Discussion Forum

Posted: 4 years ago 2020年9月3日 GMT-4 11:54

Updated: 4 years ago 2020年9月3日 GMT-4 11:55

Just a few suggestions. 1. Don't create a highly complex geometry all at once, especially if you are not expert in the use of this software. Build up your geometry gradually and make sure it can mesh satisfactorily as you go along. Avoid configurations where tiny (relative to the computational space) volumes, surfaces, or edges are formed needlessly. Align your parts carefully. It matters. 2. Be careful with all your geometry specifications. Use unions and other Boolean operators to combine components, and when needed, change the tolerance values so that the geometries are properly interpreted. 3. Avoid having surfaces become very, very close tangentially to one another. E.g., a ball resting on a surface. Instead, make clean *intersections* or clean *non-intersections*. Choose one or the other based on the physics. 4. Lofted and swept surfaces can create trouble (although they work better than they used to). Add extra points and careful mappings, choose parameters carefully, etc., to keep them well controlled. Consider rotating some parts to realign their defining points better, before lofting/sweeping. If you can use simpler shapes/curves or straight lines/planes to model the same physical behavior with sufficient accuracy, do so. 5. De-feature your geometry to make it as simple as practical, to accomplish the job. The "design module" may be helpful. 6. Consider creating geometries with other CAD tools and importing them into Comsol Multiphysics. This can sometimes help, but sometimes it just makes things worse. 7. Consider creating your geometries as sub-components in separate Comsol files, and saving them as .mphbin files, then importing them into your overall more-complicated model. It is sometimes easier to work with/assemble complicated systems from invididually simpler geometries that are separately debugged. 8. Don't include and mesh parts of your model that are irrelevant to the physics (e.g., the interior volume of a *perfect* conductor in an EM problem!) If you already *know* a solution within some region is "zero," then you shouldn't be meshing or computing any values in that region. Instead, such volumes or areas should be represented only by their boundaries/edges. 9. Take control of the mesh, personally. The automatic meshes generated may be good enough, but they are seldom optimal. Use your own knowledge of the physics to decide where to apply fine meshes (where they are needed) and coarser meshes where you can get away with them. 10. There is always more than one way to build a geometry. And there may also be more than one way to represent the relevant physics. You mentioned *paper*. Do crinkles in that paper matter? If not, don't try to include them. Does the thickness of that paper matter? How about its EM material properties? Might it possibly be adequate to represent the paper itself by simply some *empty space* around the wires?! (If so, I would model the wires in space, but not "include" the paper at all!).