Uniaxial tensile simulation of polycrystalline model based on crystal plasticity

[PengWei97]

Dear MOOSE experts,
I recently wrote an input file based on the files prop_grain_read.i and crysp_user_object.i to simulate plastic uniaxial stretching of polycrystalline crystals, and established a two-dimensional model of 0.21*0.21mm. There are 200 grains in it. Stretching in the y-axis direction.
284 time steps, time = 0.327438, stretching by 0.03%.
But the calculation is very slow. Is there any suggestion to speed up the calculation?

The content of the written input file and output terminal is as follows,
prop_grain_read_cp.i
prop_grain_read_cp.txt

Any suggestions or recommendations to fix these problems would be greatly appreciated.
Thank you
Wei Peng

[jiangwen84]

@PengWei97 I need to let you know that we are currently switching the UO-based CP (the one you used) to stress-update based CP. One on-going PR is #17405. The new version might give you a bit speedup.

For your current input file, I suggest you add maximum_substep_iteration = 4 in your FiniteStrainUObasedCP block. It will allow you to take larger dt. With sub stepping, each time step will take longer time, but it can speed up your simulation by taking less number of time steps.

@sapitts @dewenyushu any other thoughts?

[PengWei97]

Hi @PengWei97 I agree with @jiangwen84 that using sub-stepping should allow larger dt thus speed up computation in general.

For the new stress-update based CP base class (#17405). As Wen mentioned, this is a new implementation which will be more actively developed and maintained than the UO-based version in the future. With this new class, we do see some improvements in terms of the total linear/nonlinear iterations for several existing models (and some of them are improved a lot!). However, we are yet to quantify the improvement or examine the consistency of this improvement among all existing models.

Hi @dewenyushu,Thank you for your advice.
I tried to clone the relevant *.C and *.h files in my development branch into the application I created, but an error was displayed when compiling, the error is as follows,
error_make.txt

Is it because this class is still under development, so we can't use it now, or is it for other reasons?

[jiangwen84]

Try one more thing:

add

petsc_options_iname = '-pc_type'
petsc_options_value = 'lu'

to [Executioner] block.

[PengWei97]

Try one more thing:

add

petsc_options_iname = '-pc_type'
petsc_options_value = 'lu'

to [Executioner] block.

hi @jiangwen84 ，follow your suggestion, this problem has been solved.
Thank you very much for your guidance.
wei

[dewenyushu]

Hi @PengWei97 I agree with @jiangwen84 that using sub-stepping should allow larger dt thus speed up computation in general.
For the new stress-update based CP base class (#17405). As Wen mentioned, this is a new implementation which will be more actively developed and maintained than the UO-based version in the future. With this new class, we do see some improvements in terms of the total linear/nonlinear iterations for several existing models (and some of them are improved a lot!). However, we are yet to quantify the improvement or examine the consistency of this improvement among all existing models.

Hi @dewenyushu,Thank you for your advice.
I tried to clone the relevant *.C and *.h files in my development branch into the application I created, but an error was displayed when compiling, the error is as follows,
error_make.txt

Is it because this class is still under development, so we can't use it now, or is it for other reasons?

Thanks for trying this out! By looking at your log file, it seems like you will need to checkout the Kalidindi class (CrystalPlasticityKalidindiUpdate.C and .h) from my PR as well since I have made some changes for it being able to run with the new base class.

The new base class is something that is still in development right now and transforming all existing CP models to the new base class will require some more work. Currently, only the Kalidindi CP model works with this new base class. So I will not be surprised if there being some glitches when transforming your model without some changes. If you come across some new issues, please keep us updated.

[PengWei97]

Hi @PengWei97 I agree with @jiangwen84 that using sub-stepping should allow larger dt thus speed up computation in general.
For the new stress-update based CP base class (#17405). As Wen mentioned, this is a new implementation which will be more actively developed and maintained than the UO-based version in the future. With this new class, we do see some improvements in terms of the total linear/nonlinear iterations for several existing models (and some of them are improved a lot!). However, we are yet to quantify the improvement or examine the consistency of this improvement among all existing models.

Hi @dewenyushu,Thank you for your advice.
I tried to clone the relevant *.C and *.h files in my development branch into the application I created, but an error was displayed when compiling, the error is as follows,
error_make.txt
Is it because this class is still under development, so we can't use it now, or is it for other reasons?

Thanks for trying this out! By looking at your log file, it seems like you will need to checkout the Kalidindi class (CrystalPlasticityKalidindiUpdate.C and .h) from my PR as well since I have made some changes for it being able to run with the new base class.

The new base class is something that is still in development right now and transforming all existing CP models to the new base class will require some more work. Currently, only the Kalidindi CP model works with this new base class. So I will not be surprised if there being some glitches when transforming your model without some changes. If you come across some new issues, please keep us updated.

Thank you very much. I will try to modify it.