcurrent comparator).
The micro was meant to gradually increment the charging current to the maximum level by gradually adjusting the LLC converter’s switching frequency and dead time..(as well as us altering the output voltage of the Boost PFC at times).

..However, we have now found out that this is totally unsuitable. In fact, the charging of an electric vehicle battery requires an extremely fast feedback loop bandwidth of the charger. This is because toward the end of the battery charge, the charge current must be pulsed on and off repeatedly. This is for the purpose of detecting damaged cells. The current pulses must get up into regulation within about 100ms. This is far faster than our slow, incremental feedback loop can handle.
We cannot find any detail on the nature of the current pulses required for this damaged cell detection phase, and our customer cannot provide any such detail, since they have irregular contact with their battery supplier.
Do you know the detail of this pulsed current regime? –eg, to what level must the current rise? (C/2 etc), and how accurate must the current be?,…and how fast must the current get into regulation during a pulse?, and how many current pulses are to be delivered.?

 

[crutschow]

Is this perhaps what you are trying to do?

 

[Flyback]

thanks very much, now we know its called "electrochemical dynamic response". Nowhere has any great detail on it. It seems to me like possibly a slight waste of time. I mean, a battery is going to die over time, so is there any point in detecting it. Also, surely just plain old measurement of cell voltage whilst charging and discharging would , I would have thought, been quite an adequate and satisfactory way of detecting damaged cells?
OK supposing a battery was going to get put in a vehicle that had to go on some crucial mission out in space or whatever, then sure, it would be worth doing electrochemical dynamic testing on it first, but for a car?...is it worth it?...does it give much more information than just measuring the cell voltages whilst charging in a non-pulsed way? And supposing one does detect cell abnormality through "electrochemical dynamic testing"...so what?....is there a way of fixing the cell?...I don't think so....whats the point in detecting something with an expensive detection method if there's no way of fixing it?

 

[Tony Stewart]

By preventing over charging in one cell using a periodic pulse desulphation & measurement means with external load cell balancing or termination of charging or reducing of charge rate, you can extend the battery life.

 

[Flyback]

thanks, I see your point, though do we need to actually do this pulsing of the lithium battery to get these results? I mean, you don't have to pulse current into a battery in order to get cell balancing done. Also, I've never heard of anyone finding out of they're overcharging a battery by looking at what happens in the battery when its pulsed with current. The "electrochemical dynamic response" is to do with finding out the cell health, not finding out whether the cell is fully charged or not?
"Electrochemical dynamic response" testing seems a waste of time for domestic cars. It's a super sensitive way of detecting the slightest defects in lithium cells, and will likely result in EV car batteries being trashed prematurely , which is bad for the environment?
Why should we make a battery charger with a super fast feedback loop bandwidth, just for the sake of a battery cell testing mechanism which is a total waste of time.?

 

[Tony Stewart]

I recall the imbalance of fresh cells is less than 1% worst case ( usually much better). Since LiPo cells have a useful range of 300mV to 600mV (500mV typ) depending on chemistry with an average in the range of (3~3.9 for most types Li-SOCl2 Li/SO2 Li/MnO2 bobbin, spiral etc. )

If the discharge range of voltage is about 1/7th of the 100% SoC voltage a 1% no load mismatch translates into 7% mismatch in ESR.
Similar to LEDs all White LEDs have the same threshold voltage ~ 1mA as LiPos may have the same 4V no load voltage after charging @4.2V.
In both cases LEDs & batteries have an ESR inverse to power capacity and ability to dissipate heat for a given small temperature rise.

As pulse discharge in LiPo and pulse drive in LEDs the Vf is totally dependent on the ESR and chemistry of the device.

• The RC time constant for the battery can determined over the life span to know what pulse width to use for this measurement.
• Capacitance, C = I *dt/dV for a DC charge or discharge for slope e.g. 2.5Ah with 1A load drops 0.5V in 2.5h= 18ksec thus C=36kF equivalent capacitance
• ESR= ΔV/ΔI for initial response due to a pulse or step. e.g. 5mΩ for a nominal cell
• thus ESR*C= is the time constant of the cell where... 5mΩ*36kF= 180 seconds, if I worked this out ok.
  
• This time constant reduces with aging since ESR rises faster than the drop in C.
• Two perfectly matched voltage cells can be tested on some LCR meters if put in series with opposite polarity.

We know when source ESR = Cell ESR maximum power transfer occurs but only 50% efficiency, so this is often compromised to 75% efficiency or limited by the temperature rise in the cell. Although capacity of the cell increases with temperature, cell life-time also degrades much more rapidly.

In the consumer area, Li/CFX (poly carbon monofluoride) and Li/MN02 (manganese dioxide) are found in cameras, calculators and watches. In the military area, Li/SO2 (sulfur dioxide) batteries are used in high power radios. Li/SOCl2 (thionyl chloride) and LiI2 (lithium iodine) are commonly utilized in industrial and medical applications. ref. Tadiran

What I suggest is consider your slow charger but rather than monitor individual cell ESR, monitor temperature with LM35's mux the signals to a serial composite signal and regulate load balance to equalize temperature rise with a setpoint 35~45'C. With forced air to improve response time and cell isolation.

What cell array sizes are you looking at? Ser/Par. Are they bridged?

 

[Flyback]

Thanks, until we get more contact with our customers battery supplier we wont know the architecture of the lithium battery, -we just currently know that its going to be charged by a 3.4kw charger with output of 300-410VDC

 

[Tony Stewart]

Get them to define the normal parameters for the battery ESR , the fault conditions and the access points for load balancing.

 

[Tony Stewart]

It must be considered an Smart Battery until then with it's own load balancing management until then, research battery managers for cars for the dynamic response to pulse charging.