7 Replies Latest reply on Feb 25, 2015 12:55 PM by Bill McEachern

    Calculation time for lift on 3D wing

    Josiah Lund

      All,

       

      I have recently been working quite a bit with SW flow sim and have gotten good results in 2D. I am working my way up and have now begun trying to tackle a 3D wing. Find my 2D simulation here: Lift Coefficient on 2D Airfoil

       

      I have been able to somewhat successfully reach convergence for the 3D case. My concern is that even on a relatively powerful computer, calculation time is in the neighborhood of 24-30 hours. How can I keep this as low as possible yet ensure I am getting accurate results?

       

      I am able to check values for this simulation, but the next step is to include the fuselage of my plane, and I will have nothing to compare with so I want to perfect my methods before moving on.

       

      The wing I am simulating is built from the same airfoil as the 2D simulation.

      XFLR5 Wing.PNG

      Rather that go into details on the setup again, I will jump straight to my results and different methods I took.

       

      The Mesh was created using a level 1 initial mesh with a level 5 refine set to refine every 1 travel starting with the 2nd full travel. The resulting mesh is pictured below.

       

      Mesh Top.PNG

      Mesh Side detail.PNG

      Mesh Front.PNG

      Mesh Front Detail.PNG

      The simulation converged after after 21 hours with a mesh of 5.75 million fluid cells + 320,000 partial cells. The converged value for CL is 0.317, off from the theoretical value by about 6%.

       

      Goal Plot.PNG

      I am okay with the accuracy, but would like to know if there is a way that I could set the problem up differently to shorten the calculation time without bringing down the accuracy of the simulation.

        • Re: Calculation time for lift on 3D wing
          Jared Conway

          rather than going straight to a refined mesh, you need to start coarse and see where accuracy starts to converge

           

          other than reducing the number of cells or throwing hardware at the problem, there is no other way to reduce the solve time

            • Re: Calculation time for lift on 3D wing
              Josiah Lund

              I may misunderstand what you are saying about the mesh, so let me just clarify. I started with a level 1 mesh and had the solver refine as it calculated. When you say to see where the accuracy starts to converge, do you mean for me to watch my CL goal plot as the solver runs to determine whether the solver is complete or if the mesh needs refinement?

                • Re: Calculation time for lift on 3D wing
                  Jared Conway

                  look at the steps in your graph, those are where the refinement happened

                   

                  what was the gain in "accuracy" at each of those levels? were any of them "good enough"? if so, you could have saved the solution time by going to that mesh level or stopping there. you would gain that knowledge by manually meshing at say lvl1, then 3, then 5 for example

                    • Re: Calculation time for lift on 3D wing
                      Josiah Lund

                      Okay, I understand what you are saying now. My concern before advancing for the full plane model is that while I can calculate the theoretical lift for an object such as this wing, and know where it should converge, I am unable to do this for the entire plane model. I have no way of knowing how accurate the simulation is with respect to true performance. If I am correct, somewhat of a standard exists for FEA when if the value that is being converged upon changes by less than 5% after a refinement, then the previous level of mesh is sufficient.

                       

                      I think the only improvement I can make now is to look at the post refinement mesh and try to cut out some of the refinements by creating local initial mesh regions to cut down in the number of refinements and therefore simulation travels needed. Would you agree with that, or is it better to just let the simulation take the time needed to automatically refine?

                        • Re: Calculation time for lift on 3D wing
                          Jared Conway

                          "If I am correct, somewhat of a standard exists for FEA when if the value that is being converged upon changes by less than 5% after a refinement, then the previous level of mesh is sufficient."

                           

                          in flow simulation, the solution always converges, the tolerance is defined by the calc control settings that you choose. you could override this and make it much tighter than 5% if you really wanted to, you would just have to wait a long time for the solution to complete.

                           

                          now if you're talking about 5% convergence from one mesh setting to another. there is no "rule of thumb", it is based on running different mesh levels to see if the results are mesh independent. IE the delta of a particular parameter with one mesh vs another isn't changing significantly. that could be a large value if you're just looking for a trend, or it could be very tight. it all comes down to your goals.

                           

                          "mesh regions to cut down in the number of refinements and therefore simulation travels needed."

                           

                          less cells = less solve time

                          refining manually vs automatically = less solve time

                          using less cells where they aren't adding "value" = less solve time

                           

                          what you need to do is find the balance between solve time and accuracy that you are comfortable with. a whole plane is going to take a lot of clock cycles. if your goal is to find the exact lift/drag, be ready for long solve times. now if you're looking to compare one plane vs another, relative changes, as long as the mesh quality is similar, you may not need to spend as much time as you are on the meshing.

                          • Re: Calculation time for lift on 3D wing
                            Bill McEachern

                            This may be of interest: I did  a job for a mod shop a couple years ago. They provided me with a model of an airplane they said they got from the manufacturer. I repaired and used this model. I have no ideaif it was accurate or not though obviously it look like the real thing. If was a light twin. The propellers were not included in my simulation but otherwise it was all there. The flaps were retracted. They prescribed the flight conditions - more or less a lowish speed high angle of attack (100knts & 10 deg AoA). The vehicle had a published GTW of 6100-6200 lb. For the conditions describe Flow sim for the unmodified A/C with not an overly fine mesh produced a lift estimate of about 6,182 lb. I thought it was pretty good. The object of the analysis was not to estimate the lift but I thought it was a good quality check on the simulation. I would post images but alas the stuff is confidential.

                    • Re: Calculation time for lift on 3D wing
                      Amit Katz

                      Josiah,

                      I just wanted to thank you for sharing all this good information and data as you progress through your project. If SW ever resumes their simulation blog this would make a great case example.