Reynolds stress model convergence issues
 

Hello all,

I am simulating the flow around two buildings next to each other. The buildings are presented by simple blocks 20x40x20m. and are 6m. apart. The goal is to determine the wind velocity at a height of 2m. on a line in the middle of the two buildings. I am comparing my results with experimental values.

I started off by using k-omega SST turbulence model and with low y+ values (the highest is 1.9, while the majority is below 1). I managed to achieve good convergence after about 2500 iterations (I actually ran it for 5000 iterations).

However, I need to run it with Reynolds stress model (RSM) too. I switched to using only 2 prism layers so that the y+ value is above 30 at most places and I am using quadratic pressure strain model which requires high y+. That is the only change I made and couldn't get a converged solution.

Some more info on my set up:
-As the model is symmetric, I am only using one of the buildings and a symmetry plane
-trimmed mesh with additional refining blocks (cell count is 5mil, while the authors of the paper I am following managed to do it with 1mil)
-inlet and outlet are far enough from the buildings
-top and side surfaces are symmetry planes while the bottom one is wall (representing the ground)
-further refinement of the mesh didn't solve my problem (I tried with up to 27mil. cells)
-also I have checked mesh quality, volume change and skewness and they all seem to be fine
-I am judging convergence by also measuring the force on the building which in the k-omega case was slightly oscillating but the velocity in the middle of the buildings was not changing at all at any of the positions for the last 2500 iterations.

I have been playing with under-relaxation factors but without much success. I really can't think of any more tricks but also this is my first time using RSM model. I would really appreciate it if anyone knows how to improve my simulation and reach a converged solution. Also let me know if you require any further information on my set up but that's all I could think of for now.

having already achieved a good k-omega solution i hope you simply changed the required physics and mesh and restarted from that solution.

there are several ways to judge convergence and in many cases unsteady flow phenomena which a good and undamped solver with better physics and finer mesh discover and so prevent a 'conventional' converged solution to happen.

i've seen cases where just a small change to the mesh or physics can suddenly result in quite different flow eg picking up new unsteady eddies, so comparing results from some published results can be problematic.

it is not common to have to turn to the unsteady solver to continue running and then do some averaging to produce 'steady' results.

Hello,

thank you for your reply. I didn't change anything apart from the turbulence model. Mesh remained the same.
I also tried elliptic blending model which is recommended in the star-ccm documentation but got an error: ""A floating point exception has occurred: floating point exception [Overflow]. The specific cause cannot be identified."

So with quadratic pressure-strain, I can't get a converged solution and with elliptic, I keep getting an error. I have now started playing with different settings but nothing works. I have read everything I could on RSM (there isn't that much compared to other models) and tried all the tips that are given but nothing has worked so far.


I also thought about running unsteady sim but my employer wants it to be a steady sim. Do you think it could work if I start running RSM from the converged k-omega solution?

that is exactly what i said at the start of my reply. almost all more advanced physics models and solver settings require babying along and any help you can give them will help so a good initial solution could be a great help and will normally mean a much quicker time to convergence