
[Sponsors] 
April 10, 2017, 10:21 
Wrong Stouhal  vortex shedding

#1 
Member
Fredi Cenci
Join Date: Dec 2016
Posts: 30
Rep Power: 2 
Hello all,
I am performing a simulation in a 2d cylinder using PISO and komega SST. I am getting good results for lift and drag, close to experimental results, however the frequency is bad. THe St should be around 0.2 but i am getting 0.13. I saw that it could be the relaxation factors that i used and PISO doesnt use, however if i remove them the st goes to 0.25 and lift and drag coefficients go up... Following is the link with my files.. https://www.dropbox.com/s/yqvex3mcld...er.tar.gz?dl=0 does anyone knows where i am messing up? THanks a lot 

April 11, 2017, 01:20 

#2 
Senior Member
Uwe Pilz
Join Date: Feb 2017
Location: Leipzig, Germany
Posts: 141
Rep Power: 3 
Dear Fredi,
I looked at your simulation. Your Re number is 10000, so the expected drag coefficient is in the area between 1.05 (Sighard Hoerner, 1965) and 1.16 (Hermann Schlichting, 1951). You got around 1.03 which is in that range. You use RANS for determination of viscosity. This method is applicable to predict total forces like drag and lift coefficients at not too high Re numbers. It is of lesser value if you need to know what happens in the boundary layer. The Strouhal number results from a flow separation in the boundary layer. To get an idea of it you should use a more advanced turbulence model. Unfortunately, the most common LES models tend to introduce to much energy in the vortices. This leads to a prediction of forces which is too high. In other words: It is very hard to predict the lift/drag coefficients and the Strouhal number with one model. ~ I saw that you use first order schemes. This is not recommended because they are too diffusive. The problems rise with the turbulence, and Re = 10000 in not a unproblematic case. ~ You should assure that your results are confident, which is a task by it's own. There is no error control in CFD, so you have to assure that your results are acceptable with all means you have. Whether the computed drag coefficient is 1.0 or 1.1 is not most important. A deviation from the experimental results is even probable, because the simulation assumes an infinite smooth surface, what is different from reality. It would be much more of interest whether the changes calculated values with the Re number are as expected. I recommend the results from Schlichting for that which cover a very wide range of Re numbers, including the drag crisis.
__________________
Uwe Pilz  Sie ahnen nicht, wieviel Poesie in der Berechnung einer Logarithmentafel enthalten ist (Carl Friedrich Gauß) Last edited by piu58; April 11, 2017 at 02:28. 

April 11, 2017, 01:49 

#3 
Member
Fredi Cenci
Join Date: Dec 2016
Posts: 30
Rep Power: 2 
Fist of all thanks for take your time and help me out. I really appreciate this.
1 According to this threat: Relaxation Factors for Transient solvers I must not have relaxation factors in PISO. So I removed them. 2 Yes, I am not trying to reach all the values of experimental results, i am just trying to perform a simulation right, and there was a disagreement between two works that I was following. First the PhD theses from Rosetti 2015, where using Refresco code and the same turbulence model he found the values attached, which are high compared from experimental. Second a paper, Stringer (2014), where using OF 1.7.1 he found much better results. Pictures attached. Conclusion: Removing the relaxation factor my results seems to go in Rosetti's direction and I was wondering why. I guess it is because of the OF version, but i am still not sure. Thanks for the tip about the numerical schemes.. what would you recommend? https://cfd.direct/openfoam/usergui...s/#x201130159 .. leastSquare,Gauss linearUpwind grad(), and which Laplacian scheame? Thanks for you time =) 

April 11, 2017, 03:37 

#4 
Senior Member
Uwe Pilz
Join Date: Feb 2017
Location: Leipzig, Germany
Posts: 141
Rep Power: 3 
The flow past a cylinder is a often used model geometry. The results differs widely, the two articles you mentioned are not the only results.
I looked at both. The Stringer works shows a drag coefficient which gets lower with Re. This is not what the experiments give: A more or less constant value around 1 between 400 and 100000, and then a sharp decreasing down to 0.3 at Re ~ 300000 followed by a continuous increasing. You find experimental values at page 121 in the work of Rosetti. Rosetti's results show a decreasing drag coefficient form Re=10000 to 1 Mio, similar to Stinger. Both results do not verify the experimental results. You may find some other articles which come to similar numerical values. Conclusion: The numeric used is not suitable for this question. I don't see any value in imitating one of the calculation procedures. I found only one work which predicts the drag crisis more or less correct: Boterill, Morvan, Owen: Investigation into the numerical modelling of the drag crisis for circular cylinders (2009). And even here the numerical results needs to be "corrected" in some way (correction for inlet turbulence, which I did not understand to full extend). The most important question is, what do you want to get from your simulation. Then you have to look for the numerical tools which may be able to give you that kind of answer. Studies of uncertainties and numerical "effects" (=errors) are a large part of such a simulation.
__________________
Uwe Pilz  Sie ahnen nicht, wieviel Poesie in der Berechnung einer Logarithmentafel enthalten ist (Carl Friedrich Gauß) 

April 11, 2017, 08:48 

#5 
Member
Fredi Cenci
Join Date: Dec 2016
Posts: 30
Rep Power: 2 
Yes, i totally agree with you. I know that using URANS and komega SST it is not gonna predict correctly the drag, lift, pressure and etc. The paper from Boterill, Morvan, Owen (2009) they used LES.
My goal here is to learn how to perform correctly the simulation. My concern is how Stringer got a Cl_rms value of close 0.4 using OF v1.7.1 and I using OF v4.1 get 1.2. I am close to Rosetti's value of 1.1. All of us is using Komega SST. So I was thinking that my values should be closer to Stringer' because it is the same software. This leads me to think that I have something wrong in my setup.. But as you look at my files and think that it is fine I will follow like this and see how the simulation goes. Look the results for my medium mesh at the pictures, they are without relaxation factors. Thanks for looking after that so much. 

April 11, 2017, 11:17 

#6 
Senior Member
Uwe Pilz
Join Date: Feb 2017
Location: Leipzig, Germany
Posts: 141
Rep Power: 3 
The first thing I recommend for change are the schemes. I am sure that Stringer did not use first order schemes.
My (personal tinted) recommendation: Use Gauss limitedLinear 1: The "1" means no limit = maximum accuracy. If you get problems with stability, reduce the 1 by 0.7 ... 0.5. If you need lower values something is wrong with your mesh.
__________________
Uwe Pilz  Sie ahnen nicht, wieviel Poesie in der Berechnung einer Logarithmentafel enthalten ist (Carl Friedrich Gauß) 

Thread Tools  
Display Modes  


Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
No Vortex shedding for diameter of 20mm in flow past cylinder  shashanktiwari619  FLUENT  0  January 11, 2017 22:03 
How to know when a vortex shedding pattern has been onset?  HectorRedal  Main CFD Forum  18  January 5, 2017 07:42 
No Vortex shedding for flow past cylinder  shashanktiwari619  FLUENT  7  December 28, 2016 10:35 
Flow Past a Cylinder, the Onset of Vortex Shedding  Nauseouslyquick  Main CFD Forum  3  October 23, 2016 16:46 
instantaneous frames of vortex shedding  jamal  FLUENT  1  January 24, 2008 10:57 