When I collect 2 channel data from prox probes, and try to look at the orbit, it isn't as clean as those I have seen from other sources. I have attached an example. I figure it is my setup that gives me the noise, and am hoping someone here can save me some work and time and tell me how they have theirs set up to give them a cleaner orbit.

Thanks in advance

orbit.doc (26 Kb, 72 downloads) noisy orbit

Try collecting data with higher time domain resolution, i.e. shorter time period collection with less cycles, equivalently at higher frequency.

You probably don't have an 'instrument noise' problem.

Remember that prox probes look directly at the shaft, measuring the distance between the probe tip and the shaft surface. They will also 'see' any surface irregularity. It seems likely that what you are seeing is very probably exactly what you have got i.e. a flat orbit of about 0.3mils pk-pk along its longest axis, which contains a significant runout or glitch content and little true vibration.

The 'noise' you are seeing is most likely due to low level glitch/surface finish.

If I understand you correctly, to get a 'clean' orbit, you should aim for a filtered (1X or 2X)orbit - this will give you the single smooth plot that you often see in the literature.

Alternatively, if your equipment allows, you could try to go for a waveform compensated plot, where the wavefrom at low speed is subtracted from the one obtained a high speed - not always successful.

You do need some sort of filtering to get a cleaner orbit. If you do as Buzz suggests and go with a 1X or 2X filter, you will only see a representation of the shaft orbit at those frequencies. While this does make a pretty picture, it is not as useful to me as a nice orbit that contains all the frequencies of interest.

With an orbit, you are looking for mostly low frequency stuff, say less than 10X or so. Your orbit plots will be nicer if you use a low pass filter (the one that lets all frequencies below 10X or 8X or whateverX go through). Maybe setting a different Fmax will work, but I'm not sure with your box. It looks like your data comes from a CSI system. I rarely used CSI to look at orbits since I had access to an ADRE that, in my opinion, did a better job with them.

Waveform compensation will help as this subtracts out all the stuff that shows up at slow roll speeds. However, you must have slow roll data and your software must have the ability to subtract it out. Also, the process is not perfect since the probe target areas on many machines changes as the machine grows and shrinks thermally. CSI could not do waveform compensation last I knew.

You can also play with the number of cycles shown on the plot. Sometimes removing a few cycles will help clear up the picture.

I would also try the CSI website to see if there is a Dr. Know or some other document available to help. I seem to remember something.

One other thing, I think that the 2120 had some settings on it that allowed you to apply a low pass filter to orbits when you were in monitor mode. I'm not sure if those same settings could be applied to orbits that you wanted to save though, and I have never used a 2130.

Michael Titone

Removing cycles may help, but collecting fewer cycles with the same number of data points does 2 things. 1) removes some visual clutter and 2) better defines the time waveforms with more points per cycle.

Filters may distort the data. If you have the data in hand, then a simple recursive filter, i.e. average nearby points will smooth the data. De-noising using wavelets works great, but you need tools for this.