Convergence issues in transonic regime
 

Dear all,

I am conducting unsteady simulations in an airfoil at transonic regime. I normally use another CFD solver that requires to have 1 cell in the y direction in order to simulate 2D flow cases (similar to OpenFOAM). However I am performing quasi-3D simulations in Fluent (by setting symmetry in the symmetry walls) in order to make sure that my mesh will work in the other CFD solver. The flow conditions are Mach 0.77, Re 13,2 Million, AoA 1 deg for a chord of 2 meters.

I have been stucked for one week with the steady phase of my simulation because my flow field presents some discontinuities at the shock-wave region that don't let my simulation to fully converge (see the flow field and residuals screenshots). The flow field at the trailing edge and at the pressure side is totally fine. I tried to run the same mesh in the other CFD solver and the results are very similar. Therefore there must be something wrong with the mesh or the solver set-up.

I have also attached a screenshot of the mesh (which has around 400k elements) at the leading edge, a contour of the y+ values and the solution methods. I am using the k-w SST turbulence model.

From my past experiences, the used mesh has reasonable sizings for these kind of simulations and I am not sure which changes I should introduce. Do you have any suggestions regarding how to face this convergence issue?

Thank you in advance!

Throwing a bunch of ideas:


Shouldn't be periodic boundary conditions? Why symmetry? Try periodic boundary conditions pls.

Do you have curvature correction ON in the k-omega SST model? If yes, turn it off. I've seen steady solver convergence improve a lot without it. (And if you don't stabilize your solution, any additional accuracy you would gain is not gained)

Which solver? pressure-based or density-based? have you tried pressure-based with "coupled" scheme?


It seems there is some unsteadiness in the interaction of shock wave–turbulent boundary layer interaction. e.g.


https://www.cambridge.org/core/journ...B139A1F83D618D


https://www.annualreviews.org/doi/ab...-010313-141346


On the other hand, I think your problem is converged. I can see the seperation and Lambda-shaped shock-boundary layer interaction. Do you have any exprimental data to validate your solution?

Thank you LoGaL for your answer. I will try to a simulation changing the BCs to periodic and deactivating the curvature correction however I don't think it will be different since I am having the same exact issues in the other CFD solver with a pure 2D simulations and chien's K-epsilon model. I will keep you update about this.

Dear Roh,

Thank you so much for your answer. I am using a density-based solver and I would like to keep using it because the solver that I have to use has a compressible formulation and therefore the scheme is coupled. I am not fully aware of the advantages of the coupled pressure-based scheme vs the density-based scheme.

You are right, in this simulation I will be studying the buffet instability which causes the SW to move back and forth. Therefore in this specific scenario the flow is prone to develop important instabilities. Another researcher has conducted the same exact simulations but in a significantly lower Reynolds number (~1-2 Million) and the steady state solution (before switching to unsteady flow) looked like a perfectly smooth flowfield with the expected SW discontinuity. This leads me to think that the issues might come from the new mesh that I have generated for this higher Reynolds number scenario.

I will take a look at the articles that you have suggested, thanks!

Dear both,

I have tried your suggestions but I am reaching the same (apparently) "unconverged" solution.

If the lambda shock is clearly defined, does it mean that the simulation is converged? I don't understand how it is possible that the SWBLI region and the shock's foot are well resolved but the upstream SW region present such waviness in the flowfield. Physically the flow is not subjected to any strong disturbance during the curvature accelaration. However it seems that (numerically speaking) the solution is a bit unstable.

Kind regards.

Hi

Put point probes around the domain and see if the quantities ( e.g velocity) oscillate. To me it’s not converged at all

By the way 400k elements for a 2 D mesh is massive, this may also give problems