Strange values after using Initial Conditions

FFD · February 5, 2016, 08:43

Hello,
I recently started to simulate my simulations with given Initial Conditions.
Because all my simualtions just differ a little bit from the others I used the solution from the previous simulation and used it as the Initial Condition on the new simulation.
In my test simulation it worked perfectly fine.

The residuals converged like two times faster and were like 2-3 times better.

So I tried to do this in the real simulation. Now the problem is that my simulation diverges. My lift is like million times higher than the real one.

What could cause this? How can I fix it?

Any suggestions?

Best ragards

-FFD

fluid23 · February 5, 2016, 14:13

If you are doing a coupled analysis you can use expert initialization. This solves a simplified system of equations on a series of increasingly dense meshes to give you a somewhat realistic flow condition to start with.

I would also caution you that your lift isn't a million times larger if your model/residuals diverge. You cannot trust those results. If expert initialization doesn't reign in your divergence issue then you have some model setup or mesh problems you must diagnose.

FFD · February 8, 2016, 12:21

I don't think I am doing coupled analysis.
In my physics I am using segregated flow instead of coupled flow if thats what you mean.

And sure I know my real lift isnt million times higher. I was just saying that to show how wrong all the values are.

I found something about it in the Steve Portal:

Quote:

Alternatively, initial solving can be ‘smoothed’ using techniques such as gradually activating models, changing boundary conditions, activating unsteady models from a steady solution, or ramping up solution accuracy.
and
Slowing the advancement of the solution by ramping Courant numbers and/or under-relaxation factors.

Now I am not very advanced in StarCCM.
how can I gradually activate models to smooth my initial conditions?
How do I change the Boundary conditions? (How do I know what values I need for the boundary conditions?)
How do I activate unsteady models from a steady solution?
and How do I ramp up the solution accuracy? (More iterations in the simulations?)
-FFD

fluid23 · February 8, 2016, 12:42

Activating models may not be appropriate for you, can you list the models you have under physics continua?

Boundary conditions should be something that you already have defined. These are things like velocity inlets, mass flow inlets, pressure outlet, walls, etc... They govern what is happening far away from your areas of interest and the boundaries of your geometry. You can ramp these up over successive iterations using field functions or you can run at one value, let it converge, run at another value, and so on... until you get to your intended value. Unless you have supersonic flow, I doubt this will be useful.

Going from steady to unsteady is a matter of changing your physics models from steady to unsteady (probably implicit unsteady in your case). Here you would run the solution out using steady model then when it converges switch over to unsteady. This gives you a flow field that is close to right for your first few time steps rather than have these first few steps take up many, many inner iterations to achieve convergence.

I assume ramping up solution accuracy either refers to increasing mesh resolution (so adding cells). Or it refers to changing discritzation schemes in your segregated solver. For example you can select 1st order accurate, 2nd order accurate, and 3rd order accurate. This refers to the order of terms truncated in the discritization of the governing equations.

Simply adding iterations won't change your solution if your model is fully converged.

FFD · February 8, 2016, 13:04

So the List of my models is:

All y+ Wall treatment
Cell Quality Remediation
Constant Density
Gas (- Air)
Gradients
K-Omega Turbulence
Reynols-Averaged Navier-Stokes
Segregated Flow
SST (Menter) K-Omega
Steady
Thee Dimensional
Turbulent

I saw my region as a windtunnel, so I do have a velocity inlet with a constant velocity.
e.g.: my simulation was: on 10m/s. so I had the velocity inlet as 10m/s and in continua I had the initial value 10m/s for all cells.
Now I extraced the solution of my last simulation. There the velocity wasn't exactly 10m/s everywhere anymore.
The outlet is simply a pressure outlet. Should I change the boundary conditions when I use Initial conditions? I mean. I still want the air to flow in 10m/s.

Quote:

Going from steady to unsteady is a matter of changing your physics models from steady to unsteady (probably implicit unsteady in your case). Here you would run the solution out using steady model then when it converges switch over to unsteady. This gives you a flow field that is close to right for your first few time steps rather than have these first few steps take up many, many inner iterations to achieve convergence.

If the computation is longer it's not a huge problem. Do you think this can solve the problem? How can I do it?

If ramping up the solution does mean a better mesh it is not possible for me now, because I have just a limited RAM for post processing.

If its the discritization scheme I gonna read in the documentation about it.

fluid23 · February 8, 2016, 13:09

Your physics selections look fine. I wouldn't bother 'ramping' in your physics models.

It doesn't sound to me like you will really benefit from changing the discritization scheme in the solver. I wouldn't even consider the unsteady case until you can get something useful in the steady case. The unsteady case will likely have the same issues. It sounds to me like a possible mesh issue. Can you send me a picture of your mesh so I can get a feel for how appropriate it is?

FFD · February 8, 2016, 13:18

Hello,
sadly I cannot send a picture of my Mesh. But can it be the Mesh if the same simulation without initial conditions was fine? (both simulations were fine without the initial conditions. the first one and the second one) But if I applied the Initial Conditions of the first one to the second simulation. I had the divergence.

I only changed a small part in that simulation.
In my test simulation where I changed a part (It was a very small simulation with like 200.000 cells). It worked fine.

fluid23 · February 8, 2016, 13:20

If you are convinced that it's an initialization issue I would avoid initializing current models to previous solutions. I would also recommend verifying your wall y+ values are appropriate. That could easily ruin your day.

FFD · February 8, 2016, 13:28

I am not convinced that it is an initilization issue. But I am not sure if it can be a mesh issue, if the simulation runs fine without my conditions.

February 5, 2016, 08:43	Strange values after using Initial Conditions	#1
FFD Member Join Date: Aug 2015 Posts: 30 Rep Power: 11	Hello, I recently started to simulate my simulations with given Initial Conditions. Because all my simualtions just differ a little bit from the others I used the solution from the previous simulation and used it as the Initial Condition on the new simulation. In my test simulation it worked perfectly fine. The residuals converged like two times faster and were like 2-3 times better. So I tried to do this in the real simulation. Now the problem is that my simulation diverges. My lift is like million times higher than the real one. What could cause this? How can I fix it? Any suggestions? Best ragards -FFD

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Problem with chtMultiregionFoam radiation boundary condition	baran_foam	OpenFOAM Running, Solving & CFD	10	December 17, 2019 18:36
Floating point exception error	lpz_michele	OpenFOAM Running, Solving & CFD	53	October 19, 2015 03:50
Cannot run the code properly: very large time step continuity error	crst15	OpenFOAM Running, Solving & CFD	9	December 14, 2014 19:17
Micro Scale Pore, icoFoam	gooya_kabir	OpenFOAM Running, Solving & CFD	2	November 2, 2013 14:58
pisoFoam with k-epsilon turb blows up - Some questions	Heroic	OpenFOAM Running, Solving & CFD	26	December 17, 2012 04:34

February 5, 2016, 14:13		#2
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 18	If you are doing a coupled analysis you can use expert initialization. This solves a simplified system of equations on a series of increasingly dense meshes to give you a somewhat realistic flow condition to start with. I would also caution you that your lift isn't a million times larger if your model/residuals diverge. You cannot trust those results. If expert initialization doesn't reign in your divergence issue then you have some model setup or mesh problems you must diagnose.

February 8, 2016, 12:42		#4
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 18	Activating models may not be appropriate for you, can you list the models you have under physics continua? Boundary conditions should be something that you already have defined. These are things like velocity inlets, mass flow inlets, pressure outlet, walls, etc... They govern what is happening far away from your areas of interest and the boundaries of your geometry. You can ramp these up over successive iterations using field functions or you can run at one value, let it converge, run at another value, and so on... until you get to your intended value. Unless you have supersonic flow, I doubt this will be useful. Going from steady to unsteady is a matter of changing your physics models from steady to unsteady (probably implicit unsteady in your case). Here you would run the solution out using steady model then when it converges switch over to unsteady. This gives you a flow field that is close to right for your first few time steps rather than have these first few steps take up many, many inner iterations to achieve convergence. I assume ramping up solution accuracy either refers to increasing mesh resolution (so adding cells). Or it refers to changing discritzation schemes in your segregated solver. For example you can select 1st order accurate, 2nd order accurate, and 3rd order accurate. This refers to the order of terms truncated in the discritization of the governing equations. Simply adding iterations won't change your solution if your model is fully converged.

February 8, 2016, 13:09		#6
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 18	Your physics selections look fine. I wouldn't bother 'ramping' in your physics models. It doesn't sound to me like you will really benefit from changing the discritization scheme in the solver. I wouldn't even consider the unsteady case until you can get something useful in the steady case. The unsteady case will likely have the same issues. It sounds to me like a possible mesh issue. Can you send me a picture of your mesh so I can get a feel for how appropriate it is?

February 8, 2016, 13:18		#7
FFD Member Join Date: Aug 2015 Posts: 30 Rep Power: 11	Hello, sadly I cannot send a picture of my Mesh. But can it be the Mesh if the same simulation without initial conditions was fine? (both simulations were fine without the initial conditions. the first one and the second one) But if I applied the Initial Conditions of the first one to the second simulation. I had the divergence. I only changed a small part in that simulation. In my test simulation where I changed a part (It was a very small simulation with like 200.000 cells). It worked fine.

February 8, 2016, 13:20		#8
fluid23 Senior Member Matt Join Date: Aug 2014 Posts: 947 Rep Power: 18	If you are convinced that it's an initialization issue I would avoid initializing current models to previous solutions. I would also recommend verifying your wall y+ values are appropriate. That could easily ruin your day.

February 8, 2016, 13:28		#9
FFD Member Join Date: Aug 2015 Posts: 30 Rep Power: 11	I am not convinced that it is an initilization issue. But I am not sure if it can be a mesh issue, if the simulation runs fine without my conditions.