peprpendicular-flap solid-openfoam case needs tuning

[MakisH]

In #507, while comparing all the solver options we have, it became apparent that the solid-openfoam (solidDisplacementFoam) gives very different results than everything else. Still, it is a nice technical test that only requires OpenFOAM. The solid-solids4foam case is mostly fine already.

I have tried the following:

• Varying the mesh size: I get completely different results
• Varying the fvSolution/stressAnalysis/nCorrectors (suggestion by @philipcardiff): I get a bit different results, but not enough.

Some pictures that might help someone pick it up later on:

For now, I just give up, and document that this is an outlier at the moment.

One case use the plot-all-displacements.sh script to generate these pictures (see #507).

[MakisH]

Even with nCorr 10000 (mesh 6x15):

[MakisH]

For solid-solids4foam, increating nCorr from 1000 to 10k given exactly the same results as before (no improvement):

[philipcardiff]

I suspect another reason for the difference with openfoam-openfoam is that the 1st-order Euler ddt and d2dt schemes are used (solids4foam uses the 2nd-order backward scheme). You could try switching openfoam-openfoam to backward.

[MakisH]

The OpenFOAM documentation states that:

Maybe the d2dt2 backward is a nice solids4foam-only feature? :smile:

with:

d2dt2Schemes
{
    default         Euler;
}

ddtSchemes
{
    default         backward;
}

and with also varying the nCorr, I get the following:

At least the ddtSchemes alone does not seem to help.

[philipcardiff]

Ah, OK, yes, we added a d2dt2 backward option in solids4foam.

In that case, using a smaller deltaT would improve the OpenFOAM predictions, if that is an option (probably also the solids4foam predictions).

[MakisH]

Decreasing the deltaT is indeed not really an option, as we wanted to have this consistent among the possible case combinations.

I compared the solid-openfoam and solid-solids4foam directories in a diff tool, and one other difference I found was the tolerances we use:

# solid-openfoam/system/fvSolution
solvers
{
    "(D|T)"
    {
        solver          GAMG;
        tolerance       1e-06;       # Made this 1e-09
        relTol          0.9;         # Made this 0.1
        smoother        GaussSeidel;
        nCellsInCoarsestLevel 20;
    }
}

# solid-solids4foam/system/fvSolution
solvers
{
    "D|DD"
    {
        solver          PCG;
        preconditioner  FDIC;
        tolerance       1e-09;
        relTol          0.1;
    }
}

I changed these tolerances, but it did not seem to improve the results.

Since we have the solid-solids4foam case, it is fine for me if this issue stays open for longer. Maybe someone else stumbles upon it and gets another idea, while it still helps as documentation.

[precice-bot]

This issue has been mentioned on preCICE Forum on Discourse. There might be relevant details there:

https://precice.discourse.group/t/the-perpendicular-flap-model-using-openfoam/1695/25