[knax]

I am using k-epsilon, 3D modelling of an internal flow pipe with many obstructions inside. My residuals seem to be almost flat after some initial slope. After which I used under-relaxation, and my residuals suddenly decrease and then flatten out again. This has happened three times upon further URFs. I wanted to know that if it flattens out at the order 10e-6, will it be a useful calculation? Or must I improve the mesh altogether? How can I know if it has or nearing convergence?

[LuckyTran]

Look at your solution values and see what is happening to them, don't only look at residuals. The change in your solution from each iteration is convergence, whereas residuals are a measure of tolerance and how well the governing equations are satisfied.

Unless you simulation is diverging, leave the urf's alone. Lowering the urf's is deceptive as it makes your solution change less each iteration and also allows residuals to locally decrease (because the non-linear update happens less often). To be sure, you should do many more iterations to make sure the non-linear update is included.

Residuals should become flat are some point, because you eventually cannot get any better a solution. The better your grid the more iterations you can do before grid limits the solution tolerance.

[knax]

Thanks a lot for your advice. So can I just put a Cl or Cd surface monitor and just see if they're changing too much? When my K and epsilon residuals were flattening around 1e-4 and continuity at around 1e-2, my static pressure surface monitor was changing between 1.866e+3 to 1.868+3.

What do you make of it? Can I make use of these values at all. Thanks for taking your valuable time.

[LuckyTran]

Yes I would check Cl and Cd and see if the changes are significant.

The results are what they are. You just have to accept that there is some variation in the pressure from iteration to iteration. Is it significant? That is a judgment call you need to make. If it is significant then you can think of ways to improve your simulation, but always you will have a similar issue, the question is how large it is.

[knax]

K and epsilon are of the order 10-5 and Cl fluctuates slightly. Between values 1.1313, 1.1312 and 1.1316.

What can you make of this?

Is this good enough for a reasonable/converged answer?

Also, is the Cl that is calculated averaged over the entire grid domain? And hence, how good a reflector is it for accuracy?

Thanks a lot again.