CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   Transient features in steady state simulation? (http://www.cfd-online.com/Forums/main/124617-transient-features-steady-state-simulation.html)

PedFr0 October 9, 2013 09:52

Transient features in steady state simulation?
 
Hi All

In the case simulating flow with unsteady features, such as vortex shedding, using steady state simulation, will the results such as drag and lift co-efficients be valid? Would they essentially be the overall averaged values of the simulation or will the simulation need to be run as transient in order to get valid results?

Your help is much appreciated

Regards
Adhikar

FMDenaro October 9, 2013 11:11

Quote:

Originally Posted by PedFr0 (Post 455936)
Hi All

In the case simulating flow with unsteady features, such as vortex shedding, using steady state simulation, will the results such as drag and lift co-efficients be valid? Would they essentially be the overall averaged values of the simulation or will the simulation need to be run as transient in order to get valid results?

Your help is much appreciated

Regards
Adhikar


If the case admits no steady solution, a steady solver would simply not converge because the residual has a physical meaning that is the time-dependent part that balances all the fluxes.

PedFr0 October 9, 2013 11:22

Hi Filippo

So essentially if I get residuals that aren't fluctuating, and eventually converge, then I have valid drag and lift coefficients? I am getting something along those lines however I also have a monitor for velocity in a point in space and this shows a fluctuating velocity with iterations, yet my drag and lift coefficient monitors seem to be converging to a constant value as well as my residuals.

In short, I am seeing only fluctuating velocities yet all my other monitors appear to be converging, thus are my drag and lift coefficients?

Thank you for your speedy response and any further help would be much appreciated.

FMDenaro October 9, 2013 11:37

Quote:

Originally Posted by PedFr0 (Post 455968)
Hi Filippo

So essentially if I get residuals that aren't fluctuating, and eventually converge, then I have valid drag and lift coefficients? I am getting something along those lines however I also have a monitor for velocity in a point in space and this shows a fluctuating velocity with iterations, yet my drag and lift coefficient monitors seem to be converging to a constant value as well as my residuals.

In short, I am seeing only fluctuating velocities yet all my other monitors appear to be converging, thus are my drag and lift coefficients?

Thank you for your speedy response and any further help would be much appreciated.

If density, momentum, energy go to a steady state any other derived quantity will be steady (drag, lift). Generally, a non-steady state of the flow will produce oscillating residuals. When the residuals appear to converge toward a constant value that can not be considered numerically "zero" then I suspect some errors...

However, laminar or turbulent flow model will drive to different meaning of the steady state.

PedFr0 October 9, 2013 11:49

Thank you for your help, and from what I can gather I basically may need to run the simulation as transient in this case if I have constant residuals yet an unsteady flow. Maybe I can attach some images when the simulation is complete and you can provide some further knowledge. Many thanks for the help.

JBeilke October 9, 2013 16:06

A converged solution does not at all mean, that you get a physically correct result. Refine your mesh and see if you still get the same behaviour. Refine it probably even more. Run it transient and compare the lift and drag again.

Apart from this there might be lots of other reasons why the physics does not correspond to the numerics very well.

FMDenaro October 9, 2013 16:18

Quote:

Originally Posted by JBeilke (Post 456017)
A converged solution does not at all mean, that you get a physically correct result. Refine your mesh and see if you still get the same behaviour. Refine it probably even more. Run it transient and compare the lift and drag again.

Apart from this there might be lots of other reasons why the physics does not correspond to the numerics very well.

yes, I agree that reaching a numerically steady solution does not mean that is accurate and physically relevant, it only means that the discrete system admits a steady solution. However, it is worthwhile remarking that if the case is statistically steady (as for simulation of turbulence with RANS) refining the grid is not necessarily an assessment of the quality because the turbulence model overwhelms the truncation error.

PedFr0 October 10, 2013 05:55

Hi, and thank you for the help. Moving forward I will certainly try your suggestions and see if it makes a difference, starting with the transient solution though as I am under some pressure for a result and this seems the most obvious first step forward. Thus far the simulation has finally complete and I have attached an image of the monitors I have in place in the case that you may provide further insight.

Many thanks for the help.

http://postimg.org/image/9nbk0sa73/full/

PedFr0 October 10, 2013 05:56

If the image does does not display properly please right click it and open it in a new tab. Thanks

FMDenaro October 10, 2013 06:34

your residuals clearly show that the solution will not converge towards a steady state. But, as I don't know details of your flow problem, I can not say if that indicates an unsteady solution or something is simply wrong in your setting.

PedFr0 October 10, 2013 06:50

Hi
My simulation is of flow over a heliostat at an angle of 30 degrees at a reynolds number of around 3x10^6. Seeing as the geometry is essentially two large, inclined flat plates that are in-line with each other, I have a strong feeling that the solution must be unsteady based on all the information you have provided. This especially so due to unsteady phenomena being associated with these kind of geometries.

Thank you very much for your help, I think I can now move forward from this problem and simply have to deal with the turn-over time associated with transient simulation.

FMDenaro October 10, 2013 06:54

Quote:

Originally Posted by PedFr0 (Post 456127)
Hi
My simulation is of flow over a heliostat at an angle of 30 degrees at a reynolds number of around 3x10^6. Seeing as the geometry is essentially two large, inclined flat plates that are in-line with each other, I have a strong feeling that the solution must be unsteady based on all the information you have provided. This especially so due to unsteady phenomena being associated with these kind of geometries.

Thank you very much for your help, I think I can now move forward from this problem and simply have to deal with the turn-over time associated with transient simulation.


The pointwise solution is surely unsteady due to turbulence but the problem is to assess if your problem admits a statistically steady state to use RANS or has a statistically unsteady stato to use RANS. But, to tell the truth, I don't see a reason why you can not reach a statistical energy equilibrium and get a steady state ...the two plated are fixed not moving, right?

PedFr0 October 10, 2013 07:04

Yes that's correct, the two plates are fixed in space.
So would you suggest then to also run the simulation as transient and to take an average of the results obtained via the transient simulation and compare them to the steady state solution in order to check if they are statistically valid?

PedFr0 October 10, 2013 07:07

http://postimg.org/image/hjzdte17b/full/

Thats the exact geometry, just to clear things up

FMDenaro October 10, 2013 07:09

Quote:

Originally Posted by PedFr0 (Post 456131)
Yes that's correct, the two plates are fixed in space.
So would you suggest then to also run the simulation as transient and to take an average of the results obtained via the transient simulation and compare them to the steady state solution in order to check if they are statistically valid?

yes, correct. Be careful to run first the URANS for a sufficient time to cancel the initial numerical transient, then storing a meaningful number of fields to perform the statistical averaging.

FMDenaro October 10, 2013 07:12

Quote:

Originally Posted by PedFr0 (Post 456132)
http://postimg.org/image/hjzdte17b/full/

Thats the exact geometry, just to clear things up


Ok, I think that this problem should be better solved by URANS/DES models ...

PedFr0 October 10, 2013 07:17

Quote:

Originally Posted by FMDenaro (Post 456135)
Ok, I think that this problem should be better solved by URANS/DES models ...

Alright, thank you for your help. I will move forward using the unsteady approach from now.


All times are GMT -4. The time now is 17:00.