Some strange things with PWM inverter filter

Hi All,

I have some inverter questions which I am hopeful someone can answer.

Basically, I have designed and built a 230V 50Hz 3KW inverter to run off a PMA (permanent magnet alternator) putting out 400-600Hz 600-1100volts. I had a lot of problems with a pre-regulator to give a stable, constant 400V link voltage but finally solved them using a three phase triac regulator. However, that is just background.

I am running at 20Khz with a full bridge. Top IGBTs PWM and bottom IGBTs commutated at 50Hz. I am performing flux-reset in the dead time by turning on both lower IGBTs when the top ones are at zero volts. The output filter is LCL with 1.0mH inductors in each leg and a 4uF capacitor (across output/load). Inductors are toroids with quad windings (to minimise skin effect) and are not saturating

My inverter is basically working but there are some strange things I do not fully understand:

a. At no or low load, I get a square wave.

b. Running at low voltage (64V peak to peak), once I draw 220ma RMS with a resistive load, I have an excellent sine wave which persists up to (and probably above) 10amps load current. Inductive loads are also okay.

c. Running at 230V with a link voltage of 390V, I again get a square wave under no load conditions. However, when I connect a resistive load (light bulb) drawing 330mA RMS I still have a square wave. It isn't until I get up to a couple of amps that I have a nice sine wave.

d. I am guessing that at low voltage and no load, the filter capacitor is being charged to the peak voltage on even the low duty cycle portions of the waveform and this persists until reasonable current is drawn.

e. But why do I have to draw even more current at the higher voltage until I get back my sine wave?

f. What can I do to get a decent sine wave at very low loads? As well as running heavy loads, I also want to run much smaller (and more waveform/voltage critical) stuff. A square wave isn't useful.

g. Of course, 2mH and 4uF do not do much filtering down at a few hunded Hz.

Does anyone have suggestions on how to get a decent sine wave at low or no load? Maybe my theory about the peak capacitor charging is wrong?

I am sure a number of you have encountered and overcome problems such as these.

Thanks,
Dave

Some more information.

I found that when I look at the bridge output BEFORE the output LCL filter, I have a square wave at the output of the inverter (but only under no/light load).

So, it looks like the problem is nothing to do with the output filter.

Then I disconnected the filter and guess what? The bridge output is a square wave at 50Hz. But the gate drives are working perfectly! When I put a small load on the inverter, I can see the PWM on the bridge output.

I am currently thinking "output capacitance" but have not found anything yet. The two low side IGBTs are on a common heatsink with insulators but I cannot imagine there being enough capacitance to get the effects I am seeing. After all, I have to take out about 200ma for the effect to disappear.

Any thoughts anyone?

Regards,
dave

---------- Post added at 09:55 ---------- Previous post was at 08:35 ----------

My theory:

When there is no load or filter on the inverter, the emitter of the PWMd high side IGBT will be floating with only stray capacitance and the leakage of the same-side low side IGBT governing its voltage.Once the first (narrow) PWM gate turn-on pulse is applied to the high side IGBT, the emitter will be pulled high (to the rail) by the gate charge. There is nothing other than leakage current to bring it down- so it stays there until commutation. Then the same thing happens on the other side of the bridge- hence a square wave- which is what I am seeing.

BUT, if this is the case, I would have thought that connecting a small load (for example, a resistor between each high side emitter and zero volts would be enough to fix the problem. Indeed, when I do this (draining 3 mA peak), the high side IGBTs start to show the switching and the bridge output looks much more like it should. However, reconnecting the output filter takes me back to a square wave. Perhaps there is winding capacitance in the inductors, worsening the problem.

I am sure this problem has occurred to others and I would welcome some thoughts,

Regards,
Dave

I think, the problem is basically caused by your way to perform "PWM". It can't drive continuous current to the inductor at light loads. This can be only achieved with a symmetrical PWM, that switches between high and low side instead of switching one transistor on and off.

Hi FvM,

So, are you suggesting to drive the "off" transistor in the bottom leg of the bridge with the complement of the PWM signal driving the high side IGBT on the same leg?

I guess I could do this if I replaced the "bottom" IGBTs with faster units. However, I chose them originally for low Vce on rather than high speed as they only commutate at 50Hz. Thus they have a smaller heatsink than the high side IGBTs.

Given that many people use the approach I chose, there is hopefully some other clever way to address the low load situation.

Regds,
Dave

If you want the output voltage to represent the PWM duty cycle independant of the load current, you have to implement synchronous switching. That means, that in each branch of the full bridge, either the low- or the high-side switch must be active, with a short dead time to prevent cross conduction. That's, how every VFD or UPS PWM converter works.

Different PWM schemes are in use, unipolar switching is most common for single-phase full bridge and three-phase converters.

As you said, the disadvantage of synchronous switching is the need for fast switches and drivers in all bridge positions. But without it, you have to implement a closed loop voltage control to make the output voltage follow the sine waveform if the inductor current can get discontinuous at light loads. And the non-snychronous PWM is only acting as a two-quadrant converter, it can't drive a correct waveform into reactive loads.

Hm,

It is a pity I was impressed by the International Rectifier idea of commutating two of the IGBTs at 50Hz, thus saving heat and cost. I did not even think about the low load case and it is now abundantly clear what is happening. However, I still have not figured out why I need to take so much current from it before things behave perfectly- still, I will hopefully figure it out in the next hour or so.

Then I want to figure out the best way to fix the problem so that I can at least get a decent wave form at 30-50watts without massive changes to PCBs, software and parts. It is working fine from 150watts to 1.8KW- just need to extend the low end down to about 40W. It is not the voltage that worries me- mainly the wave form. I will do some calculations on different output filters.

By the way, I did get my IGBT pre-regulator working properly but have now replaced it with a three phase pre-regulator using phase controlled triacs working at up to 750Hz. This has survived a couple of short circuits and other abuse and removes the problem of inductive spikes by using zero voltage commutation of the alternator. It runs nice and cool.

Regds,
Dave

I was impressed by the International Rectifier idea of commutating two of the IGBTs at 50Hz, thus saving heat and cost

Click to expand...

Which IRF AN/reference design you're referring to?

Your PWM basically works like an asynchronous (transistor/diode) buck converter. If the operation gets discontinuous at low loads, the output voltage no longer represents the duty cycle. It's no problem with DC/DC converters, because they implement a voltage control loop. Only the change of loop gain may be an issue. You can reduce the DCM limit current by increasing the filter inductance, you should be also able to predict the behaviour by calculating the inductor current waveform.

By the way, I did get my IGBT pre-regulator working properly

Click to expand...

very good.

P.S.: regarding saving heat and cost. The switching losses don't necessarily increase by changing to synchronous switching. The diode of the low-side IGBT is switching anyway. By activating the transistor, you force the current to continue after zero crossing. This only happens with low or reactive loads. In addition, you can achieve the same waveform by operating both half bridges at half the frequency, which also won't increase switching or conduction losses, but levels switching losses and thus reduces thermal stress to the IGBTs. The only problem is the fast gate driver requirement for the low side switches.

With MOSFET bridges, synchronous switching is mandatory anyway to reduce conduction losses.

Yes, I agree with what you are saying.

I have the original IRF paper printed out, but cannot find it at the moment. They were advocating this particular approach for efficiency and cost reasons, which I think is okay with certain loads and obviously not very low currents. It came from their IGBT design page.

Anyway, off to do some calculations which hopefully will give me some ideas.

Dave

Hi FvM,

As my uP only has two sets of hardware PWM and I do not want to butcher my PCB, I am doing the full synchronous in software. This involves rewriting the PWM code from C to assembly language because of timing constraints. At the same time, I am implementing Magic Sine Waves which will cut down on the switching somewhat (but are very demanding upon correct timing).

I'll let people know how it goes- I am 80% complete (but about to go on holidays for a week).

I did have one more question. For dead time in the same leg, I was thinking of allowing double the time from datasheets (i.e. worst case rise times, currents, delays etc times 2). From your experience, what do you think? Or should I allow even more just to be safe? Previously, I was only switching diagonal IGBTs and even then, only the high side ones were going fast. Now I am switching fast within the legs and do not want shoot-through.

Regards,
Dave

With IGBT, it's no problem to add a microseconds extra. It would be a problem with MOSFET, increasing conduction and switching losses.

I see your point, that many GP microprocessors, e.g. PIC don't have PWM units well suited for synchronous PWM. The specialized motor controller devices have it.

Hi FvM, thanks for that. I'll add a couple of extra uSecs to the dead time.

At 90% complete now but need to go away for a week so will report progress upon return. The precision timing has been relatively easy using 16 bit hardware comparators in the uP.

I'll be very interested to see how the Magic Sinewaves perform. Will also do quite a bit of low voltage and high current testing before turning on the generator.

Thanks,
Dave

Well, I completed the transition to fully synchronous Magic Sinewaves. I have kept the fast Hi Side IGBTs and the slower Lo Side ones because there is less switching with the Magic Sinewaves. They run slightly cooler than with the 20KHz PWM and as a bonus, there are fewer spikes floating around. However, despite the hype for Magic Sinewaves which says you do not need as big filters as PWM, I have found this to be incorrect. 2* 1.2mH toroids with 4uF capacitor in a low pass configuration was fine for 20KHz but with the Magic Sinewaves, there is (both in theory and practice) a big 61st harmonic (i.e. 3KHz). So, I need a 15mH inductor which is not too huge and can take 12 amps RMS without saturating. So, if someone has suggestions of part numbers or particular products, I would be very grateful. Alternatively, I could go with 20uF AC capacitors but they tend to be rather huge.

Notes on Magic Sinewaves:

a. I used 19 points per quadrant. I generated tables in 2% steps from 75% to 99% amplitude to handle voltage regulation. Going to the maximum of 23points only shifts the harmonic mentioned above by a relatively small amount and is not worth while for a 50Hz inverter. THD is low.

b. The Sine waves produced are better than with the PWM solution because of the zero switching.

c. I used an 18F series PIC @ 40MHz & used the hardware compare function to generate the precise timing needed. Pulses and delays need to be accurate to about 1uS.

d. Contrary to the Magic Sinewave hype, I need a BIGGER output filter than with regular PWM because there is a dominant harmonic at 3KHz. It is there "for real" as well as in the Magic Sinewave calculators on Don Lancaster's site. On the other hand, once I get a suitable inductor, I can use thicker wire rather than the four wire windings currently that I used for the regular PWM solution (because of skin effect).

Regards,
Dave

I didn't see yet a straightforward explanation of "Magic Sinewaves" referring to known power electronics standards, but as far as I understand, it's just another name for low frequency PWM with harmonic cancellation. In so far, it's clear, that it will have harmonics starting from a certain number. The trade-off switching losses versus harmonic content respectively filtering effort must be found for each particular application and can't be cut by magic, I fear.

I find Lancaster's explanations reasonable. I agree that there is no magic! The harmonic cancellation is rather nice but the downside is that you WILL have one or two strong harmonics. Their amplitude is greater at lower modulation indices. He mentions that some work has been done on using different pulse series for each quadrant enabling cancellation of the strong harmonics. However, storage and complexity grow considerably and I am not sure I agree with the mathematics involved in showing cancellation.

Regds,
Dave

More helpful advice needed!

Having made a suitable output filter, I completed the trials at low voltage and high current, I moved onto testing with the motor-alternator. I was mostly happy with the results. I replaced the original slow Lo-side IGBTs with faster units. Turn off was 400nSecs worst case and turn-on 300nSecs so I went for a dead-time of 1.3uSecs. That all seems fine.

Then I started running with increasing loads of power tools. All went fine until I went to about 1100W with an angle grinder. After turning it on and off a few times, I suddenly blew a pair of IGBTs when turning it off. This was not shoot-through as it is one high side and the diagonally opposite low side. I am at a loss as to what might be the cause. The output filter is 9.6mH (up from the original 2.4) and the capacitor is 15.4uF (up from the original 4.4uF). This is because of the 3KHz 61st harmonic which was quite big prior to filtering.

My first thought was the inductive current spike through the diagonal IGBTs given I had beefed up the lowpass output filter. However, with a characteristic impedance of about 25ohms, the max current should only be about 16A with the DC link volts at 386V and this should be easily within the capabilities of the IGBTs (irgb4064).

Any thoughts about what might be happening here?

Many thanks and Happy Christmas,
Dave

Can you be sure, that the DC bus voltage doesn't rise above safe limits when removing the load? What is the amount of Vce inductive voltage spikes in normal operation?

No, I cannot be sure of this and perhaps it is the problem.

I think I need to improve the time constant of my pre-regulator. Certainly, it is not good enough at maximum load but I can improve that. In theory, the inductive spike upon removing the load is only about 36volts. However, given this is four times greater than before (when the output inductors were smaller), added to the pre-regulator spike might have just tipped the voltage over the limit.

So, I am going to (1) speed up the pre-regulator and (2) add a fused crowbar at around 500VDC on the inverter board and (3) do some measurements at 800watts to verify inductive spike and pre-regulator overshoot conform with theory.

Many thanks,
Dave

Well, I have just blown (short circuited) my last two spare IGBTs so there will be a delay until the replacements arrive. It is still a diagonal pair blowing (ie one high side and the opposite low side) rather than shoot-through.

The problem does not appear to have been voltage. I had improved the performance of the pre-regulator so that now, taking an 800W load on and off line repeatedly, the pre-reg overshoot plus inductive spike are maximum around 30volts. On top of that, the fused crowbar (response time around 2.5uSecs) is not triggering.

This time, I was monitoring with an oscilloscope and found out more information.

1. I believe I was wrong in my original statement that the problem happened when I removed the load (of the angle grinder). In fact, this (and probably last) time, it happened at turn on but then this put 390VDC across the output so that a connected light bulb and the angle grinder ran, albeit at elevated voltage/current. So, even though I thought the inverter was running, it was not doing so- just the 2 shorted IGBTs driving DC to the loads.

2. Now (as opposed to when I was using non-synchronous PWM), the PWM is all driven under high speed interrupt. When I want to turn it off electronically, I set a flag and the interrupt code turns off all IGBTs at the next voltage zero cross. For an inductive load, this is probably bad because the out of phase current will presumably be at close to maximum. Originally, I would turn off the IGBTs no matter where they were in the cycle.

3. What I think is happening is this. When I turn on the angle grinder, there is a large start-up current. This also pulls down the link voltage somewhat. My uProcessor is monitoring both voltage and current and turns off the PWM if the current is >24amps RMS or >12A RMS with DC link below 250DC. It might even be triggering erroneously as I have not been able to test with really large currents. Anyway, what ever the reason, as soon as I press the angle grinder switch, I am quite sure the uProcessor main loop is signalling the PWM to stop (at the next zero cross) and it is at the moment when the IGBTs are turned off that the Hi-Lo pair is blown up. Significantly, if the PWM was still running after this, the other pair would also be blown up due to shoot-through and this is not happening. A fault LED on my uP board confirms the above.

4. I am probably blowing up the diodes rather than the IGBTs and it probably did not happen originally because the inductors were 1/4 the size and I turned off the IGBTs wherever they were in the cycle. It is probably dI/dT related or just too much current.

Questions:

a. What do you think is the best way to turn off the inverter IGBTs electronically? i.e. where in the voltage waveform and with what timing? Perhaps I should turn on both low side IGBTs together and dissipate the energy from the inductors that way? I could also turn off the IGBTs at max voltage (across the load) but perhaps that is also problematic?

b. The IGBTs can take 40A pulsed. I will re-calculate the likely inductive current spike because I thought it was okay but perhaps I am wrong.

c. Any other thoughts?

Regds,
Dave

I fear, I don't know enough about your design to guess specifically about most likely failure mechanisms.

The Co-pak uses separate IGBT and diode chips, so you are able to determine which device is damaged by dissecting the package. Of course, if the gate is shorted too, it would be clear, that the IGBT is burnt.

At least three ways of IGBT failure exist:

- overvoltage. If the bus voltage can't exeed safe limits, it's still possible due to circuit inductance in combination with high di/dt Mainly a matter of PCB design and bus blocking capacitors. If circuit inductances can't be sufficiently reduced, controlled switching speed is the only way to handle overvoltages.

- overcurrent & IGBT desaturation. Happens beyond the specified peak currents. Somewhat alarming, the IRGB4064 datasheets indicates desaturation with Vge of 12V around 25 A! (See figure 7 - 9). Advanced gate driver circuits usually have desaturation detection with automatic shutdown.

- unintended operation in linear area with high power dissipation due to unsuitable driver circuits. Also ringing control signals causing multiple switching belong to this category.

Overcurrent shutdown should act immediately, if the safe current level is exceeded.

AArrggghhh! You are right. I had been using IRGB4062s and changed to the 4064s because I can get them more easily. I had thought they were almost the same but indeed they are totally different with respect to desat. I think that is what is happening because I have now measured the startup current of the angle grinder and it is 20amps (RMS) for 200msecs so the peak current will be 28 amps so it seems very likely this will be killing them. The 800 watt load has a startup current of 10A RMS (14peak) and that is fine. On top of that, my gate drive is 12V. Whereas the 4062 will give me 90amps at 12V

Regrettably, I had just ordered 20 more 4064s from the US. What a pain!

For now, I might increase the drive to 15V and just boost the minimum alternator speed a bit as well which with a LDO regulator will still give me the 15V without too much trouble.

Regds,
Dave