Modifying a DC-DC buck converter

Thread Starter

anishkgt

Joined Mar 21, 2017
529
Hey all,

I came across the LT8390, a DC-DC buck converter for charging 3s2p supercapacitor at 100A using an HP server supply rated 12v @ 82A. Would this be possible using this IC ?
LT8390_2.png
 

Irving

Joined Jan 30, 2016
957
What voltage/size supercap?

Unless you have experience of designing & building high-power electronic systems this isn't a beginner project. A 100A charger needs some serious engineering on the PCB to be remotely reliable. It may be possible with that chip, I'd need to do some serious research to determine that, which suggests, with respect, you don't have the expertise to build such a device from scratch.
 

ronsimpson

Joined Oct 7, 2019
831
You chose a buck/boost and you want a buck. (I think) A buck has less parts.

What is the part number on the capacitor?
 
Last edited:

Thread Starter

anishkgt

Joined Mar 21, 2017
529
The Supercaps are rated 2.7v 360F. The source is rated for 82A only but am hoping to leave some extra room hence designing with 100A in mind. I am still researching on the options, buck-boost or just buck. The buck-boost seemed more efficient but are costly. The cheaper ones don't have mosfets to do the switching. From what i had read using a diode to do the blocking is ideal when dealing with lower currents but not very efficient with high currents due switching losses moreover switching at higher frequencies would reduce the size of the inductor hence allowing for smaller size which is something i don't have much on the PCB. To handle more current adding 4 inductors in parallel would keep the overall series resistance low and allow for higher current.

The LTC3895 and LT3812 from Linear Technology are buck converters that i am looking at. Correct me if am wrong, since the datasheets has not mentioned about the current limit, i presume the only component that would restrict it would is if the inductor is under sized.

@Irving Appreciate your concern. Doing things myself and some help from the community has always helped me on. As with power electronics I've completed a SPOT Welder using MOT earlier and that went well. It was my first project using the mains and a learning curve. The CD spot welder is something i was looking forward to after the MOT. As of now the PCB layout and traces for the CD Spot welder is completed. The spot weld current is limited to 2000A. Its just the power supply that is remaining. I had earlier thought of using an opamp to control the voltage and use several mosfet to dissipate the heat but that was not very efficient i guess. So headed this way. So far with the help of the community things are going well.
 

Irving

Joined Jan 30, 2016
957
OK... Let's examine this in more detail.

3s2p = 3 * 2.7v = 8.1v
So to charge them from 12v you need a buck (down) converter as @ronsimpson suggested & it will be much simpler.

But...

What is your application? What do you plan to do with these supercaps? Give a complete explanation, as this may have specific implications on the charging regime.

Next,...

I understand your 12v supply is capable of 85A but if I connect a 12 ohm resistor to that supply, how much current is drawn?

One last point. Your spot welder's weld current came from the MOT secondary winding. It went nowhere near the PCB which handled nothing more than about 5A. And you didn't design that from scratch. All that fancy electronics has nothing to do with the weld current, only the weld time.

A 100A charger is a completely different ball game. Ignoring the fact that it's unlikely you actually need that capability, this is like saying I've followed the instructions to build a model RC car from a kit, so now I want to design & build my own full-size car from scratch!
 

Thread Starter

anishkgt

Joined Mar 21, 2017
529
Well the project is a Spot Welder using supercapacitor. The supercapacitors get charged and dumps all the energy on to a nickel strip which is welded on to 18650 cells for making packs. I use 6 MOSFETS to switch the capacitors controlled via a independent drivers from an Atmel328.

The whole idea of using a server power supply is because it has the amps to charge the supercapacitors. These server power supplies has all the bells and whistles so no extra effort to manage over current and protection but may to tweak a bit as the server charging a supercap from 0 would see it as a dead short.
 

Irving

Joined Jan 30, 2016
957
OK, I get your plan. Is this an existing design or something you're working up yourself?

Here, FWIW, is my take on a design for a spot welder (based in some part on the design ideas proposed by the designer of the existing device you already built).

The supercaps you've linked to each have a energy density of 0.36Wh and a power density of 3.6kW/kg.

Taking energy density first, 6 * 0.36 = 2.16Wh. At 50% discharge that's 2.16/4Ah = 0.54Ah or 1900As. So, on the face of it there's plenty of weld current on offer. The only issue is they should be limited to 225A maximum discharge per cell, or 450A for a 2P configuration, and thats a non-repetitve value, ie not something you should be doing as a matter of course. Their max discharge amps is suggested as 72A for a 2P configuration, to meet the estimated 500,000 cycles lifetime. But lets say you limit the weld current to 600A and/or 0.5sec.

Looking at power density, 6 * 3.6kW/kg = 21.6kW/kg and 6 * 71g =0.42kg so 21.6 * 0.42 = 9kW @ 50% discharge = 9000/4 = 2250A in theory, though in practice your caps life is measured in a few hundred cycles at that current.

Basically the weld capacity is there, but unless you limit the current flow, your cap life is going to be measured in months rather than years - though obviously it depends on how much welding you do, if several times a day, then weeks, but if only a couple of times a month, then maybe a couple of years, there's no easy way to assess that.

From a charging regime, they suggest 1/10 the discharge current, ie 7.2A for a 2S pack, which is 135 seconds, say 2.5min, from fully discharged to fully charged. Or, given you've already compromised the caps life by overdischarging, you might as well increase the charge rate to, say, <15sec at 72A.

So what you don't need is a buck converter. What you need is a constant current source at 80A with a charge cutoff when the pack voltage gets to 8.1v or when the weld starts. With an interlock to stop you welding when pack voltage drops below 5v say, until charged back up. And a controlled discharge current, or uncontrolled but timed (which is easier, but more brutal on the caps).

Now you have a fledgling specification, you can start assessing the 'how' of it...
 

Thread Starter

anishkgt

Joined Mar 21, 2017
529
Thanks for the input. The design is my own. There are a couple of designs out there but most rely on shorting LiPo batteries. LiPo's can output that much current without a problem, well most of them, some will start puffing halfway through. So i've decided to go with Supercapacitors. Doing a little research and measuring the current on my previous welder, the average current to weld 0.2 nickel strips are between 1.2kA to 1.8kA. Yes the weld current has to be limited and the only way i see to do it is to increase the resistance in the current path (weld cable length and electrodes) and interrupt the weld cycle when the current exceeds the set limit.

That is correct its the constant current that is vital here but i lack the knowledge to come up with something of this spec. I could roughly get one done with an opamp and a 5 x 50A MOSFETs but that would have a lot of heat to dissipate. Not sure if i should go that way. This was why i went with a buck converter and constant voltage.
 

Irving

Joined Jan 30, 2016
957
A switch mode approach is the low loss option, with a current controlled feedback loop, though more complex than a simple linear approach. Given the <15sec charge time I'd personally go with a linear option for simplicity.

If you want to go the switched approach, the topology is sort-of buck, but with a current rather than voltage feedback and since the output voltage is driven by the load impedance, the output volts are below the input volts. So while technically not a buck converter, you'll need a buck controller that uses an external MOSFET, or better, two in synchronous rectification to reduce diode losses. So, some research needed across Analog Devices, STM, Ti, etc product ranges for some candidate devices, and much simpler than that LT8390. You definitely need a slow-start for the current when the caps are close to fully discharged (say < 2v) but ideally not let them get that low. And some high-current MOSFETS for both the charging and weld control side with Rds(on) < 3milliohm ideally. You'd also better check if your PSU -ve side is connected to mains earth or not as it may need to be isolated - we won't know till further into the design.

Sunshine and family calls - later...
 

Thread Starter

anishkgt

Joined Mar 21, 2017
529
If a taking linear approach then how is the load resistance determined to decide up on the Shunt Resistance. Is it

Lr = (Vcc/Vs - 1)Rs
Where
Lr - Load Resistance
Vcc - supply voltage
Rs - Shunt Resistance

Just curious if need to go this way. The MOSFET would not be ON for very long except initially when the caps are close to 0v. Secondly wouldn't the Short-circuit protection kick in during weld ? After a weld the voltage drops to 6v so again charging at 82A to 8.1v

On a second thought, how does controlling current keep the caps from being over charged above 8.1v ? Shouldn't be there a CV as well ?
 
Last edited:

Irving

Joined Jan 30, 2016
957
For the linear option I'd have a series sense resistor (2milliohm) and an opamp monitoring the voltage across it controlling the series transistors/mosfets. You then also control those devices to switch off the charging current a few mS before a weld.

Separately you need a voltage monitor on the caps to shut off the charge at 8.1v and reinstate it at <5v or so.

Whether the latter is done under MCU control or via a crowbar is up to you.
 

Irving

Joined Jan 30, 2016
957
? Shouldn't be there a CV as well ?
No, caps are not batteries, they behave very differently. There is no absorption phase or concept of trickle charging, so no CV needed, you just shut off the charge when they hit 8.1v.

Neither do you need to balance SoC across individual caps, as the series connected capacitors will equalise charge across themselves automatically assuming all the same value.
 
Last edited:

Thread Starter

anishkgt

Joined Mar 21, 2017
529
Thanks, I was always thought the CV was also needed to as caps fall into the energy deivce category but with the exemption that they can't hold the charge for very long and can dump the stored energy unlike a battery. Now i am kinda confused about which to go. I already have a capacitor balancer using the ALD810027.

Since CV is not a problem i could probably ommit the buck-converter altogether but without a voltage regulator wouldn't the capacitor see the 12v at first or is it only when it goes past the nominal voltage that would be a problem.
 

Thread Starter

anishkgt

Joined Mar 21, 2017
529
Moreover the caps have an ESR of 3.6mOhm which would seem like a dead short to the power supply which would trigger the over-current protection feature.
 

Irving

Joined Jan 30, 2016
957
The problem with relying on the PSU's over-current feature is that its probably not a constant current but a voltage fold-back which will cause the supply to continuously cycle in and out of protection, therefore you have no control of what is happening.

A capacitor charged from a constant current I, charges linearly such that I * t = C * V.

You have 3 x 390F in series = 130F x 2 in parallel = 260F. To charge from zero to 8v at 80A = 8 * 260/80 = 26 sec, or from 4 to 8v = 4 * 260/80 = 13sec (as noted previously).

With an linear 80A constant current regulator the linear element initially drops nearly 12v, dissipating 12 v* 80A = 960W initially, dropping to 4v * 80A = 320W after 26sec and shuts down, the average being 640W and storing ~2.3Wh in the caps. That's going to require something like 18 HEXFETs in parallel on a forced air cooled heatsink...hmmm, a switched solution is looking more attractive...
 

Thread Starter

anishkgt

Joined Mar 21, 2017
529
yea its the watts that is not feasible. Decided to go with the buck LTC3824. How do we narrow down to THE buck converters ? i just went along with what in the datasheet and made sure the IC are not limited to a specific amps and has a design example to follow along. LT seems to have one for most of the buck-converters. When i find one that meets these then i choose one that is cheaper and withing voltage and just follow the datasheet and add parallel inductors, resistors and capacitors to share the load. So after an hour and a half found a Vishay inductor for the calculated inductance of 210nH rated for 50A, would add two in parallel. The datasheet suggests multiple layers when laying-out the PCB. Would it be possible in the slightest way of having them in a single layer-double side ? If current is a problem would it help with leaving the trace unmasked and later add solder for that extra current ?

Any suggestions/tips on how to narrow down the search ?
 
Last edited:

Irving

Joined Jan 30, 2016
957
For 100A even 4oz copper isn't enough. You need a minimum of 6mm wide, 1mm thick copper sheet, maximum 25mm long. I use a copper busbar which also serves as a heatsink.

If you parallel up inductors you need twice the inductance on each. Its usually better to run two separate switching transistors/inductors driven from the same controller, or even better, a 2 or more phase switch as this evens out demand on the source. A 2-phase is probably sufficient here.
 

Thread Starter

anishkgt

Joined Mar 21, 2017
529
How is the duty cycle related to the current in the buck-converter, i mean i don't see it being used in any equations for inductance or for calculating max current. This maybe why i am not able select a suitable inductor. For the inductance value i calculate i get wierd current ratings on digikey.

here is excel sheet i use
Buck converter Calculations
Freq (Hz)200000200 kHz
Vin(MAX) (V)24
Iout(MAX) (A)90
Vout (V)8.1
Ripple current (A)36
Inductance (H)1.9875E-070.19875 uH198.75 nH
Rsense (Ohms)
Duty cycle
0.375​
 

Attachments

Irving

Joined Jan 30, 2016
957
How is the duty cycle related to the current in the buck-converter, i mean i don't see it being used in any equations for inductance or for calculating max current.
It's not - using the calculations in the datasheets for most buck converters - because those calculations are for CV output as, for most applications, that's the norm. They are useful for setting the basic components, but then you're on your own. Ths is where you need to do a lot of researching parallel ideas.

With many of the Analog Devices and Ti parts you can run a LTSpice/pSpice simulation to experiment with CC regulation. This article, from Ti, gives some ideas about CC buck converters. Here's a similar AD one that looks more generally at making CC buck converters, mainly for LED drivers, but it gives some ideas. Finally, here's an AD article on building a UPS based on supercaps - nowhere near the current you're aiming for, but gives some ideas. I was looking for articles on charging supercaps for EV use, but while there's been a fair bit of research, there's relatively little published info - probably because its all a bit commercially sensitive at the mo.
 
Top