Incise TriangularFEMForceField depending on stress

[TheGreatLudini]

Hi everyone,

I would like to simulate the tearing of a 2D FEM mesh, using a TriangularFEMForceField (SOFA v21.12 on Windows). For that purpose, I am extracting the principal stress in every element and compare it to a threshold. If the stress is exceeding the threshold inside a triangle, I would like to incise that triangle along the principal stress direction, which includes splitting the triangle or snapping the incision to an edge, snap to a border, and so on.
So essentially, I want to recreate the behavior that happens when you interact with Shift+wheelClick with the mouse, but instead of 'cutting' (checking if there is a collision between mouse and object and then incise the mesh), I want to implement 'tearing' (checking the stress and then incise the mesh at a desired triangleID).

My question is now: What is the best way to do this? I saw that there are already some functions implemented in InciseAlongPathPerformer, TopologicalChangeManager and TriangleSetGeometryAlgorithms that handle these kinds of topological changes when interacting with the mouse. Maybe if I could somehow use these functions (for example TriangleSetGeometryAlgorithms::SplitAlongPath or TriangleSetGeometryAlgorithms::InciseAlongEdgeList), but without the Performer from the mouse interaction inbetween... Or is there another way to do it?
Furthermore, I would like to use SofaPython3 for my scene, so that I can implement different tearing criteria and experiment a bit.
As I heard that you are an expert in the topic of topological changes, maybe you can help me with this problem, @epernod? Thanks in advance!

Best regards,
Ludwig

[hugtalbot]

Dear @TheGreatLudini

Thanks for this interesting post. As you know, topological change is not a straightforward topic, but you identified correctly the components to look at mostly the InciseAlongPathPerformer, TopologicalChangeManager.

In your case, you would like the physics (some ForceField corresponding to a constitutive model) to provide the stress info and from here compute a incision / tearing / rupture. I would do it as follows:

• create a controller which at each end of time step would check the stress values given (as a Data) by the ForceField. This controller would define the path to be cut, if any.
• from here, two solutions: either you use the API of InciseAlongPathPerformer to create your cut, or you want to customize it (make it different that just lines between two points of incision), and then you should implement your own functions calling directly the TopologicalChangeManager

To make it in python to have different tearing criteria, which is an interesting move, you can:

• write the controller part (computing the points of cutting from the stress info) as a python controller
• pass the information to a modified InciseAlongPathPerformer which would have input Data (points of cut) and process the topological information
  As written, the InciseAlongPathPerformer will have to be modified to define the incision points as a Data ➡️ it should inherit from BaseObject to be in the scene and be able to declare Data

I hope this helps.
Maybe @epernod will correct my post.

[epernod]

Hello,

just quickly to complete @hugtalbot answer. As you mention, the blackbox code is inside the class TriangleSetGeometryAlgorithms. TopologicalChangeManager is a good input to see how to use the methods but you can also have a look at TopologicalChangeProcessor which works nearly like an SOFA engine. Instead of using the mouse, you can give him information on where to incise at what time and it will perform it.

And to give a proper answer, if you look at the case INCISE in this processor. You can see the order of calls:

1. TriangleSetGeometryAlgorithms::computeIntersectedObjectsList is called first to //Computes the list of objects (points, edges, triangles) intersected by the segment from point a to point b and the triangular mesh.
2. TriangleSetGeometryAlgorithms::SplitAlongPath is called using the output of step 1 to: //Split triangles to create edges along a path given as a the list of existing edges and triangles crossed by it.
3. TriangleSetGeometryAlgorithms::InciseAlongEdgeList is finally called using previous info to: //Duplicates the given edges

[epernod]

Otherwise, I think what you a looking for is this behavior based on TriangularFEMForceField stress information. This was done during an internship I supervised, but not fully integrated.
If you are ready to give a hand and finish the integration as an open-source feature inside a SOFA plugin, I'm ready to share this work.

[BrunoB81HK]

This is extremely interesting! Can this behavior be adapted for a TetrahedronFEMForceField or a TetrahedronHyperelasticityFEMForceField?

[epernod]

Hi,
yes this is possible as you can compute the stress inside your TetrahedronxxxFEMForceField. However you need to be able to incise inside a tetrahedral mesh and this feature is not available straight forward in SOFA.
This was also started in the internship but not finished and it is using this plugin: https://www.sofa-framework.org/applications/marketplace/cutting-mesh-refinement/

[TheGreatLudini]

Hello @hugtalbot and @epernod,

Thank you very much for your answers, this helps a lot. I will look into your suggestions and keep you updated on my results.

Regarding the work that was done during an internship that you supervised, @epernod: Thank you for the offer, this looks exactly like the behavior I want to implement. However, because of limited capacity, I cannot guarantee to finish the integration of your work as an open-source plugin for SOFA, so therefore I will try to create a tearing plugin on my own. Maybe I can share my work as a fully integrated plugin at a later stage.

Best regards,
Ludwig

[hugtalbot]

Hey @TheGreatLudini

I would definitely think about it twice regarding this cutting algorithm. If you have any issues working on the integration we will be there to support you. We always prefer avoiding re-implementation of existing codes.

Thanks a lot @epernod for this generous proposal 🎖️ !

[TheGreatLudini]

Hi @hugtalbot and @epernod,

I made a lot of progress in the tearing simulation, using the TopologicalChangeProcessor. Now, I am facing some issues regarding the tear propagation. My tear algorithm now looks like this:

1. Around the end point of the last tear, I am analyzing each triangle for its stress
2. Of all triangles around this point that are tearable (i.e., all triangles whose stress is higher than a threshold), I calculate the principal stress direction
3. These principal stress direction vectors are averaged, and the tear then propagates from the last end point in the direction of this averaged vector, with a fixed vector length. Tearing is done by splitting the triangles along the path, and subsequent remeshing

I am only analyzing the triangles around the tear end point because in reality, a tear would rather propagate from an existing end point. If I analyzed all triangles in the mesh, the tear would sometimes "jump" to a completely new location, leading to unrealistic tears.

I have to main problems with this algorithm: First of all, the tear can only propagate from one single end point. But usually, a tear has multiple locations from where it could propagate. But if I let the tear occur in all locations of the mesh, it looks unrealistic. It would be best if the tear could propagate from all of its end points, but I don't know how to determine if a triangle is located at an end point of an incision.
Secondly, I would like to propagate the tear in very small steps. Right now, after calculating the tear direction, I am tearing with a fixed length. So for example, the tear always goes 3mm further. But it would be better to change this step size dynamically, for example by tearing only so far that the next triangle in the mesh is reached. But I don't know how to implement that. As it seems, the method TriangleSetGeometryAlgorithms::computeIntersectedObjectsList does not allow an incision that begins and ends in neighboring triangles.

Coming back to the video that you posted, @epernod: In this video, the tearing looks really smooth and can also propagate from multiple ends. Could you specify how the tearing propagation was implemented? Were all triangles in the mesh analyzed for their stress? How was the length of the tear propagation calculated? I would really appreciate it if you could give some hints on that! Thank you very much.

Best regards,
Ludwig

[sin3point14]

We're currently checking out the various frameworks to achieve the simulation of tearing a thin tetragonal membrane. If this plugin is close to what we want to achieve, we can surely take it ahead!

[epernod]

Going through the discussion... Indeed I just answered in #3161 that the Tearing plugin has been released and should help you for the simulation you want to achieve: https://github.com/InfinyTech3D/Tearing

Hope it helps and any feedback is welcomed to improve it.

[RickYang1]

Hello,@TheGreatLudini
Sorry to bother you. I’m currently troubled by the issue of tear propagation too. Knowing that you’re highly experienced, I’d like to take the liberty of asking if you’d be kind enough to share some valuable suggestions on this matter. If you could spare some time to offer your guidance, I would be extremely grateful.
Thanks in advance!

[hugtalbot]

have you taken a look at the Tearing plugin @RickYang1 ?

[RickYang1]

@hugtalbot ,yes, I've already checked the Tearing plugin, but it doesn't seem to have effectively resolved the issue of tear propagation.

[hugtalbot]

This topic is directly related to this one #3161

[hugtalbot]

with no further discussion, here I will close it