[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Getfem-users] mesh_deformation

From: Zhenghuai Guo
Subject: Re: [Getfem-users] mesh_deformation
Date: Sat, 1 Dec 2018 20:15:54 +0000

Dear Andriy,


I try your suggestions - using im_data for the coefficients - in python which are working fine. Thank you very much.


But I still have problem in deforming the mesh in Python – simply don’t know how to call the ‘temporary_mesh_deformator’ in Python. I guess I should do it in C++.


I don’t really use C++ too much. Could you please let me know where the demos are for the C++?


I can see that in the folder ‘getfem-5.3/tests’, there are some ‘.cc’ test files – are they the demos? I try to compile it : ‘g++  -o elastostatic’, but it did not go through (give the message as attached). I am not sure if I did it right or simply because I did not install the getfem right in the first place.


Could you please advise?


Thank you very much

Kind regards

Zhenghuai Guo



From: Andriy Andreykiv <address@hidden>
Sent: Wednesday, November 21, 2018 9:53 PM
To: Zhenghuai Guo <address@hidden>; address@hidden
Subject: Re: [Getfem-users] mesh_deformation


Dear Zhenghuai Guo,


For 1 you don't need to do anything. All the variables in the next step remain from the last step. In some cases, if your variable has two versions (old and current, for instance) you need to call model::next_iter() to copy all the old versions of variable to the current.

But if the variable doesn't have the two versions you don't need that. Normally we want to have two versions of variables when our formulations use their increments.

For 2 you don't really need to add the linear elastic term with your own code, as the actual function add_isotropic_linearized_elasticity_brick uses high-level assembly internally. Here is it's source code 


size_type add_isotropic_linearized_elasticity_brick
(model &mdconst mesh_im &mimconst std::string &varname,
 const std::string &dataexpr1const std::string &dataexpr2,
 size_type regionconst std::string &dataname3) {
  std::string test_varname
    = "Test_" + sup_previous_and_dot_to_varname(varname);
  std::string expr1 = "((("+dataexpr1+")*(Div_"+varname+"-Div_"+dataname3
    +")-Sym(Grad_"+dataname3+"))):Grad_" +test_varname;
  std::string expr2 = "(Div_"+varname+"*(("+dataexpr1+")*Id(meshdim))"
  ga_workspace workspace(mdtrue);
  workspace.add_expression(expr2, mimregion);
  model::varnamelist vl, vl_test1, vl_test2, dl;
  bool is_lin = workspace.used_variables(vl, vl_test1, vl_test2, dl, 2);
  if (is_lin) {
    pbrick pbr = std::make_shared<iso_lin_elasticity_new_brick>
      (expr2, dataname3);
    model::termlist tl;
                         sup_previous_and_dot_to_varname(varname), true));
    if (dataname3.size()) dl.push_back(dataname3);
    return md.add_brick(pbr, vl, dl, tl, model::mimlist(1, &mim), region);
  } else {
    return add_nonlinear_generic_assembly_brick
      (mdmimdataname3.size() ? expr1 : expr2, regionfalsefalse,
       "Linearized isotropic elasticity (with nonlinear dependance)");

It's c++, but python logic would be similar.

Regarding the source code you've attached, I see that you still add the elastic constants as a fixed size data. As I mentioned in the last email, you need to have different constants for every integration point

of your model. Hence they have to be im_data.


Best regards,



On Sun, 18 Nov 2018 at 20:47, Zhenghuai Guo <address@hidden> wrote:

Dear Andriy,


Thank you for the advice. You understand correctly about what I am trying to achieve. The problems I am having now are to how to (1) use an updated deformed mesh and (2) consider the displacement/stress value of previous timestep as the initial condition at each new time step.


Could you please advise about the following?


  1. If I use Isotropic_linearized_elastic_brick, how can I pass the displacement and stress values from one time step to another? At each timestep, I would like the initial condition to take into account the state (e.g. displacement and stress) of last timestep. I am not sure maybe it does it automatically or not.
  2. I am trying to use high level assembly string language to carry out linear elasticity. I am guessing, by doing this I can be more free. And maybe I can consider the displacement/stress value of previous timestep as the initial condition at each new time step.

Could you please have a look at the attached small code? I try to add the assembly string of isotropic linear elasticity by md.add_linear_term(mim,'(clambda*Trace(Grad_u)*Id(qdim(u)) + cmu*(Grad_u+Grad_u’)):Grad_Test_u ')


but, when I run the code, it appear with some problem – although it does not give error, but it does not proceed and the python shell is restarted. Could you please advise about this issue?


Thank you very much


Zhenghuai Guo


From: Andriy Andreykiv <address@hidden>
Sent: Friday, November 16, 2018 11:03 PM
To: Zhenghuai Guo <address@hidden>
Cc: address@hidden
Subject: Re: [Getfem-users] mesh_deformation


Dear Zhenghuai Guo,


If I understand correctly, you're trying to model creep with contact. If so, then you can probably assume that your elastic properties are weakly coupled with stress. In that case you can probably just update your elastic properties every time step.

So, you need to write a loop where you change the properties every step. Here are some ideas


1)  Your elastic properties lambda and mu cannot be fixed sized constants, as they will depend on stress values throughout the domain. The logical way to handle this is

     by setting them as a so-called im_data, which is a field defined on Gauss points (on mesh_im class). Please look up documentation or source how to add im_data.


2) Isotropic linearized elastic brick does not compute and store stress field. You will need to define stress field as im_data too and compute it from your solution

    using probably high-level assembly syntax


ga_interpolation_im_data(md, "lambda*Trace(Grad_u)*Id(qdim(u)) + mu*(Grad_u+Grad_u'))" , im_data &imd_stress, base_vector &stress_result_vector);


3) The whole calculation should look like this:


initialize the elastic properties as im_data

create im_data for the stress

add contact bricks

while(t < T)



   compute the stresses from the displacement field using high-level assembly syntax interpolations

   compute the new values for the elastic properties with high-level assembly syntax interpolations


   t = t + delta T;



Best regards,



On Fri, 16 Nov 2018 at 04:42, Zhenghuai Guo <address@hidden> wrote:

Dear Andriy,


I am planning to do a time dependent deformation with consideration of the contact and friction. I would like to apply the elasticity theory with Young’s modulus being a function of time and local stress value.


Could you please advise about the following?


  1. If  I use add_isotropic_linearized_elasticity_brick (mim, varname, dataname_lambda, dataname_mu, region=None), how can I make the Young’s modulus time and stress dependent? In the test examples e.g. in, Young’s modulus is only added by the method add_initialized_data as a constant scalar value.
  2. If I carry out elasticity formulation by a approach like the one in using low level approach to building the linear system by hand, can I in the meantime apply together the bricks framework e.g. add_master_contact_boundary_to_large_sliding_contact_brick(indbrick, mim, region, dispname, wname=None)?



Thank you very much

Best regards

Zhenghuai Guo




From: Andriy Andreykiv <address@hidden>
Sent: Thursday, November 8, 2018 8:56 PM
To: Zhenghuai Guo <address@hidden>
Cc: address@hidden
Subject: Re: [Getfem-users] mesh_deformation


Dear  Zhenghuai Guo,


You can build getfem and getfem based programs using either GCC c++ compiler for Linux based systems

or with Microsoft Visual Studio (you can use free Community edition). You can find MSVC solution in msvc directory of the distribution.

Unfortunately, the solution for MSVC is not kept up-to-date and you would need to re-add all getfem sources to it to make it work. 


You can deform your mesh with a simple call: 

    auto deformator = temporary_mesh_deformator(mf, U, true, false);

    //the first true means "deform on creation", the second false means "do not restore the mesh back when temporary_mesh_deformator reaches the end of life"

    mf - is the mesh_fem for your displacement field and U is the displacement vector with gmm::vect_size(U) = mf.nb_dof();


Best regards,




On Thu, 8 Nov 2018 at 01:15, Zhenghuai Guo <address@hidden> wrote:

Dear Andriy,


Thank you for your explanation.


Regarding creep, at the moment I only use liner_elasticity_brick with young’s modulus being changed on each time-step. This is just to start with. I am new in Getfem in fact.


Could you please advise about the follows?


  1. Are  you using c++ to run getfem? If so,  can you give some hints how I can to it? I can’t see any instruction about setting up for c++.
  2. If you don’t use c++, how do you normally use the getfem::temporary_mesh_deformator  (from getfem_deformable_mesh.h) or other C functions?


I was trying to see if it is possible to use this function in Python interface by using SWIG or Python.Boost. But it is far beyond my knowledge.


Thank you very much



Zhenghuai Guo


From: Andriy Andreykiv <address@hidden>
Sent: Wednesday, November 7, 2018 8:32 PM
To: Zhenghuai Guo <address@hidden>
Cc: address@hidden
Subject: Re: [Getfem-users] mesh_deformation


Dear Zhenghuai Guo,


I don't use Python interface much, but your assumption is correct, using getfem::temporary_mesh_deformator  (from getfem_deformable_mesh.h) you can apply displacement field to the mesh.

By default temporary_mesh_deformator  will deform the mesh and un-deform it in the destructor, unless you build it with the argument to_be_restored=false.

I only assume that you can do it with Python too.   


I'm not really experienced with creep, but intuitively I would assume that you can also use large deformation formulation to account for the change in geometry.

Or it's not how you intend it?


In your follow up email you are asking about the usage of mesh slices. From what I know it's used primarily for post-processing, not calculation. If you intend to use it solely for

post-processing than you can easily achieve it nowadays with Paraview, were you import a vtk file, warp the result with a displacement field and take a desired slice.


Best regards,



On Sat, 3 Nov 2018 at 12:33, Zhenghuai Guo <address@hidden> wrote:

Dear Sir or Madam,


Could you please advise how I and deform a mesh according to a displacement field?


I am trying to simulate a time dependent deformation of a cylinder like object using python-interface. After applying stress the object creeps with time.


I think I can just go with many small time steps. In each time step, I would like to update and deform the mesh according to the displacement calculated as a function of time. And the deformed mesh will be an input for next time step.


I can see there is some related information such as (1) ‘getfem_deformable_mesh.h’ in page 18 in  (2) ‘getfem::slicer_apply_deformation’ in . But I can find details examples.


Thank you very much

Zhenghuai Guo

Tyree Xray CT network facility, School of Minerals and Energy Engineering Resources, UNSW Sydney




Attachment: e_out.txt
Description: e_out.txt

reply via email to

[Prev in Thread] Current Thread [Next in Thread]