[CaptainCombo]

Hi,


I am simulating flow (Re = 525) in a channel that it's upper wall is grooved, it's length is 1.2m and the gap between upper and lower walls is 2.32cm. I made channel width 2.5mm and applied periodic boundary conditions in spanwise directions. Almost 5M hexa mesh was used for this geometry. LES (with dynamic stress) was employed for turbulence modeling. Time step size was equal to 0.0001. My purpose in using LES was to see the change of the instantaneous flow field in time but obtained the same flow field data at every instant. Also, I monitored velocity data at different points and it stopped to change remained constant after a while.


Can you give me some advice? What do you think he problem is?

[FMDenaro]

Originally Posted by CaptainCombo

Hi,


I am simulating flow (Re = 525) in a channel that it's upper wall is grooved, it's length is 1.2m and the gap between upper and lower walls is 2.32cm. I made channel length 2.5mm and applied periodic boundary conditions in spanwise directions. Almost 5M hexa mesh was used for this geometry. LES (with dynamic stress) was employed for turbulence modeling. Time step size was equal to 0.0001. My purpose in using LES was to see the change of the instantaneous flow field in time but obtained the same flow field data at every instant. Also, I monitored velocity data at different points and it stopped to change remained constant after a while.


Can you give me some advice? What do you think he problem is?

Have you evaluated first the Re_tau number of your case? If your case reach a steady state you could first try to perturbate the initial condition. Then, if you still get a steady state check the velocity profile and see if it is a parabolic profile (approaching the Poiseuille solution).

Have also a check to the values of the dynamic SGS model, it could happen that you have too eddy viscosity.
I suggest also to perform a run on the same grid without any SGS model.

[CaptainCombo]

Originally Posted by FMDenaro

Have you evaluated first the Re_tau number of your case? If your case reach a steady state you could first try to perturbate the initial condition. Then, if you still get a steady state check the velocity profile and see if it is a parabolic profile (approaching the Poiseuille solution).

Have also a check to the values of the dynamic SGS model, it could happen that you have too eddy viscosity.
I suggest also to perform a run on the same grid without any SGS model.

Thanks for the answer. Before proceed I wonder that even if the Reynolds number is low, don't I have to see a change in instantaneous data? How every instant can be exactly the same as other? We made PIV experiments for similar channels(not exactly same geometry) and low Reynolds numbers. We can observe for example a variation in velocity vector field in time.



According to the paper(experimental study) that I take the geometrical model, natural frequency is 0.31 for the same flow conditions. I checked velocity profile and confirmed parabolic velocity profile. I disabled dynamic stress and have just started the simulation. I will share developments here.

[FMDenaro]

Originally Posted by CaptainCombo

Thanks for the answer. Before proceed I wonder that even if the Reynolds number is low, don't I have to see a change in instantaneous data? How every instant can be exactly the same as other? We made PIV experiments for similar channels(not exactly same geometry) and low Reynolds numbers. We can observe for example a variation in velocity vector field in time.



According to the paper(experimental study) that I take the geometrical model, natural frequency is 0.31 for the same flow conditions. I checked velocity profile and confirmed parabolic velocity profile. I disabled dynamic stress and have just started the simulation. I will share developments here.

If the Re number is too low you could reach a steady state. A quantitative measure cannot be the comparison of the fields between two time steps but you need to evaluate the time derivatives in any node.

[CaptainCombo]

Originally Posted by FMDenaro

If the Re number is too low you could reach a steady state. A quantitative measure cannot be the comparison of the fields between two time steps but you need to evaluate the time derivatives in any node.

If field variables do not change with respect to time what is the purpose of evaluating time derivatives of them? Time derivatives will be same for every two-time step right?

[FMDenaro]

Originally Posted by CaptainCombo

If field variables do not change with respect to time what is the purpose of evaluating time derivatives of them? Time derivatives will be same for every two-time step right?

Have you measured that the difference of the velocity in all the nodes is exactly zero? Otherwise a small variation that is divided by the small time step appears as a non vanishing time derivative.

[LuckyTran]

Is this a laminar case or turbulent case? That kind of Reynolds number is laminar isn't it? Then you shouldn't expect any change in time.

Look at the instantaneous flow field. Does it look laminar?

Also the channel width is way too small to support any turbulence.

[CaptainCombo]

Originally Posted by FMDenaro

Have you measured that the difference of the velocity in all the nodes is exactly zero? Otherwise a small variation that is divided by the small time step appears as a non vanishing time derivative.

It seems I need to implement UDF to do that. It will take a while.

Originally Posted by LuckyTran

Is this a laminar case or turbulent case? That kind of Reynolds number is laminar isn't it? Then you shouldn't expect any change in time.

Look at the instantaneous flow field. Does it look laminar?

Also the channel width is way too small to support any turbulence.

The paper I refer reported the 0.31Hz natural frequency that's why I am expecting change. I kept channel width so small to reduce solution time. As I mentioned, periodic boundary conditions are applied in a spanwise direction. Just like solving 2D simulation.

[LuckyTran]

Originally Posted by CaptainCombo

The paper I refer reported the 0.31Hz natural frequency that's why I am expecting change. I kept channel width so small to reduce solution time. As I mentioned, periodic boundary conditions are applied in a spanwise direction. Just like solving 2D simulation.

Ok... but is this the result of a unsteady laminar effect or turbulence?

If it's not turbulence then LES doesn't provide anything over unsteady laminar simulation (but it doesn't hurt either). But if it is turbulence... and you provide a domain width which is too small and does not support the length scales (in 3D) of this effect, then you obviously won't see it. If you want to do 2D LES then do so. But you constrain the domain such that it doesn't support the physics then you're out-of-luck.

If it is supposed to be turbulent but the solution relaminarized accidentally, then you need to give it a good kick (perturbations) and help drive it towards a turbulent state.

If this natural frequency is laminar... then you have other unknown problems. But this could also be because it doesn't have the right perturbations to initiate the instability you are looking for.

[FMDenaro]

Originally Posted by CaptainCombo

As I mentioned, periodic boundary conditions are applied in a spanwise direction. Just like solving 2D simulation.

Not at all, periodicity in spanwise direction is not a 2D model.