C mesh is not converging for transonic airfoil

cspode · July 2, 2015, 00:05

Hello,

I'm facing some problems while running simple transonic airfoil flow using C mesh like the attached one ("C_grid.jpg"). The trailing edge is sharp. The y+ is OK, around 1.

When I run the same case using an O grid the residuals goes smoothly to convergence even using Multi Grid. See "Cl_Omesh.jpg".

Under C mesh, the solution dit not converge and after more than 1000 iterations the residual and lift are still oscillating. See "Cl_Omesh.jpg". In the C mesh case, if I turn on the Multi Grid the run just diverges after few iterations. I've also tried to reduce the CFL, but it doesn't help.

Do I need to set up something special for C type mesh? Some one knows if is there any restriction for very stretched elements in the farfield boundary like the case of C type meshes?

All the suggestions are welcome.

Thanks,

cspode.

DLuo · July 6, 2015, 12:44

Have you tried plotting the point of Maximum resdidual as reported in the LOG file? Also have to run it out past 1000 iter? Just because the O Grid converges in 1000 iter doesn't mean the C grid will as well. It's likely that given more compute time you'll see the C-Grid start converging as well.
If it still doesn't converge take a look at the maximum residual locations and consider plotting the flow-field over several different time slices to see how it's varying. My guess is that something around the trailing edge is causing havoc in the solution for the C-grid.

cspode · July 6, 2015, 21:55

Thank you for your suggestions DLuo. Even with more iteration (50k) I've got only two order reduction in the residual, but, at least, the lift converges almost flat (after 15000 iter). Using SST turbulence model seems to help on numerical stability. The points of maximum residual are always on the wake, near y=0 but not specifically at the trailing edge, it varies during the simualtion. I think it's not a mesh problem, as I've run the same mesh in other solvers and it just ran smoothly, with 6 order residual reduction in much less iterations.

DLuo · July 7, 2015, 01:00

Agreed your C-grid mesh will run perfectly fine in other solvers but not necessarily SU2. From my experience SU2 is more of a raw CFD solver with less behind-the-scenes magic than some of the other commercial solvers out there. One reason commercial codes charge as much as they do in licensing fees is because of how much effort they've put into it so that it can handle less than ideal meshes.
That's not to say the quality of your C-Grid mesh is bad, but the propagation of the high-aspect ratio cells in your wake may not be 100% good, I can't really say since I don't have your mesh. I also can't really tell from the image but it also appears that this airfoil has a sharp trailing edge. I know for a fact that SU2 is not friendly with sharp TEs unless you split the local mesh at the TE using something like a "pole" in pointwise terms.
There are various other tips I've seen on this forum that are go-to pointers for troubleshooting. Turn of multi-grid. Change CFL to ~1 or lower.
As far as why the O-Grid is much easier to converge than the C-grid given that the rest of your setup is identical, all I can say is that it's down to the mesh quality. Since you said the max residuals occur in the wake region of the mesh try smoothing the wake or even diffusing it out so that you have better control of the quality there.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
2D FFD Optimization	RLangtry	SU2	2	August 5, 2014 10:48
Mesh motion with Translation & Rotation	Doginal	CFX	2	January 12, 2014 07:21
mesh size requirement for NACA0015 airfoil	zhengjg	Main CFD Forum	0	March 20, 2013 22:58
[Other] 2D hex mesh on multi element airfoil	Verfblikje	OpenFOAM Meshing & Mesh Conversion	0	January 19, 2012 11:55
getting airfoil surface to be recongized for tri mesh	josip76	ANSYS Meshing & Geometry	4	June 9, 2011 23:48

July 6, 2015, 12:44		#2
DLuo Member D L Join Date: Jun 2012 Posts: 49 Rep Power: 14	Have you tried plotting the point of Maximum resdidual as reported in the LOG file? Also have to run it out past 1000 iter? Just because the O Grid converges in 1000 iter doesn't mean the C grid will as well. It's likely that given more compute time you'll see the C-Grid start converging as well. If it still doesn't converge take a look at the maximum residual locations and consider plotting the flow-field over several different time slices to see how it's varying. My guess is that something around the trailing edge is causing havoc in the solution for the C-grid.

July 6, 2015, 21:55		#3
cspode New Member Cleber Spode Join Date: Mar 2013 Posts: 5 Rep Power: 13	Thank you for your suggestions DLuo. Even with more iteration (50k) I've got only two order reduction in the residual, but, at least, the lift converges almost flat (after 15000 iter). Using SST turbulence model seems to help on numerical stability. The points of maximum residual are always on the wake, near y=0 but not specifically at the trailing edge, it varies during the simualtion. I think it's not a mesh problem, as I've run the same mesh in other solvers and it just ran smoothly, with 6 order residual reduction in much less iterations.

July 7, 2015, 01:00		#4
DLuo Member D L Join Date: Jun 2012 Posts: 49 Rep Power: 14	Agreed your C-grid mesh will run perfectly fine in other solvers but not necessarily SU2. From my experience SU2 is more of a raw CFD solver with less behind-the-scenes magic than some of the other commercial solvers out there. One reason commercial codes charge as much as they do in licensing fees is because of how much effort they've put into it so that it can handle less than ideal meshes. That's not to say the quality of your C-Grid mesh is bad, but the propagation of the high-aspect ratio cells in your wake may not be 100% good, I can't really say since I don't have your mesh. I also can't really tell from the image but it also appears that this airfoil has a sharp trailing edge. I know for a fact that SU2 is not friendly with sharp TEs unless you split the local mesh at the TE using something like a "pole" in pointwise terms. There are various other tips I've seen on this forum that are go-to pointers for troubleshooting. Turn of multi-grid. Change CFL to ~1 or lower. As far as why the O-Grid is much easier to converge than the C-grid given that the rest of your setup is identical, all I can say is that it's down to the mesh quality. Since you said the max residuals occur in the wake region of the mesh try smoothing the wake or even diffusing it out so that you have better control of the quality there.