morphogen.homogenization.nonlinear

Nonlinear homogenization along loading paths: the periodic element.

The unknowns of a step are the fluctuation w of the displacement u = E x + w (periodic, or zero on the boundary for kinematic conditions) and the components of the macroscopic strain E that are not prescribed: the macroscopic degrees of freedom of the periodic element of Z-set. Their equations are

  • equilibrium, R_w = int B^T sigma dV = 0 on the reduced unknowns of w;

  • the macroscopic stress, R_E = int sigma dV - V Sigma = 0 on the components of E prescribed as stresses (V: the volume of the box, voids included). The reactions of the other components of E are V Sigma.

The stresses come from the behaviours of the phases (von Mises elasto-viscoplasticity, morphogen.homogenization.behaviours; crystal plasticity, morphogen.homogenization.crystal), integrated implicitly over each step. Newton’s method solves the equations with the consistent tangent:

[ K_ww K_wE ] [dw] [R_w] [ K_Ew K_EE ] [dE] = - [R_E], K_ww = int B^T C B, K_wE = int B^T C, K_EE = int C

whose macroscopic block is condensed: K_ww Y = K_wE (one solve per unknown component of E, from the solutions of the previous iteration) gives the 6 x 6 (at most) system (K_EE - K_Ew Y) dE = -R_E + K_Ew K_ww^-1 R_w. The linear solvers are those of the linear problems (sparse LU, or conjugate gradients with algebraic multigrid, whose hierarchy is built once per step). The consistent tangents of the points are symmetrized for them when the laws of all the phases have symmetric ones; otherwise (tangent, TANGENTS) they are kept whole, K_Ew is no longer K_wE^T and the tangent is solved by LU or by GMRES (with the multigrid of its symmetric part), the Krylov solvers symmetrizing them again at the linearizations where their measured asymmetry is below ASYMMETRY. A step that does not converge, whose local integration fails, or whose corrections by GMRES stagnate at _STALLED Newton iterations in a row, is divided by two (at most max_cuts times), and the step grows back after two converged steps.

In finite strains (kinematics="finite", morphogen.homogenization.finite) the equations are written on the reference configuration with the first Piola-Kirchhoff stress P and its tangent dP/dF; the macroscopic gradient F = I + E is upper triangular, its components E (nominal strains and simple shears) conjugate to those of the mean of P (nominal stresses). The outputs add the Cauchy stress and the logarithmic strain.

For linear behaviours in small strains the solutions are those of the condensation of morphogen.homogenization.core.superpose().

The time stepping and the Newton iterations (_PathSolver) are shared with the FFT solver of voxel images (morphogen.homogenization.fft_nonlinear), which gives its own discretization and linear solver.

morphogen.homogenization.nonlinear.KINEMATICS = ('small', 'finite')

Kinematics of the nonlinear problems

morphogen.homogenization.nonlinear.TANGENTS = ('auto', 'symmetric', 'full')

Consistent tangents of the points in the Newton iterations of the volume elements: symmetrized and solved by conjugate gradients, or full and solved by GMRES (sparse LU by finite elements); auto symmetrizes them when the laws of all the phases have symmetric tangents, and otherwise, for the Krylov solvers, where their measured asymmetry is below ASYMMETRY

morphogen.homogenization.nonlinear.ASYMMETRY = 0.02

With tangent="auto" and laws whose tangents are not all symmetric, the volume elements solved by Krylov methods (voxels; finite elements with multigrid) measure the asymmetry of the consistent tangents of the points of each phase at each linearization (asymmetry()) and symmetrize them where the largest is below this value (conjugate gradients, MINRES); above it they solve the full tangents by GMRES

morphogen.homogenization.nonlinear.kinematics_of(value) → bool[source]

Whether the kinematics "small" or "finite" is finite.

class morphogen.homogenization.nonlinear.Behaviours(kind: ndarray, pmat: ndarray, data: tuple, width: int, crystals: list = <factory>, mfront: list = <factory>, symmetric: bool = True)[source]

Bases: object

Behaviours of the elements of phases: the kind of each element (elastic, von Mises, crystal, MFront), its row of parameters (pmat: of params, of the crystal arrays or of mfront; -1 for elastic elements), the data of the integration kernels (see morphogen.homogenization.behaviours.update_points()), the number of internal variables of a point, the crystal plasticity of each crystal row and the laws of the points integrated outside the kernels (mfront: the MFront behaviours loaded by MGIS, MFrontLaw, and the learned laws, LearnedLaw). symmetric: whether the consistent tangents of all the laws are symmetric (see behaviour_table()).

property nonlinear: bool
morphogen.homogenization.nonlinear.behaviour_table(phase, materials: dict[int, Material], table: MaterialTable, *, finite: bool = False) → Behaviours[source]

Behaviours of the elements of phases phase with their materials table (morphogen.homogenization.materials.material_table()): the slip systems of the crystals turn with the orientation of their material.

The consistent tangents are symmetric (Behaviours.symmetric) for elasticity, for von Mises and Hill plasticity without Armstrong-Frederick terms (D = 0), for crystal plasticity in small strains without isotropic hardening (Q b = 0), for the learned potentials and standard models and for the MFront behaviours that declare it; not for the Armstrong-Frederick hardening (the backstresses turn the direction of flow), the isotropic hardening of the slip systems and the multiplicative crystal plasticity of finite strains, porous plasticity (the growth of the porosity), the learned perceptrons and recurrent networks, and the other MFront behaviours.

morphogen.homogenization.nonlinear.external_points(laws, kind, pmat, point_mat, rotations, finite: bool, *, symmetric: bool = True)[source]

The points of the behaviours integrated outside the kernels (of the kind MFRONT, of the laws pmat): those of the MFront behaviours by MGIS (MFrontPoints), those of the learned laws by their models (LearnedPoints), together, their tangents symmetrized with symmetric (for the conjugate gradients); None without them.

morphogen.homogenization.nonlinear.asymmetry(tangent, wts, groups=None) → float[source]

The asymmetry of the consistent tangents C_p of the points (points x nc x nc, weights wts) that decides tangent="auto" (ASYMMETRY): |C - C^T| / |C| for the block-diagonal operator C of the points, sqrt(sum_p w_p |C_p - C_p^T|^2 / sum_p w_p |C_p|^2) (Frobenius norms), the stiffer points weighing more; with groups (the phase of each point), the largest of those of the phases, so that a small phase whose tangents are far from symmetric is not diluted by the others (a learned law at 0.6 % of the voxels: 1.3e-2 for the whole volume element, 0.2 for its phase). Point by point, |C_p - C_p^T| / |C_p| reaches 0.1 to 0.2 at a few points of morphoGen’s own laws (porous points near their softening, crystals whose slip systems harden each other strongly) whose tangents are soft, where the operator stays nearly symmetric.

class morphogen.homogenization.nonlinear.StepState(time: float, strain: ndarray, q: ndarray, unknown: ndarray, disc: object, grads: ndarray, wts: ndarray, sigma: ndarray, tangent: ndarray)[source]

Bases: object

A converged step of a loading path, given to the monitor of _PathSolver.run(): its time, the macroscopic strain (6), the unknowns of the fluctuation, the indices of the macroscopic components solved for, and the discretization with the arrays of its integration points (measures of strain, weights, stresses and consistent tangents at the end of the step).

morphogen.homogenization.nonlinear.simulate_point(material: Material, path, *, name: str = 'point', orientation=None, kinematics: str = 'small', tol: float = 1e-10, max_newton: int = 30, max_cuts: int = 12, tangent: str = 'auto') → PathResult[source]

Response of one material point (homogeneous strain and stress) to a loading path: the components prescribed as stresses are solved for by Newton’s method with the consistent tangent (a material point simulator, as Zsim). orientation: Bunge angles (degrees) of a crystal; kinematics: "small" or "finite" strains (the strain then holds the components of the upper triangular displacement gradient, the stress those of the first Piola-Kirchhoff stress). In finite strains, and where its twins turn it, the result has the orientation of the lattice of a crystal at each step: its Bunge angles g R_e^T, with the rotation R_e of the polar decomposition of F_e (the material lines along the first axis stay along it, and the planes normal to the third stay normal to it: the grips of a tension along the first axis, the platens of a compression along the third). tangent (TANGENTS): Newton’s method solves a dense system, with by default ("auto") the exact tangents of the laws that are not symmetric (those of the learned laws and of the MFront behaviours too) and the symmetrized ones of the others (which only lose their round-off); "symmetric" symmetrizes them all, "full" keeps them all exact.

class morphogen.homogenization.nonlinear.Extended(nodes: ndarray, node=None, offset=None)[source]

Bases: object

The nodes of a cell and the images of some of them across its faces: node the node of each extended node (the nodes of the mesh first), offset its translation (zero for the nodes themselves) and reference its position; fold, the dofs of the extended nodes from those of the nodes. The image of a node of displacement u translated by L is displaced by u + H L, H the macroscopic displacement gradient: the cell deforms with it.

positions(u, macro) → ndarray[source]

The positions of the extended nodes for the displacements u of the nodes and the macroscopic displacement gradient macro.

class morphogen.homogenization.nonlinear.NonlinearFEProblem(mesh, materials: dict[int, Material], bc: str = 'periodic', *, solver: str = 'auto', tol: float = 1e-08, maxiter: int | None = None, newton_tol: float = 1e-06, max_newton: int = 25, max_cuts: int = 10, kinematics: str = 'small', tangent: str = 'auto', integration: str = 'full')[source]

Bases: _PathSolver, FEElasticProblem

Nonlinear behaviours of the phases (von Mises elasto-(visco)plasticity, crystal plasticity) on a tetrahedral mesh, loaded along paths: periodic conditions (the periodic element, whose macroscopic strain components are degrees of freedom) or kinematic ones (kubc: u = E x on the boundary, E likewise), in small or finite strains (kinematics).

newton_tol: relative residual of convergence of the equilibrium (to the norm of the nodal forces before they cancel) and of the macroscopic stress (to the integral of the norm of the stress); max_newton iterations per step; max_cuts divisions of a step by two. tangent: the consistent tangents of the points symmetrized ("symmetric": conjugate gradients with multigrid), full ("full": GMRES with the multigrid of their symmetric part; LU either way for small systems) or, by default ("auto"), symmetrized when the laws of all the phases have symmetric tangents (behaviour_table()), and otherwise kept full by LU and, with multigrid, symmetrized at each linearization where their asymmetry is below ASYMMETRY.

integration: "full", the behaviours integrated with the strain of each point, or "bbar", with the volume change of each 10-node tetrahedron as its mean (B-bar; F-bar in finite strains: morphogen.homogenization.dilatation), for nearly incompressible phases — elastomers — and isochoric plastic flow.