reducing inverter idle current

[joestue]

Thread here:
http://www.dutchforce.com/~eforum/index.php?showtopic=42471

I managed to cut the idle current down from 1.45 to .96 amps by adding a ferrite cored inductor in series with the 8 khz inverter.
the inverter is a 1400 watt APC ups.

I suspect most of the heat is the iron loss now, the transformer is always sitting about 10C above ambient.

[boB]

Yep.  That's the way to do it.

So, that dutchforce forum looks like some good stuff.

Where are you ?  You got some great stuff on your site.

Your prototyping methods looks kinda like mine used to be.

[oztules]

Joe,
I think if you look at your numbers, the 12-17uh is irrelevant in a lot of ways at that frequency.
Some of what you say I agree with, and it does attenuate some of the 8khz carrier.

I think it has a lot more to do with fet switching losses. The tiny inductance allows the switch to come on  a fraction before the current hits.... and the current spike was caused by the distributed capacitance of the secondary..... ie a really just a turn on snubber.

Further evidence to this is the type and style of transformer makes a huge difference to the outcome of any input inductance added.... ie a torroid has a 6 fold effect with the 15uh ferrite inductor, where as , poor quality high leakage transformers have almost no effect, quality EI transfromers have an effect better than I see in yours, but not much.

A torroid is definitely the way to go. My 60 lbs core home wound torroid has only 500ma magnetising current with only 15uh inductor This is a 48v 240v transfromer.
To further lower idle, I used more secondary and primary turns to keep the magnetising current down as well... so I wound the secondary as 280v, and the primary at 8:1 from that. Without the inductor it is closer to 4amps idle current.

This brings the idle power down dramatically from over 5kwh/day  to just over  1/2kwh/day.

The inverter is useful for 10kw for 10 mins, and 6kw continuous, with surges over 24000w.... and it draws less magnetizing current than yours, so there is room for improvement.

This is prototype 1 with 44lbs transformer and 50mmsq primary, the later one has 70mmsq primary, and over 60 lbs of core.


The box complete


Total cost about $400.
....................oztules

[joestue]

The following photos are voltage measured across a 16.7 milliohm shunt. 4 terminal type, reasonably low inductance i suppose, it is in series with the 13vac side of the transformer. I was able to dig into the switching waveforms and see the current spike charge up the parasitics, but I don't believe its significant.
In any case, compare with and without the inductor.
note the time base changes.



oh, and there is a dc current on the order of 2-3 amps that varies periodically. these inverters have a 1:100 current transformer in series with the 20uF output capacitor, so i presume that it is measuring switching waveforms with that coil. there does not appear to be any circuitry to measure current flowing through the fets.

two more images, should be self explanatory.
with an inductor, there are about 160 switching cycles per 8 milliseconds.
without an inductor there might be 10 times that, i'm not certain, don't really want to count them.



with and without inductor at 750 watts.

[oztules]

So, what do you think the real reason that such a small inductance makes such a big difference Joe.
I have seen up to 8 times lower power drain with the right transformers, and almost no change with lossy transformers.

Those transformers with lots of leakage do very badly, those that are tight and have good laminates do far better.... without the inductor, they are about the same.... add inductor, and the low leakage ones shine.... does that help at all?

..............oztules

[joestue]

Eddy current losses in the lamination stack is the primary loss. hysteresis loss is the second.
Eddy current follows flux squared and frequency squared.
It is clear that adding the external inductor reduces the rms current, but in my case, the switching frequency is decreased, which increases the rms ripple current.
If your inverter keeps the switching frequency the same, then you would see a bigger difference.


The second issue is the core is saturated with a dc bias.
The fets in my H bridge have an output impedance of 8 milliohms i think.
The transformer primary has a resistance of 5 milliohms, the 10 awg wires add another 2 milliohms.
thus the 2-3 amps of dc that i'm measuring, while a 16.7 milliohm resistor is added in series, is formed by a voltage of about 60millivolts appearing across the combined resistance of 32 milliohms. if i wind a custom high VA toroid, replace the fets with modern fets of 1/10th the resistance, that 2-3 amps of dc could easily increase 10 fold, which would completely saturate a toroid, which has very low leakage inductance and nearly zero ability to handle dc current biasing the transformer.

thus its possible that the dc resistance of the external inductor helps out.
in my case, the external inductor adds maybe 2 milliohms.

[joestue]

EDIT:

I think its a tie between high frequency ripple current conduction losses in the dc resistance of the H bridge + primary, and the losses associated with eddy currents and hysteresis loss in the leakage inductance of the transformer.

Take a look at those middle two photos showing the nasty 120hz waveform at 2.5ms/div.
100mv hf ripple means 5.8 amps of peak to peak high frequency ripple with an inductor, without an inductor we see 300mv of ripple which is 17-18 amps of hf ripple..
there is more hf ripple than there is 60hz exciton current without an inductor, and that high frequency ripple flows through the secondary and the output capacitor, causing copper losses in both primary and secondary
60hz excition current doesn't cause copper losses in the 120v winding, it stays on the primary side, thus any attempt to reduce hf ripple will reduce copper losses in both coils.

[joestue]

so i should explain those last two photos better.

the 500mv per division is the voltage dropped across the 16.7 milliohm shunt.
the peak voltage dropped across the shunt is about 1.25 volts, or 74 amps.
and we note that the core, is still saturated by the dc bias, as evidenced by the asymetrical switching cycle.

also, the oscope shot without an inductor, isn't really without an inductor, the jumpers i'm using to short out the inductor have a resistance of a few milliohms. so it is probably much worse than that.

[oztules]

Only picture of my scope on this project. was about 1.5kw or 30 amps rms into the primary with inductor.

Now the camera is not talking to the new linux... lacking the back end it claims.... which is crap.... will solve soon...

My Freq seems not to change, will monitor it closer over then next few days if I get time to get all the gear together at the same time and place.


John

[oztules]

Couldn't wait... so went out and tested a new board and transformer.

Same result/s. Without small inductance 1.1a, with small inductance...  .45a

There is no changing in frequency like yours, there is no noticeable change in any parts of the wave form.... except, if we go to 1us/div we see a ringing lasting for about 1us.
It has about 8-9 wobbles before it zero's out.
When I short the inductor out, this ringing has the same number of wobbles, but about half the magnitude, and the current drops by over 1/2 an amp... that looks to be the culprit.

There is no other discernible change to any other  parts of the wave that I can ascertain, at any timing I can get up to with this scope.


Knowing this now, it is time to see if I can find where the ringing is coming from, and maybe stop it at the source perhaps.
But it explains the things i am seeing... very small inductor, and very big changes to current draw... sometimes up to 8 or more times. Tonights was only twice.
Will now change transformers and see the difference/s, as they are very very varied from the looks of it.

Tonights test was on  small 1.5kw torroid from a  name brand inverter....(because I could at least carry it) very average transformer if I was to comment on the big name brands... and poor performance for a lousy 4kg core, I can get those figures with 60 lbs of core. not impressed.... but it did for the first test....
Now it has peaked my attention, will have to hump out the big trannies, and see why/ where is ... the difference... it's huge.

Looking at this, I don't currently buy your assertion for this set up at least. The big cores seem to get as good a figures as the little ones... so not hysteresis or iron loss, as six times the core can use the same idle current as a small core.... this I don't understand at this point... at all.

Looks like switching losses are mitigated, and nothing else.



...............oztules

[joestue]

i might agree with you, that hysteresis loss is negligible.
however, not for my transformer. its probably wasting 5 watts per kilogram at 60 hz sine wave excition, a good toroid should be less than 1 watt per kilogram.
i'll have to hook my analog watt meter up to the grid and see what it measures.

I will have to spend some time with small toroidal cores to isolate the various issues.. also, i'm going to hook a capacitor in series to kill the dc component. however, i don't have any large enough capacitors to perform any significant load tests.

oh, for what its worth, i had 7 turns of 10 awg wire looped through no less than 20 of those T-106 mix 26 toroids. i added that inductor in series with the ferrite cored inductor i have installed. idle current increased from .96 amps to 1.04 amps. (because those cores are very lossy)
the thought occurs to me that i can take the 20uf capacitor off the output, and loop it through the current transformer twice. this should fool the inverter into thinking ripple current is much higher than it is, and the switching frequency will increase.

[oztules]

More testing this morning.... now I don't know what to think.

Here is the plot across the transformer without the inductor.


And here is the input to the transformer with the inductor:


Now what to think??


......oztules

[joestue]

I cooked up a ~35,000 uf capacitor in series with the transformer primary and all manner of nonsense started happening. Evidently, it does have a way of zeroing the dc current but whatever method its using, it isn't zeroed properly.
So that capacitor which should have had a negligible 1vac at 60hz across it, ended up with even double digits of dc voltage across the cap, combined with sub harmonic oscillations, on the order of .5 to 2 hz. fixing a 100mohm resistor across the 35mF capacitor removed the sub harmonic oscillation but the 120 and 180 hz harmonics remained, and the capacitor had on the order of 100mv dc across it.
Either I need a much larger capacitor or inserting a capacitor in series with the transformer fundamentally breaks something.
A one farad capacitor would be ideal, but i'm not going to spend the money on it as it would need to handle 100 amps of ripple current.

In the hour i was making these measurements, the transformer had warmed up to its usual no load idle temperature. Seems further attempts to decrease idle current are offset by dc losses in the primary circuit, which indicates that the most significant loss at this time is 60hz iron losses.

i wonder if i can replace the ac current transformer in series with the 20uf output capacitor with something dc coupled on the primary side.
i also suspect the 20uF cap can be replaced with 5uF, which would remove ~90VA of circulating current

replacing the transformer with a toroid will magnify the saturation problem caused by dc current, so i'm not going to do that. but i do have a suitable core from a burned out 500VA variac.

[oztules]

"replacing the transformer with a toroid will magnify the saturation problem caused by dc current, so i'm not going to do that. "

I really think you should try the torroid. Every time i have, it woks better than anything else by about 1/2  or lots more.... Dont know why still, but I would try that and reduce the 20uf to about 4uf.... or less with an output inductor.

boB probably knows why.


.................oztules

[mab]

I'm watching this thread with great interest as I'm using a 1400va 24v APC as my primary inverter and would like to reduce the idle current.

I have to admit that when I got the inverter several years ago (0.99p off that well known auction site) that I assumed it was MSW (big Tranny)); it was quite recently that I looked at the output and found out my error.

[boB]

oops
I thought there was a delete function.

[boB]

The saturation from one transformer to another is not necessarily an apples to apples comparison.

Unless the area of the iron and type of iron and wire size and number of turns and all that are equal.

A toroid ~should~ have higher permeability ((higher inductance compared to E-I laminations) given everything else equal ...   But there are so many variables it's not always easy to tell what is going to be better.

Idle losses (50/60Hz), copper losses and size are certainly compromises.  High power inverters are usually
balanced towards higher proportion of copper losses at high power.  The higher ripple current and core
losses at idle are certainly a problem if the transformer is small enough to have good copper losses at
full power (shorter mean turn wire length) and the 50/60 Hz core and ripple losses can be made better
by a larger transformer and more turns but then you have higher copper losses at full output power.

That's where your ferrite helps at least with the high frequency switching currents.
It's still kinda early for me.  I hope I have had enough coffee to say that right.

This is all what makes this stuff fun and mysterious to most.

[joestue]

Changing the capacitor doesn't seem to have much effect on frequency.
Reducing it from 20 to 8 uF but looping it through the current transformer three times reduced idle current from 1 amp to .93 amps.
reducing it to 1 uF and looping it through the current transformer 4 times didn't further reduce the idle current much, perhaps down to .91 amps. but at only 1uF the main carrier started showing up on the output. so i left it at 8 uf.

and then promptly blew it up.

i think i shorted one side of that capacitor to the chassis. saw a flash of light near the fan, but i haven't found any bad components.
it appears one side of the H bridge is dead. fets are good though, so the fault is somewhere between the uproc and the fets.

i'm going to continue these experiments with a VFD however.

[joestue]

more info on the transformer:
high side winding is as follows
 
white is ground, the voltages are relative to each other. seconds is the amount of time to make one turn of the wattmeter.
blue    --125 --142  --164   --125 vac to this winding=42 seconds
yellow--114 --125  --145   --125 vac to this winding =32 seconds
black  --97   -- 109 --125   --125 vac to this winding=23 seconds
white  --0     -- 0v    --00

20 watt lightbulb = 25.6 seconds. = 126.6 vac x .173 amps = 21.9 watts.


125 to the blue wire makes 13.68 volts on the other end, and pulls .266 amps from the line.
125 to the 114 volt line makes 15.15 volts on the other end, and pulls .466 amps from the line.
125vac to the black line makes 17.4 volts on the other end, and pulls .96 amps from the line.
this is the mode during normal operation.

so, allegedly what we see here is that during normal operation, the transformer pulls 17.6 watts from the line during nominal rated flux.
13.4 watts if we drop the flux down to 89% of nominal.

[boB]

.......and then promptly blew it up.

.....
i think i shorted one side of that capacitor to the chassis. saw a flash of light near the fan, but i haven't found any bad components.

This is called "STS" ...

Something Touched Something.  Happens to me all the time.

[oztules]

"so, allegedly what we see here is that during normal operation, the transformer pulls 17.6 watts from the line during nominal rated flux.
13.4 watts if we drop the flux down to 89% of nominal. "

If you look at reply #2, I mention this as a way to pull down magnetizing current. I use both extra turns on the primary and secondary, and the inductor.

I have not found a better combination thus far, but certainly lowering the operating flux makes the difference you would expect..... the inductor I stumbled on makes a massive difference...... but DEPENDS on the transformer type as to how much.

So I know it works, and works splendidly, but I still don't exactly know why.

Both boB and I have alluded to the switching losses, but why this is dependent on the transformer leakage I can only assume as the capacitive distribution of the secondary being ameliorated by the pre snubbing effect of the inductor..... after about 30uh, the advantage seems to drop off, and more even brings the current back up.

Those transformers that have high reactance, cant get the benefit of the inductor. The less leakage, the better the result......... ie try with a torroid .

If I knew what I was talking about.... I'd be dangerous



............oztules

[joestue]

yes, i know reducing the flux reduces no load losses. but at the expense of more copper losses.
I don't have that option without making a new transformer, or adding a second transformer in series.. which "could" be an option.

i think the more important observation is most of the no load loss is the iron core. I know that the circuit board draws about .134 amps when its "on" even when not switching.
so that's about 3.5 watts. 17.6 watts at 125 volts, 13.4 at 114 vac sine wave means the core is probably wasting 16 watts at 120vac, the output voltage of the inverter.
this makes a total of 19-20 watts. just 4-5 watts shy of the total, 24 watts, so that's not bad. i'm pretty sure 90% of that 5 watt discrepancy is in the transformer..
but i'm not about to rig up a calorimeter. 


In the past I have tried to take the core out (non welded core) and put it back in, after modifying the windings, but i can never get but 98% of the core back in, and you have to destroy at least one or three of the "E" pieces in order to the core out of the bobbin.  the few times I did try this, magnetizing loss went up due to additional eddy currents from broken lamination insulation.

[oztules]

"i think the more important observation is most of the no load loss is the iron core."

No....That is  not quite what I have found.
With a torroid, I can get a 200w to 20 watts drop, and they are big ones.... thats 10:1

Without the inductor, the figures are about the same ... ie an EI transformer with high leakage, and a torroid with the same ratings, will be about the same 200 watts loss at idle. These are 40 pounders or more.

The inductor will lower the toroids idle by 10:1 in this case, it will not lower the EI by more than 2:1 for good cores, and virtually NO difference for poorer cores.

So I figure it is the leakage making the difference in how the core will react with the inductor... not the mass of the core...lowering the flux by over winding is a given.. so can be ignored for this case...... although with big torroids, I just rewind with bigger copper and so the copper loss does not go up for the extra turns..... and we are talking 3v/turn or more@50hz, so it does not take much extra length to add a few volts here and there.

If thats true, and the leakage is the key... then what would that effect, and why only 18uf or so makes so much  difference, not 2.7mh ( tried that too ), which really would have squashes a bit of the HF from seeing the cores.

Thats why I think the switching losses are being altered, and that they are altered to the extent that they can help tight transformers, and make little or no difference to poorer EI transformers..... once mine gets down to 20 watts or so, I think it is iron losses from the massive core/s.

Would be interested in boBs take on this .... but thats what I see as the key to low idle  losses..... otherwise use a small transformer until the power rises and switch in more in parallel to handle the loads...... BUT this won't help overall efficiency with higher power.

The headline idle losses are lower, but switching losses will occur when the second bigger transformers would start to conduct.

Using a  good big one gives efficiency across the whole range.

............oztules

EDIT: from post #9.. this:
"
There is no changing in frequency like yours, there is no noticeable change in any parts of the wave form.... except, if we go to 1us/div we see a ringing lasting for about 1us.
It has about 8-9 wobbles before it zero's out.
When I short the inductor out, this ringing has the same number of wobbles, but about half the magnitude, and the current drops by over 1/2 an amp... that looks to be the culprit.
"

should read nmore like this "
There is no changing in frequency like yours, there is no noticeable change in any parts of the wave form.... except, if we go to 1us/div we see a ringing lasting for about 1us.
It has about 8-9 wobbles before it zero's out.
When I short the inductor out, this ringing has the same number of wobbles, but twice the magnitude, and the current increases by over 1/2 an amp... that looks to be the culprit.
"

[joestue]

test setup 1:
original transformer, with 8 uf capacitor across white and yellow wire.
mitsubishi vfd outputting 120vac at 60 hz at 10 khz carrier. this is dumped into a 120:15vac autotransformer. auto transformer feeds ups transformer low side.
8uf cap has 128 vac across it according to my oscope.

watt meter is measuring the power drawn from mains into the vfd. the same inductor used in the ups experiments is on the main transformer.

--19.5 seconds, with inductor.
--14 seconds, without inductor.

It did not take long for the auto transformer to warm up. the copper warmed up more than the core did. the 120vac winding is about 1.8 ohms, the auto transformer was originally a buck/boost from a msw ups. 1 amp of 60hz ripple current is only 2 watts,

120:15vac auto transformer excited from 125vac mains: 64 seconds.
Auto transformer excited from mains, 15vac feeding ups transformer with 8 uf output capacitor. 123vac on output capacitor: 18.4 seconds.
I attribute this discrepancy, 18.4 seconds is less than 19.5, to two things
19.5 seconds includes the vfd losses, however, the watt meter may be insensitive to the poor power factor drawn by the vfd.
Inaccurate vrms figure from my oscope due to $#|+ty firmware.
thus the actual voltage on the output was probably less than 120vac, maybe as low as 110vac rms.

further testing:

280v variac excited from the vfd at 120vac with wiper at 280 volt position. 106 seconds.
280v variac excited from the vfd with the wiper at 140 volt position. 53 seconds.
280 volt variac excited from vfd at 140vac postion, with 20uf capacitor connected from wiper to free end of the variac. 19.5 seconds.
280volt variac excited from 250vac mains at 280volt position, 280 seconds