Torsional vibration measurement requires exotic instrumentation, such as a strain gage or a rotary encoder.

Assuming a power train which includes motor, do you think torsional resonance will manifest itself also in motor current signature and if so, in which way?

Torsional resonance may probably also manifest itself in a form of lateral mechanical vibration at the frequency of torsional resonance or by modulating some other frequency measured on the motor in radial direction, if you agree, since torque variation causes motor stator to rock around shaft axis.

Sharing your experience will be appreciated.

Dave

IF the torsional resonance is excited, then of course you get torsional oscillation at the resonant frequency.

Oscillations in motor torque should show up as:
1 - oscillations in current TWF envelope magnitude
2 - sidebands around line frequency in the motor current spectrum spaced at the oscillating frequency.

If the resonant/oscillating frequency is relatively high, then of course you have to look far away from Fline to find them. If the frequency is above LF, then one of the sidebands would appear at (Foscillating - Fline).

pdma has a paper about detecting torque oscillations on theiwr website, although I don't think in that case the oscillations were related to excitation of a resonance.

I had measure torsion vibration on a synchronous motor. In that case, I believe, the speed modulation is directly proportional to the pole / rotor alignment. There is ways of translating current variation to rotor angle modulation (torsion vibration) but one has to know number of poles, rotor excitation, stator impedance and few more. Maybe our friend Pete could light our path?
Best regard, Marcel

Whoops. I seem to have destroyed my post trying to quote/reply to myself.

What I had said was (as Marcel said) sync machines can have torsional oscillation at pole pass frequency during startup. Here is the followup I was trying to add.

quote:

electricpete
Someone posted a waterfall of a sync machine startup which showed this behavior here on the forum (I guess the torsional was coupled to lateral by assymetric stiffness). I searched a little but couldn't find the link.

I found that posted example of pole-pass frequency torsional oscillation of sync motor during startup. It is here:
http://maintenanceforums.com/e...651072763#3651072763

(Sorry it is not particularly relevant to current analysis).
=========
As far as translating motor current modulation to magnitude of angle oscillation, I have no idea.

This message has been edited. Last edited by: electricpete,

I agree with the ways a torsional resonance will manifest itself in motor current, I just wanted to see some real life examples.

Attached is an example of current modulation, although not due to torsional resonance but rather likely due to driven machine shaft misalignment.

Slight_motor__amps__modulation..doc (40 KB, 22 downloads)

David,

If the mixer has 2 blades or paddles, then I would expect the torsional modulation of motor current to be from pulsation rather than shaft misalignment. Shaft misalignment would probably have to be extreme to cause torsional vibrations.

Walt

Walt,
The mixer has 6 sets of pins, each forming a curved paddle shape, and exhaust path. Misalignment ( which is set up rather by eyeballing and could be real bad ) was the best I could come up trying to explain this modulation. Again, we are shifting towards misalignment and its effects. As a response for a particular coupling ( or may be not necessarily) it could produce 1x as it has been proven experimentally.

David

The old EMPATH instrument had a feature called Phase Demodulation; which was supposed to be sensitive to torsional vibration.

If you really want more info on it, send a mail to don.ferree@areva.com. He should be having some case studies.

Regards,

Aditya

Torsion vibs sure show up in active power spectra but with a lot of arttenuation due to the machine electrical time constants.
Same holds true for armature currents.
However, the attenuation of the torsion modes in thee active power spectrum sharply increase with the mode frequency. Only modes with a contribution along the air gap show up.
Calibrating the power active power in terms of torsion angle is too tricky in my view.
See enclosed spectrum for an example with a large turboset facing sharply varying loads in the immediate electrical vicinity..
There are quite a few other techniques than mentioned to measure torsion, all with their shortcomings and advantages.
Gérard

Gerard,

This is very interesting. But how do you know that these peaks are due torsional resonances as oppose to torsional vibration?

Also measuring active power is less practical then just electrical current. How did resonance exhibit itself in current spectrum in the case above?

David

David,
We are dealing with large turbosets whose torsional frequencies are computed by finite elements techniques with quite a good accuracy.
With large trubosets, the prime mover is a turbine whose torque variations depend on the turbine speed governor. Such governors are limited in frequency bandwidth (due to steam expansion, valving lmitations,etc.).
The only practical way to excite torsion in large turbosets is through the air gap torque following grid disturbances. And only torsion modes with a large dfelction (contribution) along the air gap.
This is in stark contrast with generators driven by marine diesel drives (or any other similar internal combustion engines) which can excite torsion modes due to cylinder firing.
For induction motors, the load may also be varying like with a piston pump.
With prime movers and loads exhibiting torque variations, torsion modes can be excited that have no contribution along the air-gap of the electrical machine and thus can go unnoticed in electrical signals.
As to measuring active power, it is directly related to torque assuming a quasi constant rpm . Currents are OK if you are sure that the voltage supply doesn't change.

Hope it helps.

I believe this FFT was developed from a power signal which sums the three phases and therefore has no line frequency component (assuming balanced phases).

If it were an FFT of an actual current TWF, I believe this would have showed up as sidebands spaced at 1hz, 9hz, 22hz around line frequency.

quote:

Originally posted by electricpete:
If it were an FFT of an actual current TWF, I believe this would have showed up as sidebands spaced at 1hz, 9hz, 22hz around line frequency.

That is unless demodulation is first applied as mentioned by Aditya. It is similar to enveloping - would move the peaks from sidebands spaced at oscillation frequency to peaks at oscillation frequency (similar to what is displayed above).

Dear Electrical Pete,
I used this phase demodulation feature to analyze the torsion modes of large turbosets.
This works fine as long as the machnine rpm remains quasi constant, which is the case when alternators remain synchronized.
Phase then works like an enhanced zero method, akin to Wheatstone and similar bridges used in accurate measurements for other physical quantities. It picks up minute variations of the rpm w.r.t. the nominal (averages) rpm.
It is also used to analyze the frequency deviations in a power grid for power quality purposes.

I have examples of such spectra, where we analyzed the phase of FFT of encoder signals with a moderate number of pulses per rev.
Later we used counters with high-precison encoders (100 tops/rev)to measure periods between tops with high frequency clocks. This really works real-time since no FFT is involved.

Trouble with large turbo sets was the axial thermal differential expansion that would damage the encoders without special features to counteract it. We found a special mounting feature to accomodate this expnasion from Hübner Berlin.

I admit I never was confronted to torsion modes involving induction motors with slip driving a load with variable torque that could excite torsion modes. In other words, I worked mostly with synchronous machines (working for a big utility until end of this month). Synchronous machines partly exhibit the same bahavior as induction motors through their damper windings, producing a torque component proportional to slip, which dampen the low-frequency oscillations w.r.t the power grid. With induction (asynchronous ) motors, one cannot be sure of the nominal rpm which is load dependent.
This can defeat the phase approach.
I guess that most people have induction motor drives in mind. So my answer may not satisfy them.
A last remark: electrical quantities cannot spot torsion modes which have no contribution along the air gap. We used active power which is closely related to torque. It was obtained by summing the products of phase currents and voltages. I have to think about your remarks about the side bands of current spectra.
Gérard

Thanks Gerard for those comments. I will think about some of that. I don't work with torsional stuff and it is always interesting to hear from folks such as yourself that are much more knowledgeable in this area.

I will try to add a little to my discussion of where I think the modulation would show up in the current and power spectra if there was oscillation due to torsional resonance which is excited for some reason (does not take much since torsional resonance in general is very lightly damped)

Start with the premise that power is oscillating (that is what your graph shows and we reach the same conclusion by physical reasoning).

Ideally, the voltage is not oscillating, so it is reasonable to assume current is oscillating (amplitude modulation).

Let A phase current be:
Ia = I0* cos(we*t)[(1+m*cos(wm*t)]
where we is electrical frequency and wm is oscillating frequency
Ia has spectral components at we, we-wm, we + wm. **

Now compute A phase power. Assume power factor = 1 for simplifity.
P = {Ia}{Va} = {I0* cos(we*t)[(1+m*cos(wm*t)]} * {V0 * cos(we*t)}
Since power is current multipled by voltage at frequency we, the power spectrum will be the current spectrum shifted both up/down by we. **

Therefore we now have spectral components at:
0, 0+wm, 0-wm **
2*we, 2*we+wm, 2*we-wm **

The components at 2*we are separated by 120 degrees. When we add them together (assuming balanced phases), they sum to 0 and the 2*we component does not appear in power. **

What is left in the power spectrum after removing 2*we is the components at 0, 0+wm, 0-wm
The component at 0 is the average real power. The component at wm is what shows up in your power spectrum plot.

** If you'd like me to show the algebra that proves the statements identified with "**", let me know and I will provide it.

This message has been edited. Last edited by: electricpete,