Unexplained torque output from large VFDs 1

[tgmcg]

I'm working on a project where we are planning to replace an older, non-supported VFD with a new VFD of similar 12-pulse LCI-type archtecture. The motor and driven equipment (centrifugal pump...>10,000 HP) remain unchanged. Reading through the original test report for the motor and VFD issued by the motor vendor, I have discovered mention of huge unexplained torque spikes exceeding 100% FLT, and other excitation of the 1st TNF. As a result, I am concerned about the likelihood of encoutering similar unexplained torque spikes after installing the new VFD supplied by another manufacturer. Ater all, how can a torsional analysis account for huge unexplained torque spikes? (rhetorical question)

I'd be grateful for any insights as to causes and possible mitigations to prevent the occurrence of such potentially damaging torque excitation after installing the new VFD. Are we planning to use the best possible VFD design to prevent such anomalies?

I have read recent technical papers on subsynchronous torsional interaction (SSTI) when using PWM-type VFDs, but nothing in those papers metions unexplained torque spikes as large as 100% FLT. I'm doubtful any of the VFD manufacturers will share this kind of information with us, so am hoping to elicit some comments from folks who have dealt with these issues first hand.

 

[jraef]

Gunnar,
Modern CSI drives are using SGCTs, Symetrical Gate Commutated Thyristors, which now allow for a PWM output instead of six step, so are no longer bound to a specific load impedance to facilitate firing. In other words they no longer as similar to LCI tecnology and don't need to be impedance matched to the motor. That facilitates easy motor selection and replacement, something that had been the realm of VSI drives for years and partially lead to the decline in use of CSI technology.

LCI drives were, however, still six step as far as I know, but apparently this one is 12 "step?" Tgmcg, in your original post, you said these are 12 pulse, now you seem to indicate 12 step. I'm unfamiliar with a 12 step output configuration, but 12 pulse would refer to the FRONT END conversion for the purpose of reducing harmonic reflections to the line source, you cannot equivocate it to output issues. So if it is a 12 PULSE drive, that would have nothing to do with torque pulsations. If it is 12 STEP, then I have no idea. I was just relating my personal experience with older 6 step CSI drives having possible issues with low speed torque stability and since the basic technology of an old 6 step CSI drive is very very similar to a six step LCI drive, I felt that information might be relevant and useful to you.

In reality, because of the nature of LCI drive technology, every drive is custom built to match the load and motor it is connected to. So to that point, the most relevant source for information on YOUR drive will be the original mfr. If that's not possible and you are going to replace it anyway, I think your concerns about torque transients are best brought up in your initial discussions with potential vendors for the replacement. Although your concerns are valid, I think it's more likely that this is an issue with your old drive specifically, and could be related to its age rather than the current technology. I have not heard of torque transients being an issue in MV drives of any sort for over a decade now.

"Will work for (the memory of) salami"

 

[tgmcg]

The acronyms are widely used industry terms and have been accurately clarified above.

The OP deals with noise and unexpected torque pulsations. This is actually a very serious issue with large VFDs, can cause catastrophic failure of drive train components, and loss of millions of $$ in lost production. The problem is becoming more apparent as VFDs are increasingly used in larger critical service applications (e.g. LNG compressor drives). To date, primary awareness of this problem appears to be among the end-users, and primarily among their mechanical engineering specialists...including myself. The adverse consequences are primarily mechanical.

I posted here in the hope that some of the more experienced electrical specialists reading this thread may have first hand experience with the phenomena.

 

[tgmcg]

12-pulse

 

[tgmcg]

jraef,

The original manufacturer is out of business. Parts and technical support are no longer available.

There's actually been a number of recent failures attributable to subsynchronous torsional ineraction (SSTI)and even white noise from PWM-type VFDs. This has been a fairly hot topic at recent Turbomachinery Symposioums in Houston, as VFD's have gained wider acceptance for critical service applications among the major oil companies.

 

[jraef]

Tgmgc,
Interesting, I had not heard this. I'll check with some of my colleagues to see if this is anything making the rounds.

"Will work for (the memory of) salami"

 

[sibeen]

The acronyms are widely used industry terms and have been accurately clarified above.

Well...no.

As I stated in an earlier post LCI can also be "Line commutated", and I still have NFI what TNF means. Skoggs has made, what I assume to be a well educated guess on the matter, but if he's still confused, then I'm none the wiser.

tgmcg, I'm not trying to have a go at you here, but when a few 'olds and bolds' ask you for a few definitions, claiming that they've already been accurately defined may not be the case. For all I now, they may be accurately defined in the mech world, but I'm not a mech person.

Skoggs asked a simple, and reasonable, request. Now, as this is not the mech section of eng tips, I humbly suggest that you cut us some slack and type out any acronyms that may cause some confusion.

 

[tgmcg]

Sibeen,

I'm an "old and bold" myself.

TNF was accurately interpreted as torsional natural frequency.

This appears to be one of those cross-discipline issues where there's actually not a whole lot of overlap between the electrical specialists and the mechanical specialists. Traditionally, that interface occurred within the electric motor, which is an electromechanical device. However, with the advent of VFD's, the problem source moves upstream further and deeper into the electrical world, but the problem manifests itself in the mechanical world. The motor is now primarily a participant, but not the cause of the problem. So there is now a bit of knowledge gap, and as you can see, it is a challenge to even find the common language to bridge that gap.

Then again, there's always been such a gap wth MEs and EEs...maybe just a little bit wider now. LOL!

 

[Skogsgurra]

Thanks to all for giving me insight.

Is there any "popular" description of the phenomenon? An article or a conference paper? I sometimes get involved in odd discussions and I definitely feel that this topic, if it is as common as it seems to be, will pop up anytime, anywhere. Like to see some text.

As to "old and bold": I am twice as old and half as bold as any of you.

Gunnar Englund

http://www.gke.org

--------------------------------------
Half full - Half empty? I don't mind. It's what in it that counts.

 

[sibeen]

jraef, i should have replied to this as well:

LCI drives were, however, still six step as far as I know, but apparently this one is 12 "step?" Tgmcg, in your original post, you said these are 12 pulse, now you seem to indicate 12 step.

I've never worked in the drives area, but did work in UPSs for many years. I must admit that I always used 12 pulse / 12 step to mean the same thing. I also worked on many systems which were a 12 pulse (step) output - a CSI (current source inverter) developed specifically for the UPS market. This UPS used 30 thyristors with a 12 pulse input rectifier, or 24 if the rectifier was a base 6 pulse model.

With the PWM world getting better at the output voltage harmonic levels, this company developed a 24 pulse inverter output, so up to 36 thyristors in the Ac - Dc - AC chain.

 

[jraef]

6/12 "pulse" refers to the input rectifier design.

6 "step" when referring to the output side of a motor drive refers to the output waveform.

Again, apples and oranges.



"Will work for (the memory of) salami"

 

[electricpete]

Google of LCI VFD torsional leads to some info:

http://turbolab.tamu.edu/proc/pumpproc/P24/10-kaiser.pdf

Any practical VFD powerful enough to run a pump cannot
operate as a linear amplifier, but must resort to simple “on” or
“off ” states of the power electronic’s switches. Because of this,
there will always be harmonics present in the output voltage and
current of the VFD, which is applied to the motor. Due to the
three-pole output structure, the harmonics are all odd, with those
divisible by three being absent (a pole is the power electronic’s
block that creates the output to one phase of the motor). The lower
order harmonics (five, seven, 11, and 13) of the fundamental
output frequency cause the stator MMF angular velocity to
fluctuate slightly. That causes the angle between flux and MMF to
vary with a characteristic pulsation frequency. The fifth and
seventh current harmonics cause a torque ripple six-times the
output frequency, and 11th and 13th harmonics cause 12th order
harmonic torque ripple and so on. Of course the amplitude of the
torque ripple is proportional to the amplitude of the current
harmonics. The poorest performing VFD in this respect is the
load commutated inverter (LCI). This VFD creates a smooth direct
current (DC) link current and switches it into the stator windings in
a prescribed sequence. The quasi-square wave current causes the
stator MMF to advance in steps of 60 degrees electrical at each
commutation (a commutation is the transfer of current from one
power switch to the next one in sequence). An LCI with two parallel
circuits operating into a dual winding motor (12-pulse output) creates
3 to 8 percent torque ripple. Another popular circuit, the neutral point
clamped (NPC) circuit has five voltage levels from line-to-line and
uses pulse-width modulation (PWM) techniques. Depending on the
frequency of switching, the current is less distorted and the torque
ripple decreases into the vicinity of 1 percent to 3 percent. A third
architecture, the multilevel series-cell circuits, can further reduce the
current distortion such that the torque ripple decreases to 1 percent
of motor base rating. Multilevel series-cell circuits also apply PWM
technology. Figure 19 displays voltage waveforms produced by
LCI, NPC, and multilevel series-cell (PWM) VFD architectures. For
the LCI drive system, it is usual practice to perform a torsional
rotordynamic analysis of the drive train, investigating critical speeds
with LCI-VFD torque harmonics.

http://www.tmeic.com/Repository/Media/Comparison of VSI versus LCI Systems FINAL.pdf

8. Torque Ripple Produced in the Motor
The main risk with output torque pulsations is they may be close to a mechanical resonant frequency and excite powerful torsional vibrations, which can lead to shaft or coupling cracks and failure. Poor output waveforms from the LCI can produce such torque pulsations or ripples [see figure]. There are not many strategies for improving the motor air gap torque pulsations, usually modifications to the mechanical drive train are the best way to avoid resonances, for example a damped coupling can be inserted.


...
LCI Torque Ripple
An LCI normally has large motor air gap torque
pulsations that are proportional to the average
motor torque. The amplitude depends on the
number of converters connected to the motor.
A 12-pulse motor (two converters) would have
about 25% peak-to-peak torque ripple. If the
ripple frequency is close to the torsional shaft
vibration frequency, resonance can occur.
The possibility of modulation torques is also
present in an LCI drive. The source converter
ripple frequency and the load converter ripple
can interact across the dc link reactor to cause
low frequency modulation of the motor
currents. This can result in low frequency
torque pulsations at high motor speeds.

Here is my attempt to digest this into something simple enough I can understand:
As mentioned, the current to the motor changes in steps. Seems counterintuitive at first, but not when we realize the current is not interrupted but switched from one phase to another (and taking advantage of machine voltage to help with commutation).

So for 6 pulse converter, the current and mmf vary through 6 discrete states in the course of one electrical cycle, so 60 degree step change in mmf at the moment of switching. The torque is given as cross product of rotor mmf and stator mmf. The rotor mmf is varying smoothly while the stator curent is varying in steps, so it is easy to see the resulting torque will vary over the course of that 60 degrees: highest just after the stator step change, lowest just before the next stator step change.

I gather 12-pulse results in 12 discrete states in the course of one electrical cyle, so 30 degree step in mmf at the moment of switching and smaller torque transients than for 6-pulse, although the two references did not agree on amount of torque oscillation expected.


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

 

[electricpete]

Here is attempt to very grossly estimate the torque oscillation above (for purposes of illustrating the idea... won't be very accurate due to stated assumptions):

Let theta = angleof stator field minus angle of rotor field = thetaS - thetaR

theta is a sawtooth function. It starts at some value theta0, then linearly increases for a period DeltaTheta = 2*Pi/6 or 2*Pi/12, then drops sharply back to theta0

(this assumes rotor has high enough inertia that rotor angle can be viewed as increasing linearly, not affected by any torque swings).


Torque is
T(theta) = K*sin(theta) ** SEE NOTE 1
where K = |V1*|V2| / (X * w)

Tave/K = int{K*sin(theta), theta = theta0....theta0+DeltaTheta} / DeltaTheta
Tave/K = { cos(theta0) - cos(theta0+DeltaTheta) } / DeltaTheta
Tpk-to-peak/K = sin(theta0+DeltaTheta) - sin(theta0)

Use small angle APPROXIMATION: sin(theta)~theta. cos(theta)~1-theta^2 / 2

Tave/K ~ { (theta0+DeltaTheta)^2 - theta0^2 } / [2*DeltaTheta]
expand and simplify
Tave/K ~ { DeltaTheta^2 +2*theta0*DeltaTheta } / [2*DeltaTheta]
Tave/K ~ { DeltaTheta + 2*theta0 } / 2 [eqn 1]


Find the Ratio pk-to-pk/average Torque:
Tpk-to-pk/Tave = 2 * deltaTheta / { DeltaTheta + 2*theta0 }
Tpk-to-pk/Tave = deltaTheta / { DeltaTheta/2 + theta0 } [eq2]

Assume we want to compare two approaches (6 pulse and twelve pulse) to give the same Tave/K. -> need to eliminate theta0. Solve for theta00 from eq1
Theta0 = Tave/K - DeltaTheta / 2

Plut into eq2
Tpk-to-pk/Tave = deltaTheta / { DeltaTheta/2 + Tave/K - DeltaTheta / 2 }
Tpk-to-pk/Tave = K*deltaTheta / Tave

The above gives us a very crude first way to estimate the torque oscillation as fraction of average torque. Note that halving the deltaTheta (using 12 pulse instead of 6 pulse) would halve the peak torque oscillation.

In addition to the crude small angle approximation, there is an even bigger approxiamtion:

Note 1: T(theta) = K*sin(theta) is a steady state relationship. We are applying it to a transient situation. The approach might be called "quasi-static". There are many dynamic effects neglected. Accuracy of the assumption is unknown. The types of phenomenon described in the thread linked by Gunnar are all things that appear from transient analysis that wouldn't be predicted by quasi-static analysis of induction motor (assuming the steady state torque speed curve applies during transient). The T(theta) = K*sin(theta) assumption is similar simplifying assumption that may lead to big errors.


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

 

[electricpete]

Tpk-to-pk/Tave = K*deltaTheta / Tave

You could get there a lot quicker using Tpk-to-pk = K*deltaTheta which simplyt comes from the small angle approximation. I thought going through the math was somewhat equivalent to drawing a picture.

It still requires some more assumptions to figure out what K*deltaTheta migh be in relation to Tave = k*theta. In other words, what is the theta corresponding to steady state operation at average torque. We know that if theta exeeds Pi/2 we experience pole slip.

Let's look at some values of Tave = K*theta

For Tave = K*Pi/3
6 pulse: Tpk-to-pk/Tave = PK*deltaTheta / Tave = {K*Pi/6} / {K*Pi/3} ~ 0.5 **
for 12 pulse: Tpk-to-pk/Tave = K*deltaTheta / Tave = {K*Pi/12} / {K*Pi/3} ~ 0.25

For Tave = K*Pi/4
6 pulse: Tpk-to-pk/Tave = PK*deltaTheta / Tave = {K*Pi/6} / {K*Pi/4} ~ 0.66
for 12 pulse: Tpk-to-pk/Tave = K*deltaTheta / Tave = {K*Pi/12} / {K*Pi/4} ~ 0.33

For Tave = K*Pi/6
6 pulse: Tpk-to-pk/Tave = PK*deltaTheta / Tave = {K*Pi/6} / {K*Pi/6} ~ 1
for 12 pulse: Tpk-to-pk/Tave = K*deltaTheta / Tave = {K*Pi/12} / {K*Pi/6} ~ 0.5

For Tave = K*Pi/8
6 pulse: Tpk-to-pk/Tave = PK*deltaTheta / Tave = {K*Pi/6} / {K*Pi/8} ~ 1.33 ***
for 12 pulse: Tpk-to-pk/Tave = K*deltaTheta / Tave = {K*Pi/12} / {K*Pi/8} ~ 0.67

** note the 6-pulse Tave = K*Pi/3 scenario would have very small margin to pole slip at peak torque: Margin = Pi/2 - Max{theta} = Pi/2 - [Pi/3 + (Pi/6)/2] = Pi/12

*** note 6-pulse Tave = K*Pi/8 scenario would involve torque going to zero since deltaTheta/2 = Tave. Any lower Tave would result in reversal of torque for 6-pulse.

The above encompasses the widest possible range of plausible values for Tave if the assumptions are corrrect.

The assumptions/caveats of previous post still apply.


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

 

[electricpete]

Correction in bold:

*** note a 6-pulse Tave = K*Pi/12 scenario would involve torque going to zero since deltaTheta/2 = Tave. Any lower Tave would result in reversal of torque for 6-pulse

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

 

[electricpete]

Last comment - take all of above with a grain of salt. That is just my attempt to communicate what I THINK the basic considerations are. I may be missing something I have not much experience with vfd's and zero with LCI vfd's.

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

 

[waross]

8. Torque Ripple Produced in the Motor
The main risk with output torque pulsations is they may be close to a mechanical resonant frequency and excite powerful torsional vibrations, which can lead to shaft or coupling cracks and failure. Poor output waveforms from the LCI can produce such torque pulsations or ripples [see figure]. There are not many strategies for improving the motor air gap torque pulsations, usually modifications to the mechanical drive train are the best way to avoid resonances, for example a damped coupling can be inserted.

This may be analogous to the "Forbidden speed" issues on the fluid side.
If the periodic pulsations excite a mechanical resonance, then the torque pulses produced by commutation may increase to the point that mechanical damage results.

Bill
--------------------
"Why not the best?"
Jimmy Carter