asdeasoft.net

Dear Sir/Madam
I have been modeling a masonry wall with a dimension of 0.99 m x 0.99 m x 0.09m with mortar thickness of 10 mm. Brick size 190 x 190 x 90 mm3. The objective was to obtain lateral force vs displacement curve. Keeping every other parameters, I observed that the peak strength depended on the mesh size, as well as the "number of "increment" or the "load step-size" in the analysis option. Following are the specific questions treating the mesh size and the number of increment independently:

1. For example if we consider only the mesh size option (structured+quad), I got the following peak strength: By Uniform Division--> 114 kN, By Uniform Size (30)-->80 kN. You can see the 34 kN difference!

2. On the other hand if we consider the "number of increment (N)" in the analysis option for displacement controlled analysis up to 1mm, keeping the mesh size same, I got the following peak strength: N= 50, dT=1/50=0.05-->108 kN; N=100, dT=1/100=0.01-->97 kN; N=200, dT=1/200=0.005-->92 kN

Can you please guide me on how to choose the appropriate mesh size and the number of increment for analysis? Any guidance on it is greatly appreciated.

Thank you

Keeping every other parameters, I observed that the peak strength depended on the mesh size, as well as the "number of "increment" or the "load step-size" in the analysis option. Following are the specific questions treating the mesh size and the number of increment independently

In FEM, it is normal that the approximated solution approaches the exact one upon mesh refinement. There are few cases in which you can use 1 or few elements per structural members, but in general your mesh needs to be fine.

Regarding the time-step size, it depends on the fact that you are using the IMPLEX algorithm. The Explicit part of the IMPL-EX, requires a small time step. You can see that upon refining the time-step, you will approach a unique solution.

If you post your model, I will show you these concepts

Dear Sir/Madam
Thank you for your reply. I hereby attach the model for your kind perusal.

As you can see here, refining the mesh converges to an (almost) unique solution.
"Almost" because:
The stable part (until the peak) converges to a unique value, while in the post-peak regime, there is a larger scatter, and this makes sense because once the system becomes unstable, the finer is the mesh, the more "freedom" the crack pattern has.

Dear Sir/Madam
Thanks a lot for your time and effort in answering my question. I understand the effect of mesh size on the response. However, do you think the time-step size in the IMPL-EX algorithm also has such an effect, and I need to do a convergence study to decide on the appropriate value of it?
Out of curiosity, just an additional question on the mesh size, though it has got nothing to do with STKO!: If there is no unique solution in the post-peak range, do you think that the model would be reliable for a cyclic analysis which usually involves the post-peak regime that has a lot scatter as per the model?

do you think the time-step size in the IMPL-EX algorithm also has such an effect

The time-step size has an influence on the IMPL-EX algorithm, due to the Explicit part of the IMPL-EX algorithm. At the moment you can do a convergence study to understand how it influences the response. Probably you will find out that, for shear walls, you can estimate the time-step size making sure the drift ratio of a shear wall is kept under something like 1/100.
We are currently working on implementing a convergence test that checks for the IMPL-EX error and adapts the time step accordingly.

Another approach that you can do, it to first run an analysis with the Implicit algorithm. It will probably have convergence issues soon after the peak point, but at least you are sure that the peak strength is not overestimated. Then, you can run the IMPL-Ex, which will converge also after the peak point. At least you have a comparison for the peak strength, so you can make sure that, with the time-step you choose, the IMPL-EX is not overestimating it.

f there is no unique solution in the post-peak range, do you think that the model would be reliable for a cyclic analysis which usually involves the post-peak regime that has a lot scatter as per the model?

This is a good question!
The problem is very simple: If you ask a pure mathematician, he will tell you that you cannot use material models with strain-softening in FEM because once a point in your mesh passes the peak point, and goes into softening, the governing partial differential equation is not elliptic anymore, and the solution will not be unique anymore.
This is because FEM is based on continuum theory, and in continuum theory, there should be no separation or interpenetration of material... which means NO CRACKS!

Unfortunately in this way, we, as structural engineers, should use FEM only with steel structures!!
However, we definitely need to use FEM for RC or Masonry structures. So we just need to be aware that whenever you have softening, you should be very careful!

That being said, you should consider another aspect. In this model, the post peak response is quite scattered, but it is also due to the fact that you have a very small model, with few micro-structural components. This happens also in real life. If you do some experimental tests on such small specimens, you will see a similar scatter in the results.
Try to do the same with a larger wall, you will see that the results will be less scattered, and probably also less dependent on the mesh size.

Dear Sir/Madam
Thanks a lot for such an elaborate answer and your time. I am now clear and convinced about how to use such models. Really appreciate STKO's support to its users.

You're welcome