Efficient power supply topology/method to convert 48VDC to 300V-400V @ 250+W?

[daqq]

Hi guys,

I've made a multiphase flyback power supply for a project that requires an adjustable, isolated power supply that charges capacitors with a voltage of 300V to 400V, with a maximum needed power of 250W. Basically a voltage limited current source. The flyback topology seems the best for me - it's pretty much a pulsed current source that I can enable/disable as I please using a simple feedback system, and I've managed to get up to a bit over 90% efficiency with three identical blocks working in parallel. That's a nice efficiency, but still there's a fair amount of heat coming of the thing - the outside temperature of the transformer maxes out at around 80-85 deg C at room temperature - and it's not going to be at room temperature, it's going to be enclosed, with restricted airflow... And I'd like to avoid adding forced air cooling. The transistors aren't very cool either.

I've checked the efficiency figures for flybacks and 90% is pretty good most of the time. Maybe I'll try using a bigger core, but that's a problem due to height. I've also been checking out toroids for flybacks - not exactly available are they? They would be great in that I could expose much more winding to the outside air and have a greater exposed surface area...


The input voltage is 48VDC, can be set to a bit higher (43VDC to 55VDC being the limits). With mundane parts, what kind of topology would you recommend for this task that would have the best efficiency? Or, better still, is there any kind of module that can do exactly this?

Thanks,

David

[oldway]

You do not say anything about the load.
If it is a low and slowly varying load, for a voltage as low as 48V, it may be advantageous to use a push pull circuit.

[schmitt trigger]

I've used iron-powder toroid cores for flybacks. They work, because such cores have a distributed air gap.

Their disadvantage? The core losses are significantly higher than ferrites. Maximum switching frequency must be definitively be kept well below 100 Khz. And/or their maximum flux swing restricted.

The company Magnetics, Inc produces a line of Kool Mu toroid cores with core losses comparable to ferrites:
https://www.mag-inc.com/Products/Powder-Cores/Kool-Mu-Cores

[nctnico]

I'd investigate where the losses are generated. Something getting hot doesn't mean it is dissipating a lot. There are formulas to calculate core and copper losses in transformers. Using litze wire (less problem with skin effect) and as many windings as possible (lower flux swing) are methods to reduce losses in transformers. Another 'problem' with flyback is leakage inductance.
A better suitable topology may be resonance combined with zero voltage or current switching.

[T3sl4co1l]

PP or H-bridge forward converter is reasonable with these numbers.

Flyback would want to be multiphase, and you'll still be spending a bit extra on caps.  If you're more concerned about development time than production cost, this would be fine, too.

Tim

[Benta]

At 250 W, you are getting outside the sweet spot for flybacks.

I'd start looking at a forward topology; whether it'll be single-switch, push-pull or half/full bridge is a case for investigation. You'll get a much smaller transformer, as it'll only have to cope with the magnetizing field. The flyback needs to store the complete transferred energy.

[jbb]

Flyback... If you're more concerned about development time than production cost, this would be fine, too.

Tim

A discontinuous mode fly back offers easy to control dynamics, so if you're in a hurry it might be a winner.

You could also look at active clamp flyback. It needs a few more components but offers soft switching.

However, soft switching schemes often only work under certain conditions (like within a fixed voltage range, and above a minimum current) so you need to be critical.

Also, remember to consider the startup behaviour when your output voltage is zero. Some converters have trouble with that.

[daqq]

Thank you for the replies guys! I'll investigate the suggested types.

You do not say anything about the load.

My apologies. I have hinted at it though  The power supply is supposed to charge a capacitor bank that is periodically partially discharged. Between the pulses (2ms spacing, less than 0.5J consumed per pulse) the converter needs to recharge the bank back up to the target voltage. That's why the flyback converter is so attractive - it's pretty much a current source.

Flyback would want to be multiphase, and you'll still be spending a bit extra on caps.  If you're more concerned about development time than production cost, this would be fine, too.

That's what I'm doing now - three units working in parallel. I'm OK with some extra cost :-)

I'd investigate where the losses are generated. Something getting hot doesn't mean it is dissipating a lot.

That's a tricky one. In a stable state the ferrite core (outer, visible bit) heats up to around 60°C, the copper part that's visible stabilizes at 80°C. As such I'm assuming that it's the winding mostly.

Using litze wire (less problem with skin effect) and as many windings as possible (lower flux swing) are methods to reduce losses in transformers. Another 'problem' with flyback is leakage inductance.

Yup, I'm using Litz wire. The leakage inductance is pretty low - also, the inductance would cause extra heating on the snubbers mostly, not in the core, would it not? As to the as many windings as possible - I'll investigate that. Didn't know about that one.

I've used iron-powder toroid cores for flybacks. They work, because such cores have a distributed air gap.

Their disadvantage? The core losses are significantly higher than ferrites. Maximum switching frequency must be definitively be kept well below 100 Khz. And/or their maximum flux swing restricted.

The company Magnetics, Inc produces a line of Kool Mu toroid cores with core losses comparable to ferrites:
https://www.mag-inc.com/Products/Powder-Cores/Kool-Mu-Cores

Good to know, thanks! Does anyone else make them? I've looked through the offerings of EPCOS and Ferroxcube, they don't seem to have that kind of material used on torroids.

[schmitt trigger]

The Kool-mu is a proprietary material, which as far as I know, it is only made by Magnetics Inc.

Which is a shame, because it is such an useful magnetic material.

Worst, I checked the distributor list, and they only have offices in US and China, and not in Europe!!! What an oversight!

[Wolfram]

Kool-Mu is a trademark but there are several other alloy powder core materials on the market from other manufacturers. Sendust (also trademarked AFAIK) is pretty much the same. MPP is less lossy, but it is significantly denser and more expensive.

[David Hess]

The power supply is supposed to charge a capacitor bank that is periodically partially discharged. Between the pulses (2ms spacing, less than 0.5J consumed per pulse) the converter needs to recharge the bank back up to the target voltage. That's why the flyback converter is so attractive - it's pretty much a current source.

It is not the only possible high compliance current source though.  I am not as comfortable as others in this forum with high power switching converters and I am not fond of flyback converters so instead of doing it in one shot, I would drive a push-pull inverter with a buck converter.  This way the transformer is not used for energy storage and the power transistors all operate at lower voltages.  The buck converter produces a continuous current output and also enforces balance in the transformer.  Some old low noise switching power supplies operate this way.

[phenol]

I've done that couple of times boosting 12v to 440VDC (~400W) and 12V to 170V (50W)using the current-fed push-pull topology. There is a synchronous buck stage feeding a PP stage via an inductor. The PP stage needs to operate with slight overlap of the switches. The good thing about this topology as i see it (apart from what David already mentioned) is that there is no output inductor on the high voltage side normally required in regulated voltage-fed PP. Therefore no ringing and overvoltage stress on the rectifier diodes.

LM5041 is a neat IC for the purpose. There's maybe another similar TI chip, UC something...

[ocset]

I  think what you should do is a simple boost converter with a variable output voltage......then a LLC converter. LLC's like high voltage and handle it well with low loss. However, LLC is only at its sweetspot at one vin/vout.....so use your boost to  be variable output to always give the llc its sweetspot vin/vout.
You can get llc drivers with fets and control all integrated on the same chip...i think fairchild do them. do a half bridge llc.

i bet with flybacks you are reflecting a lot of voltage back to primary, and then getting big switching losses on your fet. Unless you are in dcm you will also get big reverse recovery losses.....something you avoid with the llc.

As you know, the LLC will be curent regulated, as its a charger. The boost would be vout regulated.
.....................................................................
sorry but you cant really use torroids for flybacks as you need to interleave wind the primary and secondary.
You need a bobbin for this

[jbb]

I've done that couple of times boosting 12v to 440VDC (~400W) and 12V to 170V (50W)using the current-fed push-pull topology...

LM5041 is a neat IC for the purpose.

I'm sold.  This type of topology should handle ugly output loads (e.g. a large capacitor bank which suddenly drops from 400V to 200V).  Even if the output is shorted suddenly, currents are limited by the buck stage inductor.  One thing to be aware of is that low transformer leakage inductance is required to manage the voltage spikes in the push-pull switches.  So remember to add some when you simulate.

You could convert the push-pull stage to a full-bridge stage but some more gate drivers would be needed.  For a 60V peak input (48V lead acid on balancing charge), your push-pull FETs would be subject to 2*60 = 120V DC + switching spikes.  So some 150V MOSFETs with avalanche capability and modest snubbers would probably fit the bill.

[phenol]

Speaking of leakage inductance, i used copper foil for the primaries and Lleak was in the <60nH range, but since it was a one-off project, i made a regenerative clamp even if a simple rc snubber would have sufficed.

[prasimix]

Maybe you should take a look at DIC (Dual inductor converter) topology advantages if you are going to use current-fed converter based on LM5041 or UC3827-1

[ocset]

I must admit i  disfavour pushpull, there is always some leakage inductance there which will need snubbing, and also the magnetising inductance of the core will need resetting. Prefer the 2 tran forward or full bridge as fullbridge resets itself, and leakage in the txfmr actually helps and acts as a turn on snubber....in the 2 tran forward, well, the great SMPS writer , Ray Ridley, rates it as the most robust of the isolated converters
-of course, a fullbridge is just two interleaved 2transistorforward converters.

Upper gate drive should be easy with your 48vin.......also, your fets can be say 80v rated and you are fine...maybe even 60v fets..really low rdson.....i woudl go fullbridge, with some charge pump drivers, no pulse txfmrs.

And please dont listen to the old myth (not from this forum) that charge pump hi side type drivers cant be used with the 2transistorforwardconverter in light load....they can be used in no load and its fine with 2TF conv.