Local mesh refinement with targeted edge lengths at specified areas

    • nwood99


      I am working on a problem that features physics that are highly localized at the uppermost surface of my geometry.  Accordingly, I am trying to implement a mesh that is very aggressively refined at the top area, shown below:

      I got this result with three successive level 1 refinements at the top area, yielding an average edge length of 122 microns at the top face.  

      My issue is that I'm seeking meshes of the above style, but with targeted edge lengths.  I was looking for a mesh with 100 micron edges at the top face.  Furthermore, I'm seeking to implement a convergence study with top-face edges having 10, 18, 32, 56, and 100 micron lengths.  I am aware that the APDL meshtool allows me to set element lengths at areas, however, if I use this setting, it doesn't refine aggressively enough.  Below is what happens if I use the meshtool setting to target 100 micron element lengths at the top face:

      You can see how the meshtool implementation causes regions of high mesh density to cascade throughout the interior.  This tendency causes the program to yield meshes with unbounded node counts as I try commanding top face elements with the 10 micron and 18 micron lengths.  

      My question is if there is a way to implement the kind of aggressive mesh refinement I get by manually commanding level 1-5 refinements (say, with the arefine command), but with targeting a specified element length?  My thanks for any help that can be provided!

    • John Doyle
      Ansys Employee

      Instead of strategic mesh refinement on initial geometry, have you consider nonlinear adaptive meshing via NLAD,,, command?  Refer also to the MAPDL Nonlinear Adaptive Meshing Guide for more details.


    • nwood99


      While I am aware of adaptive meshing, I was looking to just assign the mesh density at the start because I'm already pretty confident in what I want the final state of the mesh to look like (the first figure in my post).  The extra expense of implementing adaptive meshing seemed unnecessary.  Its just that I wasn't sure exactly how small to make the final edge length.  The convergence study would be identifying the point of diminishing returns in simulation accuracy within the 10 um to 100 um edge length range, since I know there's no disk of divergence for these parameters.  If there is really no support for this kind of area-wise local refinement with a targeted edge length, I could fall back on NLAD routines, it just seems kind of wasteful from my perspective.  If there is any support for doing it in advance like I've outlined, please let me know!


      Nathaniel Wood     

Viewing 2 reply threads
  • You must be logged in to reply to this topic.