[Time4Tea]

Hi, I have a general CFD question: what are the telltale signs that convergence problems with a steady solution may be due to transient effects being present (which might indicate the need to switch to a transient solver)?

I am trying to run some incompressible simulations of internal flow in valves using FLUENT and I am having some problems with getting good convergence. The flow patterns are complicated and there is quite a bit of separation, so I am wondering if I should switch to the transient solver. Thanks.

[FMDenaro]

Originally Posted by Time4Tea

Hi, I have a general CFD question: what are the telltale signs that convergence problems with a steady solution may be due to transient effects being present (which might indicate the need to switch to a transient solver)?

I am trying to run some incompressible simulations of internal flow in valves using FLUENT and I am having some problems with getting good convergence. The flow patterns are complicated and there is quite a bit of separation, so I am wondering if I should switch to the transient solver. Thanks.

It depends both on the physics of the problem and on the formulation you are using. Are you working with laminar or turbulent flows?
In the latter case, steady state is only statistical (RANS) and cannot be simulated if you do not have an energy (statistical) equilibrium

[Time4Tea]

Hi FMDenaro,

Thanks for your reply. It is turbulent and I am trying to use an SST k-omega model. I'm looking at water flow through a valve with 200 psig inlet and 50 psig outlet. How would I know if I have statistical energy equilibrium, and what would you recommend that I do if I don't?

[FMDenaro]

Originally Posted by Time4Tea

Hi FMDenaro,

Thanks for your reply. It is turbulent and I am trying to use an SST k-omega model. I'm looking at water flow through a valve with 200 psig inlet and 50 psig outlet. How would I know if I have statistical energy equilibrium, and what would you recommend that I do if I don't?

is the valve fixed? are the BC.s steady? then you should get a statistically steady solution using RANS.

[Time4Tea]

Yes, the mesh is fixed and b/cs are steady. However, is it not possible that there could be sources of transient effects in the Reynold's-averaged flow, which are not necessarily related to turbulence? For example, vortex shedding or unsteady separation effects? Even if RANS is applicable, how can I tell whether unsteady effects in the Reynolds-averaged flow might be affecting my solution?

Basically, I am trying to determine whether my convergence problems are down to inherent transient effects in the flow or something else (i.e. the mesh).

[FMDenaro]

Originally Posted by Time4Tea

Yes, the mesh is fixed and b/cs are steady. However, is it not possible that there could be sources of transient effects in the Reynold's-averaged flow, which are not necessarily related to turbulence? For example, vortex shedding or unsteady separation effects? Even if RANS is applicable, how can I tell whether unsteady effects in the Reynolds-averaged flow might be affecting my solution?

Basically, I am trying to determine whether my convergence problems are down to inherent transient effects in the flow or something else (i.e. the mesh).

no, not at all, RANS equations you are solving are steady by definition and your flow problem has vortex shedding from the valve which effects are all modelled on the averaged steady velocity.

Your are just facing with numerical problems of convergence of the modelled equations.

[Time4Tea]

Originally Posted by FMDenaro

no, not at all, RANS equations you are solving are steady by definition and your flow problem has vortex shedding from the valve which effects are all modelled on the averaged steady velocity.

Your are just facing with numerical problems of convergence of the modelled equations.

My understanding is that it is possible to conduct unsteady RANS simulations, which can be used to model unsteady phenomena such as vortex shedding, which are not turbulent effects:

https://www.google.com/url?sa=t&rct=...24088155,d.cWw

Perhaps I have some misconception here? Is FLUENT not a suitable tool to use for unsteady RANS?

I thought the RANS equations still include a time-dependent term?

[FMDenaro]

Originally Posted by Time4Tea

My understanding is that it is possible to conduct unsteady RANS simulations, which can be used to model unsteady phenomena such as vortex shedding, which are not turbulent effects:

https://www.google.com/url?sa=t&rct=...24088155,d.cWw

Perhaps I have some misconception here? Is FLUENT not a suitable tool to use for unsteady RANS?

I thought the RANS equations still include a time-dependent term?

what you are referring to is the unsteady RANS, called URANS. It is used when the non-steady effects are due to external forces, for example a piston moving in a cylinder or, in your case, the valve movement.
Fluent allows for solving in URANS formulation but the problem is not in the code you use but in the flow problem you solve.
I suppose that if you switch the code to run in URANS, the time derivative of the velocities would be nothing else that the residuals you are not able to drive to convergence in RANS. I don't see an evidence of a physical meaning in the vortex shedding with URANS. As a counterpart, you can find in litarature some so-called time-filtered LES that retains a different meanng of the solution.

[FMDenaro]

I would add that I do not agree to what is written in the paper you linked. The case of the cube mounted on a wall is statistically steady if no external forces are present and a steady RANS has is correct meaning

[Time4Tea]

Ok, it seems that perhaps I have some level of confusion over unsteady simulations and RANS. I have to admit, it does seem somewhat contradictory to talk about unsteadiness in a quantity that is time-averaged. However, the RANS equations do contain a time-varying component, don't they? So, you are saying that term is only applicable in situations where the b/cs are unsteady, but RANS cannot be used to simulate unsteady phenomena that are caused purely by internal flow instability, with steady b/cs?

So, if I was to use (unsteady) RANS to try to simulate a flow situation where there would (in reality) be unsteadiness (e.g. vortex-shedding), those unsteady features should be averaged out, in the same way as the turbulence?

So, essentially, I would need to look at something like a LES to try to simulate vortex shedding?

[FMDenaro]

Originally Posted by Time4Tea

Ok, it seems that perhaps I have some level of confusion over unsteady simulations and RANS. I have to admit, it does seem somewhat contradictory to talk about unsteadiness in a quantity that is time-averaged. However, the RANS equations do contain a time-varying component, don't they? So, you are saying that term is only applicable in situations where the b/cs are unsteady, but RANS cannot be used to simulate unsteady phenomena that are caused purely by internal flow instability, with steady b/cs?

So, if I was to use (unsteady) RANS to try to simulate a flow situation where there would (in reality) be unsteadiness (e.g. vortex-shedding), those unsteady features should be averaged out, in the same way as the turbulence?

So, essentially, I would need to look at something like a LES to try to simulate vortex shedding?

no at all, the RANS are steady equations by definition.....the formulation is based on the Reynolds average

<f>(x) = lim T-> + Inf (1/T) Int [t0, t0+T] f(x,t) dt


if you would use a different definition in which <f> has dependence on time it results no longer true that <<f>>=<f> as in RANS. That leads to a different unresolved tensor and a different modelling that is more related to the time-filtering LES.

In conclusion, for your flow problem I strongly suggest to perform an LES simulation using the dynamic model.

[Time4Tea]

Hmm, I have to admit I am getting quite confused. You say that the RANS equations are 'steady by definition'; however, the RANS equations that are given in the FLUENT theory guide clearly include unsteady terms for the averaged quantities (I would post a screenshot of that section, but I'm not sure if that would be infringing copyright). So, are we talking about different RANS equations here?

[FMDenaro]

do not use the user guide of Fluent like a textbook of fluid dynamics...

you can see a clear explanation at page 11 here
https://www.scribd.com/doc/50191605/...ing-CFD-Wilcox


As you can see, the key is in the definition of averaging....

[Time4Tea]

Thanks very much for your help with this question, FMDenaro, I do very much appreciate it. However, regarding the reference that you linked and going back to my previous point, you can see on page 16 that the Reynolds Averaged momentum equation (2.23) contains an unsteady term (dUi/dt), which is similar to the equation in the FLUENT theory guide. So, surely this equation is not steady?

[FMDenaro]

Using the ensemble averaging You get the URANS equations

[FMDenaro]

In brief, I have the following idea. Assuming the ensemble average (instead of the Reynolds one)

<f>(x,t) = lim N->+Inf (1/N) Sum [k=1...N] f_k(x,t)

the unsteady RANS has this specific meaning:
For example, the cyclic flow in the in-cylinder engine (compression/expansion cicles), the value at each time means that the function is representative of the average (at that time) of N different realizations realized at that crank angle. Therefore, you can simulate the complete crank angle cycle associate to a period. The simulation is unsteady but the value at a time has a very specific meaning.
In your flow problem, there is no meaning to think that the ensemble average leads to an unsteady solution, you should see the time derivatives of the velocity field converging toward vanishing values.

[Time4Tea]

Ok, so it seems that FLUENT must be using the (ensemble averaged) URANS equations with their transient solver, as the RANS equations they provide include the unsteady term.

So, if the URANS equations are ensemble averaged, shouldn't they be able to resolve deterministic unsteady flow features, such as vortex shedding? If it is deterministic, then wouldn't those features be the same in every realization, so they wouldn't be averaged out?

[FMDenaro]

consider your flow problem: the inflow is steady and the valve is fixed therefore in any of the N (going to infinite) realization you have the same configuration. Now immagine to have really N realizations produced by experiments and at each time you perform the ensemble average. For N going to infinite what should be the reason to produce an averaged velocity field changing in time? I would consider this possibility if N is finite and lesser than the required number of samples.
On the other hand, the appeareance of unsteadiness in a vortex shedding is typical in case of unfiltered (DNS) or filtered (space-time LES) fields.
I can immagine that URANS can produce a numerical unsteady solution in your case but this is due to numerical and modeling reasons but is not necessarily representative of a physical pattern.
See also the book of Ferziger and Peric where URANS is introduced.

I added a link of simulation of vortex shedding with URANS:
http://www.tfd.chalmers.se/~lada/pos...sis_vagesh.pdf