On 06/03/2015 12:46 AM, derj...@web.de wrote:
Am Dienstag, 2. Juni 2015 20:22:29 UTC+2 schrieb Robert Cimrman:
The material parameters can be given in general by a function of space and time, and are evaluated in the Gauss quadrature points, not in the FE nodes. It is possible to project the parameters given as the function into the FE space (the nodes). It is also possible to have an auxiliary FE field defining the parameters and use it to evaluate the parameters in the quadrature points.
Thank you, that was very helpful. Considering that I'm new to FE, and it feels like I already can almost do what I want, really speaks in favor of SfePy.
What exactly do you want to achieve?
Well, it's not originally my idea, so I'm not sure where it might lead to. As a starting point, I try to reproduce [1] any may explore [2].
Interesting!
I basically want to update the material properties (basically elastic modulus) of a node dependent on the stress of the surrounding nodes, according to cellular automata like rules.
I think I found in the examples how to do the first approach ("project the parameters given as the function into the FE space"), and I guess this should work somehow. But better suited seems the second approach ("an auxiliary FE field defining the parameters and use it to evaluate the parameters in the quadrature points"), which probably allows to stick closer to the SfePy framework. I try to figure out how that works, but I'd also be grateful for any hint.
I guess it's not uncommon to use SfePy for iterative processes? Maybe it has even been used for some kind of topology optimisation?
Maybe it was, but I am not aware of that :)
Anyway, As the stresses as well are evaluated in the quadrature points (and have to be projected to the FE space), I suggest the following: define the parameter (elastic modulus) in the centre of each cell (element) - here you can average a general function over the cell, or use a constant function over each cell - and use it with the average stress in the cell. ('el_avg' mode in term evaluation, like it is done in post-processing functions). So the cellular automata grid would be in the cell centers, instead of the cell vertices. No projections needed.
If there's any interest, I'd be happy to keep you updated on this!
Sure, it is nice to see a new application! Do not hesitate to ask more.
r.