[Structural] Adding SBM based on extension operators #12120

rubenzorrilla · 2024-02-28T16:53:06Z

📝 Description
This PR adds a new variant of the Shifted Boundary Method (SBM) based on extension operators for structural mechanics (so far only static small displacement problems). This implementation leverages the utility previously merged in #10346.

Details can be found in here.

...StructuralMechanicsApplication/custom_conditions/displacement_shifted_boundary_condition.cpp

loumalouomega · 2024-02-29T08:24:14Z

...StructuralMechanicsApplication/custom_conditions/displacement_shifted_boundary_condition.cpp

+    const double w = GetValue(INTEGRATION_WEIGHT);
+    const auto& r_N = GetValue(SHAPE_FUNCTIONS_VECTOR);
+    const auto& r_DN_DX = GetValue(SHAPE_FUNCTIONS_GRADIENT_MATRIX);
+    array_1d<double,3> normal = GetValue(NORMAL);


I would add a check of the norm being different of 0

Fair enough. But rather than putting it in here (it will be evaluated at each iteration), I'd put it in the ShiftedBoundaryMeshlessInterfaceUtility that calculates the normal values, so we only check it once.

loumalouomega · 2024-02-29T08:33:19Z

...StructuralMechanicsApplication/custom_conditions/displacement_shifted_boundary_condition.cpp

+    const double w = GetValue(INTEGRATION_WEIGHT);
+    const auto& r_N = GetValue(SHAPE_FUNCTIONS_VECTOR);
+    const auto& r_DN_DX = GetValue(SHAPE_FUNCTIONS_GRADIENT_MATRIX);
+    array_1d<double,3> normal = GetValue(NORMAL);


Idem for the norm check

(see my answer above)

loumalouomega · 2024-02-29T08:33:57Z

...StructuralMechanicsApplication/custom_conditions/displacement_shifted_boundary_condition.cpp

+    const double w = GetValue(INTEGRATION_WEIGHT);
+    const auto& r_N = GetValue(SHAPE_FUNCTIONS_VECTOR);
+    const auto& r_DN_DX = GetValue(SHAPE_FUNCTIONS_GRADIENT_MATRIX);
+    array_1d<double,3> normal = GetValue(NORMAL);


(see my answer above)

...StructuralMechanicsApplication/custom_conditions/displacement_shifted_boundary_condition.cpp

loumalouomega · 2024-02-29T08:35:55Z

...s/StructuralMechanicsApplication/custom_conditions/displacement_shifted_boundary_condition.h

+    ///@name Operations
+    ///@{
+
+    Condition::Pointer Create(


Missing doxygen

Shouldn't we rely on the base class one as this is override?

...s/StructuralMechanicsApplication/custom_conditions/displacement_shifted_boundary_condition.h

...StructuralMechanicsApplication/custom_elements/small_displacement_shifted_boundary_element.h

loumalouomega · 2024-02-29T08:42:22Z

...StructuralMechanicsApplication/custom_elements/small_displacement_shifted_boundary_element.h

+///@name Kratos Classes
+///@{
+
+template<std::size_t TDim>


No doxygen. Why Template in element but not in condition?, why only dimension and not the number of nodes?

The condition uses the base Kratos Geometry, for which the dimension and number of nodes make no sense. Indeed, the number of nodes is completely dependent on the cloud of points used for the calculation of the extension operator.

About the Doxygen, I'm adding it right away.

loumalouomega · 2024-02-29T08:52:28Z

...StructuralMechanicsApplication/custom_elements/small_displacement_shifted_boundary_element.h

+    ///@name Operations
+    ///@{
+
+    Element::Pointer Create(


Missing doxygen

Shouldn't we rely on the base class one as this is override?

...StructuralMechanicsApplication/custom_elements/small_displacement_shifted_boundary_element.h

loumalouomega · 2024-02-29T08:54:58Z

...StructuralMechanicsApplication/custom_elements/small_displacement_shifted_boundary_element.h

+
+    static constexpr std::size_t BlockSize = TDim;
+
+    static constexpr std::size_t LocalSize = NumNodes * TDim;


Okay you assume simplex (triangle + tetrahedra)

Yes, at least for the moment since our main levelset calculation routines assume this.

perhaps add some compiletime checks for the geometries it is used with? Or at least runtime

loumalouomega

Comments

…displacement_shifted_boundary_condition.cpp Co-authored-by: Vicente Mataix Ferrándiz <vmataix@altair.com>

…ub.com/KratosMultiphysics/Kratos into structural/new-shifted-boundary-method

loumalouomega

BTW, very nice job. I guess this is for FSI.

philbucher · 2024-02-29T10:33:21Z

...alMechanicsApplication/python_scripts/structural_mechanics_static_shifted_boundary_solver.py

+        for node in self.GetComputingModelPart().Nodes:
+            dist = node.GetSolutionStepValue(KratosMultiphysics.DISTANCE)
+            if abs(dist) < tol:
+                node.SetSolutionStepValue(KratosMultiphysics.DISTANCE, 0, tol)


variable utils

The problem in here is the if statement (as far as I understand we cannot pass the function to be evaluated). Indeed, I'd do an SBM version of the distance modification process (I left this for a future PR).

philbucher · 2024-02-29T10:33:29Z

...alMechanicsApplication/python_scripts/structural_mechanics_static_shifted_boundary_solver.py

+    def GetDefaultParameters(cls):
+        this_defaults = KratosMultiphysics.Parameters(r"""{
+            "conforming_basis" : true,
+            "extension_operator_type" : "MLS",


Suggested change

"extension_operator_type" : "MLS",

"extension_operator_type" : "mls",

I put it in capital letters as it is an acronym. Note that changing it to lower case would imply changing the ShiftedBoundaryMeshlessInterfaceUtility settings as well.

i still recommend to change it, so far we do it for most acronyms too
Up to you ;)

Fair enough. If we're changing it, I'll do it at once (note that this involves another PR as well as things that are already merged to the master).

@philbucher if this makes sense for you I'll proceed with the merge.

this one is ok for me, but I would still add the check regarding the geometry (see my other comment)

philbucher · 2024-02-29T10:34:55Z

...StructuralMechanicsApplication/custom_elements/small_displacement_shifted_boundary_element.h

+
+    static constexpr std::size_t BlockSize = TDim;
+
+    static constexpr std::size_t LocalSize = NumNodes * TDim;


perhaps add some compiletime checks for the geometries it is used with? Or at least runtime

rubenzorrilla · 2024-02-29T14:51:16Z

BTW, very nice job. I guess this is for FSI.

Thanks. Well, not necessarily for FSI, I'd say is more suitable for rapid prototyping, optimization, or any application involving large displacements/rotations of the boundaries.

loumalouomega · 2024-02-29T14:55:59Z

BTW, very nice job. I guess this is for FSI.

Thanks. Well, not necessarily for FSI, I'd say is more suitable for rapid prototyping, optimization, or any application involving large displacements/rotations of the boundaries.

Indeed, you could use it for topological optimization avoiding remeshing...

philbucher

did you try to make this work at compile time?
Just curious if it is possible 🤷

rubenzorrilla · 2024-03-08T10:26:57Z

did you try to make this work at compile time? Just curious if it is possible 🤷

I did not even try because, as far as I understand, it is not possible. The point is that the geometry is assigned when constructing the element in runtime (or when declaring the prototypes).

rubenzorrilla added 3 commits February 28, 2024 16:08

Bringing structural app files from dev branch

7d3ffb5

Clean up test

22743d8

Final clean up

55c161b

rubenzorrilla added Applications Feature labels Feb 28, 2024

rubenzorrilla self-assigned this Feb 28, 2024

rubenzorrilla requested a review from a team as a code owner February 28, 2024 16:53