HM process. #1508

endJunction · 2016-11-01T23:08:56Z

Two other changes outside 'ProcessLib/HM' are:

Move Process::_process_variables into protected section; direct access in the HMProcess is needed for creating specialized dof-table.
Add getBaseNodes() is added to 'MeshLib/Elements/Utils.h'; it's used in creation of the mesh subset for the pressure variable which is not using the quadratic elements' nodes.

Theory questions to @nagelt.
A confined compression test is included.

TODO:

Documentation:
doxygen
conifg parameters
docs.opengeosys.org/Selected Benchmarks.

ogsbot · 2016-11-04T00:15:23Z

Jenkins: OGS-6/Linux-PRs-dynamic failed: https://svn.ufz.de:8443/job/OGS-6/job/Linux-PRs-dynamic/1584/

norihiro-w · 2016-11-07T08:30:36Z

ProcessLib/HydroMechanics/CreateHydroMechanicsProcess.cpp

+            "%d components.",
+            process_variables[0].get().getName().c_str(),
+            process_variables[0].get().getNumberOfComponents(),
+            DisplacementDim);


remove DisplacementDim

norihiro-w · 2016-11-07T08:31:46Z

ProcessLib/HydroMechanics/CreateHydroMechanicsProcess.cpp

+        "intrinsic_permeability",
+        parameters, 1);
+
+    DBUG("Use \'%s\' as hydraulic conductivity parameter.",


hydraulic conductivity -> Intrinsic permeability

norihiro-w · 2016-11-07T08:34:53Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+        _material_state_variables;
+
+    typename BMatricesType::KelvinMatrixType _C;
+    double integration_weight;


need prefix _ as you do for other variables

norihiro-w · 2016-11-07T08:35:58Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+    }
+};
+
+struct HydroMechanicsLocalAssemblerInterface


why do you need this struct?

norihiro-w · 2016-11-07T08:37:41Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+    : public HydroMechanicsLocalAssemblerInterface
+{
+public:
+    using ShapeMatricesType =


I would rename this to ShapeMatricesTypeDisplacement

norihiro-w · 2016-11-07T08:38:20Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+
+    // Types for displacement.
+    // (Higher order elements = ShapeFunctionDisplacement).
+    using NodalMatrixType = typename ShapeMatricesType::NodalMatrixType;


I prefer NodalMatrixTypeDisplacement

norihiro-w · 2016-11-07T08:41:49Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+
+    typename ShapeMatrixTypeDisplacement::NodalRowVectorType _N_u;
+    typename BMatricesType::BMatrixType _b_matrices;
+    typename BMatricesType::KelvinVectorType _sigma, _sigma_prev;


the stresses here are actually effective stress. please rename the variables such that we can easily distinguish them from total stress.

norihiro-w · 2016-11-07T08:43:38Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+                                                            pressure_size>>(
+            local_rhs_data, displacement_size + pressure_size);
+
+        typename ShapeMatricesTypePressure::NodalMatrixType laplace_p;


why not store these laplace, storage, and Kup matrices as the member variable of this class or the integration point data?

norihiro-w · 2016-11-07T08:46:06Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+            auto const body_force =
+                _process_data.specific_body_force(t, x_position);
+            assert(body_force.size() == DisplacementDim);
+            auto const b =


is it necessary to evaluate the body force at each gauss point and each time step?

norihiro-w · 2016-11-07T08:47:02Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+                    displacement_size>::Zero(DisplacementDim,
+                                             displacement_size);
+            for (int i = 0; i < DisplacementDim; ++i)
+                N_u.template block<1, displacement_size / DisplacementDim>(


N_u can be precomputed and be stored at the integration point data.

This point and storing the laplace_p and storage_p matrices I'll check later.

The big N_u matrix is easy to create (just three copies of shape_matrices_u.N) contrary to the BMatrix which creation needs additional computations, not only copies. So the additional storage of big N_u matrix might be counterproductive.
Probably for the laplace_p and storage_p it is better to preallocate the space.

Both variants need to be benchmarked. The fastest solution will be chosen.

Storing laplace_p and storage_p in the class is more expensive (in time) than creation in the function.
Storing the large N_u in the integration points gains some speedup.
(Tested on square 1e2 example running approximately 6.7s)

norihiro-w · 2016-11-07T08:50:02Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+
+            storage_p.noalias() += N_p.transpose() * S * N_p * w;
+
+            local_rhs.template block<pressure_size, 1>(pressure_index, 0)


can be written as local_rhs.template segment<pressure_size>(pressure_index)

norihiro-w · 2016-11-07T08:52:24Z

ProcessLib/HydroMechanics/HydroMechanicsProcess.h

+private:
+    using LocalAssemblerInterface = HydroMechanicsLocalAssemblerInterface;
+
+void constructDofTable() override


indentation is wrong

norihiro-w · 2016-11-07T08:54:02Z

ProcessLib/HydroMechanics/HydroMechanicsProcess.h

+
+    _local_to_global_index_map.reset(new NumLib::LocalToGlobalIndexMap(
+        std::move(all_mesh_subsets), NumLib::ComponentOrder::BY_LOCATION));
+    //std::cout << *_local_to_global_index_map << "\n";


norihiro-w · 2016-11-07T08:55:35Z

ProcessLib/HydroMechanics/HydroMechanicsProcess.h

+                new MeshLib::MeshSubsets{_mesh_subset_all_nodes.get()}};
+        });
+
+    _local_to_global_index_map.reset(new NumLib::LocalToGlobalIndexMap(


you need to pass std::vector<unsigned> const vec_n_components{1, DisplacementDim}; to the LocalToGlobalIndexMap constructor

norihiro-w · 2016-11-07T08:59:18Z

ProcessLib/HydroMechanics/LocalDataInitializer.h

+{
+namespace HydroMechanics
+{
+/// The LocalDataInitializer is a functor creating a local assembler data with


can you comment here what is a difference from ProcessLib/Utils/LocalDataInitializer.h?

norihiro-w · 2016-11-07T09:00:26Z

ProcessLib/HydroMechanics/LowerDimShapeTable.h

@@ -0,0 +1,66 @@
+/**


why not move this file to NumLib/Fem/FiniteElement?

norihiro-w · 2016-11-07T09:01:39Z

ProcessLib/Process.h

@@ -185,12 +185,12 @@ class Process

    unsigned const _integration_order;

-private:
-    GlobalSparsityPattern _sparsity_pattern;
-
    /// Variables used by this process.
    std::vector<std::reference_wrapper<ProcessVariable>> _process_variables;


you can access process variables via the public function getProcessVariables().

norihiro-w · 2016-11-07T09:07:58Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+    // Types for pressure.
+    // (Lower order elements = Order(ShapeFunctionDisplacement) - 1).
+    using ShapeFunctionPressure =
+        typename LowerDim<ShapeFunctionDisplacement>::type;


I would add ShapeFunctionDisplacement to template parameters of this class and use LowerDim in LocalDataInitializer, because algorithms in the local assembler don't need this information.

wenqing · 2016-11-07T08:26:57Z

ProcessLib/HydroMechanics/CreateHydroMechanicsProcess.cpp

+         solid_density.name.c_str());
+
+    // Fluid density
+    auto& fluid_density = findParameter<double>(


Move this line (the fluid density) to the place after that of fluid viscosity.

For the single phase flow in this process, I think, you can replace these Parameter fluid data with the models of fluid properties, e.g.

auto& viscosity = MaterialLib::Fluid::FluidProperty::createViscosityModel(consig); auto& density = MaterialLib::Fluid::FluidProperty::createFluidDensityModel(consig);

where the auto type is std::unique_ptr<MaterialLib::Fluid::FluidProperty>.

Of course, you have to change the access of fluid density and viscosity in the local assembly accordingly with the above changes.

I'd like to skip the usage of fluid properties for now. When the material lib is set into a general framework we can start updating the processes to use that framework.

wenqing · 2016-11-07T08:31:05Z

ProcessLib/HydroMechanics/CreateLocalAssemblers.h

+            break;
+        default:
+            OGS_FATAL(
+                "Meshes with dimension different than two and three are not "


-->Meshes containing one dimensional elements ...?

In general correct, but at this place only the mesh's dimension is checked. The line elements and their error messages are handled ("not created") in the HydroMechanics::LocalDataInitializer.

wenqing · 2016-11-07T08:38:15Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+            auto const porosity = _process_data.porosity(t, x_position)[0];
+            auto const body_force =
+                _process_data.specific_body_force(t, x_position);
+            assert(body_force.size() == DisplacementDim);


This assert can be moved to this place where specific_body_force is read.

wenqing · 2016-11-07T08:41:16Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+            auto const body_force =
+                _process_data.specific_body_force(t, x_position);
+            assert(body_force.size() == DisplacementDim);
+            auto const b =


This can be moved outside the integration loop. Besides, there is a tool in MathLib, MathLib::toVector, for this conversion.

wenqing · 2016-11-07T08:45:25Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+        laplace_p.setZero(pressure_size, pressure_size);
+
+        typename ShapeMatricesTypePressure::NodalMatrixType storage_p;
+        storage_p.setZero(pressure_size, pressure_size);


storage_p --> mass_p

TBD;
In that particular matrix specific storage is used, no mass, so I'd like to keep it like it is.

In the FEM, int (N^T m N) dA, which associated with time (here for int (\dot p m N) dA), is normally termed as mass matrix.

wenqing · 2016-11-07T09:50:15Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+            //
+            laplace_p.noalias() += dNdx_p.transpose() * K_over_mu * dNdx_p * w;
+
+            storage_p.noalias() += N_p.transpose() * S * N_p * w;


storage_p --> mass_p

Besides, if the general fluid property models are use for fluid viscosity and density, please refer to LiquidFlowMaterialProperties::getMassCoefficient(...) to get rho and drho_dp. With the changes, S can be replaced with porosity * drho_dp / rho + S.

wenqing · 2016-11-07T09:50:55Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+        local_Jac
+            .template block<pressure_size, pressure_size>(pressure_index,
+                                                          pressure_index)
+            .noalias() += laplace_p + storage_p / dt;


storage_p --> mass_p

wenqing · 2016-11-07T09:56:19Z

ProcessLib/HydroMechanics/CreateHydroMechanicsProcess.cpp

+    // Storage coefficient
+    auto& storage_coefficient = findParameter<double>(
+        config,
+        //! \ogs_file_param_special{process__HYDRO_MECHANICS_storage_coefficient}


storage is not a coefficient, it is physical parameter.

ogsbot · 2016-11-08T18:50:42Z

Jenkins: OGS-6/Linux-PRs-dynamic failed: https://svn.ufz.de:8443/job/OGS-6/job/Linux-PRs-dynamic/1620/

ogsbot · 2016-11-08T18:51:02Z

Jenkins: OGS-6/Linux-PRs-dynamic failed: https://svn.ufz.de:8443/job/OGS-6/job/Linux-PRs-dynamic/1621/

endJunction · 2016-11-08T19:17:48Z

Sorry, I had to rebase on ufz/master because of merge conflicts in Tests/Data.

norihiro-w

looks good

norihiro-w · 2016-11-08T19:49:07Z

ProcessLib/HydroMechanics/HydroMechanicsProcess.h

+            std::move(all_mesh_subsets), vec_n_components,
+            NumLib::ComponentOrder::BY_LOCATION));
+    }
+    void initializeConcreteProcess(


please add an empty line

wenqing · 2016-11-09T08:33:07Z

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

+        laplace_p.setZero(pressure_size, pressure_size);
+
+        typename ShapeMatricesTypePressure::NodalMatrixType storage_p;
+        storage_p.setZero(pressure_size, pressure_size);


In the FEM, int (N^T m N) dA, which associated with time (here for int (\dot p m N) dA), is normally termed as mass matrix.

The vector is filled with pointers to base nodes.

Used to derive shapefunctions of lower order for example as needed for pressure given displacement shape functions in HydroMechanics process.

ogsbot · 2020-06-19T12:37:04Z

OpenGeoSys development has been moved to GitLab.

See this pull request on GitLab.

endJunction force-pushed the HM branch 4 times, most recently from 5ca81fe to b16a496 Compare November 4, 2016 00:03

endJunction force-pushed the HM branch from b16a496 to c0b613a Compare November 4, 2016 17:18

norihiro-w reviewed Nov 7, 2016

View reviewed changes

ProcessLib/HydroMechanics/HydroMechanicsFEM.h

}

};

struct HydroMechanicsLocalAssemblerInterface

Copy link

Collaborator

norihiro-w Nov 7, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why do you need this struct?

norihiro-w reviewed Nov 7, 2016

View reviewed changes

ProcessLib/HydroMechanics/LowerDimShapeTable.h

@@ -0,0 +1,66 @@

/**

Copy link

Collaborator

norihiro-w Nov 7, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why not move this file to NumLib/Fem/FiniteElement?

norihiro-w reviewed Nov 7, 2016

View reviewed changes

wenqing reviewed Nov 7, 2016

View reviewed changes

endJunction added the needs an update label Nov 8, 2016

endJunction force-pushed the HM branch from 0ea7fd4 to 6893eb6 Compare November 8, 2016 18:45

norihiro-w approved these changes Nov 8, 2016

View reviewed changes

wenqing approved these changes Nov 9, 2016

View reviewed changes

endJunction force-pushed the HM branch from 69b533b to 1fcb208 Compare November 9, 2016 23:37

endJunction added please rebase and removed needs an update labels Nov 9, 2016

endJunction force-pushed the HM branch 2 times, most recently from f451836 to 1d2045a Compare November 10, 2016 15:42

endJunction and others added 3 commits November 10, 2016 17:49

[MeL] Add getBaseNodes() collecting nodes from elements.

b3b7d82

The vector is filled with pointers to base nodes.

[App] Add HydroMechanics to ProjectData.

9712c32

[NL] Add lookup table for lower order elements.

8725d9c

Used to derive shapefunctions of lower order for example as needed for pressure given displacement shape functions in HydroMechanics process.

endJunction force-pushed the HM branch from 1d2045a to d14561c Compare November 10, 2016 16:51

endJunction and others added 3 commits November 10, 2016 18:40

[PL] Implemention of HydroMechanics process.

b0f189e

[T] Add confined compression HM ctests.

762aacc

[doc] ConfigTree documentation stubs.

eda2ffd

endJunction force-pushed the HM branch from d14561c to eda2ffd Compare November 10, 2016 17:40

endJunction merged commit 73e300f into ufz:master Nov 10, 2016

endJunction deleted the HM branch November 10, 2016 23:24

endJunction removed the please rebase label Nov 10, 2016


		storage_p.noalias() += N_p.transpose() * S * N_p * w;

		local_rhs.template block<pressure_size, 1>(pressure_index, 0)

HM process. #1508

HM process. #1508

Conversation

endJunction commented Nov 1, 2016 • edited Loading

ogsbot commented Nov 4, 2016

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

wenqing Nov 7, 2016 • edited by endJunction Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ogsbot commented Nov 8, 2016

ogsbot commented Nov 8, 2016

endJunction commented Nov 8, 2016

norihiro-w left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ogsbot commented Jun 19, 2020

endJunction commented Nov 1, 2016 •

edited

Loading

wenqing Nov 7, 2016 •

edited by endJunction

Loading