Flow over 2D cylinder (Reynold 10^6)

mineralwater7 · June 16, 2018, 10:13

Hi all,

I am working on the simulation case which describe the flow of water over a 2D cylinder. Diameter of cylinder is 1m. The Re=10^6. Velocity=1m/s.
The goal is to find out the Cd and Cl on the cylinder.
Running with diffferent mesh refinement, I got Cd about 0.24, which is different from experiemental result (0.4)

I run transient with the setup:

viscous model: k-ep realizable.
nearwall treatment: standard wall function
reference value get from inlet (area=1,depth=1, length=1)
solution method is PISO with all the pressure, momentum, turbulent and transient formulation is 2nd order

Timestep 0.05 with 120 iteration/timestep

Could anyone pls let me know where is my mistake

Thank you

RaiderDoctor · June 16, 2018, 15:43

Okay, there may be a few things wrong with this setup.

First: we need to see what your mesh looks like. You say that you've refined it, which is fine, but is there inflation on the circle? Is the wake region properly meshed, how big is the fluid domain, etc. Just a picture with a mesh count would be fine.
Next, although you set your reference values from the inlet, you need to manually input the length, depth and area. According to the manual, "For example, the reference length and area will not be set by computing the reference values from a boundary condition; you must set these manually", "Area - sets the reference area, which is used to compute the force and moment coefficients"
"Depth - sets the reference depth used for computing cell volumes in 2D"
"Length - sets the reference length, which is used in the computation of the moment coefficient"
https://www.sharcnet.ca/Software/Ans...task_page.html

So, if your area is actually 0.785 m^2, and length (diameter) is 1 m, then you are technically not setting your values correctly.

Last comment, you say this is a transient simulation, so is your step size the correct size? Is it too big, too small, etc.? When you monitor your Cl and Cd, are they converged for each time step, or are they still changing?

mineralwater7 · June 16, 2018, 21:43

Thanks RaiderDoctor.

_ For the mesh side, I use ICEM for structured mesh
Domain: Front is 10D, back: 20D. top and bot: 6D
Using Ogrid for the cylinder. Pls see attach pics

_ For the reference value, as I think, the default numbers (Area=1, Depth=1, Length=1) is for my case.

"The reference area depends on what type of drag coefficient is being measured. For automobiles and many other objects, the reference area is the projected frontal area of the vehicle. This may not necessarily be the cross sectional area of the vehicle, depending on where the cross section is taken. For example, for a sphere A=pi*r^2 , (note this is not the surface area = 4*pi*r^2)"
https://en.wikipedia.org/wiki/Drag_coefficient

For 2D cylinder, the reference Area would be the area in the flow direction, which is the rectangular. The depth for 2D case in Fluent should be 1 by default. And the length (diameter) is 1. So the area would be depth*length =1.
May I ask how do you get the number 0.785 m^2? Pls correct me if am wrong

_ For the time step size, since Reynold 10^6, I use Strouhal number 0.22. So the timestep calculated as 1% of the period is 0.05. However, I use pressure based solver (implicit) so the timestep size would not affect much on the solution I guess?

And yes, they did converge within each timestep (criteria I set 10^-6)

RaiderDoctor · June 17, 2018, 15:17

You are right in that I was accounting for the cross-sectional area, not the frontal area. That's my bad, sorry haha.
I got the value of 0.785 m^2 by solving for the area of the circle. As you correctly pointed out, this is wrong as area is supposed to represent the front of your object, not the cross section. Since the diameter is 1 m, then the area would be 1 m^2.

The depth, from what I can make out from others

(Volume in fluent 2d)
(Depth for 2D problems )
is simply the unit depth associated with the mesh. So now it's my turn to ask the question; is this still applicable even if the real depth is not 1 m? What I mean to say is, if the mesh you generated has a depth of less than 1 m (say 0.5 m), then would you have to change the depth to correctly reflect that? I'm not sure.
While doing a bit of research on my own, I came across this thread that is very similar to yours (Flow over circular cylinder). While I'm sure you've probably already seen it, it did get me thinking about your time scale. How long are you simulating this for? If it's only for one period, then you are not allowing the simulation to achieve a steady-state. You were right about the implicit solver, as well. I usually work with total times of less than 1 s and a step size of 0.5 ms, which is why I was concerned about yours haha.

And lastly, your mesh is quite beautiful. You should think about posting a tutorial on how you did that in ICEM. I know it might be common knowledge, but it might give some others practice in working the software.

mineralwater7 · June 17, 2018, 22:35

Haha. Yeah. I got a good news that my results is within the range of experimental results. So it is valid now. Such a big release.

For the depth, if you consider the real depth in here, then I guess 3D case is a must. Fluent will make 1 as a default value for 2D.

Time period I did do it for a long period, just to make sure the vortex shedding. Normally I would run for more than 30s.

Lastly thanks very much for the comment on the mesh. And yes, I would love to show ppl who new to Icem to do it.

Thanks for the help

RaiderDoctor · June 18, 2018, 09:55

Awesome, glad to hear it! What was it that got you the correct result, if you don’t mind me asking? Was it the increased time length?

mineralwater7 · June 19, 2018, 11:19

No. I mean my Cd value in valid range right at the beginning. Just that I do not know and try to figure out what did wrong.

RaiderDoctor · June 19, 2018, 11:42

Okay, I misunderstood, sorry. So your Cd value (of 0.24) did not change, but you are going to take it as an acceptable value (although your experimental result was 0.4). You didn't change anything with the setup at all, you are just taking what your previously got to be correct. Well, good for you! Sorry I couldn't help out more.

mineralwater7 · June 19, 2018, 21:35

Yea. We discuss we learn. Thanks man

Leonardo.flores · June 22, 2018, 02:02

Dont forget checking your Courant number (CFL). For a stable simulation it should be around 1. (But for some cases it could work with 100 or even more). This might affect your time step, which looks big. If you refine your timesteps and notice Cd and Cl are changing you should keep refining until no variance is noticed.

120 iterations per ts is too much... When you refine your time-steps you should get about 10-20 iterations per ts. Also PISO is an expensive method, and works well with CFL=1. For higher values SIMPLE gives same results and less CPU demanding.

June 16, 2018, 10:13	Flow over 2D cylinder (Reynold 10^6)	#1
mineralwater7 New Member Longism Join Date: Mar 2018 Posts: 7 Rep Power: 8	Hi all, I am working on the simulation case which describe the flow of water over a 2D cylinder. Diameter of cylinder is 1m. The Re=10^6. Velocity=1m/s. The goal is to find out the Cd and Cl on the cylinder. Running with diffferent mesh refinement, I got Cd about 0.24, which is different from experiemental result (0.4) I run transient with the setup: viscous model: k-ep realizable. nearwall treatment: standard wall function reference value get from inlet (area=1,depth=1, length=1) solution method is PISO with all the pressure, momentum, turbulent and transient formulation is 2nd order Timestep 0.05 with 120 iteration/timestep Could anyone pls let me know where is my mistake Thank you

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Drag force coefficient too low for a flow past cylinder at Re= 1e05	Scabbard	STAR-CCM+	2	June 5, 2020 15:44
Flow past a cylinder at Re 1e05 using LES, drag force coefficient is to low	Scabbard	Main CFD Forum	21	June 19, 2018 14:58
Drag force coefficient too high for a flow past a cylinder using komega sst	Scabbard	OpenFOAM Running, Solving & CFD	37	March 21, 2016 17:16
benchmark: flow over a circular cylinder	goodegg	Main CFD Forum	12	January 22, 2013 12:47
Flow over a cylinder	Anna	Main CFD Forum	9	March 24, 2006 15:32

June 16, 2018, 15:43		#2
RaiderDoctor Senior Member Join Date: Dec 2016 Posts: 152 Rep Power: 10	Okay, there may be a few things wrong with this setup. First: we need to see what your mesh looks like. You say that you've refined it, which is fine, but is there inflation on the circle? Is the wake region properly meshed, how big is the fluid domain, etc. Just a picture with a mesh count would be fine. Next, although you set your reference values from the inlet, you need to manually input the length, depth and area. According to the manual, "For example, the reference length and area will not be set by computing the reference values from a boundary condition; you must set these manually", "Area - sets the reference area, which is used to compute the force and moment coefficients" "Depth - sets the reference depth used for computing cell volumes in 2D" "Length - sets the reference length, which is used in the computation of the moment coefficient" https://www.sharcnet.ca/Software/Ans...task_page.html So, if your area is actually 0.785 m^2, and length (diameter) is 1 m, then you are technically not setting your values correctly. Last comment, you say this is a transient simulation, so is your step size the correct size? Is it too big, too small, etc.? When you monitor your Cl and Cd, are they converged for each time step, or are they still changing?

June 17, 2018, 15:17		#4
RaiderDoctor Senior Member Join Date: Dec 2016 Posts: 152 Rep Power: 10	You are right in that I was accounting for the cross-sectional area, not the frontal area. That's my bad, sorry haha. I got the value of 0.785 m^2 by solving for the area of the circle. As you correctly pointed out, this is wrong as area is supposed to represent the front of your object, not the cross section. Since the diameter is 1 m, then the area would be 1 m^2. The depth, from what I can make out from others (Volume in fluent 2d) (Depth for 2D problems ) is simply the unit depth associated with the mesh. So now it's my turn to ask the question; is this still applicable even if the real depth is not 1 m? What I mean to say is, if the mesh you generated has a depth of less than 1 m (say 0.5 m), then would you have to change the depth to correctly reflect that? I'm not sure. While doing a bit of research on my own, I came across this thread that is very similar to yours (Flow over circular cylinder). While I'm sure you've probably already seen it, it did get me thinking about your time scale. How long are you simulating this for? If it's only for one period, then you are not allowing the simulation to achieve a steady-state. You were right about the implicit solver, as well. I usually work with total times of less than 1 s and a step size of 0.5 ms, which is why I was concerned about yours haha. And lastly, your mesh is quite beautiful. You should think about posting a tutorial on how you did that in ICEM. I know it might be common knowledge, but it might give some others practice in working the software.

June 17, 2018, 22:35		#5
mineralwater7 New Member Longism Join Date: Mar 2018 Posts: 7 Rep Power: 8	Haha. Yeah. I got a good news that my results is within the range of experimental results. So it is valid now. Such a big release. For the depth, if you consider the real depth in here, then I guess 3D case is a must. Fluent will make 1 as a default value for 2D. Time period I did do it for a long period, just to make sure the vortex shedding. Normally I would run for more than 30s. Lastly thanks very much for the comment on the mesh. And yes, I would love to show ppl who new to Icem to do it. Thanks for the help

June 18, 2018, 09:55		#6
RaiderDoctor Senior Member Join Date: Dec 2016 Posts: 152 Rep Power: 10	Awesome, glad to hear it! What was it that got you the correct result, if you don’t mind me asking? Was it the increased time length?

June 19, 2018, 11:19		#7
mineralwater7 New Member Longism Join Date: Mar 2018 Posts: 7 Rep Power: 8	No. I mean my Cd value in valid range right at the beginning. Just that I do not know and try to figure out what did wrong.

June 19, 2018, 11:42		#8
RaiderDoctor Senior Member Join Date: Dec 2016 Posts: 152 Rep Power: 10	Okay, I misunderstood, sorry. So your Cd value (of 0.24) did not change, but you are going to take it as an acceptable value (although your experimental result was 0.4). You didn't change anything with the setup at all, you are just taking what your previously got to be correct. Well, good for you! Sorry I couldn't help out more.

June 19, 2018, 21:35		#9
mineralwater7 New Member Longism Join Date: Mar 2018 Posts: 7 Rep Power: 8	Yea. We discuss we learn. Thanks man

June 22, 2018, 02:02		#10
Leonardo.flores Member Leonardo Join Date: Nov 2017 Posts: 37 Rep Power: 8	Dont forget checking your Courant number (CFL). For a stable simulation it should be around 1. (But for some cases it could work with 100 or even more). This might affect your time step, which looks big. If you refine your timesteps and notice Cd and Cl are changing you should keep refining until no variance is noticed. 120 iterations per ts is too much... When you refine your time-steps you should get about 10-20 iterations per ts. Also PISO is an expensive method, and works well with CFL=1. For higher values SIMPLE gives same results and less CPU demanding.