Terminator networks for VFD 3

[xj25]

Hi,

We have a problem of voltage spikes in a VFD installation (220V) with 7 induction motor fans in parallel (about 8 amps in total).

We have measured spikes in PWM pulses (50ns rise time) about 800V peak in the farthest motor from VFD.

We are assesing what to do (add reactors, dV/dt filters etc. etc.) and I found a "termination network" discussion about using RC network between phases matching the cable impedance like this:

R about 100ohms
C about some tenths of nF

Supposedly, at the pulse rise the R matches the cable impedance so spike is "cancelled", after that C charges to line value so no permanent power is wasted.

Has any sense for you guys? any experience or supplier with this kind of terminators?
I have thought that might try with three 100ohms 47nF snubber between phases to see what happens...

 

[electricpete]

Pete: The waveforms were taken without filter and transformer.

Ah yes thanks - that explains why it wasn’t working! Lol.

A mildy interesting article here:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.15.3129&rep=rep1&type=pdf

Interaction of drive modulation and cable parameters on AC motor transients
by Kerkman, R.J.; Leggate, D.; Skibinski, G.L.;

It addresses some of the things we have been talking about here:
1 -oscillation
2: reflection at > factor of 2 amplification

This author discusses oscilation on page 2 in a manner which I would say is very similar to what I described previously. They identify that the oscillating period of the cable corresponds to 4 times the cable length. The example figure is figure 1 where it can be seen tp is the travel time – first observed between inverter pulse and motor response….. then 4tp is the period of the resulting oscillation. Why does this frequency not hold for the waveforms posted in this thread? Beats me…. There’s got to be more to the story somewhere.

They discuss two possibilities for reflection magnitude > twice the inverter pulse
First: IV.A on page 5 discusses “double pulsing effect” which corresponds to the case of figure 3 (a few pages earlier) where the incoming pulse changes direction at the same time that the reflected negative wave hits the motor. In this case, we would expect to see the final voltage end up zero (due to short pulse), but in the waveforms posted in this thread it ends up at 310volt, so this double-pulsing phenomenon does not apply.


Second: IV.B on page 6 discusses “Polarity Reversal” – this is where the pulse doesn’t change to zero, but changes to an opposite polarity again at about the same timeframe when the first negative reflection hits the motor. There is no evidence of this behavior in the one inverter pulse that was zoomed in, so we might be inclined to discount this possibility. BUT, there are some interesting clues that this might be occurring.
First clue: the data posted 14 May 11 4:47 in this thread on page 5/7 seems to show a very regular pattern of alternating magnitudes of the peaks at the motor: 800 / 620 / 800 / 620 / 780 / 620 / 760. It is a distinct enough pattern that we suspect it may not be coincidence (if it is just a characteristic of system reacting to the same pulse every time, we expect roughly the same result every time). So maybe there is some slight difference among pulses …. What difference in pulse can cause this…. Something like the oddball reversal of figure 9.
Second clue: This is probably a bigger clue. look at figure 9 of the linked paper and see there is an interesting feature that the waveform first goes in one direction and then reaches the high peak in the opposite direction. That feature is shared by all of the high peaks on page 5/7 and by none of the low peaks.

So, there is at least a few clues that this “polarity reversal” of the inverter pulse might possibly be going on. It doesn’t particularly show anything like that evident on page 2/7, although there are a few tiny positive peaks evident above zero on the right side of that figure where everything should be at/below zero. Since it is so far zoomed out, I don’t particularly trust it to show a small polarity reversal. Perhaps further inspection zooming in on a few pulses might show some clues. You showed us one…. Perhaps look at a few more… maybe only 2 in a row are required if some anomaly is occurring every other one.

Just a thought.
=======
I'll be interested to hear like if this is a normal induction motor or a ceiling fan type motor.

=====================================
(2B)+(2B)' ?

 

[electricpete]

Figure 5 of the linked article shows oscillation frequency vs cable length. There is one cable type (Hypalon) with extraordinarily high dielectric constant (EpsilonRelative=11), such that the resonant frequency of 60 feet of cable would be right about 1000khz = 1mhz, which was what was seen in this post. Do you know the cable type by any chance?

=====================================
(2B)+(2B)' ?

 

[ScottyUK]

A question ePete:

Would you have to take into account motor winding capacitances, or are they sufficiently decoupled from the cable capacitance by the series inductance of the winding that they wouldn't contribute much?


----------------------------------

If we learn from our mistakes I'm getting a great education!

 

[xj25]

The motor power is about 300-400W at current conditions (near nominal).

It is a ceiling axial fan type, and I think it is external rotor type.

Insulation class I don´t know but EBM seems worried about overheating due to working with VFD, so maybe is not very good. I will check on catalogue on monday and will ask them details about the filter.

I will check also details about cable type installed (should be 1.5/3kV insulation, halogen free type).

Regards

 

[electricpete]

Would you have to take into account motor winding capacitances, or are they sufficiently decoupled from the cable capacitance by the series inductance of the winding that they wouldn't contribute much?

I think you're asking why I didn't include motor winding capcitance in the LTspice model. The answer there is that I viewed it as a small effect and I was building a quick crude model. But you're right, the EPRI model of the motor for purposes of predicting voltages at the motor terminals is an R/C circuit as I mentioned 14 May 11 19:29. This might possibly be the explanation for the spreading of the pulse (from 100 nsec rise time to 400nsec as mentioned 14 May 11 19:29). It would be interesting to add that capacitance to the model..... I'm just not sure what value to use (EPRI has only values for large motors). Maybe someone knows a good value for these tiny motors ? (which for all we know could be celing fan type.... still waiting to hear about that).... I'll poke around and see what I can find on small motors.

While we're talking about the model of the motor, I'd like to backtrack to my statement about the model and whether it includes any L to create any ringing. (. I have heard it referred to on the forum as LC ringing. It may be the case, but I can tell you in large volumes of surge study by EPRI, I don't recall them every referring to anything like that. The model of the motor for purposes of predicting voltages within the system up to the terminals of the motor is an R / C circuit). I do remember now reading several places L-C ringing associated with the TRV accross the contacts during circuit interruption. But I still am pretty sure it does not end up playing any role in surge seen at the motor terminals under the EPRI model. I'm not sure why it would be that L-C ringing applies in one case and not the other.

=====================================
(2B)+(2B)' ?

 

[electricpete]

Here I have re-run the simulation by adding 2 features:
1 - add capacitance into motor model.
2 - use a reversing pulse.

Slides 1 and 2 show the model. As noted on slide 1, I did not pick the capacitance value based on any empircal data.... I picked it based on the value I thought would recreate the measured waveform best (that might be considered cheating, depends how you look at it).

Slide 3 shows the reversing pulse that I used. The negative peak is -0.5 with 100nsec rise time, followed after a delay of 500nsec by change of 1.5 to value of +1 with 100nsec rise time. The 500nsec was again chosen to try to recreate the measurements. 500nsec is roughly worst case because the opposite polarity reflection of initial negative pulse arrives at motor 7 at the same time as the delayed positive pulse (for motor 7).... all of this under the assumption that I will create a 1usec period, which I did (it was a result of changing the capacitance, not changing the cable parameters).

Slide 4 and 5 are results. Slide 4 shows that motor 7 is the worst of all... probably because time delay was selected to create worst case at motor 7. Slide 5 shows chracteristic of the waveform. It matches xj25's waveform in the following respects:
* Peak of 2.85 times initial pulse.
* Oscillation frequency of 1 Mhz (adding the caps slowed down the oscillation frequency... did not have to resort to the Hypalon cable assumption).
* Rise time in the neighborhood of 400nsec, depending on how you count it. (**)
* Initial negative deflection before large positive peak. (**)

Some factors that didn't match well:
** The initial negative deflection here is much larger magnitude than in xj25's data.
** The rise time in the simulation is relatively smooth, whereas xj25's seemed to have a distinctive kink, as if 2 slopes: one in the first 100nsec and another lower slope in the last 300nsec of the rise. The simulation did not recreate this.

=====================================
(2B)+(2B)' ?

 

[xj25]

ElectricPete, the motor is ceiling fan with external rotor type.

I think that don´t uderstand very well why is not used any motor winding inductance in the model. Could you just comment it?

I added to the model an inductance of 0.7mH in each branch representing the MCB coil (I made up the value from an inductance of MCB that should be something like 1cm x 1cm x 3/4turns) but then all the effect of the ringing dissappears :-(.

About referenced EPRI 5862 Volume 2, Section 5, is possible to check it somewhere?

Reading in detail the VFD manual, it explicitly says that it should be used a output inductance filter if motors are cabled in paralel with star topology (however in bus it accepts it).

Looking for info I found this document that maybe someone finds useful.

I am having a quite busy time now so I will return when having news to tell you how it ends.

Regards

 

[electricpete]

I think that don´t uderstand very well why is not used any motor winding inductance in the model. Could you just comment it?

I added to the model an inductance of 0.7mH in each branch representing the MCB coil (I made up the value from an inductance of MCB that should be something like 1cm x 1cm x 3/4turns) but then all the effect of the ringing dissappears

About referenced EPRI 5862 Volume 2, Section 5, is possible to check it somewhere?

The EPRI document is copyrighted. It is a very valuable document imo, but prohibitively expensive if you’re not an EPRI member.

A less expensive less-detailed article by the same authors can be obtained from IEEE here:
“Equivalent Circuits for Simulating Switching Surges at Motor Terminals” by Sharma, Gupta, Pillai, et al from IEEE Transactions on Energy Conversion, Vol 3 , No. 3 , September 1988

Attached is an excerpt from that article. Figure 2 is the model that these authors use for purposes of calculating surge at the motor terminals (the same model as the EPRI report). It has resistance and inductance at the source, a transmission line representing the cable, and resistance/capacitance at the motor. Even though both L and C are present, I don’t believe this would give traditional L-C ringing because of the separation of these components by a time-delay transmission line. I tend to think in your case where there are many parallel cables connected to the source bus, then the source acts low impedance for purposes of reflections, since the parallel cables have low impedance. Therefore I have assumed that neglecting the source inductance is not a big error here. I haven’t tried it out.... does your simulation show different results?

Also I will mention that the multiple series L / parallel C is sometimes used to represent a transmission line. The intent is that the discretization must be fine enough such that each L and C corresponds to a length of transmission line far less than the wavelength of interest (for example less than 10%). In that case when the transmission line model is properly applied/discretized these particular L-C elements generate no oscillation.

As to the model of the motor itself, I have seen (not necessarily understood) a wide variety of models used to represent motors for a variety of high-frequency purposes. Many of them do include inductance elements. Some of them distinguish differential mode vs common mode. I believe the EPRI authors focus on common mode. Which model is right I’m sure may be slightly subjective and certainly depends on the purpose of the analysis. Personally I tend to put a lot of weight on the EPRI report because it has more details than any other reference I have on the subject. But I’m no expert.


=====================================
(2B)+(2B)' ?

 

[electricpete]

The filter is a kind of low pass plus isolation transformer. Do you think that really is needed something like that?. I don´t see the point of the isolation transformer (unless it helps with bearing currents?).

From what I have read recently, the purpose of the sin filter is generally to minimize audible noise. (That might make sense if these fans are around people). Other less expensive filters can protect the motor pretty well (dv/dt filter was mentioned) from vfd in general.

(Of course don't lose track of Gunnar's comment... maybe a vfd is not even needed for ceiling fans which can be controlled by other simpler means)


=====================================
(2B)+(2B)' ?

 

[LionelHutz]

A delta:wye isolation tranformer on the output of a VFD makes a very good filter by itself. I'd suspect that any extra filtering they are describing might be just to remove a bit of the RF that still gets through.

Of course, that document says a sinewave filter and the drawing may just be a generalization not really represting any specific filter. There are sinewave filters that don't use an isolation transformer.

 

[xj25]

I build on Pete simulation including TL model for ground return and MCB inductances and adapted impedance values for motor & separate single phase cables found in other references. Results remain more or less the same.
In Pete model the small MCB inductance will flat out the transients. Here not, but not sure if it is correctly done.

The difference with actual results is the duration of the transients, the real case is about 4us. I coundl´t get to adapt it without modifying quite much the "ringing" freq.
Using a serial inductance of 2mH eliminate the transients in the simulation.

Regarding varying speed "by other means" i.e. voltage, how much is the speed variation range than can be expected without overheating too much the motor? We are working with speed(freq.) changes from 60 to 40Hz more or less.
We have been recomended to use a multistep, stepdown transformer.

Regards!

 

[hacksaw]

get some ferrite line filters, Radio Shack sells a clip-on toroid, as do a few electrical supply houses. It may take three or four in series on the hot lead to dissabate the transients.

interesting that you are seeing turn-on & turn-off transients at the motor and not at the vfd.

good luck

 

[LionelHutz]

EBM-Papst says to use a sinewave filter and ozmosis is confirming this is required for those kinds of fans. It's an interesting exercise to try and filter out the spikes but it just doesn't make much sense (at least to me) to spend much time trying to engineer a filter to only eliminate the ringing when you can buy an output filter or a small delta:wye isolation tranformer and be done with the project.