How do the sampling during run up of machines?

[kmgary]

Hello everyone!
I am a student, and do some research on time-frequency analysis of vibration.So, the speed-depend vibration data during run up of machines is important to me.Recently, I did a vibration test about run up of machinetools.What I wanted to attain was the sampling data of non-stationary speed-depend vibration data.I used the BK 2635 charge amplifier and BK 4370 accelerometer which mounted on the shell of the gearbox in test.But the test was failed, the signal related to run up was not find in time-frequency plane.Please tell me how I can get a good vibration data relate to run up.Thanks!

Your Sincerely!
Gary

 

[GregLocock]

OK, so you've described the accelerometer and the charge amp. What did you analyse the signal with? You would normally ned a two channel analyser, using the other channel for the speed (or tacho) signal from the gearbox, perhaps a visit to

http://www.prosig.co.uk

might clarify your questions?

Cheers

Greg Locock

 

[kmgary]

Thanks for your reply.Yes, your suggest is good. But, in my research I can get the instantaneous speed of shaft by some time-frequency analysis algorithm,such as STFT etc. if the non-stationary vibration data is good (I did some analysis to run up/run down vibration data provided by others, but I am lack the experience on actual test). I am developing a PC based signal analysis software myself, which focuse on non-stationary vibration signal analysis, I am sure the time-frequency analysis algorithms are work well.Where I can find some information on skills for run up /coast down vibration data sampling and the mounted of sensor.

Your Sincerely!

gary

 

[DePauw]

Hello kmgary

First to know is the total run-up time.
Then you divide it in +- 20 to 50 pieces
In each part you take a mesurement, attention there is a relation between number of lines, max freq and sampling time.

The problem is the FFT, this algoritme thinks that the speed is constant so it works badly with a variabel speed.

Good luck
DePauw

 

[GregLocock]

Ok, you want a single channel runup analyser?

Get round the FFT smearing problem by making a lot of short frames of data, overlapped by 50-75%, then calculate the spectrum of each. Then take the spectrum of the magnitude of the spectrum (this is called cepstrum analysis). This will give you a time domain function with some peaks in. Hopefully the tallest peak will be at the inverse of the frequency of your instantaneous rotational speed. This works because it detects the regular harmonic structure of your signal.

This will allow you to build a map of rotational speed against time for your entire sweep. Obviously it may need to be smoothed or massaged.

You can then re-analyse the original data in any manner you see fit, you might like to heavily oversample the original data and then synchronously sample it using your speed map, for instance.

A google search on cepstrum may be educational (or it may be a complete blank)

Cheers

Greg Locock

 

[kmgary]

First, I thank DePauw and Greg Locock for your good suggestion.My sampling rate was set to 50KHz in test. It is a very heavily oversampleing to the original data,but is useful for synchronously sample in later.And I do a downsampling before the STFT analysis for a good frequency resolution. The winlen(window length) for data cutting is shorter than FFT length(For example, winlen can be set to 64,128,...,when FFT length is set to 1024), I search the peaks along the 'ridge' of STFT spectrogram of a selected order component. Because every frame is very short,it can be consider as a qusia-stationary signal in a short time, the frequency smear is avoid.The peaks are instantaneous speed.The cepstrum analysis,Greg Locock mentioned, is a another good way to get the instantaneous speed,I will try it.
I guess the reasons lead to my test failed were as following:
1.The vibration during run up is too weak.I do the test in a factory,the machinetool would do a cutting work when run-up is end,the vibration was get the max when cutting.And I set the gain of amplifier according the max vibration.
2.The mounting position of the accelerometer is bad.

Thanks again! Any new suggestion are welcome!

Your Sincerely!

Gary

 

[GregLocock]

Massive oversampling is a very handy technique, not least because it simplifies the design of your anti aliasing filters. If you can afford the disk space then I would carry on with it. I believe the HP Paragon front ends used to use it and they were very successful. Cheers

Greg Locock

 

[DePauw]

Hello kmgary

Sampling rate is 50 kHz, thus this mean fmax=20kHz?
If fmax=20kHz and you have 3200lines the resolution is 6Hz, this is 180/min, so your first line is 180/min, the second 360/min , you will have only 15 significant lines for an run-up 0-3000/min, this looks no good.

Good luck.
DePauw

 

[kmgary]

Thanks again for your replies!
Dear GregLocock
Do you means I can sent the run-up sampling data to you for a help? The data file not very big about 1.2Mb.If you like, I can do it and send you some plots of my analysis result at the same time.But tell me how I send the data to you. Thanks!

Your Sincerely!
Gary
E-mail address:km_gy@yahoo.com.cn


Dear DePauw
In my test, fmax<1KHz,and I do a downsampling for a better frequency resolution. For example, I can set the down sampling rate is 128,you can calculate the resolution, I think it can stand for instantanous speed.Of couuse, a digital anti-aliasing filtering need be done when down sampling. The hardware anti-aliasing filtering is simplyfied, just as GregLocock mentioned.

Your Sincerely!
Gary
E-mail address:km_gy@yahoo.com.cn