Homec4science
Diffusion µSpectre (feat/mpi) Browse

muSpectre/doc/dev-docs/source/NewMaterial.rst
57e889b22b09feat/mpi

Last Change
Raw File

NewMaterial.rst
View Options

Writing a New Constitutive Law
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The abstraction for a constitutive law in **µSpectre** is the ``Material``, and new such materials must inherit from the class :cpp:class:`muSpectre::MaterialBase`. Most often, however, it will be most convenient to inherit from the derived class :cpp:class:`muSpectre::MaterialMuSpectre`, as it implements a lot of the machinery that is commonly used by constitutive laws. This section describes how to implement a new constitutive law with internal variables (sometimes also called state variables). The example material implemented here is :cpp:class:`MaterialTutorial`, an objective linear elastic law with a distribution of eigenstrains as internal variables. The constitutive law is defined by the relationship between material (or second Piola-Kirchhoff) stress :math:`\mathbf{S}` and Green-Lagrange strain :math:`\mathbf{E}`
.. math::
:nowrap:
\begin{align}
\mathbf{S} &= \mathbb C:\left(\mathbf{E}-\mathbf{e}\right), \\
S_{ij} &= C_{ijkl}\left(E_{kl}-e_{kl}\right), \\
\end{align}
where :math:`\mathbb C` is the elastic stiffness tensor and :math:`\mathbf e` is the local eigenstrain. Note that the implementation sketched out here is the most convenient to quickly get started with using **µSpectre**, but not the most efficient one. For a most efficient implementation, refer to the implementation of :cpp:class:`muSpectre::MaterialLinearElastic2`.
The :cpp:class:`muSpectre::MaterialMuSpectre` class
***************************************************
The class :cpp:class:`muSpectre::MaterialMuSpectre` is defined in ``material_muSpectre_base.hh`` and takes three template parameters;
#. ``class Material`` is a `CRTP <https://en.wikipedia.org/wiki/Curiously_recurring_template_pattern>`_ and names the material inheriting from it. The reason for this construction is that we want to avoid virtual method calls from :cpp:class:`muSpectre::MaterialMuSpectre` to its derived classes. Rather, we want :cpp:class:`muSpectre::MaterialMuSpectre` to be able to call methods of the inheriting class directly without runtime overhead.
#. ``Dim_t DimS`` defines the number of spatial dimensions of the problem, i.e., whether we are dealing with a two- or three-dimensional grid of pixels/voxels.
#. ``Dim_t DimM`` defines the number of dimensions of our material description. This value will typically be the same as ``DimS``, but in cases like generalised plane strain, we can for instance have a three three-dimensional material response in a two-dimensional pixel grid.
The main job of :cpp:class:`muSpectre::MaterialMuSpectre` is to
#. loop over all pixels to which this material has been assigned, transform the global gradient :math:`\mathbf{F}` (or small strain tensor :math:`\boldsymbol\varepsilon`) into the new material's required strain measure (e.g., the Green-Lagrange strain tensor :math:`\mathbf{E}`),
#. for each pixel evaluate the constitutive law by calling its ``evaluate_stress`` (computes the stress response) or ``evaluate_stress_tangent`` (computes both stress and consistent tangent) method with the local strain and internal variables, and finally
#. transform the stress (and possibly tangent) response from the material's stress measure into first Piola-Kirchhoff stress :math:`\mathbf{P}` (or Cauchy stress :math:`\boldsymbol\sigma` in small strain).
In order to use these facilities, the new material needs to inherit from :cpp:class:`muSpectre::MaterialMuSpectre` (where we calculation of the response) and specialise the type :cpp:class:`muSpectre::MaterialMuSpectre_traits` (where we tell :cpp:class:`muSpectre::MaterialMuSpectre` how to use the new material). These two steps are described here for our example material.
Specialising the :cpp:class:`muSpectre::MaterialMuSpectre_traits` structure
***************************************************************************
This structure is templated by the new material (in this case :cpp:class:`MaterialTutorial`) and needs to specify
#. the types used to communicate per-pixel strains, stresses and stiffness tensors to the material (i.e., whether you want to get maps to `Eigen` matrices or raw pointers, or ...). Here we will use the convenient :cpp:type:`muSpectre::MatrixFieldMap` for strains and stresses, and :cpp:type:`muSpectre::T4MatrixFieldMap` for the stiffness. Look through the classes deriving from :cpp:type:`muSpectre::FieldMap` for all available options.
#. the strain measure that is expected (e.g., gradient, Green-Lagrange strain, left Cauchy-Green strain, etc.). Here we will use Green-Lagrange strain. The options are defined by the enum :cpp:enum:`muSpectre::StrainMeasure`.
#. the stress measure that is computed by the law (e.g., Cauchy, first Piola-Kirchhoff, etc,). Here, it will be first Piola-Kirchhoff stress. The available options are defined by the enum :cpp:enum:`muSpectre::StressMeasure`.
Our traits look like this (assuming we are in the namespace ``muSpectre``::
template <Dim_t DimS, Dim_t DimM>
struct MaterialMuSpectre_traits<MaterialTutorial<DimS, DimM>>
{
//! global field collection
using GFieldCollection_t = typename
GlobalFieldCollection<DimS, DimM>;
//! expected map type for strain fields
using StrainMap_t = MatrixFieldMap<GFieldCollection_t, Real, DimM, DimM, true>;
//! expected map type for stress fields
using StressMap_t = MatrixFieldMap<GFieldCollection_t, Real, DimM, DimM>;
//! expected map type for tangent stiffness fields
using TangentMap_t = T4MatrixFieldMap<GFieldCollection_t, Real, DimM>;
//! declare what type of strain measure your law takes as input
constexpr static auto strain_measure{StrainMeasure::GreenLagrange};
//! declare what type of stress measure your law yields as output
constexpr static auto stress_measure{StressMeasure::PK2};
//! local field_collections used for internals
using LFieldColl_t = LocalFieldCollection<DimS, DimM>;
//! local strain type
using LStrainMap_t = MatrixFieldMap<LFieldColl_t, Real, DimM, DimM, true>;
//! elasticity with eigenstrain
using InternalVariables = std::tuple<LStrainMap_t>;
};
Implementing the new material
*****************************
The new law needs to implement the methods ``add_pixel``, ``get_internals``, ``evaluate_stress``, and ``evaluate_stress_tangent``. Below is a commented example header::
template <Dim_t DimS, Dim_t DimM>
class MaterialTutorial:
public MaterialMuSpectre<MaterialTutorial<DimS, DimM>, DimS, DimM>
{
public:
//! traits of this material
using traits = MaterialMuSpectre_traits<MaterialTutorial>;
//! Type of container used for storing eigenstrain
using InternalVariables = typename traits::InternalVariables;
//! Construct by name, Young's modulus and Poisson's ratio
MaterialTutorial(std::string name, Real young, Real poisson);
/**
* evaluates second Piola-Kirchhoff stress given the Green-Lagrange
* strain (or Cauchy stress if called with a small strain tensor)
*/
template <class s_t, class eigen_s_t>
inline decltype(auto) evaluate_stress(s_t && E, eigen_s_t && E_eig);
/**
* evaluates both second Piola-Kirchhoff stress and stiffness given
* the Green-Lagrange strain (or Cauchy stress and stiffness if
* called with a small strain tensor)
*/
template <class s_t, class eigen_s_t>
inline decltype(auto)
evaluate_stress_tangent(s_t && E, eigen_s_t && E_eig);
/**
* return the internals tuple (needed by `muSpectre::MaterialMuSpectre`)
*/
InternalVariables & get_internals() {
return this->internal_variables;};
/**
* overload add_pixel to write into eigenstrain
*/
void add_pixel(const Ccoord_t<DimS> & pixel,
const Eigen::Matrix<Real, DimM, DimM> & E_eig);
protected:
//! stiffness tensor
T4Mat<Real, DimM> C;
//! storage for eigenstrain
using Field_t =
TensorField<LocalFieldCollection<DimS,DimM>, Real, secondOrder, DimM>;
Field_t & eigen_field; //!< field of eigenstrains
//! tuple for iterable eigen_field
InternalVariables internal_variables;
private:
};
A possible implementation for the constructor would be::
template <Dim_t DimS, Dim_t DimM>
MaterialTutorial<DimS, DimM>::MaterialTutorial(std::string name,
Real young,
Real poisson)
:MaterialMuSpectre<MaterialTutorial, DimS, DimM>(name) {
// Lamé parameters
Real lambda{young*poisson/((1+poisson)*(1-2*poisson))};
Real mu{young/(2*(1+poisson))};
// Kronecker delta
Eigen::Matrix<Real, DimM, DimM> del{Eigen::Matrix<Real, DimM, DimM>::Identity()};
// fill the stiffness tensor
this->C.setZero();
for (Dim_t i = 0; i < DimM; ++i) {
for (Dim_t j = 0; j < DimM; ++j) {
for (Dim_t k = 0; k < DimM; ++k) {
for (Dim_t l = 0; l < DimM; ++l) {
get(this->C, i, j, k, l) += (lambda * del(i,j)*del(k,l) +
mu * (del(i,k)*del(j,l) + del(i,l)*del(j,k)));
}
}
}
}
}
as an exercise, you could check how :cpp:class:`muSpectre::MaterialLinearElastic1` uses **µSpectre**'s materials toolbox (in namespace ``MatTB``) to compute :math:`\mathbb C` in a much more convenient fashion. The evaluation of the stress could be (here, we make use of the ``Matrices`` namespace that defines common tensor algebra operations)::
template <Dim_t DimS, Dim_t DimM>
template <class s_t, class eigen_s_t>
decltype(auto)
MaterialTutorial<DimS, DimM>::
evaluate_stress(s_t && E, eigen_s_t && E_eig) {
return Matrices::tens_mult(this->C, E-E_eig);
}
The remaining two methods are straight-forward::
template <Dim_t DimS, Dim_t DimM>
template <class s_t, class eigen_s_t>
decltype(auto)
MaterialTutorial<DimS, DimM>::
evaluate_stress_tangent(s_t && E, eigen_s_t && E_eig) {
return return std::make_tuple
(evaluate_stress(E, E_eig),
this->C);
}
template <Dim_t DimS, Dim_t DimM>
InternalVariables &
MaterialTutorial<DimS, DimM>::get_internals() {
return this->internal_variables;
}
Note that the methods ``evaluate_stress`` and ``evaluate_stress_tangent`` need to be in the header, as both their input parameter types and output type depend on the compile-time context.
c4science · Help