Liquid flow process #1468

wenqing · 2016-10-06T14:39:58Z

Added classes for modelling liquid flow process based on the previous implementations about fluid and porous medium properties. The implementation is verified with the following
~~three~~ four tests from OGS5:

LineDirichletNeumannBC ~~sat1D~~
PressureBCatCornerOfAnisotropicSquare ~~sat2D~~
GravityDriven
AxiSymTheis (with axisymmetry)

In the current ogs6/master, the calculations of velocity and flux were implemented in wrong places for groundwater, heat conduction. A PR about the correction of such calculations for all non-deformation processes comes immediately once this one is merged.

wenqing · 2016-10-06T14:46:11Z

ProcessLib/LiquidFlow/LiquidFlowLocalAssembler-imp.h

+                    fac * sm.dNdx.transpose().col(GlobalDim - 1) * rho_g;
+            }
+
+            // Compute the velocity


The velocity calculation will be moved to a new member function in the next PR in order to perform it after the linearization.

wenqing · 2016-10-06T14:57:14Z

ProcessLib/LiquidFlow/LiquidFlowLocalAssembler-imp.h

+        local_b_data, local_matrix_size);
+
+    // For velocity calculation:
+    // TODO: move velocity calculation to a member function, and call the


Done in my branch for the next PR.

endJunction · 2016-10-06T15:16:45Z

Applications/CLI/Tests.cmake

@@ -520,6 +520,52 @@ if(NOT OGS_USE_MPI)
    #     EXECUTABLE_ARGS tes-inert-wedge.prj
    # )

+    # Liquid flow
+    AddTest(
+        NAME LiquidFlowProcess_sat1D


Is it possible to name the tests, s.t. its general setup is clear from the name.
Otherwise I've to look into the prj files every time to find out what that could be.
Something like LiquidFlow_1D_line100_saturationZone_quadraticPressureLeft?

Changed the name of two tests.

endJunction · 2016-10-06T15:19:15Z

ProcessLib/LiquidFlow/CMakeLists.txt

@@ -0,0 +1 @@
+


Why do we need these empty files?
APPEND_SOURCE_FILES(SOURCES LiquidFlow) in ProcessLib/CMakeLists.txt is sufficient. Are there other reasons?

endJunction · 2016-10-06T15:23:19Z

ProcessLib/LiquidFlow/LiquidFlowLocalAssembler-imp.h

+        const double mu = _material_properties.getViscosity(p, _temperature);
+
+        // Assemble Laplacian, K, and RHS by the gravitational term
+        if (perm.size() == 1)  // Save the computing time for isotropic permeability.


I'd prefer no code duplication here unless this saves more than 1% runtime.
Did you measure how much computation time is saved, I'm just curious about the trade off btw. a branch and additional multiplications/additions.

I think for 3D elements, such as hexehadra, it could save the computing time for a homogeneous medium. Image to compare the computing times of

Matrix with a size of (20, 3) multiplied with a matrix with a size of (3, 3)

Matrix with a size of (20, 3) multiplied with a scalar number.

To make this function concise, the computation of velocity, which will be in the next PR, is removed.

I cannot imagine any runtime numbers. But the code duplication is error-prone and more difficult to maintain. If there is no obvious difference in the runtime I recommend to use code w/o duplication.

endJunction · 2016-10-06T15:27:34Z

ProcessLib/LiquidFlow/LiquidFlowMaterialProperties.cpp

+}
+
+double LiquidFlowMaterialProperties::getMassCoefficient(
+    const double var4porosity, const double var4storage,


The 4 means fourth position or some other index?

Replaced it with a more meaningful name.

endJunction · 2016-10-06T15:29:07Z

ProcessLib/LiquidFlow/LiquidFlowMaterialProperties.h

+{
+    typedef MaterialLib::Fluid::FluidProperty::ArrayType ArrayType;
+
+    explicit LiquidFlowMaterialProperties(BaseLib::ConfigTree const& config);


Usually a create function is used for better testability of the class and to separate parsing from actual construction.

I've tried. To do that, not only this struct needs a copy constructor, but also classes of fluid properties and porous medium properties need their copy constructors in order to move the unique_ptr type instances of these classes between the creator and receiver. Besides, the copy of eigen matrix of permeability is a concrete copy. I also noticed that TESProcess has parsring inside its ctor too. Therefore, I would like to keep the parsing in the ctor.

endJunction · 2016-10-06T15:33:23Z

ProcessLib/LiquidFlow/LiquidFlowMaterialProperties.h

+     * \param material_group_id  Material ID of the element
+     */
+    double getMassCoefficient(const double p, const double T,
+                             const double var4porosity, const double var4storage,


Instead of interesting shortcuts, simply call the variables porosity_variable and storage_variable.

endJunction · 2016-10-06T15:37:41Z

ProcessLib/LiquidFlow/LiquidFlowProcess.h

+        NumLib::LocalToGlobalIndexMap const& dof_table,
+        MeshLib::Mesh const& mesh, unsigned const integration_order) override;
+
+    void assembleConcreteProcess(const double t, GlobalVector const& p,


Sometimes the argument of GlobalVector type is called p, sometimes x, and there is some assignment of p = x in postTimeStep. Better to choose one name and use it consistently.
I'd prefer the original x, because it can have more then pressure variable and is consistent with the caller.

endJunction · 2016-10-06T15:39:33Z

ProcessLib/LiquidFlow/createLiquidFlowProcess.cpp

+    auto const gravitational_term =
+    //! \ogs_file_param{process__LIQUID_FLOW__gravitational_term}
+    config.getConfigParameter<std::string>("gravitational_term");
+    const bool has_gravitational_term = (gravitational_term == "enabled") ? true : false;


Simply:

const bool has_gravitational_term = (gravitational_term == "enabled")

endJunction · 2016-10-06T15:41:14Z

ProcessLib/LiquidFlow/createLiquidFlowProcess.h

+    MeshLib::Mesh& mesh,
+    std::unique_ptr<ProcessLib::AbstractJacobianAssembler>&& jacobian_assembler,
+    std::vector<ProcessVariable> const& variables,
+    std::vector<std::unique_ptr<ParameterBase>> const& /*parameters*/,


Why is only parameters commented?

Removed the comment.

endJunction · 2016-10-06T15:41:59Z

Tests/ProcessLib/LiquidFlow/TestLiquidFlowMaterialProperties.cpp

+    const char xml[] =
+        "<material_property>"
+        "    <fluid>"
+        "        <fluid>"


Why is fluid repeated here?

Removed the repetition.

wenqing · 2016-10-06T16:07:41Z

@endJunction Thanks for the comments, and I will revise the source code late on. BTW, just captured the error in MaterialLibSolidsKelvinVector4.Inverse, and you can find it by clicking above building "Details".

[       OK ] MaterialLibSolidsKelvinVector4.SelfTestMappingKelvinToTensor (19 ms)

[ RUN      ] MaterialLibSolidsKelvinVector4.DeviatoricSphericalProjections

[       OK ] MaterialLibSolidsKelvinVector4.DeviatoricSphericalProjections (0 ms)

[ RUN      ] MaterialLibSolidsKelvinVector4.Inverse

/home/travis/build/ufz/ogs/ThirdParty/autocheck/include/autocheck/reporter.hpp:91: Failure

Value of: AUTOCHECK_SUCCESS

  Actual: false

Expected: true

Falsifiable, after 194 tests:

(  20.715

 88.8961

-17.4266

-60.6874)

[  FAILED  ] MaterialLibSolidsKelvinVector4.Inverse (5 ms)

[ RUN      ] MaterialLibSolidsKelvinVector4.Determinant

endJunction · 2016-10-06T17:27:06Z

Thanks for the inverse test failing pointer; I didn't see any for some time now. I'll relax the tolerance a little more.

ogsbot · 2016-10-07T13:47:02Z

Jenkins: OGS-6/Linux-PRs-dynamic failed: https://svn.ufz.de:8443/job/OGS-6/job/Linux-PRs-dynamic/1344/

wenqing · 2016-10-07T14:57:50Z

@endJunction ctests excluding those for this PR failed because that branch IntegrationOrder of the data repository is not merged to the master.

endJunction · 2016-10-10T16:21:04Z

ProcessLib/LiquidFlow/LiquidFlowLocalAssembler-imp.h

+ */
+
+#ifndef OGS_LIQUIDFLOWLOCALASSEMBLER_IMP_H
+#define OGS_LIQUIDFLOWLOCALASSEMBLER_IMP_H


Until now everybody named the implementation files ending with '-impl.h'; don't break that please.

endJunction · 2016-10-10T16:23:02Z

ProcessLib/LiquidFlow/LiquidFlowLocalAssembler-imp.h

+        _material_properties.intrinsic_permeabiliy[mat_id];
+
+    // Note: For Inclined 1D in 2D/3D or 2D element in 3D, the first item in
+    //  the assert must be changed to perm.rows() == _element->getDimension()


Can you do this please.

Do you mean Inclined 1D in 2D/3D or 2D element in 3D? I can do it, but it should be in done the general way in the integration of MathLib.

I can get time for that after finish TH and a project.

endJunction · 2016-10-10T16:31:19Z

ProcessLib/LiquidFlow/LiquidFlowLocalAssembler.h

+    LiquidFlowLocalAssembler(
+        MeshLib::Element const& element,
+        std::size_t const /*local_matrix_size*/,
+        bool is_axially_symmetric,


Can be const, as the other arguments are.

endJunction · 2016-10-10T16:32:46Z

ProcessLib/LiquidFlow/LiquidFlowLocalAssembler.h

+
+    const bool _compute_gravitational_term;
+    LiquidFlowMaterialProperties const& _material_properties;
+    double _temperature;  ///< Temperature.


no need to comment the obvious.

endJunction · 2016-10-10T16:34:38Z

Tests/ProcessLib/LiquidFlow/TestLiquidFlowMaterialProperties.cpp

+        "        </viscosity>"
+        "    </fluid>"
+        "    <porous_medium>"
+        "        <porous_medium>"


why the duplication?

It is for patch-wise heterogeneous porous medium properties. For example, a layered 3D domain or a domain with different material sub domains.

endJunction · 2016-10-10T16:37:18Z

ProcessLib/LiquidFlow/LiquidFlowMaterialProperties.h

+    {
+        ArrayType vars;
+        vars[0] = T;
+        vars[1] = p;


Just an understanding question. How do you know that T has index 0 and p has index 1 and not vice versa, like the arguments are?
There must be a better way to designate such implicit knowledge (maybe using enums).

Replaced the index with enum item.

endJunction · 2016-10-10T16:38:15Z

ProcessLib/LiquidFlow/LiquidFlowMaterialProperties.h

+        return viscosity->getValue(vars);
+    }
+
+    std::unique_ptr<MaterialLib::Fluid::FluidProperty> density_l;


what does the '_l' mean?

Renamed it with liquid_density. The struct may get solid density in future development.

endJunction · 2016-10-10T16:38:54Z

ProcessLib/LiquidFlow/LiquidFlowMaterialProperties.h

+    /// Porous medium properties of different material zones.
+    /// The vector is left empty if the property data are given in vtu file,
+    /// e.g for heterogeneous medium.
+    std::vector<Eigen::MatrixXd> intrinsic_permeabiliy;


typo in permeability.

endJunction · 2016-10-10T16:40:58Z

ProcessLib/LiquidFlow/LiquidFlowProcess.cpp

+    DBUG("New time step");
+
+    _p_previous_timestep =
+        MathLib::MatrixVectorTraits<GlobalVector>::newInstance(x);


endJunction · 2016-10-10T16:41:47Z

ProcessLib/LiquidFlow/LiquidFlowProcess.cpp

+void LiquidFlowProcess::preIterationConcreteProcess(const unsigned /*iter*/,
+                                                    GlobalVector const& /*x*/)
+{
+    // TODO


What is todo? If nothing concrete, please remove.

Removed this function for current step.

endJunction · 2016-10-10T16:42:55Z

ProcessLib/LiquidFlow/LiquidFlowProcess.h

+    std::vector<std::unique_ptr<LiquidFlowLocalAssemblerInterface>>
+        _local_assemblers;
+
+    std::unique_ptr<GlobalVector> _p_previous_timestep;


Where is this used?

endJunction · 2016-10-10T16:47:15Z

Tests/ProcessLib/LiquidFlow/TestLiquidFlowMaterialProperties.cpp

+        "            </porosity>"
+        "            <storage>"
+        "                <type>Constant</type>"
+        "                <value> 1.e-4 </value>"


Maybe I'm on a wrong train, but shouldn't be the properties used for this instead of reimplementation of Constant, MeshElement/NodeProperties, etc?
I ask it differently: Is it possible to have heterogeneous storage?

As mentioned above, currently I only consider the patch-wise heterogeneous porous medium properties, which occur in most applications of subsurface modelling. If element-wise heterogeneous porous medium property is used, we can easily added by using MeshElementProperties.

This Contant is just a type of storage/porosity models, which may be defined by some empirical formulas.

I suggest to use the properties that have been already merged to the master.

TomFischer · 2016-10-11T08:50:18Z

ProcessLib/CMakeLists.txt

@@ -14,6 +14,8 @@ APPEND_SOURCE_FILES(SOURCES Parameter)
 add_subdirectory(GroundwaterFlow)
 APPEND_SOURCE_FILES(SOURCES GroundwaterFlow)

+append_source_files(SOURCES LiquidFlow)


Why not APPEND_SOURCE_FILES to stay consistent within the file?

Since I think that cmake command should be in small letters. Now, it is changed accordingly.

TomFischer · 2016-10-11T11:06:36Z

ProcessLib/LiquidFlow/LiquidFlowLocalAssembler-imp.h

+        const double mu = _material_properties.getViscosity(p, _temperature);
+
+        // Assemble Laplacian, K, and RHS by the gravitational term
+        if (perm.size() ==


The line break is at a strange place.

It was formatted by clang-format. Now, the break is removed (by amending the last commit 964056f).

and for axisymmetical problems.

and made some other minor changes.

rebased the data repository.

…po in one variable name, and made some other minor improvements according to the comments by Dima and Tom.

and gravitational_term

… cases With this improvement, the if-condition for isotropic anisotropic is moved outside the integration loop.

…ed a file

Where g is the gravitational acceleration.

wenqing · 2016-10-26T11:57:42Z

Rebased.

endJunction · 2016-10-26T18:47:37Z

Thanks for repeating the test. The new results look more trustful (although around 40% are not accounted for and most difference is in output, which should be identical) . When you have time please add that test to the ctests "LARGE", so other can experiment with it.

@wenqing Could you please check where the ufz/ogs/Tests/Data points to. It seems to me that in the last commit on ufz/ogs/master (dd97ac9) you have set the submodule wrong.

wenqing · 2016-10-27T08:33:38Z

@endJunction The pointer to Data is corrected. I will commit the large tests this afternoon.

wenqing · 2016-10-27T08:40:07Z

@endJunction push -f is not allowed, I will add a new commit for this.

wenqing · 2016-10-27T08:44:31Z

Done with git amend

ogsbot · 2020-06-19T12:35:58Z

OpenGeoSys development has been moved to GitLab.

See this pull request on GitLab.

wenqing commented Oct 6, 2016

View reviewed changes

wenqing changed the title ~~Liquid process~~ Liquid flow process Oct 6, 2016

wenqing commented Oct 6, 2016

View reviewed changes

wenqing force-pushed the liquid_pcs branch from a02cb3b to 6e147c7 Compare October 6, 2016 15:01

endJunction reviewed Oct 6, 2016

View reviewed changes

wenqing force-pushed the liquid_pcs branch from 9d20101 to 4d81e39 Compare October 7, 2016 13:44

wenqing added the please review label Oct 7, 2016

wenqing force-pushed the liquid_pcs branch 3 times, most recently from cc148c1 to b8f08f7 Compare October 10, 2016 14:05

endJunction reviewed Oct 10, 2016

View reviewed changes

TomFischer reviewed Oct 11, 2016

View reviewed changes

wenqing force-pushed the liquid_pcs branch from 5941dd3 to 542b006 Compare October 11, 2016 10:19

TomFischer reviewed Oct 11, 2016

View reviewed changes

wenqing force-pushed the liquid_pcs branch from 542b006 to 964056f Compare October 11, 2016 11:41

wenqing added 20 commits October 26, 2016 13:56

[LF] Added a flag for gravitational term

3630025

[LF] Removed material IDs

fdc2c35

[LF] Enabled the option for computing the gravitational term

3e7420c

and for axisymmetical problems.

[TEST] Added two tests for liquid flow

2b83cef

[Data] Updated input data for liquid flow process

531a9ca

[LF] Add a prefix to the safe include guard

28478ea

[TEST] Added an axisymmetric test about the Theis' pumping problem

2d3d0b2

[Test] Renamed some tests for liquid flow process

8fdcc5d

[LF] Changed some variable names to meaningful ones

bcd4a60

and made some other minor changes.

[LF] Updated for the newly added integration_order

e043990

[Test] Added an assert to the dimension of permeability tensor and

6f3998d

rebased the data repository.

[LF] Changed the names of one file and some variables, corrected a ty…

5c8f13a

…po in one variable name, and made some other minor improvements according to the comments by Dima and Tom.

[Test] Correct a test name

19adccb

[Mat] Alternative implementation of porous medium properties.

c9afadc

[LF] Redesigned material property class

24bb0e2

[LF] Added attributions for xml tags of porous_medium

fe54078

and gravitational_term

[LF] Introduced template members to distinguish isotropic anisotropic…

e2b69b5

… cases With this improvement, the if-condition for isotropic anisotropic is moved outside the integration loop.

[LF] Fixed a memory leakage, changed the gravity term input and renam…

2a50015

…ed a file

[LF] Improved the way to input the gravity term of the Darcy velocity

e0a951b

[LF] Added const to a variable and corrected the sign of g.

27a104d

Where g is the gravitational acceleration.

wenqing force-pushed the liquid_pcs branch from 0b9077f to 27a104d Compare October 26, 2016 11:57

wenqing merged commit 52a4e44 into ufz:master Oct 26, 2016

wenqing mentioned this pull request Oct 26, 2016

Added computeSecondaryVariable to class Process #1501

Merged

endJunction removed the please rebase label Oct 26, 2016

Liquid flow process #1468

Liquid flow process #1468

Conversation

wenqing commented Oct 6, 2016 • edited 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

wenqing Oct 7, 2016 • edited 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

endJunction Oct 6, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

endJunction Oct 6, 2016 • edited 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

wenqing Oct 7, 2016 • edited Loading

Choose a reason for hiding this comment

wenqing commented Oct 6, 2016

endJunction commented Oct 6, 2016

ogsbot commented Oct 7, 2016

wenqing commented Oct 7, 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

endJunction Oct 10, 2016 • edited 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

endJunction Oct 10, 2016 • edited 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

TomFischer Oct 11, 2016 • edited by wenqing Loading

Choose a reason for hiding this comment

wenqing Oct 11, 2016 • edited Loading

Choose a reason for hiding this comment

wenqing commented Oct 26, 2016

endJunction commented Oct 26, 2016 • edited Loading

wenqing commented Oct 27, 2016

wenqing commented Oct 27, 2016

wenqing commented Oct 27, 2016

ogsbot commented Jun 19, 2020

wenqing commented Oct 6, 2016 •

edited

Loading

wenqing Oct 7, 2016 •

edited

Loading

endJunction Oct 6, 2016 •

edited

Loading

endJunction Oct 6, 2016 •

edited

Loading

wenqing Oct 7, 2016 •

edited

Loading

endJunction Oct 10, 2016 •

edited

Loading

endJunction Oct 10, 2016 •

edited

Loading

TomFischer Oct 11, 2016 •

edited by wenqing

Loading

wenqing Oct 11, 2016 •

edited

Loading

endJunction commented Oct 26, 2016 •

edited

Loading