What is categorization of a grain type?
How do you categorize the grains and what are the consequences of that categorization?
Does it decrease the accuracy of the results?
Categorization
Re: Categorization
Dear mtoloui,
The main reason why categorization was introduced into MICRESS is the performance gain which can be achieved by reducing the number of effective grains.
As soon as the number of grains is exceeding about 1000 the performance of MICRESS is going down due to the administrative effort for the possible grain interactions. At the present stage, it gets very hard to get beyond 10000 grains.
In many simulation cases, the number of grains or precipitates can get very high, while the individuality of each single grain is of minor importance, for example in the case of the precipitation of a secondary phase. Then a huge perfomance gain can be obtained if several grains can be "categorized" in the sense that they are treated as if they would be one single grain. One prerequisit is of course that they have identical properties, i.e. phase, orientation, reX energy etc. Furthermore, if they have been nucleated as "small grains", they need to have already reached "full size", otherwise they still have there individual analytical or stabilized curvature and hence cannot be categorized.
The consequences of categorization are the following:
- the grains loose their identity, they appear with the same grain number in the ".korn" output
- the grains will not form an interface if they are touching afterwards due to further growth
- the grain statistics output (mean radius etc.) will be meaningless or at least more difficult to interpret
- if global TC coupling is used (only versions 5.404 and above) the approximation of one set of linearisation parameter per grain may be worse if this "grain" exists in several places
- the orientation distribution may get less smooth if one throws orientations into categories (see below)
The categorization option is controlled in the following way:
- Specifying "categorisation" in the phase data allows to switch on/off this option for the given phase. All grains of this phase will be checked during runtime with respect to the possibility of giving them a common grain number.
- If "categorization" is activated for a given phase which has been defined as anisotropic, then one is prompted to swtich on/off "categorisation" for each grain or seed type with this phase in the initial grain setting and nucleation input, respectively. The meaning now is that the grain orientation may be "categorized" such that there is a chance of setting grains with identical orientation so that "categorization" in the above sense can take place. As optional parameter, one can set the number of orientation categories to be produced.
So you can see that, unfortunately, the keyword "categorize" is used with two meanings, namely merging several grains into one, and concentrating continuous orientations into few distinct ones. The exactness of the results should not be affected other than by the choice of discrete grain orientations and in the special case of global TC coupling.
Bernd
The main reason why categorization was introduced into MICRESS is the performance gain which can be achieved by reducing the number of effective grains.
As soon as the number of grains is exceeding about 1000 the performance of MICRESS is going down due to the administrative effort for the possible grain interactions. At the present stage, it gets very hard to get beyond 10000 grains.
In many simulation cases, the number of grains or precipitates can get very high, while the individuality of each single grain is of minor importance, for example in the case of the precipitation of a secondary phase. Then a huge perfomance gain can be obtained if several grains can be "categorized" in the sense that they are treated as if they would be one single grain. One prerequisit is of course that they have identical properties, i.e. phase, orientation, reX energy etc. Furthermore, if they have been nucleated as "small grains", they need to have already reached "full size", otherwise they still have there individual analytical or stabilized curvature and hence cannot be categorized.
The consequences of categorization are the following:
- the grains loose their identity, they appear with the same grain number in the ".korn" output
- the grains will not form an interface if they are touching afterwards due to further growth
- the grain statistics output (mean radius etc.) will be meaningless or at least more difficult to interpret
- if global TC coupling is used (only versions 5.404 and above) the approximation of one set of linearisation parameter per grain may be worse if this "grain" exists in several places
- the orientation distribution may get less smooth if one throws orientations into categories (see below)
The categorization option is controlled in the following way:
- Specifying "categorisation" in the phase data allows to switch on/off this option for the given phase. All grains of this phase will be checked during runtime with respect to the possibility of giving them a common grain number.
- If "categorization" is activated for a given phase which has been defined as anisotropic, then one is prompted to swtich on/off "categorisation" for each grain or seed type with this phase in the initial grain setting and nucleation input, respectively. The meaning now is that the grain orientation may be "categorized" such that there is a chance of setting grains with identical orientation so that "categorization" in the above sense can take place. As optional parameter, one can set the number of orientation categories to be produced.
So you can see that, unfortunately, the keyword "categorize" is used with two meanings, namely merging several grains into one, and concentrating continuous orientations into few distinct ones. The exactness of the results should not be affected other than by the choice of discrete grain orientations and in the special case of global TC coupling.
Bernd
Re: Categorization
Hi, Bernd,
I am now using MICRESS_4.08 LINUX version. When I try to use the "categorize" to model precipitation of numbers of grains, no subsequent number is needed for the .dri file:
#
# Simulation of recrystallisation in phase 1?
# Options: recrystall no_recrystall
no_recrystall
# Is phase 1 anisotrop?
# Options: isotropic anisotropic faceted antifaceted
isotropic
# Should grains of phase 1 be reduced to categories?
# Options: categorize no_categorize
categorize
#
In my case, 200 grains are set for phase 1, and phase interaction between different grains of the same phase is considered. So, I want to use ""categorize to make "the grains will not form an interface if they are touching afterwards due to further growth".
Another question: What is the meaning of the the number after "categorize"?
Thank you!
Sincerely yours,
Sunny
I am now using MICRESS_4.08 LINUX version. When I try to use the "categorize" to model precipitation of numbers of grains, no subsequent number is needed for the .dri file:
#
# Simulation of recrystallisation in phase 1?
# Options: recrystall no_recrystall
no_recrystall
# Is phase 1 anisotrop?
# Options: isotropic anisotropic faceted antifaceted
isotropic
# Should grains of phase 1 be reduced to categories?
# Options: categorize no_categorize
categorize
#
In my case, 200 grains are set for phase 1, and phase interaction between different grains of the same phase is considered. So, I want to use ""categorize to make "the grains will not form an interface if they are touching afterwards due to further growth".
Another question: What is the meaning of the the number after "categorize"?
Thank you!
Sincerely yours,
Sunny
Re: Categorization
Dear Sunny,
In order to throw several grain to the same grain number, they have to have identical properties. If the phase in question is anisotropic, you could nucleate them with a random orientation distribution, and categorisation would be practically impossible because they have all different orientations.
For this case, the categorize keyword in the nucleation input allows you to throw the random orientations into categories, say 10°, 20°, 30° etc. In the grain input or nucleation input, the number after the categorize keyword is the number of categories (e.g. 36 for 36 categories all 10 degrees in 2D). Then, grains of the same orientation category can be "categorized" to the same grain number. In the case of isotropic phases, this number is meaningless, because all grains can be assigned to the same grain number.
In the phase input, the keyword "categorize" means that for the given phase you want to assign grains with identical properties to the same grain number. Here, no additional parameter is needed/allowed.
Bernd
In order to throw several grain to the same grain number, they have to have identical properties. If the phase in question is anisotropic, you could nucleate them with a random orientation distribution, and categorisation would be practically impossible because they have all different orientations.
For this case, the categorize keyword in the nucleation input allows you to throw the random orientations into categories, say 10°, 20°, 30° etc. In the grain input or nucleation input, the number after the categorize keyword is the number of categories (e.g. 36 for 36 categories all 10 degrees in 2D). Then, grains of the same orientation category can be "categorized" to the same grain number. In the case of isotropic phases, this number is meaningless, because all grains can be assigned to the same grain number.
In the phase input, the keyword "categorize" means that for the given phase you want to assign grains with identical properties to the same grain number. Here, no additional parameter is needed/allowed.
Bernd
Re: Categorization
Is there any way to categorize the initial microstructure manually i.e. assigning grains which are far from each other the same order parameter?
For instance when the boundaries are not periodic, can we assign the same order parameters to the grains close the boundaries which are far from each other?
For instance when the boundaries are not periodic, can we assign the same order parameters to the grains close the boundaries which are far from each other?
Re: Categorization
Dear mtoloui,
If you are using a deterministic or random initial grain structure, there is presently no direct way of manually assigning identical grain numbers to different grains.
But if you are reading in the initial grain structure from file, you can of course assign the grain numbers like you wish, including identical numbers for areas which are not connected and far away from each other. Such, effectively, you can always start from a deterministic or random initial structure created with MICRESS, save the grain output as ASCII and read it in in a second run after having modified the grain numbers according to your wishes!
Bernd
If you are using a deterministic or random initial grain structure, there is presently no direct way of manually assigning identical grain numbers to different grains.
But if you are reading in the initial grain structure from file, you can of course assign the grain numbers like you wish, including identical numbers for areas which are not connected and far away from each other. Such, effectively, you can always start from a deterministic or random initial structure created with MICRESS, save the grain output as ASCII and read it in in a second run after having modified the grain numbers according to your wishes!
Bernd