6 Replies Latest reply on Jun 27, 2011 11:47 AM by Bill McEachern

    Air flow accuracy based on computational domain size

    Karim Alame

      Hi all,

       

      I'm currently running  a simulation past an airfoil at a high Reynold number of 6*10^6.

      I have messed with two diffferent settings, whereby I used the default comp. domain and then another with a much larger domain ( so I dont get flow interaction with the boundaries) .

      The resultant forces are considerably different.

      Now I do want more accuracy but with a 1.8 GHz processor and 3 GB of RAM I'm pretty limited, a typical large domain calculation took me around 30 hours the other day.

       

      Also, Im running it for a physical time of 1 sec, would increasing that time lead to a better accuracy ?

       

      Thanks for your time, any input would be appreciated !

        • Re: Air flow accuracy based on computational domain size
          Bill McEachern

          If it is an essentially a 2D problem I would suggest you use that function and your compute times would drop to mere minutes even on your hardware. However if you are after the induced drag a big domain may or may not be required depending on the velocity of the flow. Tuning the mesh and using the autoadaptive functions might be a good idea to reducing computational times. It is relatively well known that the Flow sim turbulence model (K-e) starts to diverge from reality in and around your R'number so you should be aware of that.

           

          Why wouldn't you be running a steady state analysis?

            • Re: Air flow accuracy based on computational domain size
              Karim Alame

              Hi Bill,

              I did set it up as a 2D problem, as far as the Reynold's number goes it was a nice head up thanks, I'll change the velocity dependency to a constant instead of a function of Reynolds number, hopefully it'll be better.

              Do you suggest I refine the mesh behind the wake of an airfoil to get a better model ?

              Also mind if I ask what is the steady state analysis ?

                • Re: Air flow accuracy based on computational domain size
                  Bill McEachern

                  Hi Karim,

                   

                  It boggles the mind that you would not know what a state analysis as you would have had to switch on the time dependency to not get one. The method used by Flow sim is that of a time averaged Navier -Stokes code. Essentially every analysis is more or less time dependent however in a steady state analysis time is not tracked explicitly. It isn't really a steady state analysis, the term quasi steady is probably more appropriate  as they don't call it fluid dynamics for nothing. In Flow sim in a steady analyis you don't track time explicitly but the solution evolves in time tracked by iterations. You still need to run it long enough to get through whatever start up condition you put it - like an impulsive start and the flow has to evolve to it gets more or less steady but with still fluctations in the wake and other disturbed areas of the flow field. If you want to see vortex shedding then a transient analysis is required and you play and animation from the output at various iteration numers or time units that you choose to save at. If you just want the drag and lift a steady state analysis should work just fine.

                   

                  If you run at initial mesh numbers of 6, 7 or 8 the program will automatically remesh and refine during the solution. Otherwise you can turn it on under calculation cotrol options.

                   

                  What is the section of the airfoil you are looking at? Is there some phenomena you are trying to capture?

                    • Re: Air flow accuracy based on computational domain size
                      Karim Alame

                      That's what I thought I just needed to make sure, I did turn on the time-dependant analysis but only because I read through one of the tutorials for a flow over a cylinder that said for Re>40 a time-dependant is required in order to take into account the unsteady effects.

                       

                      I used the mesh setting of 7 at the beginning.

                       

                      Thanks for the clarification I've been reading up on CFD ( still a junior and havent taken that class yet) and I thought it would be a fun project to validate CFD with actual wind tunnel data I've obtained. I want to compare lift and drag so I guess I'll stick to steady-state analysis assuming I disregard that unsteady effects for Re >40 ?

                       

                      Also the only reason I wanted to capture the wake was so I can compare the vortex shedding with smoke visualization of an actual airfoil at a high angle of attack. Specifically the NACA 0012.

                       

                      Thank you so much for your time ! I really appreciate it.

                       

                      Best Regards.

                      • Re: Air flow accuracy based on computational domain size
                        Karim Alame

                        Hi Bill,

                        I finally got the simulation to work smoothly and it only took about 2 minutes of running time.

                        The results I got from the 2D airfoil where almost spot on to my wind tunnel data, however not surprisingly it did divergre at high angles of attack. As for the simulations I used a steady state analysis, can I achieve better accuracy at higher angles of attack or is that the limitation of the code? Would running a time dependant analysis give better results at those conditions ?

                        Thank you for your time.

                          • Re: Air flow accuracy based on computational domain size
                            Bill McEachern

                            Hi Karim,

                            At high alpha you get increasing areas of separation in the aft section up intil stall. Stall can be from the leading or trailing edge depending on what the nose looks like. Small nose radius typically of laminar foils tend to have leading edge stall. The NACA 00XX tend to have trailing edge stall. These flow separations are a highly transient affair - fluctuations. The other thing is that they tend to be 3D phenomena as well. You can try a transient analysis but you will need to "average the results" over some time inteval once the initial start up trainsient has died out. Give it a shot with a 2D analysis and after the start up transient has died out you would need to save the resutls quite frequenctly and then average out the fluctuations to get something analogous to what a manometer would see in a wind tunnel measurement. then again the 3D feild of hte separation may make getting a decent number problematic and you would have to run a 3D analysis and that would require knwing hte other dimensions of the test set up.

                            Good luck.