Dear all,
I plan to use MICRESS for the solidification of a welded microstructure. Is there a general number for the computation time for a 2D simulation of a 500 x 500 µm^2 area with 1 µm spatial resolution for 1 second of solidification time wiht 1 µs timestep?
Up to now I have not found any literature mentioning the actual CPUTIME of the calculated problems.
Thanks for your help.
Cheers
Constantin
CPUTIME of Solidification
CPUTIME of Solidification
---- Constantin
Re: CPUTIME of Solidification
Dear Constantin,
Welcome to the MICRESS Forum!
You raise a question which has no simple answer, because CPU time of a simulation problem always depends on a lot of parameters.
The most critical parameter is certainly grid resolution and the size of the simulation domain. Whether the problem can be solved with the grid specification you assume cannot be said beforehand with certainty. The main question here is how fine are the expected microstructures and diffusion profiles. This depends as well on the cooling parameters as on the thermodynamics of the system: A low-alloyed steel for example has no strong segregation, and will form much coarser dendrites compared to a Ni-based alloy. For the latter case, I recently simulated a welding process with [gr]Delta[/gr]x=0.16 µm which is considerably smaller than what you assume.
Naturally, the grid spacing has a very strong impact on CPU time. If for a given total size Δx is reduced by a factor of 2, the factor on CPU time is somewhere between 4 and 16 in 2D-simulations, depending on whether interface related operations like redistribution or solving the phase-field equation or whether solute diffusion is limiting performance. Thus, in case you would e.g. need to switch from [gr]Delta[/gr]x=1 µm to 0.16 µm, the factor in CPU-time would be between 36 and 1300 - probably it would be impossible to do the simulation for the whole size of the domain under these conditions...
A second very important factor is the complexity of the system. The more elements and phases are involved, the longer will be the simulation time. Furthermore, strongly segregating phases require the use of small phase-field time steps, which makes simulation much slower. The span in CPU time between a simple binary two-phase system and a 10-component system with 10 solid phases is certainly above 100!
A further very important question is the focus of the simulation. If you want to get the fully solidified microstructure with precipitation of phases in the interdendritic region, it takes much longer as if you e.g. want only to evaluate the secondary dendrite spacing, which can perhaps be obtained already after about 90% solidification. Typically, the last 10% need 90% of CPU time!
So, as you see, there is no simple answer to your question. However, there is another approach to the problem: If you do what is possible (and this you should always do), then you will design the simulation such that it takes not more time than you can practically afford (with all designing, testing etc.). Most of our applications (i.e. one production run after the setup is perfect) take between about 3 days and 3 weeks of running time...
Bernd
Welcome to the MICRESS Forum!
You raise a question which has no simple answer, because CPU time of a simulation problem always depends on a lot of parameters.
The most critical parameter is certainly grid resolution and the size of the simulation domain. Whether the problem can be solved with the grid specification you assume cannot be said beforehand with certainty. The main question here is how fine are the expected microstructures and diffusion profiles. This depends as well on the cooling parameters as on the thermodynamics of the system: A low-alloyed steel for example has no strong segregation, and will form much coarser dendrites compared to a Ni-based alloy. For the latter case, I recently simulated a welding process with [gr]Delta[/gr]x=0.16 µm which is considerably smaller than what you assume.
Naturally, the grid spacing has a very strong impact on CPU time. If for a given total size Δx is reduced by a factor of 2, the factor on CPU time is somewhere between 4 and 16 in 2D-simulations, depending on whether interface related operations like redistribution or solving the phase-field equation or whether solute diffusion is limiting performance. Thus, in case you would e.g. need to switch from [gr]Delta[/gr]x=1 µm to 0.16 µm, the factor in CPU-time would be between 36 and 1300 - probably it would be impossible to do the simulation for the whole size of the domain under these conditions...
A second very important factor is the complexity of the system. The more elements and phases are involved, the longer will be the simulation time. Furthermore, strongly segregating phases require the use of small phase-field time steps, which makes simulation much slower. The span in CPU time between a simple binary two-phase system and a 10-component system with 10 solid phases is certainly above 100!
A further very important question is the focus of the simulation. If you want to get the fully solidified microstructure with precipitation of phases in the interdendritic region, it takes much longer as if you e.g. want only to evaluate the secondary dendrite spacing, which can perhaps be obtained already after about 90% solidification. Typically, the last 10% need 90% of CPU time!
So, as you see, there is no simple answer to your question. However, there is another approach to the problem: If you do what is possible (and this you should always do), then you will design the simulation such that it takes not more time than you can practically afford (with all designing, testing etc.). Most of our applications (i.e. one production run after the setup is perfect) take between about 3 days and 3 weeks of running time...
Bernd