14 Replies Latest reply on Jan 15, 2015 2:11 PM by Jared Conway

    grid convergence study question

    Dan Hofstetter

      Hi all,

       

      Does anyone have any tips or tricks for performing a grid convergence study in Flow Simulation?  I need to try to determine the grid convergence index (GCI) for my simulation.  It seems that to do it right, I have to disable all of the automatic settings.  I have an airflow problem where I have set up the following parameters in the manual mesh settings:

       

      Case123456
      Nx70140168202242280
      Ny183643526272
      Nz224453637688
      Total Cells27,720221,760383,201662,1721,144,2331,774,080

       

      However, the total cell count never works out to those numbers due to some of the mesh refinements that occur, such as finer mesh in corners.  I solved steady-state simulations for each case, using global goals for average static pressure, turb dissipation, turb energy, Vx, Vy, Vz, and Mass of Fluid.  I left all of the convergence criteria set to auto.  For a grid study, would it be better for me to manually set the convergence criteria so each simulation solves to the same level of accuracy?  Or are the automatic criteria good enough?

       

      After all simulations finished, I compared Velocity at several points (using point parameters) and plotted the values from each case.  What I am finding is that the values don't seem to be converging, and in many cases are oscillating as much as 20% from one case to the next even at the highest mesh levels used so far. 

       

      Graph.jpg

       

      The room I am simulating is 36' long x 9 ft wide x 8 ft high, and the total flow rate is less than 2,000 cfm (see attached model).  Should I really need more than 1/2-million cells in a room of that size?

       

      The next steps I need to accomplish are to perform a time-step sensitivity study, then run a transient simulation and try to achieve a time-accurate gas decay solution that I can compare to measured data.  Any help is appreciated!

        • Re: grid convergence study question
          Jared Conway

          can you define how you want to define your GCI?

           

          a few things i'm not following:

          1. in your table, you say that the cell count doesn't work out to what is in the table because of refinements. what is the number in the table? the basic mesh? does having more or less cells cause an issue with the way you are defining the GCI?

          2."...set the convergence criteria...solves with the same level of accuracy" what do you mean here? you could manually stop the solution at a certain point, number of iterations or some % of the goal but i'm not sure what you are looking to gain with respect to accuracy. it could still be on the high or low side.

          3. regarding oscillation, are you talking about the goals plots or are you talking about the results between studies? ie 1 to 2, 2 to 3 and then 3 to 4?

            • Re: grid convergence study question
              Dan Hofstetter

              Thanks for replying so quickly!  I'll answer in the order you listed:

               

              1.  The number in the table is the basic mesh (simply Nx * Ny * Nz), so it is the ideal mesh size.  After Flow Simulation creates the basic mesh and does some refinements, the total number of cells increases.  Some of those are fluid, partial, or solid cells.  It causes an issue with calculating the GCI, but that could be OK as long as the solution is close to being asymptotic from one case to the next.

               

              2.  I guess I was thinking that I might need to manually set the convergence criteria in each study, so that  the solution stops at the level of convergence in all simulations.  I'm not sure how much this matters, though, I was just wondering if anyone had a better suggestion. 

               

              3.  Regarding oscillation, I am talking about the results between studies as you mentioned:  1 to 2, 2 to 3, and 3 to 4.  Once the solution is nearly mesh-independent, the difference in variables should be less than 5% or so from one simulation to the next as the mesh gets finer.

                • Re: grid convergence study question
                  Dan Hofstetter

                  I added two more configurations with more mesh cells and ran the simulations last night, and I am getting even more divergence.  Do I need 10 or 20 million cells to solve this problem? 

                   

                  My plan was to run somewhat uniform grids (using manual settings) at a few levels, compare results and show what level of grid resulted in less than 5% difference in velocity values, then run the simulation again using automatic settings.  Then I would compare the velocitty values from the manual and automatic grids to show that the automatic settings produce results that are just as good as the manual settings (with hopefully many fewer cells in the overall mesh).  So I would have a GCI value to report for the manual mesh settings, and I would be able to state that the automatic settings produced results that were just as good.

                   

                    

                  Case78
                  Nx290348
                  Ny7590
                  Nz91109
                  Total Cells1,977,2343,416,661

                   

                  Graph2.jpg

              • Re: grid convergence study question
                Amit Katz

                I would be interested in seeing what you got if you go up to 10-20 million cells. Not necessarily because I believe you would have better results, but because from my personal experience Flow Sim starts to have serious stability and convergence issues past the first few million cell count.