Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!


  1. Oct 28, 2008 #1

    I m a student using COMSOL Multiphysics. I have to admit that I am new to this software but it is giving me loads of problems. I am trying to model an eddy current sensor by modelling the coil and a plate (mild steel). The size of the plate is 17mmx17mmx3mm but I had no problem with the geometry.
    The skin depth at 100khz is 0.1mm and therefore the mesh size has to be at least 0.05mm. Unfortunately when I mesh the plate, a warning is popping up stating that it is out of memory. I am using a 1.5ghz processor(intel) and 1.2G ram. do you think that the problem is the laptop?

    Any replies are greatly appreciated
  2. jcsd
  3. Oct 29, 2008 #2


    User Avatar
    Science Advisor
    Gold Member

    The mesh size only has to be small in the vicinity of the surface effect. Meshing the entire part using tiny elements is a waste of memory and WAY overkill. I estimate that if you use 0.05mm elements across the entire part, you'll end up with a mesh of about 7 million elements- far outside the capability of a standard laptop or desktop computer. You'll probably want to keep your model down to 100,000 elements or so to ensure it solves in a reasonable amount of time and doesn't run out of memory.

    I would recommend resizing the mesh so that the elements are small within the surface phenomenon's region, and you can use larger elements in the middle of the part.
  4. Oct 30, 2008 #3


    User Avatar
    Science Advisor
    Gold Member

    Also, comsol by default typically uses higher order elements --> reduce the order. Partition the model appropriately if possible to enable higher mesh gradients. Use the comsol's mesh density options for point/boundary/subdomain. If it still presents problems, try the mapped meshing and interactive meshing tools to accurately specify element numbers at critical areas.

    Also, it may be an option to do opposite what suggested at first and use "much higher" order elements and model larger areas with "a couple" of elements still with fair accuracy due to the complexity of the test/interpolation functions. In some cases yields best accuracy with fewest dof.
Share this great discussion with others via Reddit, Google+, Twitter, or Facebook