[siw]

Hi,

Just done a fluid-porous-fluid steady-state RANS simulation (hybrid fluid mesh structured porous mesh) and ran the first 100 iterations using the Upwind advection scheme just to get the simulation going since it is more robust. I then ran the next 100 iterations using the High-Resolution scheme to improve the accuracy.

The image shows that convergence was better with the Upwind scheme, see the image. Why would that be? I'd conclude to use Upwind in all similar cases going on what this has shown.

The monitor point (mass flow rate at the outlet) was constant from about iteration 10 to the end.

[Possa]

Originally Posted by siw

Hi,

The image shows that convergence was better with the Upwind scheme, see the image. Why would that be? I'd conclude to use Upwind in all similar cases going on what this has shown.

Hello siw,

If I understood correctly you are solving a fluid flow inside a porous medium, is that so?

If that's the case, the Upwind scheme is suppose to make the convergence better. This numerical scheme is based on the upstream flow conditions (just like the flow in a river - the upstream condition push the flow). If you take a look in the theory behind the advection schemes used in CFX you will probably understand.

If your problem is a fluid flow inside a porous medium then the UDS scheme is a good estimation of the physical phenomenon.

The High-Resolution scheme is not always better than the Upwind, it depends on the conditions of the problem. Each one has its limitations and applications. To know which one to use it is necessary to understand what they do mathematically.

Regards,
Possa

[Possa]

Originally Posted by Possa

If you take a look in the theory behind the advection schemes used in CFX you will probably understand.

A little more about the advection schemes can be found here http://www.cfd-online.com/Wiki/Appro...onvective_term

Regards,
Possa

[siw]

My simulation could be thought of like the CFX catalytic converter tutorial but without the heating features. There is fluid domain followed by a porous domain and then another final fluid domain.

[Possa]

Hi,

Then I stand for my previous post. Your problem is a upstream problem. The Upwind scheme is physically consistent. Using HR scheme doesn't necessarily mean it will provide better results.

[siw]

As I see it the High-Resolution scheme is an upwind scheme as well and it just includes a blending term to vary between 1st order (i.e. the same at the Upwind scheme) and 2nd order. Trying to keep it as close to 2nd order as possible to avoid wiggles.

[Possa]

Yes, you are correct, but the 1st order scheme (what CFX calls Upwind) is more robust than the High Resolution because of the order.

What I'm trying to say is that for some problems higher order schemes doesn't bring a better result, but it will probably show worse convergence. You have to analyse what scheme is better for the problem.

Regards,
Possa.

[ghorrocks]

Upwinding is a first order scheme and contains lots of numerical diffusion. This means the flow is damped artifically more than it should be which results in easy and tight convergence, but inaccurate results.

High Res uses a second order scheme when possible and blends to a first order scheme to maintain boundedness. It contains far less numerical diffusion. This means the flow has less artificial damping and therefore closer to the correct amount of damping. This results in slower and less tight convergence (as more little perturbations are present) but higher accuracy.

I do not agree with Talita's comment - there is no such thing as a simulation where first order is more suited than second, any time you choose a first order scheme you are sacficising accuracy for easy convergence.

THe CFX advection schemes "Upwinding" and "High res" are both upwinding schemes, just "upwinding" is first order and "High res" is second order, and blending to first order where the flux limiter cuts in.

And finally, do not be confused into thinking that simulation error is proportional to the residual. The residual is simply how tightly the linear solver has achieved, and the link of this to accuracy is complex. The original post's second order simulation is probably more accurate than the first order bit, even though the residuals are not as tight.

[tyw7]

Sorry for necroing an old thread. So is high resolution generally more accurate than first order (assuming it converges)?

[Opaque]

Definitely.

In the worst case scenario, high resolution reverts to Upwind; therefore, it can only be more accurate. Whenever the field is fairly constant, the gradient is nearly 0; therefore, then both are accomplishing the same discretization.

[ghorrocks]

If you do a simulation using the Hi Res scheme you will have a new variable ".beta" for the equations it is applied to. This shows the regions where the flux limiter has determined it cannot use the second order scheme and is ramping in the first order upwinding.

So you can see in the post processor the regions where it is using first order, second order or a blend of them.

[rellis2012]

Hi All,

this question might be a slight tangent to this thread in which case I apologise, but I am struggling to find a logical answer at the moment.

When looking at results for a highly turbulent flow field using DES the results obtained are dramatically different when using the High Resolution scheme and the Central Difference (Bounded CDS) scheme. Namely the turbulence in the high resolution case does not seem to be carried downstream and so the wake behind the object is not being broken up. However for the Central Difference scheme simulation i am seeing what i would associate as 'normal' behaviour (ie. the wake is being broken up and matching experimental data far better) Could this be to do with the fact that the blending function in high resolution scheme is selecting the Upwind scheme for certain regions of the mesh and as such some of the flow resolution is being lost downstream due to the dissipation?

Thanks in advance for any help,

Rob

[ghorrocks]

Yes, that is likely. Don't forget that the second order scheme in Hi-Res also is dissipative, just a lot less dissipative than the first order scheme. Central difference schemes have very low or no dissipation.

So I think your logic is on the right track. LES and DES models require extremely low dissipation advection models in order to work properly.

[rellis2012]

Thanks for the reply Glenn, very valid point on the dissipation also present in the second order scheme in High Res.

I guess with further mesh refinement and time step control then the High Res model would possibly become second-order dominated and produce a more similar results to the Bounded CDS?

I have been flicking through the ANSYS Best Practice for Scale Resolved Simulations which seems to recommend the High Res model for models like mine (simulation of wake behind a tidal turbine). Although from what i seem to be getting the Bounded CDS is the 'better' option, so I will most likely go forward using that. Just looking for any other comments you might have on the matter or does this seem logical based on what i have found so far?

Thanks again in advance, the help is always appreciated!

[ghorrocks]

I am glad to hear you read the best practices guide. But note it is a guide, if you are getting better results with CDS rather than HiRes then I would go with CDS.

If you want to explore this is little more then consider the Hybrid differencing scheme with the blend factor set to 1.0. This will force it to use second order everywhere. And CFX also has a lot of other differencing schemes available but hidden to the user, such as QUICK and others. Have a look in the <CFX ROOT>/etc/RULES file to see what is available.

[rellis2012]

Thanks very much for your comments Glenn, it is greatly appreciated! I will be sure to look at the other schemes, i was quietly hoping this might be possible but as per ANSYS don't make it to obvious.

Thanks,

Rob