[Muzz]

The question in the title is something that I've been trying to grasp at a reasonably high level to answer some general modelling questions.

I've been doing some atmospheric boundary layer modelling for urban flow problems and I've been using RKE with good success. For simple building configurations (single/twin building setups, where the buildings are just cuboids), RKE residuals decrease by five orders of magnitude and the relevant engineering quantities settle.

For realistic urban environments (modelling real cities), I get residual decrease of 9 orders of magnitude and the relevant engineering quantities settle.

So far, so good. However, when I run a URANS simulation (I initialise the URANS with RANS result), the simple building configurations exhibit large scale vortex shedding and is clearly unsteady (which is fine). However, for the realistic urban environment, there is almost no temporal variation in the results.

Of course, considering the physical nature of the problem, transient behaviour is surely expected. Why is there a difference? My time step in each case is relatively large (0.5 s) but the inner iterations demonstrate the same, if not improved, levels of convergence. Could it be that the time step must be reduced? However, wouldn't a larger time step impact residual convergence or does it only impact the solution?

To this end, I'm not entirely sure how URANS is making my simulation unsteady.

[FMDenaro]

Originally Posted by Muzz

The question in the title is something that I've been trying to grasp at a reasonably high level to answer some general modelling questions.

I've been doing some atmospheric boundary layer modelling for urban flow problems and I've been using RKE with good success. For simple building configurations (single/twin building setups, where the buildings are just cuboids), RKE residuals decrease by five orders of magnitude and the relevant engineering quantities settle.

For realistic urban environments (modelling real cities), I get residual decrease of 9 orders of magnitude and the relevant engineering quantities settle.

So far, so good. However, when I run a URANS simulation (I initialise the URANS with RANS result), the simple building configurations exhibit large scale vortex shedding and is clearly unsteady (which is fine). However, for the realistic urban environment, there is almost no temporal variation in the results.

Of course, considering the physical nature of the problem, transient behaviour is surely expected. Why is there a difference? My time step in each case is relatively large (0.5 s) but the inner iterations demonstrate the same, if not improved, levels of convergence. Could it be that the time step must be reduced? However, wouldn't a larger time step impact residual convergence or does it only impact the solution?

To this end, I'm not entirely sure how URANS is making my simulation unsteady.

The topic about what URANS really means is still debated. The key is that if the flow problem has no external time-depending forcing, URANS should finally converge to a steady RANS state. What is the physical meaning of the time derivatives if no external forcing exists, it is a numerical error (as same as RANS solution does not converge) or there is some physical aspect? My personal opinion is that if a physical aspect exists is because the URANS is actually to be seen as a sort of "hybrid" LES.

Have a look to my discussion here



https://www.researchgate.net/post/UR...ompared_to_LES

[Muzz]

Originally Posted by FMDenaro

The topic about what URANS really means is still debated. The key is that if the flow problem has no external time-depending forcing, URANS should finally converge to a steady RANS state. What is the physical meaning of the time derivatives if no external forcing exists, it is a numerical error (as same as RANS solution does not converge) or there is some physical aspect? My personal opinion is that if a physical aspect exists is because the URANS is actually to be seen as a sort of "hybrid" LES.

Have a look to my discussion here



https://www.researchgate.net/post/UR...ompared_to_LES

Thank you for your input. So, does that mean that it's entirely possible to have a URANS simulation, when animated, showing no temporal variation in the flow? My situation with the complex urban geometry seems to show convergence on the same solution that I initialised URANS with, at each time step. Therefore, when played back, nothing appears to be changing.

Are there any good papers that you could recommend that touch upon this subject? Thank you again.

[FMDenaro]

Originally Posted by Muzz

Thank you for your input. So, does that mean that it's entirely possible to have a URANS simulation, when animated, showing no temporal variation in the flow? My situation with the complex urban geometry seems to show convergence on the same solution that I initialised URANS with, at each time step. Therefore, when played back, nothing appears to be changing.

Are there any good papers that you could recommend that touch upon this subject? Thank you again.

If you run a RANS solution, the residuals can be seen as the time derivatives present in URANS. Thus, one could deduce that the convergent RANS solution is congruent to a steady state URANS solution. The key of the question is if the URANS has a real turbulence model that effectively switches from that of RANS in such a way an unsteady behavior is physical when no external forcing is present. In my opinion that can be debatable, the turbulence model in URANS are practically derived from the RANS.
Your RANS and (steady) URANS solutions produce the same fields? Are you using a 3D geometry in the URANS case?

If you read the post on RG you will find some papers to read.

[Muzz]

Originally Posted by FMDenaro

If you run a RANS solution, the residuals can be seen as the time derivatives present in URANS. Thus, one could deduce that the convergent RANS solution is congruent to a steady state URANS solution. The key of the question is if the URANS has a real turbulence model that effectively switches from that of RANS in such a way an unsteady behavior is physical when no external forcing is present. In my opinion that can be debatable, the turbulence model in URANS are practically derived from the RANS.
Your RANS and (steady) URANS solutions produce the same fields? Are you using a 3D geometry in the URANS case?

If you read the post on RG you will find some papers to read.

Yes, my RANS and URANS results are identical (with no time-variant behaviour in URANS). It is a 3D problem for both RANS and URANS.

I should add that unsteady behaviour is observed in different 3D problems, when the geometry in question is that of a single bluff body, not a built up and congested 3D environment.

[FMDenaro]

Originally Posted by Muzz

Yes, my RANS and URANS results are identical (with no time-variant behaviour in URANS). It is a 3D problem for both RANS and URANS.

I should add that unsteady behaviour is observed in different 3D problems, when the geometry in question is that of a single bluff body, not a built up and congested 3D environment.

It is a case where the congested environment can suppress oscillation? What about the numerical and turbulence model setting? While refining the grid and reducing the time step the results are still steady? In principle the inflow for URANS should not be as same as in RANS, it should have some unsteady behavior.

[Muzz]

Originally Posted by FMDenaro

It is a case where the congested environment can suppress oscillation? What about the numerical and turbulence model setting? While refining the grid and reducing the time step the results are still steady? In principle the inflow for URANS should not be as same as in RANS, it should have some unsteady behavior.

Suppressed oscillation is what I was considering, with limited expertise in the area. I'm using an implicit unsteady, segregated solver with RKE as turbulence model. Spatial and temporal discretisation is 2nd order (FYI, I'm using STAR-CCM+). The grid refinement was done with steady RANS.

I've not experimented with reducing the time step yet (I will do so now). I started off with a relatively large time step and the solution would still converge. I was working under the assumption that a time step that was too large would cause the solution to diverge and I could use that as a measure of how close I was getting to an appropriate time step.

The inflow BCs are steady and are the same as I'd used for steady RANS.

As an aside, would you say that the computational resources required for URANS are intrinsically greater than steady RANS? Is it situation dependent?

[FMDenaro]

Originally Posted by Muzz

Suppressed oscillation is what I was considering, with limited expertise in the area. I'm using an implicit unsteady, segregated solver with RKE as turbulence model. Spatial and temporal discretisation is 2nd order (FYI, I'm using STAR-CCM+). The grid refinement was done with steady RANS.

I've not experimented with reducing the time step yet (I will do so now). I started off with a relatively large time step and the solution would still converge. I was working under the assumption that a time step that was too large would cause the solution to diverge and I could use that as a measure of how close I was getting to an appropriate time step.

The inflow BCs are steady and are the same as I'd used for steady RANS.

As an aside, would you say that the computational resources required for URANS are intrinsically greater than steady RANS? Is it situation dependent?

Well, it could be questionable that the grid resolution used in RANS is sufficient for a good URANS. Some comments are in the post I linked.

Could you show the geometry and BCs of your problem?

[Muzz]

Originally Posted by FMDenaro

Well, it could be questionable that the grid resolution used in RANS is sufficient for a good URANS. Some comments are in the post I linked.

Could you show the geometry and BCs of your problem?

I had saw in reading some of those posts that there was a discussion whereby grid resolution for ILES and URANS should be comparable. Grid resolution could certainly be an issue as well.

I'll go back and try increasing grid resolution and decreasing time step. It may be that I've come with my question without full exhausting my options yet. Thank you for your input.

[FMDenaro]

Originally Posted by Muzz

I had saw in reading some of those posts that there was a discussion whereby grid resolution for ILES and URANS should be comparable. Grid resolution could certainly be an issue as well.

I'll go back and try increasing grid resolution and decreasing time step. It may be that I've come with my question without full exhausting my options yet. Thank you for your input.

Furthermore, try to monitor in time the total kinetic energy integrated in the whole domain, you could need also a long period of time integration to let oscillations developing...