[CFXer]

Hello.

I am trying to simulate heat exchanger including porous domain.

The turbulence model is SST.

In fluid domain parts, I set the y plus value between 2 and 30, but it doesn't converge well.

So I have a question, is it essential to set y plus value under 30 in porous domain?

I use wall boundary condition as free slip wall, so I didn't consider it's y plus value.

Anybody has an idea, please notify me.

Thank you in advance.

[ghorrocks]

Mesh resolution and convergence are totally different things. A fine mesh does not mean your simulation will converge - in fact, quite the opposite - a fine mesh is harder to converge than a coarse one.

It sounds like this FAQ is relevant to you: https://www.cfd-online.com/Wiki/Ansy...gence_criteria

[Stel]

Originally Posted by CFXer

Hello.

I am trying to simulate heat exchanger including porous domain.

The turbulence model is SST.

In fluid domain parts, I set the y plus value between 2 and 30, but it doesn't converge well.

So I have a question, is it essential to set y plus value under 30 in porous domain?

I use wall boundary condition as free slip wall, so I didn't consider it's y plus value.

Anybody has an idea, please notify me.

Thank you in advance.

Low yplus values usually comes with a high computational cost and convergence problems. Fortunately, you don't always need that, it all comes down to your requirements. If you are really interested in solving the structures inside viscous sublayers, or if a macro quantity you are interested in is highly sensitive to wall refinement, so go for it. If your problem is not that sensitive to wall refinement, or if you are not interested in boundary layer flow altogether, then you could use a very regular mesh distribution as it tends to give so much better convergence. The ideal thing to do is to perform a careful mesh sensitivity test to investigate that.

[CFXer]

Thank you for the advice.

I have a question.

As you said, I tried out the mesh modify to set yplus value under 1, it's residuals became higher than before : about 1e-1, with huge oscillating.

How can I reduce the residuals when the small yplus applied? I thought the courant number is the problem so I reduced the timescale factor, it doesn't very effective.

And one more, how can I determine the residual criteria for convergence?

I have found a post about convergence criteria (https://www.engineering.com/DesignSoftware/DesignSoftwareArticles/ArticleID/9296/3-Criteria-for-Assessing-CFD-Convergence.aspx), but it is general situation, I want to know how I can determine 'my' simulation converged.

[ghorrocks]

Did you read the FAQ I linked to? It answers your exact question. I don't care to type it out again.

To determine the convergence criteria you should be using you should do a sensitivity analysis. Try a range of convergence criteria and look at the effect on a simulation output of interest to you. Choose the convergence criteria which gives the accuracy you require.

[CFXer]

I'm sorry that I cannot understand that link perfectly, I'm going to read it again.

And according to your advice, I'm going to do mesh sensitivity analysis, checking the important parameter increasing mesh number.

Thank you!

[ghorrocks]

If you don't understand the FAQ then feel free to post a question asking for clarification. If you don't understand something then lots of other people probably don't understand it either. Make sure you explain what you don't understand, of course.

[Stel]

Originally Posted by CFXer

Thank you for the advice.

I have a question.

As you said, I tried out the mesh modify to set yplus value under 1, it's residuals became higher than before : about 1e-1, with huge oscillating.

How can I reduce the residuals when the small yplus applied? I thought the courant number is the problem so I reduced the timescale factor, it doesn't very effective.

And one more, how can I determine the residual criteria for convergence?

I have found a post about convergence criteria (https://www.engineering.com/DesignSoftware/DesignSoftwareArticles/ArticleID/9296/3-Criteria-for-Assessing-CFD-Convergence.aspx), but it is general situation, I want to know how I can determine 'my' simulation converged.

Convergence problems for low yplus values can happen due to a number of things (carefully read the FAQ), but usually: 1 - your mesh is so fine that it starts to capture very small flow structures; it is hard to tell if they can be physical or not without a greater understanding of your problem. If so, you should try solving the problem as transient with some loops between timesteps. 2 - Numerical "noise"; using double precision solver could help, but you could also try the upwind scheme just to check if the problem is related to discretization. 3 - poor grid quality. No further advice here except to increase mesh quality, you could try a regular mesh distribution and see what happens.

In any case, your residuals are so high right now (1e-1 as you say) that I think that it is a result of multiple problems. Revise your setup, mesh and numerical aspects very carefully.

[CFXer]

Originally Posted by Stel

Convergence problems for low yplus values can happen due to a number of things (carefully read the FAQ), but usually: 1 - your mesh is so fine that it starts to capture very small flow structures; it is hard to tell if they can be physical or not without a greater understanding of your problem. If so, you should try solving the problem as transient with some loops between timesteps. 2 - Numerical "noise"; using double precision solver could help, but you could also try the upwind scheme just to check if the problem is related to discretization. 3 - poor grid quality. No further advice here except to increase mesh quality, you could try a regular mesh distribution and see what happens.

In any case, your residuals are so high right now (1e-1 as you say) that I think that it is a result of multiple problems. Revise your setup, mesh and numerical aspects very carefully.

Can you notice me how the "upwind" scheme can check if the problem is related to discretization? I don't know the exact meaning of discretization also... sorry.
Or can you link me that I can refer to the meaning of discretization or upwind scheme?

[Stel]

What I was suggesting you was to run the same problem but using the Upwind scheme instead. It is a very "smooth" discretization scheme so most of the time it is numerically stable, but also inaccurate for most problems. Thus, I don't recommend you to move on with it, but just to get a solution with it and see if it shows better convergence (a pure exercise to find where the problem is).

Sorry, I don't know an internet link where you can find some 101 about discretization, but a regular Google search should help you with the basics. You could also try any CFD book.