I try to simulate the 2D unsteady air flow through a rectangle. I always met the problem 'turbulence viscosity ratio exceed 10e5'. I search this problem in the Forum, and I make sure the boundary conditions(simple: velocity-inlet and pressure-outlet), scale, viscous model is Realizable k-epsilon and Non-Equilibrium wall function. The mesh is very dense(smallest is 0.5mm and ratio is 1.01). Under-relaxation factors are all 0.1. I also try other viscous models but it didn't work. The time step is small enough(0.0001). I check the region where this number is 10e5, it's in the wake behind the rectangle. The shape of the flow is beautiful like Karman vortex street and it didn't divergent after 60000 iterations. I think it's perhaps because it's very turbulent in the wake. So should I use LES in this case? Thanks a lot.

Hi

I'd suggest you use adaption with velocity gradients first before you think of LES. This message sometimes disapears given enough iterations, and it is mainly due to poor mesh or assigning the wrong material properties, so please check those two too.

AAA

Thanks. I've forgot to mention this. I've used adaption with velocity gradients or pressure gradients or turbulence viscosity ratio, the number of mesh which exceed 100000 inceased...(I waited for more than 1000 iterations). I used structural mesh in this case and also checked the mesh where the problem happened. The dimension of mesh in the region is about 1mm*1mm, I think it's small enough(The rectangle is about 4m*4m, total domain is about 70m*80m). So how should I do to assign the mesh? Until now the number of iterations is more than 400000. Thank you again.

Those are pretty small relaxation factors aren't they? Aren't you concerned about getting worthless data?

Thanks. I learned from others that small relaxation factors can make convergence better. Why it can get useless data?