[Julian121]

I am doing a FFT analysis of a pressure signal recorded by a probe.

In the attached pressure signal, to exclude the effect of transients should the full data be analysed or some initial data should be removed?

I found out that if I exclude the first 0.002s or 0.005s data, the frequencies and their powers change significantly.

What is the correct method to capture accurate frequencies?

[evcelica]

That is up to you.

If the flow was developing in the beginning of the simulation, and not true to reality, then I would exclude it.

If that will be part of reality, then include it.

I'm guessing in the beginning you just started from some simplified initial conditions? So the flow field there was still developing, and not actual reality?

[Julian121]

Originally Posted by evcelica

I'm guessing in the beginning you just started from some simplified initial conditions? So the flow field there was still developing, and not actual reality?

Yes, I started from an unconverged steady state solution.

How should I judge whether the flow is still developing or is actual reality?

Convergence to 1e-4 is achieved in every time step except the initial 10 time steps.

[evcelica]

I would have at least started from a converged solution.

Your flow was most likely developing if it skews your FFT badly. Run the simulation for more time, and see if that changes the FFT greatly. If it doesn't,then it sounds like the beginning has different dynamics than the rest, So it should be excluded.

From your graph, it looks like You finally have a stable pattern after 0.018s. I would continue your simulation for more time, just start from your previous results, and see if that pattern continues, then cut off everything before 0.018s or so.

[Julian121]

This time, I started with a better initial condition at a different mass flow.

The FFT at the end of 0.0409474s and 0.0492807s including full data were taken.

The FFT at the end of 0.0492807s excluding initial 0.02s is shown.

There is slight difference between the frequencies.

Do you think any initial time should be excluded here?

[TENE]

Hello

please, what is your process (script) to caculate the PSD

Patrick.

[Julian121]

Originally Posted by TENE

Hello

please, what is your process (script) to caculate the PSD

Patrick.

You can do it in CFD-Post or matlab.

[TENE]

Thank you for your answer,

I did it on matlab, but I’m not sure my script is good.

N = length(x);
Fn=Fs/2
FTx= fft(x-mean(x)); %fft(x);
FTx= FTx(1:fix(N/2)+1); %uniquement les N/2+1 valeurs sont nécessaires pour la PSD
PSDx = (1/(Fs*N)).*abs(FTx).^2; % calcul de la PSD
PSDx(2:end-1) = 2*PSDx(2:end-1);
freq = 0:Fs/N:Fn;

figure
semilogx(freq,10*log10(PSDx))

x represent my data and Fs is the sample frequency.

Patrick.