Velocity & Displacement

[MachineryWatch]

I am working on a project to monitor the structural vibration of a hydro electric dam. Sensors are located on the dam and the penstocks which carry water to the turbines. The purpose is to identify changes in the structural response of the dam rather than condition monitoring of the machinery. Frequency of interest is <10 Hz, which includes the running speed of the turbines and several harmonics. The two main concerns are long term effect of &quot;steady state&quot; vibration AND transient movement (displacement) during gate opening, turbine and pump start-up, etc.

Accelerometers are &quot;seismic response&quot; 10 volt per g models. The data acquisition system has 1 level of hardware (analog) integration. A second level of integration is achieved digitally in the firmware for displaying spectral data. This works fine for the &quot;steady state&quot; conditions and has been checked with shakers on-site at each accelerometer and proves to be very accurate at several frequencies <10 Hz.

The desired low and high alarm levels for for the transient data is specified as 30 mils (pk-pk) and 72 mils (pk-pk), respectively. The alarms for steady state conditions are not a problem. These can be incorporated using a &quot;spectral band&quot; after the FFT is calculated because the movement is periodic. My problem is the transient displacement alarms. The transient movement is not periodic, therefore, the FFT is meaningless. The transient alarm must, therefore, be established in the waveform. Problem is, I'm stuck with a velocity waveform.

With regard to specific frequencies the specified 72 mils (pk-pk) alarm value represents a velocity level that varies from only 0.024 ips (pk) at 0.1 Hz all the way up to a whopping 2.4 ips (pk) at 10 Hz! Is there a valid way to come up with an equivalent velocity alarm for the transient since the movement is not periodic?

Skip Hartman

http://www.machinerywatch.com

 

[tomirvine]

For displacement measurements, I recommend a Bruel & Kjaer charge mode accelerometer along with a Bruel & Kjaer 2635 signal conditioner. This signal conditioner provides analog integration with both velocity and displacement outputs. A low noise cable is required, however.

Your equivalent velocity method is challenging. You might be able to bandpass filter the velocity signal into several bands over the frequency domain of interest. The bandpass filtering would be performed in parallel manner on the original signal.

A separate velocity alarm could be set for each band, depending on the band center frequency.

There are some software products such as DasyLab that should be able to perform this operation.

(I am not a salesman for Bruel & Kjaer or DasyLab).

Tom Irvine

http://www.vibrationdata.com

 

[MachineryWatch]

Thanks for the response, Tom. I am kind of stuck with the instrumentation I have for this application. I appreciate the suggestion regarding filtering, but I'm still not sure that would get me what I need.

The real issue is what amounts to a thrust motion and a return to normal position. This motion may take around 4 seconds to complete and is really just movement in 1 direction and return. Not what I would really classify as vibration. Just movement away from a reference point and the eventual return to that reference point.

Skip Hartman

http://www.machinerywatch.com

 

[MikeyP]

Some Laser Vibrometers measure down to DC frequencies and provide both displacement and velocity outputs.

Michael

 

[vgarzani]

This may not be specifically practical to your objectives, but with regard to the transient nature of what you will be looking at, consider browsing discussions and examples of time-frequency analysis. Look at UMich and Rice electrical and computational science schools. I think the academic types would be interested in discussing practical applications of some pretty neat characterization tools. If you have access to Matlab there are some downloads available that are worth looking at.

 

[vorwald]

Here are my suggestions

a) Put a low pass filter to remove all signals above 25Hz to reduce the high frequency input from the trasient.

b) Increase the lenght of your sampling time. This should allow the transient spectrum to average out. It also gives a refined frequency spacing. Also, increase the number of samples that are averaged to produce the spectrum. This also has the effect of smoothing out the transient (nonharmonic) input.

c) Divide the sprectrum into bins and reset the warning based on the values in the bins. For example, assume the natural frequencies are .5, 2.5, and 4 Hz. You could have three bins 0.25 to 0.75, 2.3 to 2.7, and 3.8 to 4.2 Hz. Integrate the spectrum (over frequency) in each bin to get an estimate of the power at that frequency. Use these values to set your high/low alarms.

d) If the above doesn't solve your problem, and you are still getting false alarms from the transient, you need to carefully examine the transient signal and develop either a filter to remove the high energy portion of it or develop a &quot;bin&quot; where you can integrate and determine a power level that indicates that a transient is occuring. When the transient is occuring, do not set off any alarms.

Hope this helps
J. Vorwald