[the_CFD_newbie]

Hello all!
Has anyone else run into Fluent taking much longer to solve when using RKE than when using kw-SST?

Typically, I've seen kw-SST and RKE take approximately the same amount of wall time to compute a certain number of iterations. But occasionally I'll have a case where RKE takes roughly 7-8x longer than kw-SST and I'm not exactly sure what the culprit is. For me, this appears to only happen when dealing with wall-resolved meshes (y+ <=1).

These jobs are running on another machine where I am not able to monitor the hardware. However, the cases are not large so the machines should have more than enough horsepower to chug through the job.

This is entirely empirical, but I've had it happen to me a couple of times recently and I'm not able to find anyone else discussing it online. Just wondering if anyone has any insight into this and if it's a known thing due to the model.

Thanks!

[LuckyTran]

You're using the density based solver I presume?

They solve the same number of transport equations so they do take the same time to solve, unless one is having trouble converging. But that would be an anecdotal occurrence and it wouldn't really be due to a particular model being used.

[the_CFD_newbie]

Originally Posted by LuckyTran

You're using the density based solver I presume?

Good question, I'm actually using PBC for these cases.

I think it may just be an anecdotal occurrence. But maybe it could somehow stem from the mesh?
For more context - if I have the solver run for 100 iterations the k-w model may take 1 hour to compute that, but the k-e may take ~7 hours. Fluent's output showing 'time/iter' starts at about 1 hour for both the k-w and k-e models, but a few iterations in the k-e 'time/iter' jumps up and it ultimately takes much longer to solve those 100 iterations.

Typically this is not the case. As you mentioned, same number of transport equations and basically same time to compute. But I've had this happen to me a handful of times recently and I was able to repeat it a couple of weeks ago. This makes me believe it's not an issue with the computer bugging out, but something with the mesh/solver?? Just some weird behavior.

Thank you for the input!

[LuckyTran]

So you are using a coupled solver... well there you go!

You should think about what are the differences between pressure based segregated solver (i.e. simple), pressure based coupled solver, and density based (coupled) solver. What exactly means coupled and why one chooses coupled over a segregated solver. What does the coupled mean? What happens when the coupled iteration fails? What happens when you have a high Courant number?


The answer really isn't that important. The takeaway is, CFD doesn't always converge!

[the_CFD_newbie]

Hey Lucky, thanks again for the input!

If I'm understanding you correctly, it's that the k-e model is taking many more 'sub-iterations' to converge on a single 'outer-iteration'? But somehow the k-w model doesn't require so many sub-iterations for this case?

I like the 'brute-force' analogy you mentioned regarding PBC methods in this post from years ago: pressure-velocity coupling scheme

[AnhDL]

What wall function did you use for RKE model?