Time history data manipulation

[Drej]

Dear all

I have some time-history data (time-acceleration) from an accelerometer placed on an engine block. Using this data I'd like to carry out an FE Harmonic analysis, however to do this I need to convert the time-accel data to freq-accel to identify the dominant frequencies and magnitudes. This will then be my input for the harmonic analysis. The problem is that I do not have software available to FFT the data to get freq-accel, and was thinking of using the RESP function in ANSYS to do this conversion instead, which generates a response spectrum from time-history data. I plan to increase the damping level in the RESP function to a super-critical level (which would give no amplification) but would give the response basically to a rigid-body motion. Would this RESP be equivalent carrying out an FFT on the data?

Many thanks.

 

[MikeyP]

Do you have Microsoft Excel? That has an FFT function. There are plenty of freeware/shareware programs that will do it too.

M

--
Dr Michael F Platten

 

[Drej]

Thanks, Michael.

Couple of quick questions:

- have you used the FFT function in Excel? is it ok?
- can you point me to these freeware/shareware programs you mention, please.

Cheers,

-- drej --

 

[tomirvine]

Please send me an Email request if you need an FFT program.

Tom Irvine
Email: tomirvine@aol.com

http://www.vibrationdata.com

 

[mkoijn]

google "octave" and "scilab" . Both are freeware math software with fft functions. Ocvtave is a great Matlab clone. possible problem with the excel fft is that it's limited to 4096 points.

Best regards

finnigan

 

[Drej]

Thanks for all your input on this.

One small point: does anyone find that scaling of the FFT, from software-to-software, is an absolute nightmare? I tried three different FFT software packages, all of which successfully identified my dominant frequencies, but all gave different magnitudes (without explanation). I know this is down to scaling, but am yet to hear a plausible explanation. And in general, the FFT magnitudes were well below the maximums in my time-history.

Cheers,

-- drej --

 

[tomirvine]

There are numerous conventions for representing FFTs.

FFTs can be displayed as magnitude and phase. FFTs can also be displayed as real and imaginary.

The FFT magnitude is symmetric about the Nyquist frequency, which is one-half the sample rate.

An FFT is usually shown extending up to the Nyquist frequency, but it could be displayed up to the sample rate frequency.

Furthermore, FFTs can give good representations of sine vibration amplitude. However, FFTs are poor at representing random vibration amplitude. The reason is that the amplitude depends on the frequency increment, which depends on the duration.

Random vibration is best represented by a power spectral density.

Aliasing and leakage errors are also concerns.

In addition, the classical FFT is based on 2^n points where n is an integer.

For example, an FFT could be calculated for 8192 points or 16384 points. Now assume a time history with 10,000 points. Either zero-padding or truncation would be required to perform an FFT.

Tom Irvine

http://www.vibrationdata.com

 

[Rob45]

drej:
Your different magnitudes might well come from the different types of windowing used by the different software packages, or as tomirvine suggested, because of differences in analysis block size.
But you're most likely simply interested in the dominant frequencies and their relative magnitudes, aren't you?

Anyway, try converting your data to a .wav file and "playing" it back through an acoustic software package like CoolEdit 97, a nice piece of freeware that will do FFT's and let you experiment with different analysis block sizes.
Have fun!
- Robert