I think because of the resolution of spatial and temporal discretization, there are some deviations from theoretical results, e.g.
<1>KINETIC. both for grain growth and recrystallization, the relationship for sharp interface: v=MP is not perfectly followed. I simulated the shrinking of a round grain (2D), the radius after some time is bigger than the theoretical value, or k is smaller than theoretical[A0-A=kt, k=2*pi*M*sigma, sigma is interfacial energy]. For interface between a unrecrystallized and recrystallized grains, the velocity is also smaller than MP.
<2>MICROSTRUCTRUE. For a round nucleus within a quite big uncrystallized grain, after some time, the growing new grain should be keeping round shape, but actually not. I think it is [/b]ue to the anisotropy of meshing and it exists as long as we use 4-node cells no matter how you refine the time step.
Actually we often deal with complex microstructure with some hardly undermined parameters, so it is not easy to know where the difference between simulation and experiments comes from. I even experienced that some people could simulated a experiment-similar results using a wrong phase field model.
Is there any suggestion of numerical tricks that could be used to diminish those errors?
some numerical consequences
some numerical consequences
- Attachments
-
- test_korn_00003000.png (7.04 KiB) Viewed 3024 times
Re: some numerical consequences
Dear zhubq,
I know from test calculations in our house that the shrinkage of a circular grain by curvature worked well and gave the correct kinetics! Most probably, there are some parameters which are not properly set. Without seeing the driving file you used I can only suspect - I have two ideas what could be wrong:
1.) Maybe, you are using a too small interface thickness. The 3.5 cells I sometimes propose in alloy examples as a compromise between speed and correct anisotropy is not a good choice if curvature is the only driving-force. Try bigger values up to 10 cells to see whether this improves your results!
2.) I know from own (painful) experiences that the automatic time stepping criterion for pure phase-field may not always be working correctly. Please try to use a prefactor smaller than 1 in the options for automatic timestepping rather than the default value. The prefactor for the phase-field criterion is the first optional parameter after the "automatic" keyword. Decrease it systematically to rule out this issue. We are currently working on that problem!
If these points do not help, please paste your driving file in the forum!
Bernd
I know from test calculations in our house that the shrinkage of a circular grain by curvature worked well and gave the correct kinetics! Most probably, there are some parameters which are not properly set. Without seeing the driving file you used I can only suspect - I have two ideas what could be wrong:
1.) Maybe, you are using a too small interface thickness. The 3.5 cells I sometimes propose in alloy examples as a compromise between speed and correct anisotropy is not a good choice if curvature is the only driving-force. Try bigger values up to 10 cells to see whether this improves your results!
2.) I know from own (painful) experiences that the automatic time stepping criterion for pure phase-field may not always be working correctly. Please try to use a prefactor smaller than 1 in the options for automatic timestepping rather than the default value. The prefactor for the phase-field criterion is the first optional parameter after the "automatic" keyword. Decrease it systematically to rule out this issue. We are currently working on that problem!
If these points do not help, please paste your driving file in the forum!
Bernd
Re: some numerical consequences
Thank you. yes, I tried 10 for thickness, and the result is good. Previously, i used 5 grids.
Re: some numerical consequences
Hi zhubq,
I am a bit astonished that you really need 10 cells interface thickness to get a reasonable shrinkage of a sherical grain! Did you check whether there are other parameters which could cause the trouble? Did you try a smaller time step or a smaller phMin?
If you wish you can paste or send the .dri file, then we will have a look at the problem here at ACCESS!
Bernd
I am a bit astonished that you really need 10 cells interface thickness to get a reasonable shrinkage of a sherical grain! Did you check whether there are other parameters which could cause the trouble? Did you try a smaller time step or a smaller phMin?
If you wish you can paste or send the .dri file, then we will have a look at the problem here at ACCESS!
Bernd