How can I achieve to show the natural frequencies with FFT on sensor data?

Hello,

I have a dataset of timeseries of digital inclinometers and I want to extract the natural frequency of the structure. Currently, my code is where sgfxM8 is the filtered x inclination :

 xi=sgfx-mean(sgfx)
   freq = 0:Fs/length(xi:Fs/2; %frequency array for FFT
   xdft = fft(xi); %Compute FFT
   xdft = 1/length(xi).*xdft; %Normalize
   xdft(2:end-1) = 2*xdft(2:end-1); 

but the result is very close to 0 Hz. Is there any other option? Thanks in advance.

\

The frequencies are between 0-0.002 Hz.

[David Goodmanson]

HI stamatia,

Could you provide Fs and the number of points in the transform? Did the data record actually go out to 2e6 seconds, which is roughly a month? Whatever the scaling, anything useful will be found by zooming out the frequency region near the origin.

The line for calculating frequencies is missing a parenthesis.

[stamatia panagiotopoulou]

Hi David,

Thank you for helping. The data are with sampling rate 1/minute and the Fs=0.0165. For your convenience attached you'll find the data.

Thank you again.

[stamatia panagiotopoulou]

Hi

[David Goodmanson]

Hi stamatia,

I have a lot more questions than I do any kind of answer. First, your time array is not uniform but appears to have 20 hours of data at 1 sample per minute and then 4 hours of data at one sample every 10 seconds. Then it repeats. 2640 data points per day. To see this, try (M is shorthand here for your actual file name)

plot(diff(M(:,1)))
ylim([1 75])
xlim([1 5e4])

which gives sample interval for each pair of data points. I put in the x and y limits to temporarily block out some bad sample spacings to make things easier to see, including a really bad one up around sample 7.6e4. Nonuniform spaciing is of course not what you want for an fft.

The data looks mostly like noise, with varying amplitude. Do you know what the 'bursts' are, which happen once each day? Human activity? Is there a range of frequencies you are expecting for the natural frequency of the structure?

[stamatia panagiotopoulou]

Dear David,

Thank you for your time and you are correct to all your comments. The noise you observed is from vechiles and the the data recording is not uniform for monitoring reasons (observe traffic periods). Yes, I am excepting low frequencies but not the exact frequency. I probably did not apply correct the median filter to remove spike noise at first.

Do you have any suggestion on that? I think I have to remove all the data are 3 times the standard deviation and then apply the median filter. Could you advise about that?

Thank you again.

[David Goodmanson]

Hi stamatia,

First, what kind of frequencies are you looking for? With a 10 sec sampling interval (and only for part of the time), the maximum frequency you can get to is, realistincally, about 0,02 Hz, if that.