Attached is my orbit plot taken from a new centrifugal pump. I'm trying to have the orbit shaft pattern at different flowarate. But y it looks spiral? Is it normal? or maybe there was a problem in setting up the proximity... can someone help me.

result1.rtf (1,822 Kb, 111 downloads) 1st Flowrate

Ok, I see it now.

It is spiral because the amplitude is growing and shrinking.

What kind of analysis equipment are you using? You are not seeing the whole cycle of the amplitude modulation in your waveform with your current settings. Can you sample for a longer period of time?

Is there another pump running in parallel with this one? This might be a beating problem.

I support the main point raised by Steve: the amplitudes of both channels are not stable.

Better to collect trends to evaluate the cycling period and the maximum readings, if there is a consistent rhythm. This is the first step I recommend.

Are you using labview? Please explain the setup. Are you using case sensors or actual prox probes.
The Orbit is spiral because the Y sensor is increasing in frequency faster then x. The waveform shows this. Also, the waveforms don't agree with the spectrums shown either.

Please explain how and what is taking the measurements. IF you are using a labview type program to collect data, you need to understand signal processing and what is required, such as unit conversions, sample rates(nyquest theorem), Lines of resolution, integration, differentiate, Hanning, flat top window and #poles with respect to phase shift, simutaneous sample and hold etc..

How are you triggering your data acquisition? Is there a keyphasor fitted. Can you do a slow roll to check for runout along your prox probe track. Have you verified you have the correct direction of rotation and orientation of your prox probes. What are the DC gaps when stationary and again whilst at rotational speed. Finally, has there been any arc welding done on the unit because this may have magnetized your shaft.

I have seen a true spiral orbit from a constant speed machine (steam turbine generator). The spiral occurs as the result of changing amplitude and phase (not frequency). The cause of the spiral vibration was a rub with rotor operating near critcal speed (balance resonance). I agree that the "home made" analyzer may not be set up correctly. A lot more info would be needed to verify if that is the case.

Walt

Spiral Orbits (not polar plots) are hard to find; probably (how else?) result from a transient event.

Bently Nevada used to show one in their courses that Don Bently obtained at the onset of an instability. Gordon Kirk had some in his early work integrating the equations of motion for rotor systems.

Given a sudden onset of imbalance like blade loss, and the theory will give these plots through integration. Actually, catching one on a real machine involves timing and luck; although with today's equipment perhaps this will be easier. Analytically, these are not too difficult to come by, these days.

Zorro,

The spiral orbit plot is purely a signal processing issue.

I am assuming you are using the LabVIEW band-pass filtering function to produce a filtered orbit plot. What you are seeing in your filtered acceleration waveforms is the settling of the digital band-pass filter function. Unless you are doing continuous (gap-free) filtering on your waveforms, the initial part of your filtered waveform will have 'distortion' due to the settling time of the digital filter. (Not actually distortion but a linear response to a step input to a filter (analog or digital) starting at an initial state of zero). Notice how the 'Acceleration (Filtered)' waveforms both start at 0. This settling time is related to the filter cut-off frequencies. For example, if you set the bandwidth of the pass-band to 1Hz it will require approximately 1 second to settle (10Hz bandwidth = approx 0.1sec). One way around this is to 'prime' the filter with waveform data. So, if the filter settling time is 500mS and you want 100mS of data to display, acquire a 600mS waveform, send it through the filter VI, then discard the first 500mS of the waveform. To estimate the settling time use the 'Filter' Express VI and open the Configure Filter window and adjust the filter settings to see the result in the preview graph.

A simpler method for plotting a filtered orbit is to use the FFTs of the unfiltered X and Y waveforms. Take the frequency, magnitude and phase data from the FFT spectral bin that you want to plot and generate the sine wave directly using the 'Sine Waveform.vi'. Make sure the FFT magnitude is set to peak and the phase units are in degrees. Plot these sine waves on the XY graph. This will produce a perfect ellipse orbit of the frequency of interest. Note: This method assumes that filtered orbit means a single frequency orbit.

Regards,
Rob Laschinger

One must use caution if using FFT data for orbits. Unless the real frequency falls directly on a 'bin' the phase will be off.

True both channels will be off the same, but if you care about the true orbit this method has potential problems.

Walt,
You are correct, the spiral is amplitude driven with constant phase( X to Y precession). At First it looked like the frequency was changing on Y but the time is slightly longer then X.

As Planet mentioned, the waveforms are filtered. Probably with an exponential type filter and not a hanning type.

You should not filter your time domains for steady state vibration . The only place you need this is before your FFT modules. You can filter time main data with impact data such as a response from a hammer impulse waveform(hammer and sensor). Also, keep the number of poles at 2 to minimize phase shift.

I remember seeing some nice 2X and altered 1X orbits from not using filtering. Aliasing can get you if you don't filter.

The 1X alteration was on some pumps in a nuclear plant. The 2X was on a hydroturbine, with a ground loop folding back to make a 18 mil figure 8 orbit.

Sorry, should have been more specific.

You should filter (DAQ hardware settings) 1~3Hz to Fmax. Your sampling rate should be 2.5 times the Fmax.

Software Filtering in the time domain creates problems such as zorro's problem. That's the problem with Labview type software. One can manipulate data with software and get all kinds of results but as Bill mentioned there are certain requirements that need to be done (signal processing) to the data for it to be valid.

quote:

One must use caution if using FFT data for orbits. Unless the real frequency falls directly on a 'bin' the phase will be off.

Bill,

You are correct. I forgot to mention a very important step in using the FFT data to plot orbits in LabVIEW.

You must truncate the time waveform data to exactly (as close as you can) an integer number of cycles before calculating the FFT.

For the best accuracy it is recommended that the waveforms are sampled at the highest rate possible. This will allow you the capture as close to exactly one cycle as possible. Starting the sampling at the tachometer trigger then ending at the next trigger for one complete revolution. Or, if you want an average value, capture multiple revolutions. Using this method the rotational frequency will be exactly aligned in a 'bin' and the phase will be accurate relative to the tach.

Also, set the FFT window to 'rectangular' (no window).

To convert the FFT magnitude values to sine wave peak, multiply them by the square root of 2. To convert the FFT phase values to correct Sine wave phase add 90 deg.

NOTE: The LabVIEW FFT function works for any number of samples, not just powers of 2 (Cooley-Tukey algorithm). It will calculate an exact FFT (DFT) for waveforms of any length.

P.S. Cooley Tukey is not to be confused with Cool Turkey which is a type of sandwich many people will be eating for the next couple of days (in powers of 2) .

Regards,
Rob Laschinger

This message has been edited. Last edited by: planet-i,