Kratos: [Fluid Dynamics] EmbeddedDragProcess

[dagmawibekel]

Hi everyone,
i was trying to compute the drag and lift forces on an embedded bluff body in a fluid flow. However, the EmbeddedDragProcess results in a segmentation fault in the first iteration if the time interval of the process starts from 0. I am advancing in time using a BDF2 discretization.
The BodyFittedDragProcess using a BDF2 scheme does work independent of the interval chosen.

With the help of @jcotela, i circumvented the problem by starting the computation of the drag forces starting from the second time step.
However, i believe that there should be a more robust solution to this problem.
Best regards!

@rubenzorrilla @KratosMultiphysics/fluid-dynamics

[rubenzorrilla]

We should find a way to check that the buffer is filled prior to the drag computation.

[dagmawibekel]

Hi,
I encountered another bug using the embedded drag process.
I was trying to compute the embedded drag for a system that has approximately 5 million tetrahedral elements. Unfortunately, i keep getting zeros for the drag forces when i use MPI. I checked the same simulation using OpenMP and there i actually get forces unequal to zero.
Furthermore, i changed the background mesh to have less elements and for less elements it again worked...

Best regards!

[rubenzorrilla]

Hi @dagmawibekel !

This is really weird... Just for me to be sure I understood it well, for MPI is working for a relatively small mesh but no for the 5M elements mesh?

[dagmawibekel]

Thanks for the quick reply.
You understood correctly.
I am running a simple benchmark example and was refining my background mesh towards my geometry. For smaller examples it is working both using MPI and OpenMP. However, if the simulation has a certain size, the embedded drag process only gives values different from zero for OpenMP. When using MPI i get zeros in each time step.

[rubenzorrilla]

Which kind of mesh refining are you using? AFAIK the mmg library doesn't work in MPI (yet).

[dagmawibekel]

I am refining my mesh in GID using refinement boxes since i know where the structure is going to be.

[rubenzorrilla]

@dagmawibekel would you mind trying after #5171 is merged?

[dagmawibekel]

works like a charm @rubenzorrilla

[rubenzorrilla]

works like a charm @rubenzorrilla

"Works" includes the zeros in some MPI cases?

[philbucher]

is this still active?

[rubenzorrilla]

I don't think so. @dagmawibekel feel free to reopen it if needed.