Fracture energy definition

[MarDL]

Dear STKO team,
I write because I am not clear on how fracture energy is defined.
I am using a very simple model simulating a uniaxial compression test on a cube. I define the material with the ASDConcrete3D bond (Preset Concrete 9P).
I enter, as required, the fracture energy, calculated according to literature formulas, with a value of 7 N/mm.

I use the tester to compare the curve and if I enter an lch equal to 100 mm (size of the cube with which I have to compare the results)in the tester I obtain a graph comparable with my experimental results.
When I run the analysis, simulating the compression test, if I use a mesh with a size (100mm) that is equal to the sample size, I get the same result as the tester. But if I change the mesh size (e.g. 5mm) I get values that are completely different.

In the following post (viewtopic.php?p=10032&hilit=fracture+energy#p10032) I read that the fracture energy is always assumed in F/L, thus the automatic regularization is ON by default.

So I don't understand why changing the mesh size the final result changes. I enclose a graph in which the dotted lines are my experimental results and the solid lines several attempts at numerical tests.


The curves are named to indicate both the fracture energy value in N/mm and the mesh size.
I don't think it's a problem with the model because I've tried it with others that work and the material seems mesh dependent.
Can you please help me clarify this doubt?
Thank you in advance for your availability.
Best regards,
Mari

[STKO Team]

Can you share your model?

[MarDL]

Thank you for your availability. Please find the model file in the attached compressed folder. The model related to the graph I posted is this one. However, I used the ASDConcrete3D material in other models and I observed the same behaviour when varying the mesh.
Thank you

01_COMPRESSION_INT.zip

Model

(115.95 KiB) Downloaded 123 times

[STKO Team]

The auto regularization is working, however when you are trying to simulate a state of almost uniform strain (in your case uniaxial compression) the model "doesn't know" where to localize. So the damage will spread over a large area before some round-off error occurs, leading to the localized solution where the regularization works (in this cases using a smaller time step may help to a certain extent):

without_imperfections.png (25.66 KiB) Viewed 2433 times

You can add some imperfections to verify this point:

GaussRandMatTableMaker2D.zip

(2.93 KiB) Downloaded 87 times

random imperfection generator.png (99.12 KiB) Viewed 2433 times

with_random_imperfections.png (20.06 KiB) Viewed 2433 times

[MarDL]

Thank you for your availability and response. Using the randomiser material wrapper the model seems more stable and the solution more accurate.
I am trying to calibrate these materials for use within models to simulate uniaxial compression tests and diagonal compression tests on masonry panels, using a 2D Continuous micro-model (Both units and mortar joints are modelled using 2D plane-stress continuum elements with nonlinear behaviour). Would you mind continuing to use the randomiser wrapper in that case as well? Or should there be no problem since it is no longer a different test on a masonry panel?

[STKO Team]

In that case it will not do a huge difference because you will have different stress states already in your model