Repository navigation
Conversation
load_driver runs one material point under mixed strain/stress control (uniaxial stress solves the five lateral strains with the material tangent) and records the Newton residuals; material_report runs the load cases of a case.json and writes CSV; reports/make_report.py turns them into figures and the results section of docs/materials/<material>/README.md.
The correction solves (P:C:P + Isym - P + Iskw) : d_eps = -P:(sigma - sigma*) with tmech::invf instead of a hand-written 6x6 elimination; Iskw makes the operator regular on the full 9x9 space. Strain, stress and tangent are tensors, converted to and from the evaluator buffers with tensor_conversion.
A report document holds the model graph, the material point, recorded
columns wired to any graph property ("material::property[i,j]"), the load
cases, the figures and the checks. Includes merge shared cases and figures;
"$name" strings take parameters. material_report writes resolved.json,
graph.json (materials, parameters, produced and consumed properties) and
CSV per case; make_report.py evaluates the checks against closed forms,
draws the figures and writes the results with the wiring diagram. Every
report is a ctest (checks only).
C++: load_control (stress/strain projectors), load_path (steps as a view), report_json (parsing into value types), case_runner (material_point and run_case over a step_sink concept), csv_sink, graph_description; the tool is now an entry point. load_driver keeps no per-step state: a step_result carries its start state, so tangent_error works for any earlier step. Produced properties are listed by name (the registry orders by address). Python: dataclasses for runs, cases and check results; expressions, checks, panel renderers (registry by kind) and Markdown sections as separate functions; unit tests in reports/test_make_report.py (ctest report_script).
The Markdown pages and committed figures are removed. make_report.py --out <dir> writes <material>.html with embedded figures and an index; nothing generated is committed. Each report starts with the parameters of report.json, where each one is used (material, case or check), the derived values and the material parameters as resolved. resolve_report records the substitutions in "parameter_usage".
- core/tensor_identities.h: sym/skew identities, volumetric and deviatoric projectors, the symmetric 3D basis in Abaqus order; used by linear_elasticity, saint_venant_kirchhoff and plasticity_utils (make_IIdev / cached_IIdev now delegate; make_isotropic_tangent keeps its K I(x)I form). - core/property_value.h: value of a plain or history property; used by material_point_evaluator::resolve_property and the report columns. - core/graph_description.h: materials with produced (history flag, sorted) and consumed properties; the report adds types and parameters. - umat/load_control.h and umat/material_point_driver.h (was report/load_driver.h): mixed strain/stress control of a material point.
make_report.py is the entry point of reports/numsim_report. Expressions own the document; Raw marks unescaped cells; ReportSummary replaces the index tuples; axis labels show tensor components as symbols. Tests per module in reports/tests (ctest report_script discovers them).
parse_point and parse_columns live in report_json.h; step_record and the step_sink concept in step_record.h, so csv_sink no longer includes the model builder. load_path::at checks its range, include nesting is capped at 16 levels, and the finite-difference step of tangent_error is a driver option (fd_step).
A plain output "iterations", so reports can record the return-map convergence of the material next to the driver's own Newton.
A check with "plot" draws value and expected per case next to its error against the tolerance (relative for relative checks); the separate closed-form figures are gone. Cases take tangent_every (default 1, 0 = never); the standard cases check every fifth step. The J2 reports record solver::iterations and show the return map next to the driver's Newton.
…n_property)
Every small-strain consumer took the strain from the fixed property "strain"
of its source. The composites that follow produce one strain per phase
("strain_matrix", "strain_<phase>"), and the phase materials have to read
those. strain_property (default "strain") names the property next to the
existing strain_source / strain_producer_name, the same way tangent_property
and nu_property do; nothing changes for existing documents.
linear_elasticity, linear_stress, j2_plasticity, drucker_prager_plasticity and
the three state functions; tests wire a material to another's stress under
the new name and check the default and a missing property.
"studies" rerun a case with several step counts; the error of a final quantity against the finest run is fitted with a power of the step size, and errors below ten times the driver tolerance count as exact. Cases take "points" (strain states in order) for non-proportional paths; the load step carries its segment and fraction. The J2 reports show the radial return exact on a proportional path and first order on tension followed by shear.
core/elliptic_integrals.h: R_F, R_C, R_D, R_J by Carlson's duplication algorithm and the complete integrals K, E, so the Eshelby tensors of ellipsoids and the orientation averaging need no Boost. core/elastic_constants.h: K, G, E, nu, lambda conversions, the isotropic stiffness, rotation of rank-2 and rank-4 tensors, rotations from axis and angle and onto a direction.
umat/deformation_gradient_driver.h drives a material point by F under mixed F / P control, Newton on the full nine components with dP/dF; deformation_control names them (21, 31, 32 included). A report whose point has a deformation_gradient_source records F and P with nine components, the Cauchy stress as stress_11..23 and J; tensors in cases may have nine components. Reports for neo_hooke and saint_venant_kirchhoff with exact checks (S = F^-1 P under uniaxial stress, simple shear, volumetric).
core/eshelby.h: sphere in closed form from the projectors; elliptic and circular cylinders along any axis or direction (Mura 11.21); ellipsoids with any semi-axes from Mura's I-integrals through Carlson's R_D, the degenerate pairs (spheroid, sphere) from the sum rules, an infinite axis dispatching to the cylinder. Tests against the closed forms, the limits, uvwmaterial's spheroid formula and the identity S_iikk = (1 + nu)/(1 - nu).
…ses, projectors of the controls, load path accessors
…n tensor, float-safe rotation_to The elliptic integrals come from Boost.Math where they are first used (the Eshelby tensors). rotation() builds [axis]_x as -(epsilon . axis); rotation_to() uses 8 eps of the scalar type instead of 1e-14, which did not compile for float and would have been c > 1 there. [[nodiscard]] on the pure helpers.
…per shape eshelby_sphere, eshelby_cylinder, eshelby_spheroid and eshelby_ellipsoid consume nu_source / nu_property and publish the output "tensor"; the engine orders and re-evaluates them with nu. Two scalar producers feed them: poisson_ratio_from_moduli (K_source, G_source -> value) and isotropic_moduli (tangent_source -> K, G, poisson_ratio through isotropize, rejecting an anisotropic tangent). Composites hold a node through eshelby_tensor_interface (tensor() at the wired nu, tensor(nu) for the self-consistent iterate). core/eshelby.h keeps the closed forms: the integrals come from Boost.Math (ellint_rd), the tensors are assembled on the eigenprojectors, equal axes are detected at sqrt(eps).
core/isotropize.h: K = Pvol::C / Dim, G = Pdev::C / (2 (Dim(Dim+1)/2 - 1)), and the isotropic part built from them. core/lebedev.h with generated tables (tools/extract_lebedev.py from uvwmaterial's rule, precisions 3 to 23, up to 194 points): spherical means and integrals of tensor-valued functions of the direction. eshelby::depolarisation for ellipsoids via Carlson's R_J, the elliptic cylinder in closed form, and the cylinder along a direction.
Phases are materials providing a tangent, with volume fractions; outputs the effective tangent, the stress and the strain concentration tensors (sum c_i A_i = Isym). Report with the closed forms of the three kernels and their ordering. Report generator: derived values resolve in dependency order, independent of the key order.
…ent producer over phases The composite consumes the phases' produced tangents, fractions and Eshelby tensors and produces the effective tangent and the concentrations; the stress is a linear_stress on the tangent. It reaches into no other node: no material_ref, no eshelby(nu), no isotropisation of the matrix. Self-consistent is a graph cycle closed by one local edge: isotropic_moduli on the composite tangent produces nu of the effective medium, the Eshelby nodes read it with nu_edge = local (from nu_initial on the first pass), and one update() is one relaxed Picard step with "residual" as its error. The matrix is a phase with its own Eshelby node, so its shape is no longer a hard-coded sphere. material_point_evaluator closes the fixed point per host call (residual_property, residual_tolerance, max_passes; the report point names them). isotropic_moduli with a non-positive tolerance projects without the isotropy check, which the effective medium needs.
…ution Spheroidal fibres in an isotropic matrix with the second-order orientation tensor: Mori-Tanaka with the orientation-averaged dilute concentration, or the averaged unidirectional stiffness. Eshelby tensors per direction are cached for the matrix Poisson ratio. Report with aligned, injection-moulded and random states against the unidirectional composite and isotropy.
… phase, passes to the fixed point A phase whose tangent depends on its own strain closes a cycle through mean_field_strain: the concentrations need the algorithmic tangent, the tangent needs the local strain. The phase node reads that tangent with tangent_edge = local (no ordering, the value of the previous pass; the initial_tangent_source on the first pass), so one update() is one pass and mean_field_stress::residual says when the passes agree. The evaluator closes the fixed point per host call through residual_property. Tests with an isotropic-damage matrix and stiff spheres: below the threshold the chain equals the linear composite; after damage the local strains partition the composite strain, the state function saw the matrix strain, the stress is the phase average, the tangent is the concentration-weighted sum and matches central differences on a vanishing step; the residual decreases over the passes of a large step; stress control through the driver converges; the Reuss kernel ends with equal stresses in both phases. The report mean_field_damage runs the standard cases with the point closing the fixed point; the card converter maps a rule-of-mixture card with a nonlinear phase to the pair for every kernel.
core/log_strain.h: ln V and d ln V / dF from an eigendecomposition of F F^T (Eigen) with the divided-difference derivative, continuous through repeated eigenvalues; tmech's log of V gives wrong values at eigenvalue splits of about 1e-7 (petlenz/tmech#37). hencky_elasticity: tau = C : (ln V - eps_eig), P = tau F^-T, consistent dP/dF, optional eigenstrain. Report with the exact uniaxial and volumetric closed forms; reports take a per-case fd_step.
docs/uvwmaterial-mapping.md: every class of uvwmaterial with its numsim-materials counterpart, parameters and status; the card formats. tools/convert_uvwmaterial_card.py turns the nested cards of the supported types into json_model documents, naming the issue of the port for the others; tested on the example cards, and the converted Mori-Tanaka card loads and gives the analytic moduli.
…se with a hardening J2 matrix mean_field_damage records the intact Mori-Tanaka next to the local pair (composite and phase stresses, damage, local strains under strain and stress control; on first loading the damaged composite is softer, after a reversal the incremental localisation leaves a residual stress, noted). The new case mean_field_reuss_plastic runs the Reuss kernel through the passes with a J2 matrix: the phases carry the same stress to first order in the step.
…he phase materials The composite is split into two nodes. mean_field_strain consumes the composite strain increment and the phases' tangents, fractions and Eshelby tensors and produces one strain per phase (history) and the concentration tensors, for the voigt, reuss, hill, dilute and mori_tanaka kernels. The phase materials consume those strains through strain_source / strain_property like any other strain. mean_field_stress averages the stresses and tangents they produce with the concentrations and reports a residual, the difference between the tangent a phase node carried and its material's current one. With linear phases every edge is global, one pass is exact and the result equals mean_field_tangent with a linear_stress to round-off, the phase strains and stresses available as properties. docs/mean_field.md describes the composite layer.
tmech main (46d8fdd, after tmech#38) is exact at repeated eigenvalues, so the material uses ln b from tmech::log(b) and its derivative, is Dim generic, and core/log_strain.h with its Eigen dependency goes away. The per-case fd_step of the reports is optional and no longer overrides the driver default; the 1e-6 workaround in the finite-strain standard is dropped (all three finite-strain reports pass at 1e-7). New tests: tangent at repeated and nearly repeated stretches (the tmech#37 symptom), objectivity P(RU) = R P(U) and P(R) = 0. Mapping document and card converter list the Hencky family.
…ls-and-constants # Conflicts: # tests/CMakeLists.txt
…eshelby-tensors # Conflicts: # tests/CMakeLists.txt
…v-depolarisation # Conflicts: # tests/CMakeLists.txt
…/rule-of-mixture # Conflicts: # tests/CMakeLists.txt
# Conflicts: # include/numsim-materials/report/report_json.h # tests/CMakeLists.txt
residual_source, residual_tolerance and max_passes are driver parameters (material_point's defaults) and keys of the report's "point"; the mean-field tangent report uses them for the self-consistent fixed point.
…averaging # Conflicts: # tests/CMakeLists.txt
…-composite # Conflicts: # tests/CMakeLists.txt
…-elasticity # Conflicts: # include/numsim-materials/umat/deformation_gradient_driver.h # include/numsim-materials/umat/material_point_driver.h # tests/CMakeLists.txt
…genstrain # Conflicts: # tests/CMakeLists.txt
…-rate # Conflicts: # tests/CMakeLists.txt
…rain # Conflicts: # tests/CMakeLists.txt
# Conflicts: # include/numsim-materials/report/report_json.h # tests/CMakeLists.txt # tests/test_report_case.cpp
The report's "point" names them for materials that publish e.g. damaged_stress / damaged_tangent (isotropic_damage report).
# Conflicts: # reports/CMakeLists.txt # tests/CMakeLists.txt
# Conflicts: # tests/CMakeLists.txt
# Conflicts: # include/numsim-materials/report/case_runner.h # include/numsim-materials/report/report_json.h # tests/CMakeLists.txt
…ield-kernel-objects
…eld-composite-global # Conflicts: # docs/materials/mean_field_composite_global/report.json
…field-strain-kernels
…eld-composite-local
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Part 3 of #168 (epic #173). Stacked on #176.
What
The inelastic materials and the solvers declare their reads:
j2_plasticity,j2_rk_plasticityanddrucker_prager_plasticityfill stress, tangent, plastic strain and equivalent plastic strain from one callback.linear_isotropic_hardeningandexponential_isotropic_hardeningread the equivalent plastic strain over a local edge.isotropic_damagereads its six inputs;scalar_damage_law(power, saturation, Peerlings) fills damage and its derivative from one callback;strain_threshold_yieldfillskappaandis_yieldingfrom one callback.autocatalytic_reaction: four callbacks, each with its reads. Residual, jacobian and rate read the history state, so they run after its update, as the creation-order tie-break ran them before.backward_euler,rk_integrator: bound withadd_callbackinstead of writingtraits().updatedirectly.vector_newton: fills every unknown and everyd_<x>_d_strainfrom its solve. Before, only the first unknown had the callback, and the others were outputs without a writer.No behaviour change in practice. The return maps, hardening laws and solvers cannot be constant, because they have history outputs or local edges.
isotropic_damageand the damage laws now fold when everything they read is constant, which no current deck does. The gain is exact per-callback reads, for #169's edges and #170's check.The dependencies these solvers hide from the graph (the time step of
backward_euler's function,material_ref) are #171.Tests
No new tests: the plasticity, damage, cure, RK and vector-Newton suites cover these materials unchanged.
Gates (build + ctest incl. ASan variants): GCC 14 1228/1228, g++-13 1228/1228, 0 warnings.