Flutter analysis: filtering data

[aa2105]

Hi
I am measuring the bending strain via strain gauges for flutter tests but need a method to filter the data. As the gauges are sensitive the data file contains a lot of 'noise' (its not noise technically, the voltage just jumps up and down by small amounts). How can I filter this data to get the trend and a smooth curve? As I eventually need to use the amplitudes to work out the damping ratio. Thanks in advance.

 

[btrueblood]

"its not noise technically, the voltage just jumps up and down by small amounts"

Hunh? Isn't that a definition of noise, random unexplained jumps up and down in your data? Presumably, if you could explain the source of the jumps, you could write a simple algorithm to calibrate it out of your data. E.g. thermal shifts, or changes in ground states due to operation of other pieces of equipment, etc.

Have you done a FFT or DFT on the data? Is the bandwidth of the noise well below or above the frequencies of interest for you?

 

[GregLocock]

It would help enormously if you could post a plot of the timehistory of the data, without units.

Anyway, assuming it is some sort of high excursion followed by a sinusoidal decay curve over several cycles, just low pass filter it to say 10 times the decay frequency.

If you are suspicious of this, use the unfiltered data to estimate the damping ratio, then LP filter it to 25% of the sampling frequency, estimate the damping ratio, and then 12%, then 6% and so on right down to the frequency of interest. Plot that trend to see how much filtering is acceptable.

Estimating damping ratios is a bit of a black art anyway, non linearities render any attempt at precision dubious, on real structures.




Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[kontiki99]

My memory is a little fuzzy but I think the rule of thumb when I worked in flight test was to use a sample rate at least 5X the highest frequency of interest and a 6 pole filter with the 3 db cutoff frequency 2x higher than your sample rate.

Straight Nyquist theory said that you only needed to over sample by 2x. Still 5x was the rule. I was told it was because we always reviewed the data in the time domain, but I never worked with the 30 lb. brains that analyzed the data so I cannot really say.

You have to supress high frequency noise because it will ailias and corrupt your data if it's sampled under 2x freq.

 

[FeX32]

6 pole really? I have always used 3 pole butterworth without trouble. Were you doing online analysis? If so, then did you have issues with phase distortion? You can reduce the aliasing by filtering before the A/D converter.


Fe

 

[kontiki99]

FeX32,

I can't give you a reason or many details. I was a civil service engineer working on the Boeing/Navy integrated test team for the F18 E/F. They had a very large catalog of standard sensors and modular instrumentation gear that was compatible with the onboard data aquisition system built into the test airplanes. All I remeber was that 6 pole LPF was the standard. I couldn't tell you what kind.

I had been an A&P since the 80s, this was my first EE job. I was told that presample filtering was essential.

 

[FeX32]

Thanks kontiki99.

Fe

 

[aa2105]

thanks guys. unfortuantely, my deadline is on 1st june so dont really have time to implement everything - i guess a rough estimate from the unfiltered data will suffice...