2 Replies Latest reply on May 15, 2012 10:28 PM by Shahrul 88

    How does Flow Simulation judge convergence?

    Dan Cook

      I'm new at Flow Simulation and am trying to get my head wrapped around some of the terminology. I'm used to CFD codes such as Fluent, StarCD or Fidap where the codes base convergence on changes in the residuals of the equations being solved. When these residuals fall below some set value for all the variables of interest, the calculation is deemed "done" and everyone cries "Hallelujah" and drinks a beer.

       

      Since FS is based on the finite volume method, I would expect that buried deep inside the code, there is a routine that is doing something similar. Could someone explain to me how what I'm used to fits in with the method used in FS? I've looked at both the technical reference and the solving engineering problems pdfs.

       

      Thanks in advance,

      dan

        • Re: How does Flow Simulation judge convergence?
          Bill McEachern

          FS runs the time averaged formulation of the NS equations which means it essentially does a transient analysis (time explicit) solution so there is no real "converging" going on. Rather, for a (quasi) steady state solution (they don't call it fluid dynamics for nothing) the program runs till the (typically) impulsive start up has died out and it gets more or less "steady" (not varying much in time). So I think your question is how does it know when this has happended? Well, if you do nothing it has preset termination criterea for given problem types - like 4 "travels" I think is the default. A travel is the number of iterations it takes for a distrubance to fully propagate across the domain. Further, you can add goals and the program can use those and will stop when the goals you have assign stop changing (the criteria is determined by the program or you can change it to whatever you like, though I would not advise this right out of the gate as a newbie). When you use goals (the recommended approach) you will be able to see the "convergence" trajectories (really they are time trajectories). When they flatten out and remain essentially invariant for some number of iterations (you can be the judge or you can let the program decide) then you can consider the particular result "achieved" or "converged". Given that FS is a "cart" code (cartesian) is suffers much reduced numerical diffusion issues compared to many body fitted codes (such as Fluent) and will typically "converge" for an incompressible flow problem in the order of low hundreds of iterations where as a similar problem in Fluent may require several thousand at very much higher cell counts - as a fluent guy once remarked "How does FS get the same asnwer as with 200,000 cells and and 2

          hours as Fluent with 2 million cells and 18 hours?" Then again FS doesn not have all the physics that Fluent supports but for what it does do it does a very fine job at it most of the time, at least in my experience.

            • Re: How does Flow Simulation judge convergence?
              Shahrul 88

              Hi All,

               

              I'm using solidworks to do convection heat transfer in the microcomputer. The problem is my understanding on the number of iterations.

               

              2 simulations are done through 2 different setting on the iteration:

                     1) goals convergence (result obtained as in Pic 1 attached - stop at iteration 56)

                     2) maximum iterations = 42 (result obtained as in Pic 2 attached - stop at iteration 42)

               

              How do I explain about the differences of data ( in this case temperature) at iteration 20 and iteration 30??

              For goals convergence; what is mean by goal convergence?? is it equilibrium state/ condition??

               

              This number of iterations are not explained about time, am I right?

               

              Thanks

               

               

              Pic 1.JPG

              Pic 2.JPG