ImplicitGenericBehaviour for poroplasticity -- difficulty with getIntegrationVariablesDerivatives_X

[GBacquaert]

Hi,

I am trying to use the @dsl ImplicitGenericBehaviour to implement a (very simple) poroplastic model. In this one, the @gradient are the strain tensor and the porosity, and their respective @ThermodynamicForce are the stress tensor and the pressure.

My difficulty relies in the use of getIntegrationVariablesDerivatives_X, to get the tangent operator blocks with the help of the Jacobian of the local non-linear system.

• For the total strain eto : getIntegrationVariablesDerivatives_eto(ddeel_ddeto, ddphiel_ddeto, ddlam_ddeto) seems to be ok.

• But for the porosity phi : getIntegrationVariablesDerivatives_phi(ddeel_ddphi, ddphiel_ddphi, ddlam_ddphi), it does not seem to compute anything right, even in the case of an elastic loading.

I attach here-under the corresponding mfront file.

PoroPlasticity.zip

P.S.: I use mfront-4.2.0

[thelfer]

@GBacquaert Let me have a look.

[thelfer]

@GBacquaert Do you have simple test with MTest at hand ? Otherwise, I'll construct one

[thelfer]

Ok, I think that I found the issue.
The problem appears for the derivatives of residual with respect to a scalar.
As a temporary hack, a workaround is to do:

//    dfphiel_ddphi = -1.;
//    dflam_ddphi = 0.;
dfzeros_ddphi(StensorSize, 0) = -1;
dfzeros_ddphi(StensorSize + 1, 0) = 0;

Can you test this ?

[thelfer]

#738

[GBacquaert]

Hi Thomas,
Indeed it fixes my problem. The computation of the tangent operator blocks seems also to be ok for both elastic and plastic loadings.

This is cool, thank's!

For those who would like to reemploy the mfront file, I put two precisions:

1. The porosity phi should be considered here as the variation from the initial one phi0, so that it results to a zero pressure when phi is zero.

2. @Theta 1 should be specified at the beginning of the file. Otherwise, you would get oscillations around the yield stress.

[thelfer]

@GBacquaert Don't hesitate to send me an MTest file if you can (with imposed thermodynamic forces to test the consitent tangent operator).

[GBacquaert]

Hi Thomas,

Here-under this is done by increasing the pressure as the @ThermodynamicForce. It results in increasing the von Mises yield stress of the mode (do not ask about the physical revelance...).

mtest_poroplasticity.zip

As a matter of fact, I have tried to consider the vice versa approach where the pressure would be the @gradient, and the porosity variation would be the @ThermodynamicForce. This is indeed convenient to deal with the case of an elastic incompressible matrix for porous media.

However, the tangent operator is therefore unsymmetric, even in the case of poroelasticity (the state equations are indeed equivalent to a min/max problem with respect to the strain and the pressure respectively). I am not sure mfront can handle this kind of problem through mtest. Would I be mistaken ? In any case, my first tries were unsuccessful ^^.

[thelfer]

@GBacquaert Thanks for the test !

It has been integrated (without the workaround of this issue) here: 854f903

However, the tangent operator is therefore unsymmetric, even in the case of poroelasticity (the state equations are indeed equivalent to a min/max problem with respect to the strain and the pressure respectively). I am not sure mfront can handle this kind of problem through mtest. Would I be mistaken ? In any case, my first tries were unsuccessful

I don't see why. There is no assumptions on the symmetry of the tangent operator. If you have an example, I can give it a look.