5 Replies Latest reply on Dec 7, 2011 3:07 PM by Bill McEachern

    Cylinder Drag... 'the need for speed'..AND accuracy!

    Karl Mohr

      Great to see Joe back!


      Motion is all around us and inside us!


      With motion flow is unavoidable!


      In Transportation, even the smallest of people movers, travelling at mere Human Powered speeds has SMALL parts experiencing flows with Re>=1.7e4 and most powered transportation experiences flows above Re=3.8e6!


      In an effort to assess accuracy of FlowSim I dropped back to 2D (external) many weeks ago and have been attempting to achieve 'standardaized' results, accurate or otherwise!


      Several have suggested that the code's accuracy drops off around Re=6e6!


      My objectives:

      Prove the degree of accuracy of FlowSim across the Re range 6e6-1.2e7 (for my craft, hump speed to cruise)


      My beliefs:

      If the accuracy is known and the test method (Re/CD/Mesh/refinement strategy) remain consistent it should be reasonable Engineering practice to apply an 'error correction'!


      I understand that there are some changes between the mesher/solver in FlowSim 2012 and 2011! (I am now using 2012!)


      On the 2D 'Cylinder Drag' tutorial:


      1. it is clear that the tutorial maintains a 'ratio' for the depth of the extrusion/boundary to the diameter, as the diameter is varied (presumably for cell mesh reasons) is there a ratio that is considered 'optimum'?
      2. in SW2012-64, with default tutorial settings, I have several Re's that will not converge (try 1e4!)
      3. if I compare 1m and 5m CYLINDER, H2O, Re=1e6, tutorial settings, (boundaries maintaining ratios) I get arount 13% difference in result! (my application, when I can learn to crawl, is a Surface Effect craft, with a wing root section of 5m!)
      4. Real testing, as per the tutorial's refference and 'real World' applications would not rescale!
      5. Joe Galliera (invaluable, thanks so much!) has said to use 'solution addaptive' mesh refinement, including for transient/time-dependent, I did believe in this, Bill MacEachern (has been thanked before! ) seemed to advise against, I am leaning towards Bill's advise now (for Transient/time dependent ONLY!) believing that one or two solution adaptive mesh refinements per travel leaves "the mesh too far behind the FlowSim time march/action"
      6. I am now using FOUR "Local Initial Meshes" (in FlowSim tree) defined by three elipses (generally concentric but offset downstream!)  plus the cylinder surface as the fourth!
      7. UPSTREAM water problem: Amazingly, Flow Sim seems to know I am in China and that the water is NEVER clean! (see images!) PLEASE ADVISE!
      8. Has anyone managed to get convergence at 'consistent settings' across the Re range for ANY model??


      Help appreciated!



        • Re: Cylinder Drag... 'the need for speed'..AND accuracy!
          Joe Galliera

          Karl Mohr wrote:


          Great to see Joe back!


          Thanks!  Although, I don't know if I went anywhere really.  I've just been very busy.  Hopeful to meet you and more at SWW San Diego 2012.

          • Re: Cylinder Drag... 'the need for speed'..AND accuracy!
            Bill McEachern

            I can' remember what the context was when I made the remarks on not using the autoadaptive but here are my qualifying comments...My thinking (at least now) on the recomendation for not using the auto adaptive was more or less if you have the mesh as dense as it would be in the autodaptive but everywhere in the downstream eddy region then the there will be no need to have the mesh changing during the time steps which,  I suspect, would provide a more stable in time solution for visulaizing and measuring the drag in an oscillating flow regime. In a 2D solution the compute times are rather short so the approach is a no big deal. In 3D on the other hand, well economy (autoadaptive) maybe the way to go. If the flow is super cirtical then the solution adpative will likely be a more economical way to go in either case as the flow oscilaltions will not occupy large spatial zones and a steady solution is likely better than the transient, all things considered.

              • Re: Cylinder Drag... 'the need for speed'..AND accuracy!
                Karl Mohr

                Thanks Bill!


                If and when I can get back to 3D I had planned to try transient though the many months of 3D analysis done to date on the craft were all done using steady state and solution adaptive Lvl2/3


                On supercritical, just in case we are not talking the same language, the supercritical region for all cylinder's (2D) is between the knee, on the tutorial's reference, attached below, at about Re=3e5 to some point after it starts to climb out of the low drag area; say 2e6, correct?


                Some old cylinder drag paper, that I was reading yesterday, refered to the plateau as 'Transcritical' has this this term been accepted for this higher drag region at higher Re's?


                (Wikipedia's article is not good, at least for my head! http://en.wikipedia.org/wiki/Supercritical_flow )


                Therefore the screen shots in the earlier post at 1e7 that I am attempting to get an understanding of are 'transcritical' (for a cylinder!)


                Yesterdays paper gives Cd of 0.7 @ Re=1e7 (large pipe in air) the tutorial about 0.6 (1m cylinder in water) and Flow Sim presently sitting at about 0.46 unresolved! What are you using to offer +-5%?? (If my memory serves me correctly! Its a couple of weeks since I looked at your site!)


                And you seem to be saying ignor the 'solution adaptive's' addition of undulating dappled mesh on the incoming stream, correct?  This is nothing that I am doing wrong, perhaps correct perhaps not, but nothing that you have not seen before!?

                Karlcylinder drag-Panton R.L.JPG

                  • Re: Cylinder Drag... 'the need for speed'..AND accuracy!
                    Bill McEachern

                    As far as I know super cirtical, in the sense that we are referring to here, is where the inertial forces start to dominate over the viscous. The other article you mentioned is a long way from a comprehensive discussion on "super critical". When viewed in terms of Froude number and waves it make a lot of sense. When thinking about R'number effects not so much. So in the plot above it is where the drag drops very noticiably and steeply at about 3e5 in the above plot. If you want a better version of this plot look in the verification section of the technical reference manual. It has more densely populated experiemental points with the points being much closer to the the experimental. I beleive the validation files are also provided on the install. It might be worth while checking those out.

                    I would call the "transcritical" range from about 3e5 to where it come out of the "bucket".