Frequency separation without any filters

[shjukheter]

That option is available in LV2011 and I did not save my current data in that code because I am away from my experiment setup. If you clearly understand my problem, Please give me a brief idea or I will attach my data tomorrow.

[shjukheter]

Here are my pictures and VI's in LV 2010.

You can see the good pictures as well as noise pictures from vibration sensor. I used "TSA periodogram" which displays only peaks but in the s(f) array has many values.

I would like to write some vaue to my serial device when there is a noise in my serial device but as far as I could not able to find a way how I can decide noise and no noise situations in my program after the FFT results. The only difference I saw noise peaks between the frequency bins. How do I can take those exactly for my nonoise and noise conditions. I have no idea. Please help me.

Thank you.

[shjukheter]

More pictures are here.

[shjukheter]

I was strucked here from two days and could not able to find a way to remove the noise. Please help me.

[johnsold]

You do not need to remove the noise, just measure it.  After looking at your dataI think that perhaps a simple signal to noise ratio measurement might work.

Rather than going through some complicated process to find the exact peaks, which we saw a few weeks ago is not easy, I decided to take the sum of the 10 highest elements in the frequency ranges 1-401 Hz and 501-901 Hz.  Then I defined the Signal to Noise Ratio as Low Freq Sum/High Freq Sum. A large number correlates with a good pump (low noise) and a lower number indicates more noise. I applied the same alogorithm to both the accelerometer and the sound data and found that the accelerometer data is about 10 times better than the sound data.  I think the background sounds are much higher than any background vibration.

Note that the cheking VIs have had all the data acquisition stuff removed and the graphs which contained the time domain data were changed to controls.  Thes changes allowed me to run with your data.

Lynn

[shjukheter]

Thank you so much lynn.

I would like to tell you one thing about my previous work. I have done the almost same thing and even could not succeed. The noise is always lies after 500Hz,So, I calculated the power of a signal after 500Hz for each pump when it is good. So from that values, I had taken some good reference value and used it as reference value for the connected device. If the calculated power value of the connected device is more than this reference value means noise and less indicates good. I am going attach power values of the good motors after 500Hz and you will see the difference. This small difference making not efficient adjustment from software. So, because of that I plan to go with some two good ideas which can be used as reference but could not able find even one idea.

Anyway, Thank you so much. Apologize for my bad English. 

[johnsold]

Can every pump be made noisy by increasing the stroke through the serial port? Or do some pumps always remain quiet? Can every pump be made quiet at a small enough stroke?

If you can always make a pump both quiet and noisy, then perhaps the method should be to compare each pump to itself over a wide range of stroke lengths.  Start very small and measure what should always be a quiet value.  Increase the stroke to a value which should always be noisy and measure again.  Then reduce the stroke until the measurements drop back to within some threshold of the quiet value for that pump.

If large strokes can damage the pump as well as make noise, start at the small end and increase the stroke length until the noise just starts to increase, then back off slightly.

The advantage of this technique is that no absolute thresholds are required. A quiet baseline is measured for each pump.  You only need to determine how much variation between successive measurements on the same pump is normal.

Lynn

[shjukheter]

I can adjust and come from small stroke to long stroke until it gets noise. Here also problems, Suppose if I fix it at certain stroke, which will not be the correct one because may be I can get nonoise situation even if I move further far from the above adjusted value.

In the before message that you posted here about signal to noise ratio, Even in that case I need some reference SNR value. Isn't it? If so, this is not going to be constant for all. So not compromised here. Is there any other way in signal processing that can take me to final journey.

[johnsold]

You are really getting into the area of knowledge about your product more than signal conditioning.  Because your pumps can vary so much from one to another, it may be impossible to establish fixed thresholds on amplitude, frequency, noise, or combinations of these (like SNR) which will work for all.  This was why I asked about varying the stoke on each pump.  The values measured at low stroke could become the reference (plus some percentage to be determined by statistical analysis of many pumps) for that particular pump as the stroke increases.  Each pump would have a different reference - itself.

When I worked on a project very much like this thirty years ago, we analyzed every unit produced over several weeks (analysis took six months), then designed and built the tester, then checked the tester against the old way of checking the equipment for about two weeks.  We had to adjust the thresholds.  A few months later we had to repeat the calibration process to convince the Quality Control department that our tester was still working properly. We spent much more time on this project learning about the product and how it behaved under normal and failure conditions than we did on building the tester.  I think you have the same type of situation.

Until you really understand your pumps and their problems, you cannot make a reliable tester.

Lynn

[shjukheter]

I have some questions related my work which I did not clear my self. Can you please help me.

 Incase of accelerometer

---------------------------------------

1) Why it is not showing clearly the noise frequencies like only nice bins at respective frequencies rather than mixing of some low bunch of peaks over the total spectrum. Is that problem with vibration sensor(I am using 4610 MEMS model currently.)

2) Is there any way to detect vibrations and same sounds(even if the sound level is too low or very high if they touch together with low force(leads to small sound) or strong force(high sound)) if two different parts hitting together. By the way, There are also existence of another noises between these two parts and they are not at all considerable here except above vibrations or sounds.

I am thinking to go with another sensor or another way. If you got right about my point, please suggest me any other sensors of your knowledge or any NI module with low price.