[ericthefatguy]

Hello everyone,

I did some simulations of a NACA64a010 airfoil in low Reynolds number air flows using SU2, and the Cds I got from SU2 are much larger than the measured ones in the UIUC low-speed airfoil data. (Here)

I also ran some cases in Fluent using the exactly same mesh, set up with identical physical & numerical settings, but the Cds I got from Fluent consist quite well with the experimental results.

Am I missing something in my configuration of SU2 for low-Re simulations?

The mesh I used for those simulations is a hybrid one with well-resolved boundary layers at y+=1, and is overall, of quite decent quality. In both SU2 and Fluent, I used Roe for convection, 2nd-order flow spatial integration, 1st-order turb spatial integration, Euler_implicit for time discretization, and Menter’s SST for Turbulence modeling. I also used the CFD-Online’s turbulence property converter (Here) to ensure that both solvers initialized the free-stream with identical turbulence properties.

Due to the file size limitations of this forum, I cannot upload the mesh file or the output file to this thread, so I will post the Cds from the experimental data and the two solvers, along with selected velocity contours at the end of this post.

One interesting observation of the velocity contours is that, the predicted separation always happens much earlier in SU2 than that in Fluent, and the “separation bubble” in SU2 is also always much larger.

Thanks,
Eric


Experimental results, from page 172, volume 1, UIUC low-speed airfoil data:
Re= 60,000 AoA= 0 Cd= 0.012
Re= 60,000 AoA= 6 Cd= 0.029
Re= 100,000 AoA= 0 Cd= 0.010
Re= 100,000 AoA= 6 Cd= 0.028
Re= 200,000 AoA= 0 Cd= 0.011
Re= 200,000 AoA= 6 Cd= 0.020
Re= 300,000 AoA= 0 Cd= 0.008
Re= 300,000 AoA= 6 Cd= 0.018

Simulation results from SU2:
Re= 60,000 AoA= 0 Cd= 0.039
Re= 60,000 AoA= 6 Cd= 0.050
Re= 100,000 AoA= 0 Cd= 0.023
Re= 100,000 AoA= 6 Cd= 0.033
Re= 200,000 AoA= 0 Cd= 0.016
Re= 200,000 AoA= 6 Cd= 0.026
Re= 300,000 AoA= 0 Cd= 0.014
Re= 300,000 AoA= 6 Cd= 0.024

Simulation results from Fluent:
With low-Re correction toggled on：
Re= 60,000 AoA= 0 Cd= 0.015
Re= 60,000 AoA= 6 Cd= 0.027
Re= 100,000 AoA= 0 Cd= 0.012
Re= 100,000 AoA= 6 Cd= 0.023
Re= 200,000 AoA= 0 Cd= 0.009
Re= 200,000 AoA= 6 Cd= 0.019
Re= 300,000 AoA= 0 Cd= 0.008
Re= 300,000 AoA= 6 Cd= 0.016

With low-Re correction toggled off:
Re= 60,000 AoA= 0 Cd= 0.016
Re= 60,000 AoA= 6 Cd= 0.026
Re= 100,000 AoA= 0 Cd= 0.014
Re= 100,000 AoA= 6 Cd= 0.023
Re= 200,000 AoA= 0 Cd= 0.012
Re= 200,000 AoA= 6 Cd= 0.018
Re= 300,000 AoA= 0 Cd= 0.011
Re= 300,000 AoA= 6 Cd= 0.016

Velocity contour of Re=200,000, AoA=6, SU2:
Re= 200,000_AoA=6_SU2.jpg

Velocity contour of Re=200,000, AoA=6, Fluent:
Re= 200,000_AoA=6_Fluent.jpg

[fpalacios]

Originally Posted by ericthefatguy

Hello everyone,

I did some simulations of a NACA64a010 airfoil in low Reynolds number air flows using SU2, and the Cds I got from SU2 are much larger than the measured ones in the UIUC low-speed airfoil data. (Here)

I also ran some cases in Fluent using the exactly same mesh, set up with identical physical & numerical settings, but the Cds I got from Fluent consist quite well with the experimental results.

Am I missing something in my configuration of SU2 for low-Re simulations?

The mesh I used for those simulations is a hybrid one with well-resolved boundary layers at y+=1, and is overall, of quite decent quality. In both SU2 and Fluent, I used Roe for convection, 2nd-order flow spatial integration, 1st-order turb spatial integration, Euler_implicit for time discretization, and Menter’s SST for Turbulence modeling. I also used the CFD-Online’s turbulence property converter (Here) to ensure that both solvers initialized the free-stream with identical turbulence properties.

Due to the file size limitations of this forum, I cannot upload the mesh file or the output file to this thread, so I will post the Cds from the experimental data and the two solvers, along with selected velocity contours at the end of this post.

One interesting observation of the velocity contours is that, the predicted separation always happens much earlier in SU2 than that in Fluent, and the “separation bubble” in SU2 is also always much larger.

Thanks,
Eric


Experimental results, from page 172, volume 1, UIUC low-speed airfoil data:
Re= 60,000 AoA= 0 Cd= 0.012
Re= 60,000 AoA= 6 Cd= 0.029
Re= 100,000 AoA= 0 Cd= 0.010
Re= 100,000 AoA= 6 Cd= 0.028
Re= 200,000 AoA= 0 Cd= 0.011
Re= 200,000 AoA= 6 Cd= 0.020
Re= 300,000 AoA= 0 Cd= 0.008
Re= 300,000 AoA= 6 Cd= 0.018

Simulation results from SU2:
Re= 60,000 AoA= 0 Cd= 0.039
Re= 60,000 AoA= 6 Cd= 0.050
Re= 100,000 AoA= 0 Cd= 0.023
Re= 100,000 AoA= 6 Cd= 0.033
Re= 200,000 AoA= 0 Cd= 0.016
Re= 200,000 AoA= 6 Cd= 0.026
Re= 300,000 AoA= 0 Cd= 0.014
Re= 300,000 AoA= 6 Cd= 0.024

Simulation results from Fluent:
With low-Re correction toggled on：
Re= 60,000 AoA= 0 Cd= 0.015
Re= 60,000 AoA= 6 Cd= 0.027
Re= 100,000 AoA= 0 Cd= 0.012
Re= 100,000 AoA= 6 Cd= 0.023
Re= 200,000 AoA= 0 Cd= 0.009
Re= 200,000 AoA= 6 Cd= 0.019
Re= 300,000 AoA= 0 Cd= 0.008
Re= 300,000 AoA= 6 Cd= 0.016

With low-Re correction toggled off:
Re= 60,000 AoA= 0 Cd= 0.016
Re= 60,000 AoA= 6 Cd= 0.026
Re= 100,000 AoA= 0 Cd= 0.014
Re= 100,000 AoA= 6 Cd= 0.023
Re= 200,000 AoA= 0 Cd= 0.012
Re= 200,000 AoA= 6 Cd= 0.018
Re= 300,000 AoA= 0 Cd= 0.011
Re= 300,000 AoA= 6 Cd= 0.016

Velocity contour of Re=200,000, AoA=6, SU2:
Attachment 36637

Velocity contour of Re=200,000, AoA=6, Fluent:
Attachment 36638

Hi Eric,
Sorry for the late reply. These kind of posts are very important for us, and we want to be sure that we provide the correct answer.

My guess is that the disagreement between SU2 and fluent is because of the parameters of the slope limiter. I have run the same problem (Re 300.000, 0 AoA and 6 AoA) without slope limiter and the results match with UIUC low-speed airfoil data better than fluent (SU2 gives 0.009 and 0.018 respectively). I have used the grid that you can find in
$/TestCases/unsteady/pitching_naca64a010_rans

Please find attached the config file and force_breakdown files (as a reference). It is critical to use the latest version in the develop branch (v3.2.7.2) https://github.com/su2code/SU2/tree/develop

Best regards,

Francisco Palacios
SU2 lead developer

forces_breakdown_0deg.txt

forces_breakdown_6deg.txt

turb_NACA64A010.txt

[ericthefatguy]

Originally Posted by fpalacios

Hi Eric,
Sorry for the late reply. These kind of posts are very important for us, and we want to be sure that we provide the correct answer.

My guess is that the disagreement between SU2 and fluent is because of the parameters of the slope limiter. I have run the same problem (Re 300.000, 0 AoA and 6 AoA) without slope limiter and the results match with UIUC low-speed airfoil data better than fluent (SU2 gives 0.009 and 0.018 respectively). I have used the grid that you can find in
$/TestCases/unsteady/pitching_naca64a010_rans

Please find attached the config file and force_breakdown files (as a reference). It is critical to use the latest version in the develop branch (v3.2.7.2) https://github.com/su2code/SU2/tree/develop

Best regards,

Francisco Palacios
SU2 lead developer

Attachment 36690

Attachment 36691

Attachment 36692

Hi Francisco,

Thank you for your reply. It turns out that my slope limiter coefficient was set as 0.1 and was kind of smallish. I was running the NACA64A010 case using the incompressible solver of SU2, and after experimenting with some larger limiter coefficient then turning the slope limiter off after 1000 iterations, the results I got with this approach is still noticeable different with the experimental data.

The compressible solver, on the other hand, gives me decent results with reasonable parameters of the slope limiter, and turning off the slope limiter after some iterations goes even better.

At mach= 0.013 ~ 0.15, it should be perfectly fine to assume that the compressibility can be neglected, so I think the main reason behind this issue might be something related to my setting of the incompressible solver.

Attached is my config file of the incompressible NACA64A010 case for the mesh in $/TestCases/unsteady/pitching_naca64a010_rans.

Am I missing something for setting up the incompressible solver?

One other question, can you give me a reference for setting up the parameters of the adjoint slope limiter (SHARP_EDGES_COEFF, REF_SHARP_EDGES)?

Thanks,
Eric

64a010_test .txt