# 3D DES simulation of turbulent jet

 Register Blogs Members List Search Today's Posts Mark Forums Read

April 10, 2017, 11:46
3D DES simulation of turbulent jet
#1
New Member

Join Date: May 2015
Posts: 22
Rep Power: 11
Hello all,

Im a beginner in Fluent and CFD so please bear with me.

Im trying to run a DES simulation for a 3d turbulent jet. I want to visualize the velocities (x,y,z), vorticities (x,y,z) etc.

I am trying to simulate a duct (0.35mx0.35mx0.25m). The duct is seperated by a wall which has a sharp-edged rectangular orifice configuration(70 mm x 7 mm). My inlet is 0.1 m before the wall that has the orifice and my outlet 0.15m after the wall. My mesh consists of tetrahedrons everywhere exccept where i used inflation. There are 1.6 million elements total.

Given data was ony the Reynolds number at the orifice (Re=23000 or Re hydrolic diamaterer=42000). It is calculated that the velocity at the orifice is v2=50m/s and so the velocity at the inlet has to be v1=0.133m/s. Also the simulation is done as incompressible.

In Fluent my inputs are:
Pressure-based solver
Transient
Energy off
Delayed Detached Eddy Simulation (DES) with the Spalart-Allmaras model

Boundary Conditions:
for walls i used default options
for outlet i used default options
for interior i used default options
for inlet i used mass-flow inlet
-mass flow rate = 0.19958 kg/s
-z component of flow direction = 0.133m/s
-turbulence specification method = turbulent viscosity ratio
-turbulence viscosity ratio=10

SIMPLE scheme was selected.
Pressure: Second Order
Momentum: Second Order Upwind
Modified Turbulent viscosity: Second Order Upwind
Transient Formulation: Bounded Second Order Implicit

Under-Relaxation factors were left default except for:
pressure: from 0.3 (default) --> 0.4
momentum: from 0.7(default) --> 0.8

All Residuals were changed to 1e-05

Using Standard Initialization, the solution was initialized as zeros (0) everywhere

For time step i choose Dt=5e-04 and max iterations/time step= 10

I use solution animations to check the results while running the simulation and the results are very close to what was expected...but the residuals are not willing to go down and converge

Do you have any changes to propose for all the above settings ?
Did i make a mistake somewhere ?

Any help will be greatly appreciated

Thank you.
Kostas
Attached Images
 3d view.jpg (18.5 KB, 30 views) xy plane.jpg (20.3 KB, 30 views) yz plane.jpg (22.5 KB, 30 views) mesh.jpg (154.1 KB, 37 views)

 April 11, 2017, 01:43 #2 Senior Member   Lucky Join Date: Apr 2011 Location: Orlando, FL USA Posts: 5,703 Rep Power: 66 Do you mean residuals are stuck at high values or are they decreasing but not enough to meet your convergence criteria? If it's the 1st, then it hints at a problem with the setup. If it's the 2nd... I think it is because you are using SIMPLE which needs many iterations to converge (because the urf's are not 1). You can verify this by increasing the number iterations per time-step to a really high number like 50 or 100 (just to check). Just do it for a few time-steps to test. You can change it back later. Also have you tried a smaller time-step? Since you don't have coupling with any other equations, I highly recommend PISO. The disadvantage is you'll need small time-steps. The advantage is, you can get away with as few as 2 iteration per time-step. You should be able to take a bunch of smaller time-steps faster. SIMPLE is great when you are also solving a lot of other equations, because coupling between equations is what slows convergence. If you're only solving continutiy & momentum, it's probaby the P-V coupling that's limiting. You could also give the COUPLED solver a whack, but it has urf's and so it's behaves more like SIMPLE. But definitely do the check using SIMPLE with 100 iterations to make sure. Btw I highly recommend you always use the bounded central differencing whenever possible (momentum especially). But maybe it's not available for DES.

 April 11, 2017, 01:48 #3 Super Moderator     Maxime Perelli Join Date: Mar 2009 Location: Switzerland Posts: 3,297 Rep Power: 41 If you are a beginner, I would: *run a steady state case first *set all under-relaxation parameters to default settings *set all numerical schema to default *set turbulent model to default (k-eps?) I would modify some parameters if you get bad results or poor convergence What is for you the default BC for outlet? I would also extend your domain , inlet and outlet are to close to the restriction __________________ In memory of my friend Hervé: CFD engineer & freerider

 April 11, 2017, 07:41 #4 New Member   Join Date: May 2015 Posts: 22 Rep Power: 11 Hello LuckyTran, Thank you for your answer When the simulation begins the residuals decrease. After a few hundreds iterations they stop decreasing or they decrease with a very slow rate. The residuals of x,y,z velocities and nut oscilate between 1e-04 and 1e-06. Only continuity stays between 1e-02 and 1e-03. For my mesh i used body of influence to create the fine areas. I put 2mm for the element sizes as an input so i calculated the timestep like this: Dx/v=Dt => 0.002/50=0.00004=4e-05. But I also tried to run a simulation with 2e-05 which is half of the above. But i couldn't get a good result. Maybe because the solution advances very slow and I didn't let it run for more than a coyple days. I will try what you said about simple and piso scheme and let you know.

 April 11, 2017, 08:13 #5 New Member   Join Date: May 2015 Posts: 22 Rep Power: 11 Hello Max, Thank you for your answer *I have already tried running a steady state simulation. And then when the residuals were flat (not converged) i changed the solution to transient and used the steady state values an initials. *Also i had the urf's at their default values. Only in the last case that i posted about had i changed them. * The schemes were also default *I can not use the k-e model because i want to visualize vorticity and the k-e doesn't show it. Only with LES or DES could i see the vorticity. And the LES is more demanding than the DES Also i used to have a bigger domain but it reached about 4-5 million elements and my computer wasn't able to solve it. So i was forced to make it smaller. The default BC for the outlet is type - Pressure outlet at momentum tab gauge pressure (pascal) - 0 constant backflow direction specification method - normal to boundary turbulence specifiction method - turbulent viscosity ratio backflow turbulent viscosity ratio - 10 constant

April 11, 2017, 10:25
#6
Senior Member

Lucky
Join Date: Apr 2011
Location: Orlando, FL USA
Posts: 5,703
Rep Power: 66
Quote:
 Originally Posted by Kostas_K The residuals of x,y,z velocities and nut oscilate between 1e-04 and 1e-06. Only continuity stays between 1e-02 and 1e-03.
This behavior is pretty normal. 1) You have your mesh which limits convergence. 2) Large time-steps also makes you lose some temporal bandwidth and also makes it harder to converge because you enter a time-step with error. 3) Over many time-steps, the error also accumulates but hopefully reaches some limit so that your sim doesn't diverge.

Quote:
 Originally Posted by Kostas_K For my mesh i used body of influence to create the fine areas. I put 2mm for the element sizes as an input so i calculated the timestep like this: Dx/v=Dt => 0.002/50=0.00004=4e-05.
So your courant number is 1 or so? Btw you should also check this with like a volume maximum monitor and/or some contours. Just visually inspect to make sure.

Honestly I think your simulation is fine and you only need to do some sanity checks. Make some point monitors at interesting points (like where you expect the shear layer) and check that the x-velocity, y-velocity, z-velocity, & pressure are converged to #.### decimals at every time step. You can also probe the vorticity since that's what you're interested in. Rather than worry about how much residuals are dropping, you should start post-processing and check whether the result makes any sense. I.e. verify that you have the expected turbulence.

April 13, 2017, 13:33
#7
New Member

Join Date: May 2015
Posts: 22
Rep Power: 11
No my courant number is not 1. It would be if i hab used for a time step 4e-05 but i used 5e-04. Because it would take too much time otherwise.
Also i just started testing what you said for PISO because for 2 days now i have been running another simulation.
I started with DES model and steady solution to get a initial value for my case (i know that DES model is only meant for transient solution). After 1600 iterations when the residuals were flat (not converged) i changed the solution to transient and used the steady state values an initials. The results in comparison with the experiment for z velocity and z vorticity can be seen in the pictures. The only problem are the residuals. Can i say that my solution is converged ?
Attached Images
 computational.jpg (87.1 KB, 29 views) residuals.jpg (52.9 KB, 25 views) experimental.jpg (57.7 KB, 23 views)

April 13, 2017, 14:08
#8
Senior Member

Lucky
Join Date: Apr 2011
Location: Orlando, FL USA
Posts: 5,703
Rep Power: 66
If you want to know if your solution is converged, then look at the solution, not residuals! Look at velocities, etc. Make a plot of velocity vs iter.

Also you need to verify that your DES is actually DES. I am worried that because you did not apply any perturbations (i.e using init-instantaneous-velocity or something equivalent) that you do not have a turbulent flow. Check the spectrum and verify that it is indeed turbulent and not laminar.

Quote:
 Originally Posted by Kostas_K No my courant number is not 1. It would be if i hab used for a time step 4e-05 but i used 5e-04. Because it would take too much time otherwise.
It is not wrong to do so, but I hope you are aware of the implications of using a very large time-step in DES/LES. Your temporal bandwidth is now extremely limited.

April 14, 2017, 07:34
#9
New Member

Join Date: May 2015
Posts: 22
Rep Power: 11
Quote:
 Originally Posted by LuckyTran Also you need to verify that your DES is actually DES. I am worried that because you did not apply any perturbations (i.e using init-instantaneous-velocity or something equivalent) that you do not have a turbulent flow. Check the spectrum and verify that it is indeed turbulent and not laminar.
How exactly can i verify that i have a turbulent flow ?
Also if i do have a laminar flow how can i apply some pertrubations using ansys in order to make it turbulent and get the right solution ?

April 14, 2017, 09:19
#10
Senior Member

Lucky
Join Date: Apr 2011
Location: Orlando, FL USA
Posts: 5,703
Rep Power: 66
Quote:
 Originally Posted by Kostas_K How exactly can i verify that i have a turbulent flow ? Also if i do have a laminar flow how can i apply some pertrubations using ansys in order to make it turbulent and get the right solution ?
Check out Paolo's blog. He has some cute scripts you can use. If you had started with a RANS as an initial condition though, you could have just typed in solve/initialize/ init-instantaneous-vel

https://www.cfd-online.com/Forums/blogs/sbaffini/

How to check... Well how do you verify that any flow is turbulent? I.e. forget that you are doing CFD but you are just an observer or experimentalist. If I set up this problem in a lab, how do you verify that it is indeed turbulent?

Well one way is to get the spectrum of the the turbulent kinetic energy and show that it is broadband and looks like a typical turbulent spectrum.