[kabat73]

Hello!
I'll try to simulate in Fluent 12.1 air flow through segmented elbow (3x30 degrees) 1.5D (1.5 times diameter) curvature. The fluid is air (treated as incompressible) because velocity range is 10 to 30 m/s. This is 3D problem. The inner diameter of a pipeline is 152mm. Before elbow there is 20D (20 diameter) length section (inflow side) to create velocity profile (upper side) and 40-50D after the elbow (outflow side - right hand).
The general problem is that after the elbow I can't get appropriate velocity profiles - they are asymmetrical through the entire outflow section (don't form to previous shape noticed before elbow - see 1st and 2nd picture). Before elbow velocity profile has very good shape (see 1st picture)! Velocity profiles after the elbow don't agree with my own experimental results (see 3rd picture) which are very accurate (and corresponds with literature data). I used k-epsilon (with non equilibrium or standard wall function), k-omega turbulence model as steady and transient form, changing mesh (about 1 million elements) made in Gambit 2.4.6 (structural hexahedral - equiangle skew below 0.5) including influence of outflow section length from 20 to 50D, checking y+ (30-60). I was also changing boundary condition - velocity inlet or pressure inlet, outflow or pressure outlet and still nothing! Approximation schemes are used as first or second order (first of all). Residuals are good converged (1e-5 for all - see 4th picture). As I mentioned I have experimental results for many lines from 3D to 20D for vertical and horizontal plane so I can experimental background to compare with CFD.
Additionally I asked Fluent support (in Poland) to help and they can't resolve this problem. I still believe that Fluent can simulate such classical flow problem.
Please help me to find where the problem can be to acquire reasonable numerical results. Any help could be VERY APPRECIATED.

Best regards for all Fluent users.

Fluent velocity profile 10m_s.jpg

Fluent velocity profile 10m_s zoom.jpg

Experiment at 10m_s vertical plane.jpg

Residuals.jpg

[nstar]

Hi,
We've done something similar before to verify the FLUENT capability, or in another word, the mesh requirements to caputure the velocity/vortex profile. With proper mesh size, the FLUENT is able to capture that.
The experimental data were obtained from a paper by a Janpanese guy. The experimental setup was to caputre the velocity profile after an elbow, which is a 90 degree one continously curved, not like the segmented one like yours.
The mesh was built in ANSA. All mesh is prism mesh, not hex mesh, which is actually better than our prism mesh. The volume mesh was created with the mapping method, basically triangle mesh was created on the circular-shaped inlet and mapped all the way to the outlet along the pipe. We estimated how many nodes were required on the diameter line of the inlet. With around 20 - 40 nodes, the fluent could be able to caputre the velocity profiles.
I don't know what exact the problem your simulation has, but hope my info will help you. By the way, do you think 600 iterations will be good enough?

Originally Posted by kabat73

Hello!
I'll try to simulate in Fluent 12.1 air flow through segmented elbow (3x30 degrees) 1.5D (1.5 times diameter) curvature. The fluid is air (treated as incompressible) because velocity range is 10 to 30 m/s. This is 3D problem. The inner diameter of a pipeline is 152mm. Before elbow there is 20D (20 diameter) length section (inflow side) to create velocity profile (upper side) and 40-50D after the elbow (outflow side - right hand).
The general problem is that after the elbow I can't get appropriate velocity profiles - they are asymmetrical through the entire outflow section (don't form to previous shape noticed before elbow - see 1st and 2nd picture). Before elbow velocity profile has very good shape (see 1st picture)! Velocity profiles after the elbow don't agree with my own experimental results (see 3rd picture) which are very accurate (and corresponds with literature data). I used k-epsilon (with non equilibrium or standard wall function), k-omega turbulence model as steady and transient form, changing mesh (about 1 million elements) made in Gambit 2.4.6 (structural hexahedral - equiangle skew below 0.5) including influence of outflow section length from 20 to 50D, checking y+ (30-60). I was also changing boundary condition - velocity inlet or pressure inlet, outflow or pressure outlet and still nothing! Approximation schemes are used as first or second order (first of all). Residuals are good converged (1e-5 for all - see 4th picture). As I mentioned I have experimental results for many lines from 3D to 20D for vertical and horizontal plane so I can experimental background to compare with CFD.
Additionally I asked Fluent support (in Poland) to help and they can't resolve this problem. I still believe that Fluent can simulate such classical flow problem.
Please help me to find where the problem can be to acquire reasonable numerical results. Any help could be VERY APPRECIATED.

Best regards for all Fluent users.

Attachment 3178

Attachment 3179

Attachment 3180

Attachment 3181

[kabat73]

Hi!
Thank you for your reply. My mesh is done very similar like yours, but in Gambit it was named Cooper. Could you send me more information about your mesh (maybe some screen shots). What was your y+ and turbulence model, steady or unsteady, ... etc.
I noticed that first steps between 500 and 1000 gives good velocity profiles, but after that profiles are getting worse and worse up to ~6000 time steps (dt=7e-5s). Velocity profiles at outflow parts of pipeline over 20D are getting strongly asymmetrical. Today I observed some problem with static pressure at that part of pipeline (12-20D and farther). I will still investigate this curious phenomenon trying to properly solve it. I will be very thankful to anyone who helps me in this matter.

Regards

[nstar]

This work was performed together with my colleague, unfortunately, I talked to him and he told me all files were not kept since it's a study done about two and half years ago.

I can volunteer to do a quick FLUENT run for you if you want to share more geomtry details with me.

Originally Posted by kabat73

Hi!
Thank you for your reply. My mesh is done very similar like yours, but in Gambit it was named Cooper. Could you send me more information about your mesh (maybe some screen shots). What was your y+ and turbulence model, steady or unsteady, ... etc.
I noticed that first steps between 500 and 1000 gives good velocity profiles, but after that profiles are getting worse and worse up to ~6000 time steps (dt=7e-5s). Velocity profiles at outflow parts of pipeline over 20D are getting strongly asymmetrical. Today I observed some problem with static pressure at that part of pipeline (12-20D and farther). I will still investigate this curious phenomenon trying to properly solve it. I will be very thankful to anyone who helps me in this matter.

Regards

[kabat73]

Hi!
Thanks for your quick reply. What kind of file formats do accept your preprocessor ANSA? I can export 3D geometry. If it will be any problem to read a converted file please prepare geometry of pipe with inner diameter of 152mm and 90 degrees elbow with curvature equal 1.5 times diameter counting to the elbow axis. Inflow section 20D and outflow 50D.

Best regards

[nstar]

ANSA takes lots of format, but I am not familiar with all of them. I normally use 'igs' and 'stl' files. I will try to create my own geometry since it's not complex anyway.

Originally Posted by kabat73

Hi!
Thanks for your quick reply. What kind of file formats do accept your preprocessor ANSA? I can export 3D geometry. If it will be any problem to read a converted file please prepare geometry of pipe with inner diameter of 152mm and 90 degrees elbow with curvature equal 1.5 times diameter counting to the elbow axis. Inflow section 20D and outflow 50D.

Best regards

[nstar]

Originally Posted by nstar

ANSA takes lots of format, but I am not familiar with all of them. I normally use 'igs' and 'stl' files. I will try to create my own geometry since it's not complex anyway.

Here's some quick run I did. the x-velocity (the direction along with the 50D out pipe) on the diameter of the pipe downstream, 3d, 4d, 5d, 7d ...

I noticed there's discripency b/w the fluent and the experiments...

[nstar]

btw, the -0.15 corresponds to inner side while -0.3 corresponds outer side. the simulation was done with 10m/s inlet air, the mesh is kinda coarse. i thought the velocity profile may not agree with your experimental results because you have ~20m/s velocity there. I later did another simulation with 20m/s inlet velocity, the profile looks pretty much identical except the magnitude.

I looked at the velocity profile at the very end of the out pipe, it still shows the elbow effect, in another word, still not symmetric

compared to CFD results your experimental data are showing the elbow effect diminished pretty fast at around 15-20D distance.

Originally Posted by nstar

Here's some quick run I did. the x-velocity (the direction along with the 50D out pipe) on the diameter of the pipe downstream, 3d, 4d, 5d, 7d ...

I noticed there's discripency b/w the fluent and the experiments...

[kabat73]

Hi,
Thank you very much for your contribution and effort. I'd like to ask you to send me txt format with results of your calculations in txt format for data you made graphs.
I recently received new results of simulations (only txt data for the moment) from Fluent support and they agree pretty well to experiment for 20D. I'm waiting for complete solution including mesh settings.
Thanks once again for your work.
Regards