In using the FFT one will have to be aware of such phenomenon as aliasing. To use the method reliably, one will have to impose a cut-off on the Fourier spectrum of the function/signal to be fast-Fourier transformed (in numerical-analysis texts this is referred to as zero-padding; in the parlance of electrical engineering, the signal must first be fed through a low-pass filter) before applying the FFT. Further, in applying the FFT, the function should be sampled at a rate consistent with the Nyquist criterion. For details, you may wish to consult Chapter 12 of the book Numerical Recipes (Cambridge University Press), by W.H. Press et al.  Applying the FFT without due care can lead to introduction of artificial features in the function being studied.

In using the FFT one will have to be aware of such phenomenon as aliasing. To use the method reliably, one will have to impose a cut-off on the Fourier spectrum of the function/signal to be fast-Fourier transformed (in numerical-analysis texts this is referred to as zero-padding; in the parlance of electrical engineering, the signal must first be fed through a low-pass filter) before applying the FFT. Further, in applying the FFT, the function should be sampled at a rate consistent with the Nyquist criterion. For details, you may wish to consult Chapter 12 of the book Numerical Recipes (Cambridge University Press), by W.H. Press et al.  Applying the FFT without due care can lead to introduction of artificial features in the function being studied.

Aliasing could be one cause of false peaks. If you used an A2D converter, check its specs to see if there is an anti-aliasing filter incorporated in the front-end. I know this is the case for most of NI's DAQs. But if you have captured the data without such a filter, then no amount of post processing (e.g. zero padding) will be able to remove the false artefacts.

The false peak(s) may also be due to side lobes caused by the finite (rectangular) time window applied when you collected the data. If the true signal spectrum contains only a single line, it will be convolved with the spectrum of the window function (the Dirichlet kernel), which is not pretty. To reduce the side-lobe level, you could apply a tapered window function before you take the FFT, but this will also broaden the main-lobe.

Note that when you take the FFT (eg using the Matalab command) you will get +ve and -ve frequencies, however the 0th frq bin (i.e. DC) will be at the first index. If your two peaks are of equal size then this could be the cause. Make sure you have mapped the FFT bins to frequency properly.

Of course there may also be unexpected harmonics in your data. Or perhaps noise if you used a short time window and did not employ any averaging.

By the way, I have no understanding of the particular physical phenomenon you are trying to understand. These are just some comments on short-time frequency analysis in general. 

The above answers are true for aliased data. Usually though Ni DAQs will have anti aliasing built in to avoid fold back. Generally when measuring cylinder wakes with a Hotwire, unless you have a very high speed flow and low sampling rate the results will be okay ( provided you have post processed the data correclty). The dual peaks are more likely to be caused by the fluid dynamics of the problem. Generally you will measure two harmonics as well as the fundamental vortex shedding frequency. This is due to oscillating drag and the harmonic of lift. Alternatively, if you are measuring the signal near an end plate, you may be measuring a different cell of vortex shedding caused an oblique shedding mode. See Williamson 1986 for details. 

You may also find some addtional explanations in my paper.

JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY

Volume:22,Issue:7 Pages:988-1003

DOI:10.1175/JTECH1746.1

Published:JUL 2005

I analyzed flow behind shield of the ultrafast thermometer.

Could you please post the figure of the spectrum? Previous answers addressed problems in doing FFT but if you are solving high Re number flows, many characteristic frequencies can also appear..

I agree with the previous answer; the double peak could result from a physical phenomena linked to the strouhal number (with harmonics)  based on the characteristic dimension of the rod. Sometimes the vortex shedding also induces an additional vibration of the rod itself which leads to an additional peak at this natural frequency

One of the peaks most probably corresponds to the most energized structures which I would suppose the vortex features. The frequency at which this peak occurs shall match the average length of the vortex because your presumably single hot wire data measure fluctuations of flow in a line. The length usually scales with the most common geometry of your installation such as the circular cylinder. The other peak might correspond to system inheritance discussed earlier. Based on experience, a hot wire can also collect noise produced by the electrical wiring, vibration in the machine, and reaction in the traverse-flow mechanism although hard-wired filters are installed. In order to overcome this, just have your FFT ready during the measurement itself, and check the possible cause one by one, for example, for the wiring problem, test if the other peak still exist if a proper ground has been provided etc.

The frequency of shedding, according to Strouhal, is S=f D/V.

The second could be the first armonic mode f=v/2L with v=sqrt(T/u) being T the tension of the wire and u the linear density of the string [kg/m].

As stated by 

Filippo Maria Denaro

we need to see a spectrum and time signal of the hot-wire in order to be aisle to see the kind of phenomenon your are concerned with. Note for example that a simple hot wire does me sure the magnitude of the flow velocity, not the direction. This implies that a vortex passing along a hot-wire will generate a lot of harmonics of the fundamental vortex shedding frequency. Next to problems with vibrations, it is well known that transversal waves in the wake of a cylinder exist and depend strongly on the length/diameter ratio of the cylinder and on how the terminations are made (simple termination or wall???). So globally the wake of a finite cylinder in cross flow is quite complex and the frequency can display jumps and hysteresis, which also can depend on the environment. Actually your hot wire can disturb the flow!

So we would need a description of the set-up in order to judge.

I need to see the velocity  spectrum and time signal.  But, there are, for circular cylinders, a fundamental frequency and the harmonical frequencies.  In rectangular cylinders is possible to measure two fundamental vortex frequencies in the same Reynolds.  In circular cylinders only one fundamental frequency is possible for a determined Reynolds.

There are many possible and correct reasons mentioned in this forum. If the question still remains unanswered then you should try to figure out if both frequency peaks move to the same direction over flow variation. If so the lower frequency peak could be the desired vortex shedding frequency because vibrations of solid materials should be higher. If only one peak moves over flow variation - congratulation!

@Filippo: Thank you for the comment, the Re is about 60000, also I attached the FFT that i got for data obtained at x/D = 2 and y/D = 0.25 downstream of the cylinder, as you can see I got wo peaks one about 38Hz and the other about 75Hz. 

As the ratio of the two frequencies is about 2 (75/38=2) I expect that 38Hz is your vortex shedding frequency. The Strouhal number Sr=f D/U should be about 0.2. If this is verified then f=38 Hz is indeed your vortex shedding frequency. The hot-wire measures the a signal determined by the amplitude  sort (U_x^2+U_y^2) of the velocity normal to the wire. Even if the signal is linearised by calibration of the hot-wire the hot-wire does not distinguishes between positive or negative velocities (in the main flow direction). Also the flow in a vortex is complex (not purely sinusoidal). Consequently the signal of the hotwire contains second harmonics of the fundamental frequency. You probably also observe peaks around 114 Hz and 152 Hz.

@Avraham: The inlet velocity was about 15m/s and the cylinder's diameter was 0.06 so I was expecting the shedding frequency of about 50Hz, however as the plot shows I got peaks around 38 and 75 which is quit strange!

Around Re=6E+4 we would expect Sr=0.19. So your Sr=0.15 based on f=38 Hz is quite strange.

Is your cylinder accurately positioned normal to the flow?

How long is your cylinder?

How is the cylinder terminated at both ends?

Is your cylinder rigid and are the fixations rigid?

@Hugh: I used no filtration/windowing on my data before applying FFT (also I checked the DAQ) also as the plot shows +/-ve is not the case since the peaks have different magnitudes.

@Avraham : it was positioned normal to the flow in the test section rigidly, the length of it was 0.5m and was terminated by end plates

I don`t think filtering before doing FFT is necessary because you can do filtering in the frequency domain. Filtering before doing FFT just leads to different amplitudes of your spectrum. And perhaps you apply phase shifting with filtering before FFT.

@Ric: thank you for the comment, in my case Re is about 60000 and the sampling rate is set to 30kHz and the measurements have been done at the x/D = 2 and y/D = 0.25

So the question for me is not why you have two peaks, but:

why the vortex shedding frequency is so low?

Possible reasons for such a behaviour might be:

-cylinder is not normal to the flow.

-cylinder attachment is elastic with a resonance frequency around 38 Hz.

-acoustic resonance of the room in the direction transversal to the cylinder (the cylinder is placed just halfway between two hard walls distant from each other by 4.5 m) close to 38Hz.

-the cylinder terminations interact with the flow.

Some suggestions:

1) use the log-log plot for the energy spectrum

2) assuming your sampling rate is 30kHz, you should have dt=pi/f = O(10^-4) s, right?

Now I think you should proceed by assuming that your signal is periodic over some period T that you should suitably estimate. Then, you have to do several FFT over blocks of signal and performing a statistical average. From the look of the plot you have considered a single sample of the time-signal

To me it looks like a harmonic of the fundamental vortex shedding frequency. Something that might help is having a look at the time domain! If you have a strongly periodic signal it will be easily identifiable. Also I agree with Filippo, perform some averaging to smooth out your spectra, it looks like your frequency resolution is too high.

@Efstathios, thank you for the comment, it seems reasonable and useful!

The Axial position of the hot wire probe depends on the Reynolds number. For laminar vortex shedding (Re

Not sure if anybody still is interested in this, but the reference attached below can answer this question very well.

https://ac.els-cdn.com/S2095034915300118/1-s2.0-S2095034915300118-main.pdf?_tid=f1a52b04-f744-11e7-a167-00000aacb360&acdnat=1515726294_ba18c249227cacd0a3ef6878d8a67625

I suggest loglog plot see base frequency and then peak. What is length of signal...