[SOLVED] Pure sinewave inverter with toroidal transformer

[Orson Cart]

Any time you turn off a mosfet and the current has no-where to go - there will be a voltage spike, for an H bridge there is a path through the other fet (its diode) to the DC bus, therefore the inductance of these paths must be low and there must be plenty on C nearby on the bus (with low internal L) to soak up the current until it stops flowing. Resonant converters use this aspect for zero volt turn on of the next mosfet. If not an H-bridge or a half bridge, then clever snubbers are the only way to tame the volt spikes (and/or slower turn off times)...

 

[red_alert]

Thanks everyone, it really was just like that: after putting another bank of polypropylene pulse capacitors across DC bus and GND (right next to the high-side MOSFETs drains) the voltage spikes have disappeared - no more snubbers at all.

Now I encounter a different problem: when I try with a larger load (an 1500W hair dryer) the transformer start buzzing very load.

When I was using the incandescent bulb (100W) I was using a 2.2uF cap across 230V output for filtering; now it seems too large. I replaced it with a 0.47uF one and the buzzing sound has decreased (yet it's still there).

How to deal with this situation? Do I have to dynamically add/remove capacitors across output based on load current?

With a lighter load (that incandescent bulb) the filtering it's not effective with the 0.47uF capacitor.

The transformer primary (30V AC) inductance is 25mH, its leakage inductance is 3uH and the primary series choke inductor is around 12uH (4 x 47uH/50A toroid chokes in parallel).

 

[FvM]

Not sure if there's an actual problem. Transformers and inductors with iron core can be expected to generate some noise due to magnetostriction if it works with audible pwm frequency. Did you check the waveforms under load?

 

[red_alert]

The waveforms look OK. Do you think the PWM frequency (6.4kHz) it's just too low for this application (5kW rated power)?

I thought it not worth increasing the switching losses (by choosing a higher PWM frequency) as there's no ferrite core to take advantage of it anyway.

Can you tell me what's the switching frequency of the power inverters you're working with (10kW range you did mention few posts back)? Are they LF transformer type, too?

Thanks in advance for your help.

 

[Warpspeed]

The buzzing transformer problem seems to be highly variable, some inverters are much more prone to this than others.

Magnetostriction is definitely possible, but usually in an inverter application the flux levels are kept fairly low to minimize magnetizing current and zero load inverter input current.
If its buzzing at zero load, its highly likely magnetostriction.

On some of the alternative energy Forums, people have resorted to coating each winding layer with a thick coat of epoxy to anchor the turns, and that appears to stop most of the noise.
If it buzzes more as the load is increased, its much more likely to be the windings flapping about.

 

[FvM]

Building high power inverters involves often a trade-off between audible noise, inductor size and switching losses. In so far there's no simple answer.

Low voltage MOSFET inverters can be usually operated with 20 kHz and higher pwm frequency, but HV MOSFET and even more IGBT inverters often end up at 3 to 10 kHz switching frequency. In an industrial enviroment, a certain amount of audible noise is can be tolerated but is still unwanted.

I forgot to mention, that you may observe a problem of DC magnetization due to inverter asymmetry. Did you check the transformer primary current for DC components?

 

[red_alert]

At the time I've ordered the toroidal transformer, the manufacturer asked me if I want him to pour some sort of resin inside the transformer ("for mounting purpose only", he said).

I've choosed not to be filled, for better cooling capabilities. Now maybe that's the downside (the increased buzzing noise).

Are those magnetostriction vibrations producing any (significant) mechanical stress?

- - - Updated - - -

I forgot to mention, that you may observe a problem of DC magnetization due to inverter asymmetry. Did you check the transformer primary current for DC components?

Actually, I only have a current sensor (ACS758) on the +48V rail. The output waveform looks perfectly symmetrical though. Is there any other way to check for DC component?

 

[Warpspeed]

Magnetostriction is actually the metal crystals changing shape due to the strong magnetic field. Its only usually a problem when you really drive a steel core alternately pretty deeply into near saturation each way. Any unfortunate mechanical resonance of nearby panels or structure can make it sound a lot worse.

Epoxy is one cure, another is dropping the whole transformer into a drum and filling the drum with oil. That can hugely increase the cooling as well as quieten it right down. Its a bit messy and not an especially elegant solution, but entirely practical for a home project.

A question you asked earlier about using a relatively low switching frequency.
The only real disadvantage is the greater difficulty of filtering it out.
The more octaves you can place between your 50/60Hz output, and the switching frequency, the easier it will be to achieve higher attenuation rate with simple filtering.

Filtering is far from easy because the load impedance of an inverter cannot be defined in any meaningful way, and unloaded mismatched output filters tend to sometimes ring and create voltage regulation problems.
Damping resistors placed across any series inductors help, but that also reduces the high frequency attenuation.
In many respects, putting together a good efficient output filter is probably the biggest challenge with building a high power PWM inverter.

 

[FvM]

Actually, I only have a current sensor (ACS758) on the +48V rail. The output waveform looks perfectly symmetrical though. Is there any other way to check for DC component?

Seeing perfectly symmetrical waveforms is a strong hint against DC magnetization and core saturation. Just wanted to mention the possibility.

Impregnating windings or even fully molding the transformer reduces audible noise by a certain extent, but never completely.

 

[Orson Cart]

Check whether the hair dryer is on half power with an internal diode in series with the heating elements, this draws half cycle currents only from your inverter and can cause problems, including the noise you mention....

 

[red_alert]

Check whether the hair dryer is on half power with an internal diode in series with the heating elements, this draws half cycle currents only from your inverter and can cause problems, including the noise you mention....

I would have never imagined one could use such a design for a home appliance: a medium power (>1kW) half wave rectifier!!

Thanks for sharing.. yes, that was actually what happened. And guess what, I didn't even bother to switch the hair dryer to full speed (paradoxically, the situation would have been improved!).

So my handy test load was actually a hard-core one. Now the fridge turns to be way more friendly!

 

[Warpspeed]

There are a lot of very unfriendly inverter loads out there, anything with SCRs in it, and florescent lights are rather nasty.

The life of an inverter is never likely to be a very happy one.

I data logged the inrush current to my 150 watt refrigerator just out of curiosity. It spiked to just over 4Kw at startup.
I was too afraid to even think about the associated power factor.

 

[red_alert]

My refrigerator rated power is 300W. I didn't log the inrush current but the inverter only "cried" (increased noise) for half a second at startup then he completely forgot about the fridge.

I've made some tests with various other loads and I came to the conclusion that the DC-link bank of capacitors is undersized.

As I'm still encountering voltage spikes at higher loads, I put some pulse capacitors (200nF) across high-side MOSFETs drains and sources. The voltage spikes have disappeared but those capacitors became sligthly warm.

So here's my next question (DC-link related): is it mandatory to have aluminium electrolytic capactors or may I rely entirely on metallized polypropylene film pulse capacitors (like WIMA DC-LINK series)?

Those pulse capacitors are rated at quite high ripple currents.
By example, a 10uF/400VDC MKP4 capacitor has Irms=6A and ESR=11mOhm (f=10kHz). If I put 20 of them in parallel I've got 120A (Irms). As my inverter maximum input current is around 100A, does it seem enough?

 

[D.A.(Tony)Stewart]

These PP plastic caps are far superior for this application. But your motor surge load may be 5 ~8x so your power source is undersized.

 

[Warpspeed]

I am slowly developing a home power system here, including the inverter, and the refrigerator start up was may main concern.
Start up is actually x26 normal running current !
I could hardly believe what I was seeing.

Polypropylene caps are the best for this type of high frequency pulse work, but if you need more surge resisting muscle on your dc bus, perhaps some low esr electrolytics may be in order.

I am using Evox Rifa aluminium electrolytics now, for any really demanding heavy duty application.
These have some awesome specifications for both esr and high frequency ripple rating.
Rather spendy new, but they often turn up on e-bay new or secondhand at much more friendly prices.
They are ideal for anything that draws very high repetitive peak currents.

 

[red_alert]

These PP plastic caps are far superior for this application. But your motor surge load may be 5 ~8x so your power source is undersized.

I have a 20kWh lead-acid battery bank to feed my inverter and it's located right beside it. The surge current of those batteries are over two thousand amps (not at this higher frequency the inverter works, obviously).

Thus I only need some high frequency (pulse) capacitors, I suppose.
For now, I have 30 x 100nF high pulse capacitors (WIMA FKP1) and 4 x 4700uF/100V electrolytic ones.

The DC value of 48V bus voltage seems not affected but I have high voltage spikes on DC-link.

- - - Updated - - -

I am slowly developing a home power system here, including the inverter, and the refrigerator start up was may main concern.
Start up is actually x26 normal running current !
I could hardly believe what I was seeing.

I was affraid of refrigerator start-up surge, too (as my APC 1500VA Smart-UPS couldn't start it) but now I have no problem at all (like I said, it's just a little increase in noise for half a second or so).

 

[red_alert]

The story continues: after changing the DC-Link capacitors with high ripple current / low ESR ones (PP film), those spikes are still there.

It's really frustrating as I've read many books/articles saying that there should be no voltage spikes in a H-bridge configuration. I've redesigned the whole H-Bridge, I've changed the PWM frequency from 5kHz to 10kHz but those spikes are still there.

In theory, the MOSFET body diode it's freewheeling thus there should be no voltage spikes over the DC bus level or bellow ground level (those body diodes should clamp any unwanted spike). The body diode turn-on time is also quite small thus the clamping should occur in no time.

Anyway, I've just read something interesting about reverse recovery overshoot of a clamping body diode. My mistake was to only think of that dead time before the synchronous switch (low-side) has to start conducting. In that situation, the body diode reverse recovery it's not important as the MOSFET it's shorting out the whole diode when it turns on.

But the worst situation arrives after the second dead time (prior high-side MOSFET turn on). At these time, the low-side MOSFET body diode it's freewheeling again and, once the high-side MOSFET is turning on, that diode will be reverse biased thus the reverse recovery process becomes important.

During this period (trr), the current through diode it's changing its sense and this is generating an overshoot at the switching node level. Well, that's what actually (keeps) happening to my H-bridge. In the article I've just read, the solution was to use a shorter dead time to deliberate generate a small shoot-through, to avoid body diode conduction.

I'm not going to try this (as it is too risky for my paralleled MOSFETs) but I'm going to buy some powerful Schottky diodes to by-pass the MOSFET body diodes (the Schottky diodes recovery time being zero - at least in theory). Is there any problem with paralleling Schottky diodes (as I can't find a single 100A/200V Schottky diode)?

 

[schmitt trigger]

International Rectifier has some "Fetkys". Integrated Mosfets with a Schotky.

They claim that the few nanohenries of stray inductance between the Mosfet and the freewheling Scotky will still create problems. By integrating both devices in the same package, the problem is avoided. Or so they claim.

Check their website and look for app notes and white papers discussing this issue.

 

[Orson Cart]

Some fets have worse diodes than other, ixys seem to be the best in this regards, any fet higher than 200V has real issues with diode recovery...

 

[red_alert]

I'm using these MOSFETs: IRFP4668 (trr is around 150ns). I'm using 4 of them for every H-bridge switch.

I've just ordered some 60A/150V Schottky diodes (two of them in parallel across every switch). In the mean time, I'm going to decrease the dead time (to ~0.5us) and check the results.

Thanks a lot for your suggestions.

@schmitt trigger: If only I could knew about those "fetkys" before.. I'll keep the current ones for now.

- - - Updated - - -

I already have some great news! After decreasing the dead time to 0.5us (from 1.5us), those spikes have been greatly attenuated (from 150V to 55V). The DC bus is 24V for now but I'm going to change it back to 48V (and I hope the spikes will stay bellow 100V).

From what I've read, you have to not give those body diodes enough time to accumulate recovery charge during forward conduction (freewheeling).

Maybe a faster turn-on time for the high-side MOSFETs could have been helpful, too (by quicly latching the switching node to DC bus voltage) but I've prefered a slower turn-on, to reduce the switching loses.