On our 3600rpm Terry turbine, we have blade pass frequency at 70x, along with harmonics and some other broadband high frequency noise.

All of the above vibration seems to vary. I believe it may be difference in mounting from one measurement to the next which causes a big difference in measurement of this high frequency content.

On a recent measurement when the high frequency stuff jumped up higher (attached), we also got a ski slope effet which caused our velocity readings to increase.

Thanks to previous discussions on the boards, I am aware of some typical reason for ski-slope effect: overloading the accelerometer by plunking down magnet without allowing settling time, cord problems, transient temperature (Thanks Jon) or just plain high acceleration.

I am not very familiar with the last one but want to understand it better. Can we judge from the acceleration spectrum whether it is likely that the magnitude of the vibration itself caused overloading the accelerometer, resultng in ski-slope effect in velocity (not shown). If we had a time waveform could we determine that?

=====================

See Ralph's message for spectrum. He was able to shrink the jpg for me (thanks Ralph)

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

Another question - is there a way to post a jpg as a link, rather than attachment, so it doesnt screw up the page width?

quote:

Another question - is there a way to post a jpg as a link, rather than attachment, so it doesnt screw up the page width?

Don't really know, but shrinking the jpg will help.

Pete,

Not sure about the ski slope, high frequency acceleration causing a low frequency ski slope in velocity? It's making me dizzy trying to figure it out. I've got some guesses, but somebody that really knows will probably come along soon.

Have you tried copying from Oddessey to Word and attaching a .doc file?

Danny

I had been using 100mv/g accels on a doctor frame and got lot's of ski sloping in the velocity spectrum. The acceleration energy was peaking out at 60 - 80g's. I went to a 10mv/g accel to get past it. The 10mv/g having a much higher dynamic range (500gpk. I'm not sure if that's your issue. It could be a faulty wire in your cable or connector also. For posting I usually paste into word or excel and then upload the file link to the board.

You can shrink images using Irfanview which is shareware ( pay optional).

http://www.irfanview.com/

We've run into this issue in the past. If your overall acceleration levels exceed the range of your accelerometer (~50 g's for a 100 mV/g general purpose accel), it's possible to get ski-slope issues. What is your overall acceleration reading?

Are you using a stud or adhesive mount? A simple test is to collect data concurrently using a magnet which acts as a mechanical low-pass filter. You won't see the ski-slope on the readings from the magnet.

Also, what kind of accelerometer and data collector are you using?

Steve

If the acceleration level exceeds 50 - 70 g peak on a 100 mV/g accelerometer, the signal will clip because it exceeds the output range of the intergral amplifier. When this happens, all bets are off. I've seen cases where some of the peaks are real and cases where nothing in the vibration signature is actually coming from the machine. To put it bluntly, overloaded data is junk.

The variations in level may be caused by the mounting of the accelerometer. If a magnet is used on anything less than a perfectly flat, smooth surface, the stiffness of the mounting will be lowered resulting in a lower resonant frequency. If any forcing frequencies happen to be near the resultant resonant frequency, overload is pretty likely. Typically, an accelerometer resonance amplifies by about 30 dB (with also happens to be a facotr of about 30 times), so a little energy at the resonant frequency can easily result in overloading.

To evaluate the effect of mounting, try taking a number of measurements, moving the acelerometer slightly between each measurement, and make measurements on bare metal if you're not doing that already. You'll probably see some big differences if you're taking measurements on a non-machined, painted surface.

For high frequency measurements...let's make that measurements on machines with high frequency forcing frequencies, a magnet mount on a non-machined surface is a poor option. A magnetic mount on a machined surface or properly installed mounting pad is a better option. The best choice is to stud-mount the accelerometer. You've probably heard about a drop of oil between the accelerometer/magnet/mounting surface? High frequency measurements are where it really counts.

If the accelerometer still overloads with a better mounting option, the acceleration levels mayt simply be too high for the accelerometer to handle, or there is a lot of energy at the resonance frequency. In this case, it's time to use a high frequency/low sensitivty accelerometer. Most data collectors allow setting the accelerometer sensitivity for each measurement location, but be sure to change accelerometers when needed. When selecting an accelerometer for an application like this, review the data sheets carefully. Some low-sensitivity accelerometers just have less gain in the amplifier stage. This may work ok depending on the cause of the problem, but the better answer is a smaller crystal, resulting in a higher natural frequency. This is the sure-fire way to remedy the problem whether it results from levels that are just too high or high-frequency forcing frequencies exciting the natural frequency. A sensitivity of 10 mV/g is usually the best choice giving a lot more head room. 25 mV/g accelerometers can also be effective but don't provide as much margin.

As in all things vibration-related, there is a tradeoff when using a low sensitivity/high frequency accelerometer....the noise floor or minimum level the accelerometer can measure will be increased. Usually not a problem for machines where this is a problem, but it does rule out using this accelerometer for everything.

Jon
http://www.spintelligentlabs.com

I beg to differ with Steve's comments regarding using a magnet as a low-pass filter. Yes, sort of true, but because it also lowers the natural frequency as I explained above, it isn't an effective test!

The idea of using a low pass filter to knock out high frequencies to evaluate the situation is a good one. B&K used to make some outrageously-priced "mechanical filters" to do this. For a quick and dirty test, a piece of innertube rubber between the accelerometer and mounting surface is effective. I used this arrangement after destroying a couple accelerometers when trying to evaluate vibration levels on a jack hammer! To avoid destroying more accelerometer, I used a mounting block hose-clamped to the handle with a piece of rubber between the block and handle. Worked like a champ to get levels are the lower frequencies that cause hand and arm damage.

Jon
http://www.spintelligentlabs.com

Jon - My point was that using a magnet will indicate that the high freqeuncy signal is overloading the accelerometer. I'm not advocating that it is the "right" method to measure the vibration. If you're interested in the high frequency component of the signal then you should considered using an accelerometer with a higher range.

Steve,

Let me explain more clearly....a picture will help clarify the point I'm trying to make.

The attached sketch illustrates the effect of accelerometer resonance. For a given input level, the curves show the output level. The peak towards the right is the accelerometer's mounted natural frequency. A little energy at this frequency results in a large output.

The blue curve shows the best case. A stiff mount provides the highest natural frequency. The red curve shows a magnet mount, or actually any less-rigid mounting situation. The natural frequency is shifted down. The response above the natural frequency is rolled off, so as you say, less high frequency energy will be measured.

On the other hand, the natural frequency is lowered from maybe 23 kHz to as low as 10 Khz with a magnet. Harmonics of turbine bucket rate, gear mesh and so on are much more likely to excite the lower natural frequency. This in fact may be exactly the situation Pete is experiencing. A magnet mount on a rough surface, resulting in a different natural frequency every time the accelerometer is placed on the machine. Sometimes closer to one of the machine's forcing frequencies, sometimes farther away resulting in levels all over the place.

Without knowing if overloading results from high levels in the "flat" part of the range, or energy exciting the natural frequency, the idea of using a magnet as a filter doesn't work. You filter some energy out, but increase sensitivity to other frequencies.

I came across an interesting application of this a few years back on a gas turbine monitoring system. It used 10 V/g accelerometers! I was puzzled because clearly signal level wasn't the reason for the high-sensitivty accelerometers. Turns out they had a natural frequency of a few kHz. About rotation rate the first few harmonics, but below all the different high frequencies generated by the 18 or so stages of the trubine. Not only was the accelerometer not overloaded by high frequency noise, the accelerometer served to filter out the stuff they were less interested in.

Jon
http://www.spintelligentlabs.com

quote:

Originally posted by electricpete:
If we had a time waveform could we determine that?

I believe the digitized TWF could easyly indicate whether or not an accelerometer has been overloaded (of course, due to a relatively slow sampling rate some peak values will not be captured when that happens). The indicaton of an overload is going to be, as Jon said, when reading is at 50-70 g's peak. Why wouldn't you post the TWF?

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

Jon - Thanks for the excellent explanation that is clear and logical. I now see why a magnet might not be an effective low-pass filter.

David - Even with the timewaveform data, you won't know if it is the machine vibration that is overloading the accelerometer or if it is the mounted natural frequency of the accelerometer causing it to overload.

quote:

Originally posted by Steve Ciesla:

David - Even with the timewaveform data, you won't know if it is the machine vibration that is overloading the accelerometer or if it is the mounted natural frequency of the accelerometer causing it to overload.

Steve,

I do agree with the above statement but regardless of the cause, at least one conclusion could be made based on the TWF: is the accelerometer overloaded or not? Then the cause could be narrowed down further.

So; where's the ski slope on the Emonitor plot above? I fail to see it! Or what are you calling a ski slope?

quote:

Originally posted by Sam Pickens:
So; where's the ski slope on the Emonitor plot above? I fail to see it! Or what are you calling a ski slope?

I believe Pete indicated that the ski-slope was in the velocity spectrum. He's showing the 1xBP and 2xBP peaks in the acceleration spectrum.

If one were using a velocity transducer? And how does 1XBP=70X & 2XBP=70X... It is confusing to me. I assumed that originally the data were taken via accelerometer and integrated to velocity?

Hi Pete,
Here are plots of the typical g's and ips spectrums for my terry turbines (we have two). Different from yours!

Here is the IPS spectrum.

Pete,

It might be helpful to discuss it with your sensor vendor, as well. I have had some sensor issues in a couple of very different applications, but they may shed some light on the subject. At one time in my career I was collecting a lot of data on Atlas-Copco screw compressors. We were aware of the high frequencies that could be generated at multiples of Gear Mesh Frequencies on these units. For this reason we would collect data on a quarterly basis with a special high frequency accelerometer that was attached with adhesive at various bearing locations on the compressor. This was primarily to get an accurate trend of the Gear Frequency Harmonic amplitudes and great care was taken in making sure the sensor was attached in exactly the same location each survey, and that the surface preparation was done correctly and that the compressor load conditions were the same, so that we could get comparable data each time at the high frequencies. The problem ended up being at a location where we were not making special high frequency measurements. On one of the normal monthly surveys our standard velocity readings with the standard 100 millivolt/g accelerometer placed on the compressor gear type oil pump began to fluctuate wildly, showing several ips-peak. The energy in the velocity spectrum was in the very low frequency part of the spectrum, but looked a little different from the standard "ski-slope" which seems to come and go from much acceleration-to-velocity integrated data. By different I mean it was not confined to the first few lines of resolution in the spectra. Instead it was shifted over to the right maybe 10 lines. We showed the data to the instrument/sensor vendor and they said it looked like the sensor internal amplifier was saturated and that we should try a less sensitive sensor. We put a 10 mvolt/g sensor on and tried again, looking at acceleration waveforms and saw that we had amplitudes of nearly 60 G's at the oil pump gear mesh frequency (around 300 Hz)!

Another project I am involved with uses 10,000 mvolt/g accelerometers to monitor the vibration on a dam (structural vibration) at a hydroelectric facility. The data has often suffered from large skislopes in the spectral data. I asked this sensor vendor to take a look at the data because I suspected the same thing was happening here. The sensor vendor told me that it was not the same problem. They looked at the waveform and said the sensor recovered too quickly for the internal amplifiers to be "saturating". The vendor indicated that the sensor's natural frequency (about 450 Hz in this high sensitivity application) was being excited. They said that this was distorting the data significantly, but did not actually "saturate" the internal amplifiers.

So, I guess there are acouple of things that can happen within an accelerometer to distort the output and cause the "ski-slope". I guess I would take a look at the acceleration waveform. You know that the true amplitude of the spikes in the waveform are often not reflected in the amplitudes seen in the FFT. Your sensor may be dealing with much higher accelerations than are indicated in the acceleration spectrum. If there is any question I would try a 10 millivolt/g sensor and make sure the attachment of the sensor is very secure. However, I would think that anything causing your sensor signal to be distorted in some way would also show something a little more revealing in the acceleration waveform than was displayed. If the overload is intermittent, depending on slight load oscilations on the turbine, you may be dealing some sort of timing issue in how your data collector sequences the collection and storage of the velocity and acceleration spectra, if you are storing it both ways rather than simply converting the acceleration waveform displayed above to a velocity spectrm through a simple integration calculation. Is the velocity spectrum you describe taken by specifying a lower frequency range, so anti-alias filtering is different from the acceleration spectrum you posted?

Skip Hartman