Hello Chirag,

from the range of your signal which ungergoes immediate changes between 0.2 and -0.2 i would think that it is hardly meaningful to compute the time average.  A negative drag coefficient would also mean  that your object is "pushed forward" by the flow.

What is the cause for the oscillations? Do you use RANS or LES for computation? Do you think the fluctuations are physically motivated? Maybe you could decrease the Reynolds number to see if you can achieve a less noisy signal.  You also should take a look at the solver settings linke timesteps and schemes.

Best regards

Bastian

Hi Chirag Patel

Can you give more details about your simulation setup? Is it a fixed case or is it sort of an oscillatory case in a spring-mass damper system ? How does the CL vs time look like. Has the CL time history reached a steady state. Is it possible to upload the the CL time history?. From the looks of it I think you have to simulate it for a longer time and have to be careful of the special and temporal resolution.

You should explain in more detail the experiment, either physical or computer.

Only then we can try to understand your results.

Good luck.

As pointed out by Bastian, the most strange behavior is the transition from nearly constant value to noisy, high frequency, fluctuation. In these fluctuations, however, I see an high frequency (most probably noise) and a low frequency, both growing.

Are you inner iteration converging? what is the time step compared to the expected oscillation period? For URANS simulations aiming at capturing large scale periodic motion (e.g. vortex shedding) I would aim at having at very least 30 physical time steps per period

Waiting for more detail ;)

Integrate with time from t1 to t2 and divide by the interval.

Hi 

When I looked at your previous works, I understood that you have developed a code for low Reynolds number regime. So I guess you are going to extend your program to higher Reynolds number. I think you are looking for Karman  vortex street in laminar flow. Am I right? If yes, then your problem has not been set up correctly.  If you are looking for help you have to clarify your problem correctly. For example, with your problem explanation here and your background, I think you are looking for Karman street. your chart shows, in my guess, you used increasing inlet velocity condition with time to capture the phenomena at your desired velocity. I think that is why in a period we have a constant Cd. If I am alright all about, I am asking about the noise region at starting time. what is it?

It shows your code and problem definition is not stable in time iterations. 

However, all are the guess and that all because of a not clearly stated issue.

Any way, in general, If you are going to extract the time intervals means and other properties from such noisy outcomes here is the method:

first of all understand the frequencies: in flow simulations we have these frequencies in data:

1- big scale fluctuations and frequencies: they are structural frequencies of the problem like Karman vortex street.  or dynamic boundary induced osilations.

2- medium scale fluctuations: there are due to in domain phenomena. for example some reverse flows due to transient pressure gradient or other forces. some turbulence fluctuations cab be categorized in this.

3- low scale fluctuations: there are due to turbulence fluctuations and noises in numerical procedure and algorithm. To make a difference between turbulence or source induced low scale fluctuations and numerical method noises , you have to analysis their PDF and Frequencies. Then filter the numerical or algorithm induced noises. 

the scale is somehow amplitude here. Since energy is conservative and always limited, high scales usually comes with lower frequencies and vise versa. 

now to analysis such data you can use the Fourier analysis to map the frequency domain. then according to your problem set up and flow structure, you can find the noise frequencies, turbulent frequencies and large scales frequencies. After then you can decompose your data in different scales. As soon you get your filtered data, you can calculate the all time averaged properties, periods, frequencies and etc.

This is the general analyzing approach. In your problem I think you have to take a look again at your simulation set up. 

I hope you find this helpful. 

It seems you need a finer time-step to see if you are getting steady drag oscillations (typically at Sr=0.2=f*D/U), at what Reynolds number is this?

It is a common mistake to see the problem of flow over a cylinder as laminar flow or turbulent flow. The correct way is to consider it as a subcritical regime (Re < 2X10E05) and supercritical regime. In the subcritical regime, the boundary layer will be laminar directly followed by separation, and in the supercritical regime, the laminar is followed by transition to turbulent boundary layer and then separation occurs. The vortex streets appear even before the boundary layer turns turbulent. These facts, though well-known, are easily forgotten, especially for a young student.

Many good suggestions have been given above. I will just add one. Is your scheme satisfying the CFL criterion? In other words, is your time step smaller than the time taken by the flow to traverse one cell?

First of all i am sorry, i realized some day before that i have uploaded cl instead of cd.

i have used Re=100.

i have uploaded correct Cd graph.

Again sorry for inconvenience caused.

Hi, some problems of this sort can only be solved by using massive computational power. You should continue until a statiscical stabilization occurs, but first of all you neee to look WHY Cd becomes negative ... A funny problem, enjoy it! Great chance to evolve your capabilities!

For presumably laminar flow at Re=100 I would expect a periodic signal with a clear shape and  frequency.

In a study about losses in periodically unsteady flow i once had a similar problem with the transient simulation using the piso-algorithm in OpenFOAM. It could be solved by increasing the number of piso-corrector-loops.