Design a better inverter- win a million bucks!

[T3sl4co1l]

More like 1/4 the volume with the cap I highlighted... or in other words, each one is about a cubic inch (square prism outline volume).

Challenge is to do a PFC in less.  You still need electrolytics in there, period, no choice on that: absolutely nothing else offers the energy density and ripple current ratings required.  A monstrous chain of supercaps, forget about it.  Film, you're out of your mind.  And if you use too few, you won't even have that, just huge plumes of magic smoke.  I wouldn't think you could get away with less than three, so you need to pack that PFC (sans output filter cap) into less than 7 in^3.  That's still a tall order.

And really, why would you go out of your way to choose bigger caps?  It only has to run 100 hours, beat them up!

Tim

[johansen]

Challenge is to do a PFC in less.  You still need electrolytics in there, period, no choice on that: absolutely nothing else offers the energy density and ripple current ratings required.  A monstrous chain of supercaps, forget about it.  Film, you're out of your mind.  And if you use too few, you won't even have that, just huge plumes of magic smoke.  I wouldn't think you could get away with less than three, so you need to pack that PFC (sans output filter cap) into less than 7 in^3.  That's still a tall order.

Suppose you boost the 450 volts up to 550 volts.
you can let the dc voltage on that 550 volt rail vary from 550 to 450 volts.
yes, that's a hell of a ripple but you can get away with a lot less stored energy.
difficult part would be finding capacitors that can handle that ripple.

another alternative is bucking the input down from 450 volts to 350 volts. suppose you use 450 volt rated electrolytics, and you let the voltage on that rail swing from 350 to 450 volts.
same ripple, but larger capacitors needed. but you can use 450volt rated capacitors, instead of trying to find 550 volt or 600 volt caps.

also, the output mosfets can be 500 volt fets instead of the 600 volt fets required for the boost topology.

problem with buck is the input high frequency ripple. CCM boost makes more sense in that regard.

[T3sl4co1l]

Although they don't seem to have any constraints over input ripple, so you wouldn't need much bypass there.  Maybe even do something shitty and RLC-quasi-resonant, using the 10 ohm source impedance to damp it.

Tim

[Riotpack]

Only guessing here but wouldn't 95% efficiency be hard to achieve? Could you use a topology based on a class D amplifier with a MOSFET H bridge to produce a sine wave to increase efficiency?

[T3sl4co1l]

That's pretty much the idea, but when you're taking margins of single watts, every single thing counts.  Example: a Qg = 100nC transistor takes about 1uJ to switch on and off; done 500,000 times per second over 4 transistors, that's 2W.  Not much out of a 100W maximum budget, but every single component in the power or control path adds up quickly.  You'll find 25 of those "nag" sources very easily.  Add in the power transistors themselves (switching and conduction loss), filter chokes and so on, and you can blow the whole thing no problem.

If you aren't going for forced air cooling, your limit is even more critical, under 30W or thereabouts.

Tim

[David Hess]

Only guessing here but wouldn't 95% efficiency be hard to achieve? Could you use a topology based on a class D amplifier with a MOSFET H bridge to produce a sine wave to increase efficiency?

I can think of existing designs which are around 93% efficient which include an active power factor correction stage before the inverter so 95% does not seem unreasonable if cost is not a factor.  They are much larger for the same power level though.

[Riotpack]

So for such a small footprint I guess you would have to use heatpipes to help pull the heat to the top of the unit and to atmosphere. My laptop CPU+GPU draws around 100W full load and gets to 80 degrees at around 25 ambient so it would be possible to get that down a bit. For the size and the 60 degree limit, I would estimate a fan would be required. Otherwise a convection cooled setup using heat sinks integrated into the case design might just do it. The Sunny Boy 3000TL is 97% efficent and uses convenction cooling for a 3kVA unit, though the size is the limiting factor for that unit.

[David Hess]

So I guess you would have to use heatpipes to help pull the heat to the top of the unit and to atmosphere. My laptop CPU+GPU draws around 100W full load and gets to 80 degrees at around 25 ambient so it would be possible to get that down a bit.

That is what I would do.  There is a major problem in this case because power is lost in very specific areas like the power transistors and it needs to be spread evenly over the whole surface of the enclosure to prevent hot spots because of the stated requirements.  Heat pipes would be more effective then an equivalent thickness of aluminum.

The alternative would be forced air cooling which is not something you find in similarly high density switching power supplies because of poor reliability and operating life.

[Riotpack]

There is a teardown of the sunnyboy model here - It even lists ICs used and shows the topology.

http://www.edn.com/design/power-management/4368876/Teardown-The-power-inverter--from-sunlight-to-power-grid

[tszaboo]

220V at 60Hz?
South Korea, Philippines, Peru and a bunch of island states.
Not exactly "nowhere in the world", but a very peculiar choice never the less.

Also the US.  US residential supplies are 240 VAC.  It is a split phase configuration and normal outlets only use one phase to get 120 VAC, but the  backup generators and PV inverters normally produce 240 V.  High power outlets used for things like electric ovens, clothes dryers, and electric cars use one of the various 240 VAC outlets.

Industrial settings don't have split phase transformers they normally have 3 phase feeds.  In that case, you normally get the 208 V phase-phase voltage on the same outlets used for residential 240V.

220 is not 240. But yes, I always forget the weird 180 degrees 2 phase configuration of the USA. I was already surprised, that they didnt went for 110V as the "remote locations of the world", now I understand.

[HackedFridgeMagnet]

Suppose you boost the 450 volts up to 550 volts.
you can let the dc voltage on that 550 volt rail vary from 550 to 450 volts.
yes, that's a hell of a ripple but you can get away with a lot less stored energy.
difficult part would be finding capacitors that can handle that ripple.

I assume you mean charge instead of energy?

   

Has anyone done some math on what kind of input capacitance you need to reach the input ripple?

Don't know why they give different figures for voltage (20%) and current (3%), but clearly the 3% figure is the problem.

Typo, It says 20% input current and 3% input voltage.

I cant see how the 20% figure comes into it either. Clearly 3% is the only problem because of the 10ohm source resistance.

I did a bit of a ball park to see how much energy needs to be stored during each half cycle to manage input ripple and provide energy for the half cycle.
8.33 joules. (very rubbery figure.)

If your pfc charged it's caps up to 400v and let them run down to 200v in the cycle. (Made these numbers up, we need to cope with 0.7 p.f. load.).

That would give 2x8.33(400^2-200^2)  = 140uF.

Just on the input.

This is vastly different to T3sl4co1l's calculations, I guess it is the 8.33 Joules figure. But I am not sure how your calcs work T3sl4co1l. Not that I have explained mine.

3% of 450V is 13.5V, and that's peak to peak over the low frequency content (excluding HF noise).  An average 2kW load goes from 0-4kW during the cycle, and varies according to the sin^2 t = (1+sin(2*t))/2 identity.  Therefore, the AC component of the current draw is sinusoidal, twice the frequency of the output, and peak-to-peak equals the (0-4kW) / (450V) spread, or 8.9A (it's actually a distorted sine, more current at lower voltages due to supply droop).  Let's say it's 10A.  Worst case frequency is 50Hz, or a ripple of 100Hz.  The reactance must be less than 13.5Vpp / 10App = 1.35 ohms, or 1179uF.

[ejeffrey]

220 is not 240.

Well, originally the US was more around 110/220 (with regional variations) and it is still quite common to call it that, even though the standard is now 120/240.  It is also within (although near the bottom) of the +/- 10% commonly accepted band.  220 V is also relatively close to the industrial 208, so it is a good compromise voltage.  No equipment designed to operate on either 208 or 240 VAC should have a problem with 220.

[T3sl4co1l]

Typo, It says 20% input current and 3% input voltage.

D'oh!

That would give 2x8.33(400^2-200^2)  = 140uF.

Just on the input.

This is vastly different to T3sl4co1l's calculations, I guess it is the 8.33 Joules figure. But I am not sure how your calcs work T3sl4co1l. Not that I have explained mine.

Sounds reasonable, given the vast difference in ripple.  You can calculate it either way, though only one method is easiest for each case.  At low ripple, one can assume nice clean sine waves, and consider a single frequency in the AC steady state.  For high ripple, the current will peak sharply (half voltage --> double the peak current draw and dV/dt -- it won't be a sine wave anymore), and an energy argument works much more nicely.

Interesting to note: your calculation uses 14 times more ripple voltage, but only 8.4 times less capacitance: this is correct, because capacitors store less energy at low voltages (indeed, going from 400V to 200V depletes 75% of the contained energy, and attempting to go lower wouldn't gain you very much), so by using a lower average voltage, you need relatively more (evidently, about double) capacitance.

Still, finding a cap that'll tolerate 50% ripple for over 100 hours, that fits in less than 10 in^3?  That's a toughie.

Tim

[HackedFridgeMagnet]

Thanks teslacoil, I was just wondering how my number was so different to yours.

I was envisioning a two stage converter-inverter with some storage caps in between.

This could give a steady current draw from the Source through a converter into the caps. This is to cope with the 3% input ripple.
This might be the equivalent of what somebody referred to as a PFC earlier.

Then you have the inverter to pump it out of these caps to the load at an instantaneous power of somewhere between 0 and 4 kWatts.
This will give me the 50% ripple on the caps. Roughly 400v down to 200v.
400 volts because on average you will never see more than 400 volts in due to the series resistance.

If only I had the expertise to make it.

[eneuro]

I think the world needs cheap and efficient inverters more than it needs tiny ones.

Many high volume manufactures needs tiny one, while they are very dificult to repair without sophisticated equipment and it will cost more than buying new, so they need small ones while BOM should be lower which means higher profits 

So, If you want something cheap and efficient DIY it and in the case of failure simply resolder failed element 

[Marco]

A lot of the size and cost is in the passives ... tiny is cheap.

[Monkeh]

So, If you want something cheap and efficient DIY it and in the case of failure simply resolder failed element 

Because we all know with high power electronics, one part just stops working and there are no consequences.

[tom66]

I had a Panasonic sustain board blow one transistor. This then proceeded to take 16 others out with it plus their driver circuits. 

[xygor]

Instead of a heat pipe I was thinking along these lines.  Assuming a H-bridge PWM topology, each leg consisting of two transistors and an inductor, use a whole bunch of those in parallel.  The parts can then more easily be placed wherever you you would prefer to deal with the heat.  The finer granularity makes it easier to find a place to fit everything.  If the parallel transistors' on and off and off times are staggered slightly, it might help with meeting the radiated emissions requirement.  Or switch a different set of transistors on each cycle (either the 60Hz or the modulation cycle) to more finely tune the heat distribution.

[T3sl4co1l]

Ya know, 600 of these things http://www.digikey.com/product-detail/en/C5750C0G2J104J280KC/445-13462-2-ND/3950698 stores 11.9J at rated voltage... in only 3.2 in^3... in an impossible assembly, of course.

Good luck with that, but an interesting observation nonetheless...

Tim

[Phoenix]

I think some of you might be missing the point of the challenge: you don't get the $1M if you meet the specs, you only win if you are the smallest! Even if meeting the maximum size requirement of 40"sq isn't THAT hard, if someone makes one 39"sq they would win. You need to build it as small as you can and hope nobody builds one any smaller (smaller is baller). I don't want to go into too much details about the challenges associated with this project, where I think you should be looking, as I'm part of a small team of power electronics people taking a serious crack at this (none of us are in it for the money and we're a completely independent team, we are simply driven by building something cool - it's fun).

BTW it is the current requirement that is the 120Hz ripple figure you need to meet. At full power (5A, 400V) you're allowed 1A of ripple with 12V ripple. At lower power say 200W (0.448A, 445.5V) you're allowed 89.7mA ripple but a whopping 13.5V ripple. If you want to use bulk DC capacitance on the bus the worst case scenario is not 100% load, you actually need more DC capacitance to meet the current ripple requirements at lower loads given the 10ohm input impedence.

Some of the topology ideas here are interesting, but that's really only the first step of a long journey. Passives/filter requirements, switching frequency, transistor selection, gate drives, auxillary circuits, cooling, packaging, EMI/EMC, feedback controllers, software... the list goes on. And all the of these factors are related to each other - a huge multivariable problem, how do you optimise it?

It's not called a challenge because it's easy to meet the minimum specifications (I'm not saying it is). It's a challenge because it's hard to build the smallest unit that meets the specs.

[Phoenix]

If anyone is still interested in this (I sure am), here is the approach taken by PES researcg group (Kolar et. al. a large/very good power electronics research group in Germany). Quite impressive, huge amount of work. (Note the use of not yet commercially available GaN devices...)

https://www.pes.ee.ethz.ch/uploads/tx_ethpublications/Keynote_Presentation_ITELEC_15_FINALFINAL_as_published_251015.pdf

Ya know, 600 of these things http://www.digikey.com/product-detail/en/C5750C0G2J104J280KC/445-13462-2-ND/3950698 stores 11.9J at rated voltage... in only 3.2 in^3... in an impossible assembly, of course.

Or 108 of these http://www.digikey.com.au/product-search/en?pv13=1529&k=ceralink&mnonly=0&newproducts=0&ColumnSort=0&page=1&quantity=0&ptm=0&fid=0&pageSize=25 in 3 in^3 (as the above team used).

Also, I'll throw up some photos of my teams unsucessful attempt (serious trouble with gate turn on due to miller capacitance and high dv/dt and no negative gate supply). My team was a team of 4, doing it independently in their free time (with lab support from where 2.5 of us work).

Did anyone else have a crack at it?

[Artlav]

So... They spent a million dollars building a $10000 dollar inverter with no practical use what-so-ever.
How is that better than the larger, but so-much-cheaper-you-can-actually-buy-it commodity inverters?

[Fungus]

So... They spent a million dollars building a $10000 dollar inverter with no practical use what-so-ever.
How is that better than the larger, but so-much-cheaper-you-can-actually-buy-it commodity inverters?

...and how exactly are the current "picnic cooler" sized inverters a big problem for household use? Even in a small mud hut you need things to sit on.

I guess you can't fit as many in a delivery truck as you could if they were 40 cubic inches, but... 

[tszaboo]

If anyone is still interested in this (I sure am), here is the approach taken by PES researcg group (Kolar et. al. a large/very good power electronics research group in Germany). Quite impressive, huge amount of work. (Note the use of not yet commercially available GaN devices...)

That is quite impressive usage of the 3D envelope.
I have the feeling that this will be a lot easyer challange when the GaN FETs will reach maturity. With DrMOS like GaN power stages. Probably is would benefit if there would be an ASIc with analog control. And there should be a way to replace the inductors and capacitors wit a hybrid inductor-capacitor.
OK, I'm hibernating myself, wake me up in 20 years.