Restart options
Posted: Thu Jul 14, 2016 4:44 pm
Hi all,
The current MICRESS version includes three different restart options which allow not only to restart a broken simulation or do parameter variations from an intermediate point of the simulation. It also allows reusing simulated microstructures as initial structures for a new simulation and thus building up simulation chains. And finally, with the newest option "structure_only", it is not only possible to read simulated microstructures with much less compatibility restrictions but also to combine different microstructures e.g. for brazing or diffusion couples.
The restart functionality originally was designed for the case that a simulation crashed at a certain time, and one wants to continue it without any changes. The user is supposed to use the same input file as before, and just replace the keyword "new" with "restart", and give the file name of the former .rest file (and perhaps chose "append" instead of "overwrite"). In this case, all the information about the initial microstructure is lost by reading the .rest file. But this does not hurt, because it would be much more effort for the user to remove this information than just to leave it there.
There is other information which is still necessary, like the definition of the phases, elements, interfacial properties etc. And some of this information, which is not conflicting with information in the .rest file, can even be changed, so that modifications of the simulation run are possible. Unfortunately, it is not very transparent to the user which information can be changed and which not. For example, the value of the interface mobility or of a certain diffusion coefficient can be changed without problem, but it is not possible to change a diffusion term from manual input to database input. The reason in this case is that the .rest file contains information about updating times for diffusion coefficients from database, therefore compatibility is lost. The same holds for the anisotropy data: It is important that anisotropy has already been defined in the first run, so that you only need to change the values of the anisotropy coefficients.
Some of these incompatibility issues can be solved by using the "restart reset_time" option which removes all timer information and restarts from time 0. This would help in case of the diffusion coefficients, but not for the anisotropy.
A more powerful option is "restart structure_only" which only reads microstructure related data and removes all numerical information. This option even allows you to read only parts of a microstructure and to combine information from different restart files. All numerical information including thermodynamic linearisation data, diffusivity, temperature and time information, etc. are lost. Only the phase and concentration distribution and the interface structure (which essentially is the microstructure like you see it in the output files) is kept, without link to the thermodynamic data (only phase and element numbers).
This means that, after reading in a restart file using the option "structure_only" the thermodynamic phase interaction data, the redistribution at the interface and all other numerical data have to be reconstructed, and thus can be adapted to a different simulation type. In principle, by changing the .ges5 file, one could completely change even the alloy system (which is nonsense in practice). Only the number of the elements has to be kept, and the new simulation needs to have at least as many phases as there are in the restart file.
In practice, to obtain highest flexibility, the new simulation creates a pre-microstructure in which each grain can be replaced by the microstructure which is stored in a restart (.rest) file. For example for brazing simulations, one could define 3 initial grains, two of which are replaced by microstructures from restart files, and the third representing the liquid braze alloy. For fitting the microstructures correctly, one can define a shift and rotation (by multiples of 90°, version 6.3 or later) for each read restart file. The most important restriction is that the set of elements used in all restart files and in the new simulation must be identical (by element number).
I personally use the restart option "structure_only" more and more also for normal process chains even if the size and format of the domain is not changing. The reason is the high compatibility which is not given for normal restart or for the option "reset_time".
Bernd
The current MICRESS version includes three different restart options which allow not only to restart a broken simulation or do parameter variations from an intermediate point of the simulation. It also allows reusing simulated microstructures as initial structures for a new simulation and thus building up simulation chains. And finally, with the newest option "structure_only", it is not only possible to read simulated microstructures with much less compatibility restrictions but also to combine different microstructures e.g. for brazing or diffusion couples.
The restart functionality originally was designed for the case that a simulation crashed at a certain time, and one wants to continue it without any changes. The user is supposed to use the same input file as before, and just replace the keyword "new" with "restart", and give the file name of the former .rest file (and perhaps chose "append" instead of "overwrite"). In this case, all the information about the initial microstructure is lost by reading the .rest file. But this does not hurt, because it would be much more effort for the user to remove this information than just to leave it there.
There is other information which is still necessary, like the definition of the phases, elements, interfacial properties etc. And some of this information, which is not conflicting with information in the .rest file, can even be changed, so that modifications of the simulation run are possible. Unfortunately, it is not very transparent to the user which information can be changed and which not. For example, the value of the interface mobility or of a certain diffusion coefficient can be changed without problem, but it is not possible to change a diffusion term from manual input to database input. The reason in this case is that the .rest file contains information about updating times for diffusion coefficients from database, therefore compatibility is lost. The same holds for the anisotropy data: It is important that anisotropy has already been defined in the first run, so that you only need to change the values of the anisotropy coefficients.
Some of these incompatibility issues can be solved by using the "restart reset_time" option which removes all timer information and restarts from time 0. This would help in case of the diffusion coefficients, but not for the anisotropy.
A more powerful option is "restart structure_only" which only reads microstructure related data and removes all numerical information. This option even allows you to read only parts of a microstructure and to combine information from different restart files. All numerical information including thermodynamic linearisation data, diffusivity, temperature and time information, etc. are lost. Only the phase and concentration distribution and the interface structure (which essentially is the microstructure like you see it in the output files) is kept, without link to the thermodynamic data (only phase and element numbers).
This means that, after reading in a restart file using the option "structure_only" the thermodynamic phase interaction data, the redistribution at the interface and all other numerical data have to be reconstructed, and thus can be adapted to a different simulation type. In principle, by changing the .ges5 file, one could completely change even the alloy system (which is nonsense in practice). Only the number of the elements has to be kept, and the new simulation needs to have at least as many phases as there are in the restart file.
In practice, to obtain highest flexibility, the new simulation creates a pre-microstructure in which each grain can be replaced by the microstructure which is stored in a restart (.rest) file. For example for brazing simulations, one could define 3 initial grains, two of which are replaced by microstructures from restart files, and the third representing the liquid braze alloy. For fitting the microstructures correctly, one can define a shift and rotation (by multiples of 90°, version 6.3 or later) for each read restart file. The most important restriction is that the set of elements used in all restart files and in the new simulation must be identical (by element number).
I personally use the restart option "structure_only" more and more also for normal process chains even if the size and format of the domain is not changing. The reason is the high compatibility which is not given for normal restart or for the option "reset_time".
Bernd