[GregCFD]

Hi everybody, another question on convergence. I'm currently doing some simulations of simple pipe flow to act as boundary conditions for a more complex simulation and I have started with the system used by Cornell for simplicity sake. The meshes I have are 30*100, 60*200, 120*400, 200*800, walls are biased with a factor of 10 and the meshing produces a high quality quad mesh. Based on flow conditions the Reynolds number is a sedate 10,000, and the pipe length is extended to ensure that the flow has reached a steady state. The software that I'm using is fluent but I'm hoping the problem I'm having is conceptual rather than the program (unlike last time...) hence posting it to this forum.

When I analyse the data I'm expecting complete convergence of the data, i.e. no change across any variable for the higher resolution meshs, but what I get is change mesh to mesh. For example at the outlet the velocity is similar except near the centre but the TKE and turbulent viscosity are quite different near the centre and the differences in the velocity are quite noticeable when you plot them along the centreline. With 48,000 and 192,000 nodes for the two largest meshes I can't imagine that the mesh needs more resolution to converge so I'm wondering where have I gone wrong. Is this kind of behavior expected from a CFD simulation, am I selecting the wrong data or locations to analyse?

[FMDenaro]

First, from the velocity profiles I see that the difference is very very small. Using RANS, the effect of the turbulence modelling overwhelms the effect of the local truncation error .

[GregCFD]

Originally Posted by FMDenaro

First, from the velocity profiles I see that the difference is very very small. Using RANS, the effect of the turbulence modelling overwhelms the effect of the local truncation error .

Thanks for the reply, the velocity profiles look fine but what I'm mostly worried about are the other three particularly the steady change in the centreline velocity down the length of the pipe and the way the turbulence variables seem to depart from each other near the centre.

[LuckyTran]

These pictures make sense. Keep refining the mesh if you're not happy with the results.