[FMDenaro]

Originally Posted by danny9019

This picture FMDenaro

I don't know why but I don't see any figure...

[danny9019]

https://www.cfd-online.com/Forums/at...1&d=1528969754

Now?

[FMDenaro]

Originally Posted by danny9019

https://www.cfd-online.com/Forums/at...1&d=1528969754

Now?

it appears:


"Invalid Attachment specified. If you followed a valid link, please notify the administrator"

[LuckyTran]

I don't see any pictures at all.

Originally Posted by danny9019

In order to do this, i have to split the sphere in the regions where y+ is less than 1 and more than 1, right? In order to construct two different boundary layer?

Thanks again

Star-CCM doesn't like to let you change the prism layer thickness to scale with the local BL thickness. Your best bet is to have a ton of prism layers everywhere. This means where the BL is thicker you will have a lower wall y+ and where the BL is thinner you end up with a higher wall y+ and you have to compensate.

What about the wake region? Do you resolve the wake sufficiently to get the proper contribution from the form drag?


I would also drop the stretch factor to 1.3 or even 1.1, but this depends on how much you need your prisms to stretch to match the core mesh.

[JBeilke]

You apply 2 very strong simplifications. Just see what you get when you

1) swich to a transient calculation
2) remove your symmetries one by one

It is completely useless to use a very fine mesh when everyting else is murky.

[FMDenaro]

Originally Posted by JBeilke

You apply 2 very strong simplifications. Just see what you get when you

1) swich to a transient calculation
2) remove your symmetries one by one

It is completely useless to use a very fine mesh when everyting else is murky.

Well, actually I do not agree. Until a RANS formulation is used, the solution is statistically steady and half sphere is justified by the symmetry in the solution.


If you want to switch to unsteady solution one should use LES (or DES)

[JBeilke]

tranisent RANS == URANS

It doen't matter how you call the numerical algoritmus (RANS, LED DNS, ...).

At first we have to think about the flow in reality and this one is very likely not steady sate.

[FMDenaro]

Originally Posted by JBeilke

tranisent RANS == URANS

It doen't matter how you call the numerical algoritmus (RANS, LED DNS, ...).

At first we have to think about the flow in reality and this one is very likely not steady sate.

It is true that the flow is unsteady in its pointwise solution but we are talking about a statistical solution, that is a solution of the statistically averaged equations. This way, the statistically averaged field is certainly steady. I don't think that URANS makes physical sense when the flow has no external time dependent force as in this case. LES and DNS are the only formulations that makes physically meaningful the numerical unsteady solutions.


"It doen't matter how you call the numerical algoritmus (RANS, LED DNS, ...). "


and this sentence is not true at all..

[JBeilke]

When the flow is transient in reality we have to use a transient approach. Thats it. Up to this point we dont need any formula or numerics. This comes much later.

We also see a karmann vortex street when using URANS. I know that some theorists like this discussion :-)

[FMDenaro]

Originally Posted by JBeilke

When the flow is transient in reality we have to use a transient approach. Thats it. Up to this point we dont need any formula or numerics. This comes much later.

We also see a karmann vortex street when using URANS. I know that some theorists like this discussion :-)

No at all... if the statistical averaging on the equations produces a modified set of equations where time disappear and the flow has a statistically steady state of the kinetic energy (production balance dissipation).
That is at the basis of all the hystorical law deduced for the turbulence like the wall laws, despite that the turbulent flow over a wall is unsteady and separated in its pointwise nature.


URANS has a specific meaning, the averaging is still time-dependent owing to an external force that introduces a time-dependent production of kinetic energy. For example the motion due to a piston.
A further meaning in the URANS can be associated to a time-filtering, that is the time averaging is at a certain time period tha filters out higher frequency. This is somehow more realted to the LES concept.

[JBeilke]

This LES discussion comes as the very last step.

When we want to find out why a simulation gives strange results we have to discuss the physics at first. At this point you use words like steady or transient. This is the most important step for any simulation.

Maybe you know the words "Gerechnet ist dann schnell".

Then we should discuss the size of the domain and what we know about the boundary conditions.

When we are fairly sure that we understand most of the effects which are to expect, we can start to think in terms of equations. Navies-Stokes might be the right one.

So when we decide that the flow is transient, we know that there is no real symmetry to expect. It might give reasonable results to use a symmetry plane but we can not be sure.

Now we might think about what type of transient approach we want to use. And instead of floating a superomputer with an LES we might start with a URANS to see what we get there. The next step would be something like an DES/LES.

[FMDenaro]

Originally Posted by JBeilke

This LES discussion comes as the very last step.

When we want to find out why a simulation gives strange results we have to discuss the physics at first. At this point you use words like steady or transient. This is the most important step for any simulation.

Maybe you know the words "Gerechnet ist dann schnell".

Then we should discuss the size of the domain and what we know about the boundary conditions.

When we are fairly sure that we understand most of the effects which are to expect, we can start to think in terms of equations. Navies-Stokes might be the right one.

So when we decide that the flow is transient, we know that there is no real symmetry to expect. It might give reasonable results to use a symmetry plane but we can not be sure.

Now we might think about what type of transient approach we want to use. And instead of floating a superomputer with an LES we might start with a URANS to see what we get there. The next step would be something like an DES/LES.

While I agree that one needs to see the physics of the problem, steady and unsteady terms are quite general as they simply state that the time derivative of a variable is or not zero.

Therefore, there is no discussion about the fact that any physical variable is unsteady and 3d in turbulence. That means also the flow over a wall, in a pipe and so on.
But hystorically, the unsteady behavior of the pointwise variables is "cancelled" by the Reynolds averaging based on the time integral. And all the hystorical theories about the law-of-the-wall are based on the steady statistical analysis because we transformed the original set of equations in a statistically averaged one. When we study the turbulent flow in a channel we can visualize the flow and it appears full of vortical unsteady structures. But in terms of the statistic, it is steady and we can study only a half of the channel height. The same happens for the flow over a cylinder, the unsteady vortex shedding disappears when we statistically average the solution in time and we can use half cylinder. That does not violate any physical principle, simply we are solving for the statistical variables.
I know that often URANS is used to approach the vortex shedding analysis. This is a trick that engineers use to produce some nice time-dependent plot. However, the existence of a time dependence in the statistically averaging is not justified if there is not external time-dependent force. Or you are not using a statistical averaging but a different definition of the variable. That is relevant when one wants to compare the numerical solution to experimental measures that are statistically averaged.


I strongly suggest to read the introduction in this paper of Spalart et al.
https://www.researchgate.net/publica...LIKE_BEHAVIOUR

[piu58]

I first wrote in the wrong thread with the same title. Soryy for double post:

There are much publications on spheres or cylinders (2D) in free stream. If you read them and compare the results with the experimental data given by Hermann Schlichting: Nearly no simulation covers such large Re numbers with sufficient accuracy.

Especially the drag crisis (which has a lot of applications, starting form a golf ball) is not captured.

[danny9019]

FMDenaro is right