To expand a bit on the question. The Reynolds number that I am using for the simulation is 22,000 and according to the literature the flow is turbulent. The paper that I am using as a reference ,which has the experimental data , measured a 2% turbulence intensity on the inlet whilst I used a 12% turbulence intensity for my simulation. I used at first a 12% turbulence intensity because I used the Star CCM+ userguide to calculate the turbulence intensity by using the turbulent velocity scale (tvs), which is assumed to be 10% of the freestream velocity. The equation is I = sqrt(1.5*tvs^2/Uinlet^2) and I also used a turbulent length scale of 7% of the inlet length for the inlet parameters. Now, when I use a 12% turbulence intensity the simulation gives me the correct results regarding the average Drag, rms Lift and frequency shedding but when i input a turbulence intensity of 2% at the inlet the simulation doesn't convergence at all. What could be the cause of this discrepancy? I am using a low y+ mesh.
There are quite a few reasons that can cause divergence in RANS simulations and it is not always easy to debug those. In general, when you want to compare your simulation to wind tunnel data you have to do the simulation with the same phyiscal flow properties that were used in the tunnel. In such, it might be pure coincidence that your simulation data using a turbulence intensity of 12% match to the wind-tunnel data measured with a turbulence intensity of 2%.
Check for mesh quality and appropriate inital conditions. When your initial conditions are very far off from your expected solution it might happen that your solution process gets unstable. Also check that you presribe your turbulence intensity not only at the inlet but also at the outlet. Watch out for numerical decay effects of turbulent variables.
Due to the low Reynolds number I might assume you are simulation rather incompressible conditions and go with the segregated flow solver (SIMPLE approach). In case you still struggle with convergence issues after checking the mesh and all physical properties you might want to try lowering the under-relaxation factors for the first iterations until the biggest instabilities have faded out
I suppose you are using a RANS formulation, convergence implying you reached the solution for the steady state (but I see you wrote " frequency shedding " that makes me doubt about that). Please specify better.
What I can suppose is that the higher intensity level force the turbulence model to produce higher turbulent viscosity. In other words the flow has more total viscosity that allow you to get a convergent solution. It is quite common that separated unsteady flows are difficult to solve using the RANS formulation. You can find in literature that URANS is often proposed to alleviate the problem (but I disagree about that).
However, other reasons can be possibile, try to change the turbulence model, to adapt better the grid in the separation region.
What kind of spatial discretization are you using? What about the method to treat the coupled equations?
Just have a look at how Star CCM uses Turbulent intensity and how does your reference paper uses it. There might be some differences.
Diffusion may be propagated where the frequency tends to be zero or near-zero value which is very prone to intense fluctuation to every tiny stimulation, that is why you get divergence in low number turbulent intensity, so in my view, you cannot use the same value 2% for your case at all. However, try to increase Y+ it may leave an effect on stabilization, though not directly.
Greetings
Did you try using different turbulence models to get a better numerical result to match the experimental value ?
Also, the turbulence intensity (%) can be specified as I = 0.16Re^(-1/8).
- Badges
- Science topic