[IngmarVanDijk]

Dear all,

I am new to this forum but have been doing CFD for a few years. Normally I calculate fairly simple quasi-one-directional systems that converge easily, but right now I am running into problems.

I am simulating a boiler behind a gas turbine. The gas turbine outlet conditions are my inlet conditions. The high inflow velocity that flows past an open area creates strong vorticity, and any model that has this vorticity seems to create convergence problems.


A figure of the model: http://i.imgur.com/uiGcLr2.png

I tried:
Refining the circulation region, has very limited effect, the circulation is captured by at least a few hundred elements across
Switching to double precision - No effect
Running for a long time - Continuity Residual flatlines at 5e-3 for at least 500 iterations (and in 50 iterations it reaches these values)


I am using a thin boundary layer, 5 elements, 1mm first layer, Omega-SST turbulence model. I understand this should go well together with any thin boundary layer?




And advice on reducing my residuals is highly appreciated. Further refinement is difficult since I'm running 1.5~2M elements already.

[IngmarVanDijk]

I should also add that I looked at the difference two different points in the flatlined convergence, say 120iterations apart.

The solutions were indeed quite different, and thus I cannot trust the calculation yet.

[IngmarVanDijk]

Switched to K-Epsilon Realizable - Enhanced Wall Treatment, which gives me much better convergence. All values are below 1e-3 and all my pressure and uniformity monitors have leveled out and also change less than 1e-3 per iteration.

Perhaps this is a simplified conclusion, but it seems that the K-Omega SST doesn't work well with large vortices in a calculation. Perhaps because it was originally written to only work well in the boundary layer? (Before adding wall functions).


Any thoughts on this are appreciated =).


Kind regards,

[ghost82]

Hi,
thanks for sharing your solution.
I think the transition model needs a transient approach: this could be the reason. It is possible that your problem is not steady and the k-omega sst could't find a steady state solution.

[IngmarVanDijk]

Indeed the location and size of the vortex kept changing when using the SST model, however the "transition model" is the 4 equation one that's heavier than k-w-SST or k-e right?

Are you sure about the transient solution requirement?


I'm going to do some more tests, and compare k-w-SST and k-e-realizable for an easier model, that does converge for both. I'm hoping they are at least the same if properly converged... I'll see what I can share here.

[ghost82]

This is a comment of Thomas Frank, from Ansys Inc.:

A k-omega SST model delivers in most cases a more exact solution, it can resolve boundary layers down to finely resolved meshes with y+ <=1 and it is usually less dissipative (i.e. produces a smaller amount of turbulent viscosity). The price for this is that a k-omega SST model usually tends to predict already transient flow behavior where a standard k-eps model produces so much of turbulent viscosity that it stays in steady-state flow regime. People find that convenient. But in fact it can hide the true nature of the investigated flow phenomenon from being discovered by the simulation, so it can be a bit dangerous to rely on (exclusively).

[IngmarVanDijk]

Wow, very interesting, definitely going to dig into this! Thanks!