Filtering Noise & Its subharmonics

[Henrik_Volkers]

and because it's so easy to realize in LabVIEW 😄

I used the  tone detection vi  (which can search and remove the major frequency)  in a loop and removed the major 18 frequencies (but also try only to remove one 😉 )

see list on the rigth. All major harmonics (base frequency ratio) are found, but also additional NOT harmonic frequencies, maybe due to ball bearings etc......

If you want to analyse the spindle you should really have a mechanical (health monitoring) background... I bed you can find books on it with a lot of math 😄 😄

[Haristhota]

Yes, you are right. The major frequency content at 33.067 Hz is due to the eccentricity of the spindle since it was rotating at 1984 RPM ( 1984/60 =33.067RPS).

This eccentricity is not considered for spindle quality analysis since this is due to assembly and mounting. That's why i want it to be eliminated including its subharmonics.

There are two types of spindle errors,synchronous (which are synchronous with spindle rotations and repeating for each rotation & asynchronous (occurs due to bearings, temperature & vibrations etc which won't repeat with rotation)

As per your explanation, i believe that shifting and subtracting method is also eliminating synchronous frequency components along with main frequency and its subharmonics

How to find out optimum shift for the particular signal? Is it always be three periods of samples?

Coming to your question, i have limited knowledge in signal processing, that you would have understood already.

[Henrik_Volkers]

So here I my thougths about the it.. (what I would try, without reading/learning about how and what others already have done in that field, what I would considder essential !! )

-Add a sync position (inputs) for the spindle AND for your test artifact. Needed to get phase relations. could be done by using one index signal as a start trigger and do a second measurement with the other index as a start trigger, the phase shift between both signals should be constsnt for one mounting setup and you save one input 😉

-Resample the (10 kHz) signal to get n samples per revolution. (See literature and help/examples of the S&V toolkit) , I would keep n as high as possible, decimation can be done in a later stage....

-mount the test artifact in different angles (if possible) ,and do remoiuntings....   do measurements

-use the method of my second vi (tone detection), or FFT (since your data is periodic in samples now , use no window (rectangular) and a datablock matching exactly multiple periodes, starting always with the spindle index)

-since you now have phase angle references you can now find out what signal is due to spindle and artifact and mounting eccenttricity 🙂

- Since you know these eccentricity values now, you can store them and subtract/eliminate these from later measurements.

Easely fit a master thesis 😄 depending of who is doing the math and programing 😉

The 'substract a n periode shifted signal' method will eliminate all periodic signal content.. the optimal shift  is a multiple in periodes. in the case of my first example one periode is 302.376.. samples   you see, you samplerate doesn't exactly match one periode.  That why I tried it with a 3 periodes shift .. 3 times 302.376 is closer to a natural number fit of samples/periode.

[Haristhota]

Thank you. I will try as per your suggestions.

I already developed an algorithm for two-probe technique. It is giving +/- 10% accurate results when compared to LION precision equipment.

This single probe technique is out of order. I have to refine my algorithm.

[Haristhota]

Hi, Henrik

The shift & subtraction technique is awesome. Using that I was able to remove most of the synchronous frequency components (I believe ) so the residual signal contains asynchronous frequency components.
This asynchronous signal is subtracted from based frequency filtered signal to get synchronous component of the signal.
Is this right technique for separation of Sync & Async components?
I tried applying some lowpass filtering before processing the raw data, but i am getting initial distortion in the signal due to filtering & i am trying to delete those initial data points, is this valid? or any better techniques are available for filtering the signal without any distortion?
The VI is attached for your perusal, pls.

http://www.mediafire.com/download/b3bddt9b191qn4e/Synch_Asynch.rar

[Henrik_Volkers]

Opening RAR files from the internet???

..

In my previous post I already wrote my suggestions... Have you read them?  Have you tried my second approach with tone detection?

The shift and substract will remove all periodic content INCLUDING periodic signals indicating periodic errors.

You need to understand what you do with your signal, and you need a good understanding of the dynamics & kinematics , otherwise you can't tell what's going on 😉

[Haristhota]

The VI, I saved it by making current values default and it became ~35MB, I couldn't directly attach to this forum so I converted to .rar, uploaded it my Mediafire account and shared the link.

Is that illegal to open .rar file in Germany?

If you check that VI, you will easily make out how I used tone detection & shift-subtract techniques for separation of synchronous and asynchronous components. (i believe that separation is happening )

Due to some technical issues,I couldn't able to do practical experimentation but I am trying to load previously acquired data and evaluating my algorithm for synch & Asynch error values.

Once I start practical experimentation, I will definitely try those suggestions.

[Henrik_Volkers]

here is a seperation for harmonic and nonharminc (since the frequencies are detected with an frequency error you have to define boundaries ...)

instead of a IFFT I used a direkt reconstruction method 🙂 

[Henrik_Volkers]

little update

If you do a FFT - IFFT approach it migth be a good idea to resample your data and/or cut your data to match the 'base' frequency. The length of your data can be used to define the frequency resolution.

I added a rescaling  to 'order'

[Haristhota]

I was breaking my head by implementing "FFT _ separation_ IFFT" approach.

I tried comb filtering, notch filtering.....and all nonsense, due to phase distortion or something i couldn't succeed in separation and reconstruction.

You solved this just like a flash.........................

I learnt many things because of you. My heartful thanks to you...!!

I personally feel that I have to learn a lot in signal processing and analysis. Could you please suggest me a best resource/book to learn it and implement in LabVIEW?