CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   Main CFD Forum (https://www.cfd-online.com/Forums/main/)
-   -   Convergence Problem (https://www.cfd-online.com/Forums/main/99440-convergence-problem.html)

 MHD_18 April 3, 2012 13:08

Convergence Problem

Dear CFD users,

Hope you are enjoying this nice day !!

I am working on steady cases and I want to check the convergence of my simulation.

The good signs is that residual pressure dropped 4 times and the mass balance is reached.

It remains to verify the satability of solution using monitor points, unfortunately, there is fluctuations in velocity :(

I would apperciate any solutions to my prolem.

Best regards,

MED

 scipy April 3, 2012 17:29

If these fluctuations are periodic/sinusoidal, then the problem you are investigating might be inherently somewhat transient. As long as the velocity (or any other monitor) oscilates around a certain value, the results can be considered as ok.

You could also try to calm these down by using another pressure/velocity coupling scheme if you're not using one already.

 MHD_18 April 4, 2012 04:25

Thanks scipy :)

I am using a cfd software which contains only one ( pressure/velocity) couling scheme.

I have another question, does the turbulence model ( LES or RANS) has any impacts on the stability of the solution ? And how ?

Good day,

Med

 sail April 4, 2012 08:13

Quote:
 Originally Posted by MHD_18 (Post 353054) Thanks scipy :) I am using a cfd software which contains only one ( pressure/velocity) couling scheme. I have another question, does the turbulence model ( LES or RANS) has any impacts on the stability of the solution ? And how ? Good day, Med
You can't run LES if you are doing steady-state.

Have you checked your mesh and plotted the residuals contours? Usually most convergence issues are due to poor meshes in certain areas.

Also different turbulences model can affect the stability: the more complex the model, the more issues you are likely to get to make it works: from a numerical side and also from the phenomenas you are able/can capture.

Another source of problems can be unrealistic turbulent BCs.

 MHD_18 April 4, 2012 09:41

:confused: why ?? Actually, I am using SGS ( LES ) model in my steady simulation.
What's wrong with it ?

In addtion, how can I check the quality of the mesh ? And, if it is poor, how can I improve it ?

For Bc, I set fixed velocity as Inlet & fixed pressure as outlet.

Thank you very much for your help :)

Med

 scipy April 4, 2012 14:12

Because LES models the large eddies that form and dissipate over time, so it requires a transient (time dependent) case ranging from t=0 seconds to t=X seconds and then the time difference split into time steps and each time step covered by some iterations. I guess :)

 sail April 4, 2012 14:42

Quote:
 Originally Posted by MHD_18 (Post 353139) :confused: why ?? Actually, I am using SGS ( LES ) model in my steady simulation. What's wrong with it ? In addtion, how can I check the quality of the mesh ? And, if it is poor, how can I improve it ? For Bc, I set fixed velocity as Inlet & fixed pressure as outlet. Thank you very much for your help :) Med
You can't do that. that won't make any sense, and you won't have any meaningful results.

to run steady simulations, you need to use Reynolds Avaraged NS: ranse.

I'm actually surprised that a softwware let you do something like that. usually commercial sw have some safety mechanism aganist such misuse. out of curiosity, what sw are you using?

 MHD_18 April 5, 2012 05:45

Les

Quote:
 Originally Posted by sail (Post 353177) You can't do that. that won't make any sense, and you won't have any meaningful results. to run steady simulations, you need to use Reynolds Avaraged NS: ranse. I'm actually surprised that a softwware let you do something like that. usually commercial sw have some safety mechanism aganist such misuse. out of curiosity, what sw are you using?
As I know, LES is transient method and we can use it to acheive a steady state by averaging such quantities ( velocity, pressure etc ... ).

Am I rignt ?

I am using Pamflow a commercial software owned by ESI .

THANKS AGAIN :)

 scipy April 5, 2012 06:02

I'm afraid you're off a bit there. LES is transient and you can get time averaged values, this part is correct, but a steady state simulation means solving just 1 "snapshot" in flow time.

Imagine wind tunnel testing as an example.. In the automotive industry they measure forces over approximately 2 minutes of flow time and then all the forces are averaged - if you wanted to match this via CFD, you'd use a transient simulation and then average the forces again. However, this is not "steady" state.. steady would be RANS snapshot in time of the flow and for this to be valid the flow has to be steady and not have any large transient effects like big flow separation, detachment, large eddies etc.

 MHD_18 April 5, 2012 07:45

Quote:
 Originally Posted by scipy (Post 353269) I'm afraid you're off a bit there. LES is transient and you can get time averaged values, this part is correct, but a steady state simulation means solving just 1 "snapshot" in flow time. Imagine wind tunnel testing as an example.. In the automotive industry they measure forces over approximately 2 minutes of flow time and then all the forces are averaged - if you wanted to match this via CFD, you'd use a transient simulation and then average the forces again. However, this is not "steady" state.. steady would be RANS snapshot in time of the flow and for this to be valid the flow has to be steady and not have any large transient effects like big flow separation, detachment, large eddies etc.
What do you mean by ( 1 "snapshot" in flow time ) ?

 sbaffini April 5, 2012 09:40

The difference is in steady vs unsteady computations. A steady computation can only be done with a RANS or, in very few cases, a laminar approach (no turbulence model at all). An unsteady computation requires an URANS, LES or laminar approach.

In steady computations, iterations are required to achieve the final steady state result, a single unique solution intended to represent the average flow quantities.

In unsteady computations, iterations are performed at each time step to achieve an istantaneous solution, representative of the flow at a certain time. When performing unsteady computations you are not interested in a single solution at a single time instant but in following in time such a solution, hence you advance your computation for several time steps, say N. These time steps, as previously noticed by someone, are like instantaneous pictures of the flow field which by themselves are usually far from the average results which usually is of interest. As a consequence, like in experiments, some statistics have to be performed on the N instantaneous pictures of the flow to obtain more meaningful average results. The "picture" term is not random, it really is like if you have N instantaneous pictures of the flow and you use them to compute averages and other statistics.

However, and don't take this bad, you really need to read some book on partial differential equations, turublence modeling, steady/unsteady numerical methods and CFD in general. Running in steady state (maybe even 2D, i don't want to know) with LES and complaining about convergence clearly shows some deficiencies which are not allowed in CFD.

 MHD_18 April 6, 2012 04:26

:)

Thanks for help :)

 Ram April 6, 2012 16:56

1 quick question..

In a transient case, where the flow is fluctuating, are the residuals (for eg: velocity residuals) expected to fall down smooth or keep oscillating?

(If the same case is run in steady mode, the residuals oscillate, right?)

Thanks
Ram

 scipy April 6, 2012 16:59

Yes, they oscilate. ANSYS says in their documentation that timestep choice is "correct" if the residuals fall by 2-3 orders of magnitude within 15-20 subiterations in a timestep. (sometimes they just can't, but if they are constantly going down in the subiterations and not up, it's ok)

 All times are GMT -4. The time now is 08:59.