Author |
Message |
Jerry Brown (jerrybrown11743)
Member Username: jerrybrown11743
Post Number: 22 Registered: 03-2004
| Posted on Monday, July 31, 2006 - 04:17 pm: | |
I am having trouble understanding how to do spatial filtering. I would like to use the technique to smooth the stress field in a steady state elasticity problem. I've tried several times to follow your instructions; but, can't seem to get it to work. Can you show me how the technique would be implemented in the "Bentbar" example? |
Robert G. Nelson (rgnelson)
Moderator Username: rgnelson
Post Number: 666 Registered: 06-2003
| Posted on Monday, July 31, 2006 - 05:24 pm: | |
"Bentbar" isn't really a good example, because the stresses don't look noticeably ratty. Here instead is a version of "anisotropic", in which the y-distribution of y-stress doesn't look very good. There are two ways to address the spatial filtering issue: 1) Define a variable to hold the filtered data and use the equation Sfilter - div((D/pi)^2*grad(Sfilter)) = Sraw This has the effect of applying a low-pass filter to the Fourier expansion of Sraw. D is a rough estimate of the spatial wavelengths to be removed, or "smearing distance". See "Tech Notes: Smoothing Operators in PDE's" for some discussion of the motivation for this. 2)The FIT function projects the computed stresses onto a finite element interpolation. This forces continutiy at cell boundaries, and retains cell integral values. You can also add an arbitrary diffusion to the fit by providing a second argument. Values for the second argument can be determined by trial and error or by the same logic as the spatial filter in (1). The attached script does both of these things and plots some comparisons.
|
Jerry Brown (jerrybrown11743)
Member Username: jerrybrown11743
Post Number: 23 Registered: 03-2004
| Posted on Monday, July 31, 2006 - 09:02 pm: | |
Thanks. I had overlooked the FIT function and didn't realize that the diffusion trick requires a separate variable. It all makes sense now. |