Effect of Young's modulus on the stiffness of a model change based on the mesh used

[BrunoB81HK]

Hi,

I'm currently working on two similar haptic simulations that use the same model parameters (Young's modulus, poisson ratio, ...) but different FEM meshes. Even when using the same parameters and the same scene, the felt stiffness of the model changes drastically when I change the mesh. Here is what I mean:

Good stiffness vs bad stiffness

The used models are defined as:

<Node name="Liver">
    <!-- Load the required meshes -->
    <MeshOBJLoader name="VisMesh" filename="../../../share/mesh/Liver/Liver.obj" handleSeams="1"/>
    <MeshOBJLoader name="ColMesh" filename="../../../share/mesh/Liver/LiverCol.obj"/>
    <MeshOBJLoader name="AttachMesh" filename="../../../share/mesh/Liver/AttachPoints.obj"/>
    <MeshGmshLoader name="FEMMesh" filename="../../../share/mesh/Liver/LiverFEM.msh"/>

    <!-- Define the model(s) -->
    <EulerImplicitSolver rayleighStiffness="0.1" rayleighMass="0.1"/>
    <CGLinearSolver threshold="1e-11" tolerance="1e-11" iterations="25"/>
    <TetrahedronSetTopologyContainer src="@FEMMesh"/>
    <TetrahedronSetGeometryAlgorithms template="Vec3d"/>
    <MechanicalObject src="@FEMMesh" name="DOFs"/>
    <PrecomputedConstraintCorrection recompute="false"/>
    <MeshMatrixMass massDensity="1000" />
    <NearestPointROI template="Vec3" name="np" object1="@./DOFs" object2="@./Attach/model" radius="0.0075"/>
    <RestShapeSpringsForceField points="@np.indices1" stiffness="1000" angularStiffness="1000"/>
    <TetrahedronFEMForceField name="FEM" method="large" template="Vec3d"
                              listening="1" youngModulus="40000" poissonRatio="0.4"
                              computeGlobalMatrix="0" computeVonMisesStress="1"
                              showVonMisesStressPerElement="1"
    />

    <Node name="Attach">
        <MeshTopology src="@../AttachMesh"/>
        <MechanicalObject name="model" src="@../AttachMesh"/>
        <BarycentricMapping input="@../DOFs" output="@model"/>

    </Node>

    <Node name="Col">
        <MeshTopology src="@../ColMesh"/>
        <MechanicalObject name="model" src="@../ColMesh"/>
        <TriangleCollisionModel contactStiffness="1000" contactFriction="0.005"/>
        <BarycentricMapping input="@../DOFs" output="@model"/>

    </Node>

    <Node name="Vis">
        <OglModel name="model" src="@../VisMesh" texturename="../../../share/textures/Liver.png"/>
        <BarycentricMapping input="@../DOFs" output="@model"/>

    </Node>

</Node>

Why is this happening and is there any way to have the stiffness match regardless of the FEM mesh?

Thanks a lot for your help!

Bruno

[hugtalbot]

Hi @BrunoB81HK

Interesting question. Are the two meshes with very different resolutions?

Anyway, this kind of issue is related to the convergence of your model.
To make sure your simulation converges properly, one shall always start by:

• running a converging simulation with
  • a very small time step dt
  • a very small space step dx (mesh size)
  • and high criterion regarding the tolerance/threshold if you have iterative solvers
    this is your "reference" simulation!
• then you can start increasing both dt anddx and comparing with the reference simulation to ensure that you remains close to the reference result

Therefore, I look in your scene and I see :
<CGLinearSolver threshold="1e-11" tolerance="1e-11" iterations="25"/>

25 iterations are very very low, simulation must be fast but nothing ensures here that your CG converges, actually I bet it doesnt.
Could you try increasing this number? (increase it a lot to start like 10000) and compare the two (slow) simulations with the two meshes

[BrunoB81HK]

Hi @hugtalbot

You guessed right, both meshes have different resolutions.

I'll try to increase the number of iterations and let you know my results. Is there a way to know if the CG converge? Perhaps with warnings or by looking at a certain variable?

Thanks a lot for your quick help!

[hugtalbot]

you may get such information activating the verbose data of components printLog
In addition, the CG also provides some graph that can be displayed but it requires QtGraph if I recall corretly

[BrunoB81HK]

I increased the number of iterations and it's now behaving as I expected it. Thanks a lot for your help!

[hugtalbot]

you're very welcome @BrunoB81HK ! 👍