10 Replies Latest reply on Sep 10, 2014 12:08 AM by Jared Conway

# Cylinder drag at Re5000 and mesh dependance

Good afternoon everybody,

I m trying to assess the effect of mesh siwe on the calculated drag of a circular cylinder in a 2D flow.

If I use :

20 mesh per cylinder diamater, I get Cd=0.90

40 mesh per cylinder diamter, I get Cd=1.2

80 mesh per cylinder diamter, I get Cd=1.4

(more or less).

Correct Cd value is very close to 1.0, as found in literature.

Meshing considered here is not refined but just gently stretch far from the cylinder.

Domain is size around the cylinder : 10D upstream, 20D downstream (in the wake), 10D across. It seems there is no big difference if I use 5D  upstream, 10D downstream , 5D across. So I guess the boundaries are far enough from the cylinder.

I tried different diameters (0.005 and 1m) with same Re, even water and air, I always get the same behaviour.

I am quite surprised about these results, considering the document I found here (page 14) where they use same kinds of meshing (but at bigger Re 3.7e5) with much better consistency.

Thanks a lot for helping !

Benoit.

• ###### Re: Cylinder drag at Re5000 and mesh dependance

check the tutorials and validation problems

this is the exact problem that they use to explain those results, my guess, vortex street

also note, in 2D, your 5D is way too big. it should be really really really thin.

• ###### Re: Cylinder drag at Re5000 and mesh dependance

Hello,

May be I was unclear about the hight of the cylinder in the 2D model, it is much smaller than 5D (5D is half the size of the domain in the cross direction). However the hight (I tried different values) has no influence on the Cd value (fortunately !).

I am aware of the effect of too rough meshing on the solution, but my question was more : in the document enhanced-turbulence-modeling.pdf how do they get so stable Cd with 20, 40, 80 mesh per diameter and why I can't get same result for lower Re. In the doc, I quote (page 15) " In all these calculations the thin-boundary-layer turbulence model was employed". As far as I know there are no option related to wall function or boundary layer modelling in FlowSin. Am I missing something ? Are we talking about the same software package ?

• ###### Re: Cylinder drag at Re5000 and mesh dependance

As far as I know there are no option related to wall function or boundary layer modelling in FlowSin. Am I missing something ? Are we talking about the same software package ? > same software, there are no differences between versions.

have you already looked at the technical reference that i suggested? you'll see they use a transient analysis and also the solidworks KB has a lot of good suggestions about this type of problem (which are probably repeated here in the SWX forum as well).

• ###### Re: Cylinder drag at Re5000 and mesh dependance

Benoit,

A.  Good work on the analysis.  Is your Cd value correct? - my books show a higher value - see attached.

B.  A wide cylinder will amplify any difference as apposed to a very narrow cylinder.

C. How do you know you have enough mesh?  You should keep increasing mesh until - increasing mesh only produces a small change in results.

D. Consider increasing Tolerance refinement.

• ###### Re: Cylinder drag at Re5000 and mesh dependance

Good afternoon,

thank you for helping.

A - Yes I believe my Cd at Re5000 is correct. My graph (the same than yours) is from Hoerner, 1958. Values of Cd between Re=10e3 and about 6.10e3 are very close to 1 (not 1.2 which is Cd between 10e4 and 2.10e5).

C - do I have enough mesh ? I have just followed the "cylinder drag tutorial". What I did :

make a copy of C:\Program Files\SolidWorks Corp\SolidWorks Flow Simulation\Examples\b2 - drag coefficient\ready to run\cylinder 0.01m\cylinder 0.01m.sldprt

select the Re1000 project

run the simulation

I get Cd history (figure CDhystory1) :

Here calculation stops because we have reached 4 travels, not because of convergence (but this is the tuto). However, convergence is correct.

Notice their is a big difference between the "Cylinder drag" tuto of 2012 and the 2014 one.

In 2012 they found Cd=1.05 which was good but obviously the mesh was not refined enough.

However,  from the 2014 tuto, if you refine the mesh at level 3, average Cd will increase to 1.5. I did not tried more meshes as it would not fit with my final application which involves many cylinders.

Level 3 result, (I think mesh refinement should be recalculated  quite often (not each travel) because solution is mesh-dependent (this is the discussion !).

So my conclusion : with 10 meshes in the initial mesh and refinement level 2 (the tuto) you get 40 meshes per diameter. This is not enough in terms of "consistency" (we have not converged yet to the solution) and this provides a Cd value far from the expected value. And from this conclusion I really wonder how they can get so good results at higher Re  in

• ###### Re: Cylinder drag at Re5000 and mesh dependance

are you saying you setup the problem exactly as described in the tutorial/validation example and as the document and you aren't getting the right results? please post the models

• ###### Re: Cylinder drag at Re5000 and mesh dependance

No problem with the tuto model, I get the same result (Cd=1.34).

• ###### Re: Cylinder drag at Re5000 and mesh dependance

so in your opinion the tutorial is incorrect relative to physical and isn't the same model as the one used for the article. is that what you are trying to point out?

• ###### Re: Cylinder drag at Re5000 and mesh dependance

Yes the Re1000 solution calculated in the tutorial overestimates the expected value of 34%. Moreover this overestimation goes to 50% if you increase the number of meshes to level 3 (instead of level 2 in the tuto) and maybe even more if you try level 4.

Regarding this article, I cannot understand how they get these good results (Cd=1.02 which is about the expected value) with 80 meshes in the diameter as the Re considered is 370 times higher.  The tuto at Re=1e5 (a Re  in the same order of magnitude) provides a Cd = 0.73 against 1.2 for the expected value (underestimation of 40%). So this article should be detailed.

• ###### Re: Cylinder drag at Re5000 and mesh dependance

have you setup the example from the article?