How to add TensorMechanics into eigenvalue calculation #22721

js-jixu · 2022-11-16T13:04:50Z

js-jixu
Nov 16, 2022

Hi, experts.

I'm doing an eigenvalue calculation. My input file clearly contains TensorMechanics, but the calculated result does not contain TensorMechanics. When I look at the non-linear iteration residuals, other variables have values, only disp_x/y/z are 0:

 0 Nonlinear |R| = 4.649184e+03
    |residual|_2 of individual variables:
                   velocity: 4647.58
                   p:        4.07466e-05
                   Tf:       8.7437e-13
                   Ts:       122.101
                   disp_x:   0
                   disp_y:   0
                   disp_z:   0

Here is my blocks about TensorMechanics(Variables and BCs are omitted):

[Kernels]
  [disp_x_diff]
    type = Diffusion
    variable = disp_x
    block = fluid
    use_displaced_mesh = true
  []
  [disp_y_diff]
    type = Diffusion
    variable = disp_y
    block = fluid
    use_displaced_mesh = true
  []
  [disp_z_diff]
    type = Diffusion
    variable = disp_z
    block = fluid
    use_displaced_mesh = true
  []
[]

[Modules/TensorMechanics/Master]
  displacements = 'disp_x disp_y disp_z'
  strain = FINITE
  material_output_order = FIRST
  generate_output = 'vonmises_stress stress_xx stress_yy stress_zz strain_xx strain_yy strain_zz'
  [mechanics]
    block = 'solid'
    temperature = Ts
    displacements = 'disp_x disp_y disp_z'
    automatic_eigenstrain_names = true
  []
[]

[Materials]
  [elasticity_solid]
    type = ComputeIsotropicElasticityTensor
    youngs_modulus = 2e7
    poissons_ratio = 0.32
    block = 'solid'
    use_displaced_mesh = true
  []
  [thermal_expansion_solid]
    type = ComputeThermalExpansionEigenstrain
    temperature = Ts
    thermal_expansion_coeff = 1e-5
    stress_free_temperature = 273
    eigenstrain_name = thermal_expansion
    block = 'solid'
    use_displaced_mesh = true
  []
  [stress_solid]
    type = ComputeFiniteStrainElasticStress
    block = 'solid'
  []
[]

And I use InversePowerMethod in Executioner block:

[Executioner]
  type = InversePowerMethod
  max_power_iterations = 50

  normalization = 'powernorm'
  normal_factor = 21230

  bx_norm = 'bnorm'
  k0 = 1
  l_max_its = 200
  eig_check_tol = 1e-3

  solve_type = 'PJFNK'
  petsc_options = '-snes_converged_reason -ksp_converged_reason -snes_linesearch_monitor'
  petsc_options_iname = '-pc_type -pc_factor_shift_type -pc_factor_mat_solver_package'
  petsc_options_value = 'lu       NONZERO               superlu_dist'

  line_search = none
  automatic_scaling = true
[]

Answered by lynnmunday

Nov 21, 2022

I don't understand your problem, you have diffusion and mechanics acting on the same variables. What kind of physics are you modeling? I haven't seen any examples for solving natural frequencies of a mechanical system in MOOSE. You should look at the electromagnetics module in moose for tests that use the eigensolver. Let me know of you get this to work, I want to solve mechanics eigenvalue problems too.

View full answer

js-jixu · 2022-11-17T01:58:32Z

js-jixu
Nov 17, 2022
Author

Is there a kind guy to take a look at my question?👀

4 replies

GiudGiud Nov 18, 2022
Collaborator

Hello

You should not need to have [Modules/ ] before TensorMechanics.
This syntax of adding Modules has been deprecated

Use the [Debug] show_action parameter to show if the action is running
Judging by the 0 residual there's nothing acting on the displacement.

Guillaume

js-jixu Nov 22, 2022
Author

Hi, Guillaume, sorry for the late reply.

You should not need to have [Modules/ ] before TensorMechanics.

I removed [Modules/ ], but the terminal reported that section 'TensorMechanics' does not have an associated "Action". It may be that my moose version is too low. I haven't updated since September. So to get it working I added [Modules/ ] again.

Use the [Debug] show_action parameter to show if the action is running.

I added show_actions = true in [Debug]. It seems that TensorMechanics is taken into calculation:

[DBG][ACT] TASK (             meta_action) TYPE (     CommonTensorMechanicsAction) NAME (          Master) Memory usage 68MB
[DBG][ACT] TASK (             meta_action) TYPE (           TensorMechanicsAction) NAME (       mechanics) Memory usage 68MB
[DBG][ACT] TASK (   check_copy_nodal_vars) TYPE (             CopyNodalVarsAction) NAME (          disp_x) Memory usage 68MB
[DBG][ACT] TASK (   check_copy_nodal_vars) TYPE (             CopyNodalVarsAction) NAME (          disp_y) Memory usage 68MB
[DBG][ACT] TASK (   check_copy_nodal_vars) TYPE (             CopyNodalVarsAction) NAME (          disp_z) Memory usage 68MB
[DBG][ACT] TASK (                  add_bc) TYPE (                     AddBCAction) NAME (          pin1_y) Memory usage 105MB
[DBG][ACT] TASK (                  add_bc) TYPE (                     AddBCAction) NAME (          pin1_x) Memory usage 105MB
[DBG][ACT] TASK (                  add_bc) TYPE (                     AddBCAction) NAME (          pin2_y) Memory usage 106MB
[DBG][ACT] TASK (                  add_bc) TYPE (                     AddBCAction) NAME (          pin2_x) Memory usage 106MB
[DBG][ACT] TASK (                  add_bc) TYPE (                     AddBCAction) NAME (          pin3_y) Memory usage 106MB
[DBG][ACT] TASK (                  add_bc) TYPE (                     AddBCAction) NAME (         axial_z) Memory usage 106MB
[DBG][ACT] TASK (                  add_bc) TYPE (                     AddBCAction) NAME (        radial_x) Memory usage 106MB
[DBG][ACT] TASK (                  add_bc) TYPE (                     AddBCAction) NAME (        radial_y) Memory usage 106MB

But as I said before, disp_x/y/z are all 0 in the calculation results. This is not appropriate, because the temperature in different parts of the system changes. Then disp_x/y/z should also change.

There are figures for displacement and temperature distribution of results:

js-jixu Nov 22, 2022
Author

I wonder if the problem will appear here. Since I have used InversePowerMethod to solve an eigenvalue problem. The ComputeThermalExpansionEigenstrain in the TensorMechanics module is also an eigenvalue solving problem. So maybe an eigenvalue problem cannot be nested within an eigenvalue problem?

When I use executioner Transient or Steady for ComputeThermalExpansionEigenstrain, everything works fine. This problem occurs only when using InversePowerMethod.

However, when I use NonlinearEigen, disp_x/y/z are not involved in the calculation in the power iteration. But at the end of the power iteration, when the nonlinear iteration is performed, disp_x/y/z participates in the calculation again. Unfortunately, the results of the calculations are still problematic. So I want TensorMechanics to be able to participate in power iteration.

lynnmunday Nov 28, 2022

ComputeThermalExpansionEigenstrain is not solving an eigenvalue problem. Eigenstrain is any stress-free strain, I don't know why someone named it eigen-strain because I don't think it has anything to do with eigenvectors.

lynnmunday · 2022-11-21T20:09:59Z

lynnmunday
Nov 21, 2022

I don't understand your problem, you have diffusion and mechanics acting on the same variables. What kind of physics are you modeling? I haven't seen any examples for solving natural frequencies of a mechanical system in MOOSE. You should look at the electromagnetics module in moose for tests that use the eigensolver. Let me know of you get this to work, I want to solve mechanics eigenvalue problems too.

46 replies

hugary1995 Dec 7, 2022
Collaborator

Hence my original suggestion: try rerunning the same experiment using small deformation (without statefulness) and see if the bug persists.

lindsayad Dec 7, 2022
Maintainer

I think because we load restart data after calling initQpStatefulProperties that we do not need any additional flags in moose for checkpoint restart of stateful simulations. That is my hypothesis. Certainly running with small strain sounds like a great test as well

hugary1995 Dec 7, 2022
Collaborator

ok, good know that dataLoad happens after initStatefulProperties! So I guess then there is no showstopper for a checkpoint restart to work with finite strain. In the code snippet I referenced above, that _step_zero is actually declared as a restartable, so it seems to me that this use case has really been thought through. Also, we run checkpoint and recover tests all the time, so this has to work.

So @js-jixu , I think our conclusion is that an exodus restart or SolutionUserObject restart won't work in your case here. However, a checkpoint restart using restart_problem_base = ... should work.

lindsayad Dec 7, 2022
Maintainer

we run recover all the time, but restarting transient calculations from steady calculations is not something we test nearly as much. I honestly wouldn't be surprised if there were some issues here ... but then again I wouldn't be all that surprised if we don't have issues either since it is fairly likely that users have wanted to do this in the past

js-jixu Dec 8, 2022
Author

Hence my original suggestion: try rerunning the same experiment using small deformation (without statefulness) and see if the bug persists.

I use strain = SMALL to substitute strain = FINITE and use ComputeLinearElasticStress to subititute ComputeFiniteStrainElasticStress. Then I get ideal results.

The displacement in Z direction calculated when I use FINITE is 0.5820, and 0.5796 when I use SMALL.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

How to add TensorMechanics into eigenvalue calculation #22721

{{title}}

{{editor}}'s edit

{{editor}}'s edit

Replies: 2 comments 50 replies

{{title}}

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{title}}

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{editor}}'s edit

{{editor}}'s edit

Select a reply

How to add TensorMechanics into eigenvalue calculation #22721

js-jixu Nov 16, 2022

Replies: 2 comments · 50 replies

js-jixu Nov 17, 2022 Author

GiudGiud Nov 18, 2022 Collaborator

js-jixu Nov 22, 2022 Author

js-jixu Nov 22, 2022 Author

lynnmunday Nov 28, 2022

lynnmunday Nov 21, 2022

hugary1995 Dec 7, 2022 Collaborator

lindsayad Dec 7, 2022 Maintainer

hugary1995 Dec 7, 2022 Collaborator

lindsayad Dec 7, 2022 Maintainer

js-jixu Dec 8, 2022 Author

js-jixu
Nov 16, 2022

Replies: 2 comments 50 replies

js-jixu
Nov 17, 2022
Author

GiudGiud Nov 18, 2022
Collaborator

js-jixu Nov 22, 2022
Author

js-jixu Nov 22, 2022
Author

lynnmunday
Nov 21, 2022

hugary1995 Dec 7, 2022
Collaborator

lindsayad Dec 7, 2022
Maintainer

hugary1995 Dec 7, 2022
Collaborator

lindsayad Dec 7, 2022
Maintainer

js-jixu Dec 8, 2022
Author