[Greg Hwang]

Hey guys! I am a newcomer to CFX.

Recently I was working on the transient simulation of a rotating seal model with tip leakage(the tip clearance is set to 0.5 mm, 2% height of the blade). I found that it is unable for the transient simulation to reach the prospectively residual (1e-4 or 1e-5, which is required by many papers I read ) since the steady simulation reached well. I tried to change the number of elements from 6.8 million to 70 million, however all the RMS Mom remained between 1e-3 and 1e-4. Then I got rid of the tip clearance in that model, the transient residual could reach below 1e-5 well.

Hence, I infer that the small tip clearance might be the very cause of bad residual in transient simulation. Nevertheless the small tip clearance is the focus point of my research, I cannot just get rid of the clearance. I would appreciate it if anyone could help

1. Is my inference rational?
2. How can I improve the residual for the transient simulation of rotating seal model with small tip clearance？

For your information:
[Time duration]:total time 0.5s
[Time step]:1e-4\5e-5\1e-5 were tried
[Coeff loops]:3 to 5
[Turbulence]:SST

[ghorrocks]

Your question is a FAQ: https://www.cfd-online.com/Wiki/Ansy...gence_criteria

It is normal for a finer mesh to be less stable than a coarser mesh. It as less numerical dissipation, so is less able to damp the motions.

[AtoHM]

To be completely sure about the residuals you can let cfx write out the residual magnitudes in the domain. Then in CFD-post you can define a point location or whatever and plot the residuals.

(Point location allows you to find the maximum value instantly) This will give you certainty about the location.

[Greg Hwang]

Thanks a lot for your reply and I will check FAQ later.

Actually what you talked about that finer mesh shall be less stable than coarser mesh is really contrary to my common sense. However when I use same case by only changing turbulence to k-epsilon or RNG k-epsilong, the residual could reach even 1e-6. What I am confused about is why just SST turbulence work so bad.

Thanks again, and wish you a good day.

[Greg Hwang]

Hey, thanks for your reply.

Actually I am not clear what you suggested me to do. The residual problem I asked happened in the process of solving, and the residual was updated with each timestep. Hence I am confused that how finding the maximum value in the CFD-post could help me improve the residual reaching 1e-4 or better?

Please enlighten me with your instructions if I missed anything. Thanks again.

[ghorrocks]

Actually what you talked about that finer mesh shall be less stable than coarser mesh is really contrary to my common sense.

Common sense? Can you please explain what you mean? Or are you just guessing?

Finer meshes have less numerical dissipation due to reduced truncation error, and reduced dissipation means reduced numerical stability which means harder to converge. A bit of knowledge of how CFD works trumps somebody's "common sense" guesswork.

What I am confused about is why just SST turbulence work so bad.

There could be many reasons depending exactly what you are modelling. But a likely explanation is that k-e and RNG k-e models are well known to over-predict the turbulent viscosity in many situations, including reattachments, low Re areas and some free shear layers. Same as before, increased dissipation means easier to converge.

And to answer your original question - please read the FAQ I linked to. I have answered this question thousands of times and I got sick of writing it over and over so wrote an FAQ on it. Please read the FAQ, it really does discuss the important issues.