Mapping a 3D mesh on a 1D beam

[peyronq1]

Hello there!

In order to model beams and rods with different cross section geometries and material properties, I would like to couple a 1D beam (from BeamAdapter) and a 3D mesh, compute the elastic forces in the mesh to reproject them later on the 1D beam, similar to the work of Paul Chaillou did on pneumatic soft robots. From what I understood, the AdaptiveBeamMapping is the component to use, but I have troubles with it. In the example of code below, I have a 1D beam and try to couple to it a 3D point cloud representing a beam with a square cross-section. The problem is at the creation of the scene, all the 3D points seem projected on the YZ plane at the base of the beam. Does someone have an idea on why, please? Do I use the right component?

Sofa and plugin version: v24.12
Python: 3.10

Thank you in advance and have a great day!
Quentin

import Sofa

def createScene(rootNode):

    rootNode.addObject('FreeMotionAnimationLoop')
    rootNode.addObject('VisualStyle',displayFlags='showVisual showBehavior')
    rootNode.addObject('GenericConstraintSolver',tolerance=1e-6,maxIterations=1000)

    modeling = rootNode.addChild('Modeling')

    rod = modeling.addChild('Rod')

    L = 100.0
    N = 10
    frames = []
    for k in range(0,N):
        frames.append([0.0,0.0,k*L/(N-1),0.0,0.0,0.0,1.0])

    rod.addObject('RegularGridTopology', name='topo', n = [N,1,1], min = [0.0,0.0,0.0], max = [L,0.0,0.0])
    rod.addObject('MechanicalObject',template = 'Rigid3', showObject=True,showObjectScale=4.0)
    rod.addObject('BeamInterpolation',defaultYoungModulus=20e3, dofsAndBeamsAligned='true', straight='1', crossSectionShape='circular', radius=1.0)
    rod.addObject('AdaptiveBeamForceFieldAndMass',computeMass = True, massDensity = 1e-3)
    rod.addObject('FixedProjectiveConstraint',indices = 0)


    conModel = rod.addChild('ConstitutiveModel')

    # As the Adaptative beam mapping seems to require mesh nodes sorted in increasing curvilinear abscissa (x axis)
    # I define manually a point cloud here with such property, corresponding to a beam with a square cross-section
    positions = []
    Nc = 3
    NL = 10
    c = 10.0
    for k in range(0,NL):
        for w in range(0,Nc):
            for z in range(0,Nc):
                positions.append([k*L/(NL-1),z*c/(Nc-1)-c/2,w*c/(Nc-1)-c/2])

    conModel.addObject('MechanicalObject',position = positions, showObject = True, showObjectScale=5)
    # When commented, the point cloud appears as expected, as a beam with a square cross section
    conModel.addObject('AdaptiveBeamMapping', interpolation='@../BeamInterpolation',useCurvAbs =True,listening=True)

    simulation = rootNode.addChild('Simulation')
    simulation.addObject('EulerImplicitSolver',rayleighMass = 0.0, rayleighStiffness = 0.0)
    simulation.addObject('SparseLDLSolver')
    simulation.addObject('GenericConstraintCorrection')

    simulation.addChild(rod)

[hugtalbot]

Hey @peyronq1

Sorry for keeping you waiting.
I am not aware of Paul's modeling approach in details. From what I understood, it was a 3D deformable beam which was mapped on a beam topology. So from what I understood back then, it was rather the 3D model which was a child of a Beam topology. To be noticed, I think the beam was not involving mechanics itself since it was already computed by the 3D mechanical model.

Sorry not to guide you further right away. Do you have any existing SOFA scene from him?
Cheers

[peyronq1]

Hello @hugtalbot ,

No worries, thank you for your inputs! In agree with everything you said. In the example I gave above, I defined the "3D mesh", which is actually just a point cloud, as a child ('Consitutive model') of the node 'Rod' where the beam is defined. And I created it following the most recent version of Paul's code, that can be found on his gitlab (I need to check before sharing it). On my machine his code gives the same problem of the 3D nodes being projected at the base of the beam instead of following its deformations. Is it well what AdaptiveBeamMapping is supposed to do?

Thank you for your time !

[peyronq1]

Hello everyone,

UPDATE: I am currently discussing this with @ChristianDuriez, who used to work with this mapping. There seems to be a problem somewhere indeed, but we have not found where yet. Concerning my initial objective, it seems I can make this 3D mesh follow the beam deformations with a SkinningMapping as well. From what I understood, this mapping does not account for an interpolation between the frames of a beam element, but taking a sufficiently high number of beam elements does the job!

[peyronq1]

Hello everyone,

I would like to restart the discussion about this problem I encountered with the AdaptiveBeamMapping. As I was saying in my previous message, according to @ChristianDuriez, this mapping is the good one to use normally to couple a point cloud (the "out" mechanical state) to a beam (the "in" state). @fredroy and @epernod , I see on the history of the mapping's code that you were both working on it recently. Any idea of where this problem could come from ? I can provide more details if needed :)

Big thanks in advance for your time :)

[hugtalbot]

Hi @peyronq1

The following mapping has been mentioned: https://github.com/sofa-framework/sofa/blob/df96d0b435b0244a7aacde4a939d142cbc6353ee/examples/Component/Mapping/Linear/DeformableOnRigidFrameMapping.scn

I will ask Mimesis for input about what they use in their usecases. We'll keep you posted

[hugtalbot]

Does a Barycentric not suffice ?

[peyronq1]

Hey @hugtalbot ,

Thank you for telling me about the DeformableOnRigidFrameMapping, I didn't have this in my Sofa Pokedex yet! I will try it at some point.

And concerning the BarycentricMapping, from what I understood it enables to make a rigid (the position part of a rigid actually) follow a vec3 point cloud but not the opposite, right?

Thanks again !