CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   Non-periodic Transient Behaviour? (http://www.cfd-online.com/Forums/main/131317-non-periodic-transient-behaviour.html)

PedFr0 March 13, 2014 04:36

Non-periodic Transient Behaviour?
 
Hi All

I would like to know if it is possible, when using RANS based turbulence models, to achieve non-periodic yet transient behaviour. I am simulating flow over a heliostat model and have run the simulation as transient using the Realisable k-epsilon, SST-k-omega and currently the RNG k-espilon models. The flow speed is about 18 m/s and the heliostat is orientated perpendicular to the flow, with a sid length of around 28cm.
When looking at the results for around 4 seconds of flow, using a timestep of 0.0001sec, I've noticed that there is no periodic behaviour in my results, however they are most definitely transient as the velocity and force monitors show no steady value.

Thanks in advance
Adhikar

FMDenaro March 13, 2014 04:50

Quote:

Originally Posted by PedFr0 (Post 479693)
Hi All

I would like to know if it is possible, when using RANS based turbulence models, to achieve non-periodic yet transient behaviour. I am simulating flow over a heliostat model and have run the simulation as transient using the Realisable k-epsilon, SST-k-omega and currently the RNG k-espilon models. The flow speed is about 18 m/s and the heliostat is orientated perpendicular to the flow, with a sid length of around 28cm.
When looking at the results for around 4 seconds of flow, using a timestep of 0.0001sec, I've noticed that there is no periodic behaviour in my results, however they are most definitely transient as the velocity and force monitors show no steady value.

Thanks in advance
Adhikar


I suppose you are using U(nsteady)RANS model ...
I think that:
1) depending on the initial velocity field, you have to wait more time to have a physical development, usually the numerical transient is long.
2) the discretization you are using, the turbulence modelling, etc produce too dissipation.

PedFr0 March 13, 2014 05:53

Hi Filipo

The initial velocity field is obtained from running a steady state simulation for around 10000 iterations. The discretisation scheme used is first order implicit for time and second order upwind for energy, dissipation and momentum.
The main issue is establishing whether or not it is valid to have non-periodic unsteady flow features in a simulation conducted using URANS turbulence models.

*edit* Revisiting my results I also noticed something rather strange in that the Realisable k-epsilon model does in fact produce periodic results over 2 seconds of flow time, however the SST-k-w model and the RNG model(currently running) appear to only produce unsteady, non-periodic results. Could you possibly provide any insight as to why this may be the case, considering all other variables between the simulations are the same. *edit*

Regards
Adhikar

FMDenaro March 13, 2014 06:29

URANS by definition is designed for have information of time evolving components only for low well-defined forcing frequencies. The simple example is the motion of piston in engine.
If you want a wider spectrum of frequencies you need to switch to LES formulation

PedFr0 March 13, 2014 06:43

So essentially, in the case that the frequency of the turbulent events are higher than URANS is capable of capturing, then my results will most likely look strange?

Many thanks for your help thus far.
Regards, Adhikar

FMDenaro March 13, 2014 06:57

what do you mean for "strange"? you should have a solution that must be congruent to the statistical averaging of URANS.
You can see some examples in Wilcox book

PedFr0 March 13, 2014 07:48

Please see below for the results for lift and drag using the Realisable k-e model and the SST-k-w model. I find it strange that the k-e shows periodic behaviour whereas the sst-k-w does not seem to display such well defined results.
Realisable results:
http://postimg.org/image/620wadkp3/
http://postimg.org/image/bc7vhodwt/
SST results:
http://postimg.org/image/q047db4cj/
http://postimg.org/image/e4cdah91l/
*edit* If images do not show properly please just right click and say open image in new tab*edit*

som87 March 13, 2014 08:43

The turbulence models you mentioned (k-omega, k-epsilon etc) are the RANS model right? In RANS/URANS, your solution is like the time average of DNS/actual solution. The time average, however, for unsteady solution will always be dependent to the time step you are considering. Upto certain limit, the decrease of time step, will help you decrease the difference between your URANS solution and actual DNS solution. However, depending on your "desired" frequency, you may or may not be able to capture the transition.
If possible (in case you really want to perform a good transient analysis) better switch to LES as Dr. Denaro suggested

Edit on your result:-The flow time is from 0..means you are considering the solution from start..? It is better if you see the convergence of your computation..(was the solution converged..?)

PedFr0 March 13, 2014 09:10

Quote:

Originally Posted by som87 (Post 479749)
The turbulence models you mentioned (k-omega, k-epsilon etc) are the RANS model right? In RANS/URANS, your solution is like the time average of DNS/actual solution. The time average, however, for unsteady solution will always be dependent to the time step you are considering. Upto certain limit, the decrease of time step, will help you decrease the difference between your URANS solution and actual DNS solution. However, depending on your "desired" frequency, you may or may not be able to capture the transition.
If possible (in case you really want to perform a good transient analysis) better switch to LES as Dr. Denaro suggested

Edit on your result:-The flow time is from 0..means you are considering the solution from start..? It is better if you see the convergence of your computation..(was the solution converged..?)

With regards to the first paragraph, the timestep I had chosen was 0.0001 sec. I am a bit uncertain as to what you mean by "desired" frequency? I essentially chose a timestep to give me a courant number of 1 in as many cells as possible whilst still attempting to get a result in reasonable time. And please can you elaborate on what transition you are talkin about?

On the edit, the transient case does start from 0 seconds, however I had first run the problem in steady state and used that result for the initial conditions on the transient case. The steady state solution however did no converge in that the values for drag and lift were constantly changing, hence driving me towards transient simulation.

Many thanks for your help, any further assistance would be much appreciated


All times are GMT -4. The time now is 11:08.